All sorts of new technologies promise to empower individuals, but the ultimate empowerer of ordinary people may well turn out to be nanotechnology, the much-hyped but still important technology of molecular manufacturing and computing. Indeed, for all the nano-hype, the reality of nanotechnology may turn out to exceed the claims. The result may be as big a change as the Industrial Revolution, but in a different direction.

Nanotechnology derives its name from the nanometer, or a billionth of a meter, and refers to the manipulation of matter at the atomic and molecular level. The ideas behind nanotechnology are simple ones: every substance on Earth is made up of molecules composed of one or more atoms (the smallest particles of elements). To describe the molecules that constitute a physical object and how they interrelate is to say nearly everything important about the object. It follows, then, that if you can manipulate individual atoms and molecules and put them together in certain configurations, you should be able to create just about anything you desire. And if technologies like computers and the Internet have empowered individuals by giving them drastically more control over the organization of information, the impact of nanotechnology—which promises similar control over the material world—is likely to be much greater. This goes well beyond home-brewing beer, though, as with making beer, nanotechnology involves letting someone else do the hard work at the microscopic level.

Richard Feynman’s first description of nanotechnology still serves:

The principles of physics, as far as I can see, do not speak against the possibility of maneuvering things atom by atom…. [I]t would be, in principle, possible for a physicist to synthesize any chemical substance that the chemist writes down. How? Put the atoms down where the chemist says, and so you make the substance. The problems of chemistry and biology can be greatly helped if our ability to see what we are doing, and to do things on an atomic level, is ultimately developed—a development which I think cannot be avoided.¹

Modern nanotechnology researchers want to go beyond synthesizing “substances” (though that has great importance) to use nanotechnology’s atom-by-atom construction techniques to produce objects: tiny, bacterium-sized devices that can repair clogged arteries, kill cancer cells, fix cellular damage from aging, and (via what are called “assemblers”) make other devices of greater size or complexity by plugging atoms, one at a time, into the desired arrangements, very quickly. Other researchers believe that nanotechnology will allow for a degree of miniaturization that might permit computers a millionfold more efficient than anything available now. Still others believe that nanotechnology’s tiny devices will be able to unravel mysteries of the microscopic world (such as cell metabolism, the aging process, and cancer) in ways that other tools will not be able to.

So far, pioneers like Eric Drexler and Robert Freitas have worked out a lot of the details, and research has produced some small devices, but nothing as exotic as those described above. But nanotechnologists are refining both their instrumentation and their understanding of nanofabrication at an accelerating rate. Will they be able to fulfill the field’s promise? Richard Feynman thought so. That raises a lot of interesting possibilities—and questions.

The digital revolution brought us a debate over the difference between virtual reality and physical reality, a distinction the courts are still trying to figure out. But we are also at the dawn of a new technological revolution—the nanotech revolution—that may challenge our definition of what physical reality is. Superman could create diamonds by squeezing lumps of coal, using heat and pressure to rearrange the carbon atoms. Nanotechnology could achieve the same transformation, with considerably less fuss, simply by plugging carbon atoms together, one at a time, in the correct manner—and without the embarrassing blue tights.

This sounds like the stuff of science fiction, and it is: In Michael Crichton’s thriller, Prey, nanotech plays the bad guy. But in real life, nanotech is already being used by everyone from Lee Jeans, which uses nanofibers to make stain-proof pants, to the U.S. military, which uses nanotechnology to make better catalysts for rockets and missiles, to scientists who are using nano-technology to develop workable artificial kidneys.²

“JUST ADD SUNLIGHT AND DIRT”

Many scientists initially doubted that nanotechnology’s precise positioning of molecules was possible, but that skepticism appears to have been misplaced. That’s no surprise, really, since living organisms, including our own bodies, make things like bone and muscle by manipulating individual atoms and molecules. Yet as criticism has shifted from claims that nanotechnology won’t work to fears that it might, there have been calls to stop progress in the field of nanotechnology before research really gets off the ground. The ETC Group, an anti-technology organization operating out of Canada, has proposed a moratorium on nanotechnology research and on research into self-replicating machines. (At the moment, the latter is like calling for a moratorium on antigravity or faster-than-light travel—nobody’s doing it anyway.)

Proponents of this line of criticism face an uphill battle. What’s attractive about devices that can be programmed to manipulate molecules is that they let you make virtually anything you want, and you can generally make it out of cheap and commonly available materials and energy—what nanotech enthusiasts call “sunlight and dirt.” Selectively sticky probes on tiny bacterium-scale arms, attached either to tiny robots or to a silicon substrate and controlled by computer, can grab the atoms they need from solution, and then plug them together in the proper configuration. It’s not quite molecular Legos, but it’s close. General purpose devices that can do this are called “assemblers,” and the process is known among nanotechnology proponents as “molecular manufacturing.”

This process raises some problems of its own, though. Assemblers that can manufacture virtually anything from sunlight and dirt might, as the result of a program error, manufacture endless copies of themselves, which would then go on to make still more copies, and so on. The fear that nanobots might turn the world into mush is known in the trade as the “gray goo problem,” the apocalyptic scenario raised in Crichton’s novel.

Nanotech’s backers, however, believe the real problem won’t be accident, but abuse. With mature nanotechnology, it might be possible to disassemble enemy weapons. (Imagine bacterium-sized devices that convert high explosives into inert substances, a technique that would neutralize even nuclear weapons, whose detonators are made of chemical high explosive.) On a more threatening note, sophisticated nanodevices could serve as artificial “disease” agents of great power and subtlety. Highly sophisticated nanorobots could even hide out in people’s brains, manipulating their neurochemistry to ensure that they genuinely loved Big Brother. Like nuclear weapons, these devices would be awesome in their destructiveness, and their misuse would be terrifying. Still, the race to harness this power is well underway: Defense spending on nanotechnology is climbing, and civilian spending is over $1 billion a year.³

In a world in which the promises of nanotechnology were realized, practically anyone could live a life that would be extraordinary by today’s standards, in terms of health (thanks to nanomedicine) and material possessions. DNA damaged by radiation, toxins, or aging could be repaired; arterial plaque could be removed; and cancerous or senescent cells could be destroyed or fixed. Organs could be replaced or even enhanced. Researcher Robert Freitas surveys many of these issues in his book Nanomedicine, which explores such topics as “respirocytes”—tiny devices in the bloodstream that could deliver oxygen when the body wasn’t able to, protecting against everything from drowning to heart attacks and strokes long enough to allow medical assistance. And this just scratches the surface in terms of potential enhancements, which might also involve stronger muscles, better nerves, and enhanced cognition—the last being the subject of an ongoing Department of Defense research project already.⁴

Most physical goods could be manufactured onsite at low cost from cheap raw materials. Imagine owning an appliance the size of a refrigerator, full of nanoassemblers, that ran on sunlight and dirt (well, solar electricity and cheap feedstocks, anyway) and made pretty much everything you need, from clothing to food. The widespread availability of such devices would make things very, very different. Material goods wouldn’t be quite free, but they would be nearly so.

In such a world, personal property would become almost meaningless. Some actual physical items would retain sentimental value, but everything else could be produced as needed, then recycled as soon as the need passed. (As someone who writes on a laptop that was cutting edge last year and is now old news, with its value discounted accordingly, I sometimes think we’re already there except for the recycling part. Don’t even ask about my MP3 player.)

Real property would retain its value—as my grandfather used to say, “They’re not making any more of it,” especially oceanfront acreage—but what would “value” mean? Value usually describes an object’s ability to be exchanged for another item. But with personal property creatable on demand from sunlight and dirt, it’s not clear what the medium of exchange would be. Value comes from scarcity, and most goods wouldn’t be scarce. Intellectual property—the software and designs used to program the nano­devices—would be valuable, though once computing power became immense and ubiquitous, developing such designs wouldn’t be likely to pose much of a challenge.

One thing that would remain scarce is time. Personal services like teaching, lawyering, or prostitution wouldn’t be cheapened in the same fashion. We might wind up with an economy based on the exchange of personal services more than on the purchase of goods. As I mentioned earlier, that’s where we’re headed already to a point. Even without nanotechnology, the prices of many goods are falling. Televisions, once expensive, are near-commodity goods, as are computers, stereos, and just about all other electronics. It’s cheaper to build new ones than to fix old ones, and prices continue to fall as capabilities increase. Nanotechnology would simply accelerate this trend and extend it to everything else. Ironically, it may be the combination of capitalism and technology that brings about a utopia unblemished by the need for ownership, the sort that socialists (usually no fans of capitalism) and romantics (no fans of technology) have long dreamed of.

PIONEERS’ PROGRESS

We’re not there yet, but things are progressing faster than even I had realized. Recently, I attended an EPA Science Advisory Board meeting where nanotechnology was discussed. What struck me is that even for people like me who try to keep up, the pace of nanotechnology research is moving much too fast to catch everything.

One of the documents distributed at that meeting was a supplement to the president’s budget request, entitled National Nanotechnology Initiative: Research and Development Supporting the Next Industrial Revolution.⁵ I expected it to be the usual bureaucratic pap, but in fact, it turned out to contain a lot of actual useful information, including reports of several nanotechnology developments that I had missed.

The most interesting, to me, was the report of “peptide [ring] nanotubes that kill bacteria by punching holes in the bacteria’s membrane.” You might think of these as a sort of mechanical antibiotic. As the report notes, “By controlling the type of peptides used to build the rings, scientists are able to design nanotubes that selectively perforate bacterial membranes without harming the cells of the host.”⁶ It goes on to note, “In theory, these nano-bio agents should be far less prone than existing antibiotics to the development of bacterial resistance.”⁷ What’s more, if such resistance appears, it is likely to be easier to counter. Given the way in which resistance to conventional antibiotics has exploded, this is awfully good news.

Another item involved the use of nanoscale particles of metallic iron to clean up contaminated groundwater. In one experiment, aimed at the contaminant trichloroethylene (TCE), the results were quite impressive: “The researchers carried out a field demonstration at an industrial site in which nanoparticles injected into a groundwater plume containing TCE reduced contaminant levels by up to 96 percent.” The report goes on to observe, “A wide variety of contaminants (including chlorinated hydrocarbons, pesticides, explosives, polychlorinated biphenyls and perchlorate) have been successfully broken down in both laboratory and field tests.”⁸ Not too shabby.

And there’s more: the development of nanosensors capable of identifying particular microbes or chemicals, of nanomotors, and dramatic advances in materials. These advances shouldn’t be underestimated.

We tend to forget this, but it’s possible for a technology to have revolutionary effects long before it reaches its maturity. The impact of high-strength materials, for example, is likely to be much greater than people generally realize. Materials science isn’t sexy the way that, say, robots are sexy, but when you can cut the weight, or boost the strength, of aircraft, or spacecraft, or even automobiles by a factor of ten or fifty, the consequences are enormous. Ditto for killing germs, or even detecting them in short order. These sorts of things aren’t as exciting as true molecular manufacturing, and they’re not as revolutionary, but they’re still awfully important, and awfully revolutionary, by comparison with everything else.

When I gave my talk at the Science Advisory Board, I divided nanotechnology into these categories:

• Fake: where it’s basically a marketing term, as with nanopants

• Simple: high-strength materials, sensors, coatings, etc.—things that are important, but not sexy

• Major: advanced devices short of true assemblers

• Spooky: assemblers and related technology (true molecular nanotechnology, capable of making most anything from sunlight and dirt, creating supercomputers smaller than a sugar cube, etc.)

I noted that only in the final category did serious ethical or regulatory issues appear, and also noted that the recent flood of “it’s impossible” claims relating to “spooky” nanotechnology seem to have more to do with fear of ethical or regulatory scrutiny than anything else. People in the industry are hoping to keep the critics away with a smokescreen of doubt as to the capabilities of the technology. That probably won’t work, especially as nanotechnology develops and is put to use in more and more ways.

Up to now, talk of nanotechnology has generally involved either the “fake” variety (stain-resistant pants) or the “spooky” variety (full-scale molecular nanotechnology with all it implies). But as what might be called midlevel nanotechnology—neither fake nor spooky—begins to be deployed, it’s likely to have a substantial effect on the nature of the debate. It’s one thing to worry about (fictitious) swarms of predatory nanobots, a la Michael Crichton’s novel Prey. It’s another to talk about nanotech bans or moratoria when nanotechnology is already at work curing diseases and cleaning up the environment.

LEARNING FROM EXPERIENCE

I think that these positive uses will probably shift the debate away from the nano-Luddites. But, on the other hand, as nanotechnology becomes commonplace, serious discussion of its implications may be short-circuited. I think that the nanotech business community is actually hoping for such an outcome, in fact, but I continue to believe that such hopes are shortsighted. Genetically modified foods, for example, came to the market with the same absence of discussion, but the result wasn’t so great for the industry. Will nanotechnology be different? Stay tuned. Whatever happens, I think that trying to stand still might well prove the most dangerous course of action.

This may seem surprising, but experience suggests that it’s true.

For an academic project I worked on awhile back, I reviewed the history of what used to be called “recombinant DNA research” and is now generally just called genetic engineering or biotechnology. Back in the late 1960s and early 1970s, this was very controversial stuff, with opponents raising a variety of frightening possibilities.

Not all the fears were irrational. We didn’t know very much about how such things worked, and it was possible to imagine scary scenarios that at least seemed plausible. Indeed, such plausible fears led scientists in the field to get together, twice, holding conferences at Asilomar in California, to propose guidelines that would ensure the safety of recombinant DNA research until more was known.

Those voluntary guidelines became the basis for government regulations, regulations that work so well that researchers often voluntarily submit their work to government review even when the law doesn’t require it—and standard DNA licensing agreements often even call for such submission. Self-policing was their key element, and it worked.

When the DNA research debate first started, scientific critics such as Erwin Chargaff met the notion of scientific self-regulation with skepticism. Chargaff predicted modern-day Frankensteins or “little biological monsters” and compared the notion of scientific self-regulation to that of “incendiaries forming their own fire brigade.” Such critics warned that the harms that might result from permitting such research were literally incalculable, and thus it should not be allowed.

Others took a different view. Physicist Freeman Dyson, who admitted that (as a physicist, not a biologist) he had no personal stake in the debate, noted, “The real benefit to humanity from recombinant DNA will probably be the one no one has dreamed of. Our ignorance lies equally on both arms of the balance. The public costs of saying no to further development may in the end be far greater than the costs of saying yes.” Harvard’s Matthew Meselson agreed. The risk of not going forward, he argued, was the risk of being left open to “forthcoming catastrophes,” in the form of starvation (which could be addressed by crop biotechnology) and the spread of new viruses. Critics like Chargaff pooh-poohed this view, saying that the promise of the new technology to alleviate such problems was unproven.⁹

Meselson and Dyson have been vindicated. Indeed, Meselson’s comments about “forthcoming catastrophes” were made (though no one knew it at the time) just as AIDS was beginning to spread around the world. Without the tools developed through biotechnology and genetic engineering, the Human Immunodeficiency Virus could not even have been identified, and treatment efforts would have been limited. Had we listened to the critics, in other words, it’s likely that many more people would have died. Meanwhile, the critics’ Frankensteinian fears have not come true, and the research that was feared then has become commonplace, as this excerpt from John Hockenberry’s DNA Files program on NPR illustrates:

Hockenberry: In those early days [Arthur] Caplan says people were concerned about what would happen if we tried to genetically engineer different bacteria.

Caplan: The mayor of Cambridge, Massachusetts, at one point said he was worried if there were scientific institutions in his town that were doing this, he didn’t want to see sort of Frankenstein-type microbes coming out of the sewers.

Hockenberry: Today those early concerns seem almost quaint. Now even high school biology classes like this one in Maine do the same gene combining experiments that once struck fear into the hearts of public officials and private citizens.¹⁰

This experience suggests that we need to pay close attention to the downsides of limiting scientific research, and that we need to scrutinize the claims of fearmongering critics every bit as carefully as the claims of optimistic boosters. This is especially true at the moment, because, arguably, we’re in a window of vulnerability where many technologies are concerned. For example, in 2002 researchers at SUNY-Stony Brook synthesized a virus using a commercial protein synthesizer and a genetic map downloaded from the Internet. This wasn’t really news from a technical standpoint (I remember a scientist telling me in 1999 that anyone with a protein synthesizer and a computer could do such a thing), but many found it troubling.¹¹

But at the moment, it’s troubling because we know more about viruses than about their cures, meaning that it’s easier to cause trouble by making viruses than it is to remedy viruses once made. In another decade or two, depending on the pace of research, developing a vaccine or cure will be just as easy. That being the case, doesn’t it make sense to progress as rapidly as possible, to minimize the timespan in which we’re at risk? It does to me.

Critics of biotechnology feel otherwise. But their track record hasn’t been very impressive so far. What’s more interesting is who’s not criticizing nanotechnology. Typically Luddite Greenpeace, for instance, has been surprisingly moderate in its response. The environmental organization has sponsored a report entitled “Future Technologies, Today’s Choices: Nanotechnology, Artificial Intelligence and Robotics; A Technical, Political and Institutional Map of Emerging Technologies”¹² that looks rather extensively at nanotechnology.

Surprisingly, the report rejects the idea of a moratorium on nanotechnology, despite calls to squelch nanotech from other environmental groups. Instead, it finds that a moratorium on nanotechnology research “seems both unpractical and probably damaging at present.”¹³ The report also echoes warnings from others that such a moratorium might simply drive nanotechnology research underground.

Though overlooked in the few news stories to cover the report, this finding is significant. With a moratorium taken off the table, the question then becomes one of how, not whether, to develop nanotechnology. The report also takes a rather balanced view of the technology’s prospects. It notes that there has been a tendency to blur the distinction between nanoscale technologies of limited long-term importance (e.g., stain-resistant “nanopants”) and build-anything general assembler devices and other sophisticated nanotechnologies, so as to make incremental work look sexier than it is. This is important: the report’s not-entirely-unreasonable worries about the dangers of nanomaterials are distinguishable from more science-fictional concerns of the Crichton variety. (Remember, Crichton rhymes with “frighten.”) Thus, it will be harder for Greenpeace to conflate the two kinds of concerns itself, as has been done in the struggle against genetically modified foods where opponents have often mixed minor-but­proven threats with major-but-bogus ones in a rather promiscuous fashion.

Indeed, it seems to me that nano-blogger Howard Lovy is right in saying, “Take out the code words and phrases that are tailored to Greenpeace’s audience, and you’ll find some sound advice in there for the nanotech industry.”¹⁴ Greenpeace is calling for more research into safety. Now is a good time to do that—even for the industry, which currently doesn’t have a lot of products at risk. Quite a few responsible nanotechnology researchers are calling for this kind of research as well. Such research is likely to do more good than harm at blocking Luddite efforts to turn nanotechnology into a political football—the next Genetically Modified Organism (GMO) derived food. Despite the vast promise of GMO foods (including vitamin-enhanced “golden rice” that can prevent widespread blindness among Third-World children), environmentalist hostility and fearmongering has kept most of them out of the market. As Rice University researcher Vicki Colvin noted in congressional testimony:

The campaign against GMOs was successful despite the lack of sound scientific data demonstrating a threat to society. In fact, I argue that the lack of sufficient public scientific data on GMOs, whether positive or negative, was a controlling factor in the industry’s fall from favor. The failure of the industry to produce and share information with public stakeholders left it ill-equipped to respond to GMO detractors. This industry went, in essence, from “wow” to “yuck” to “bankrupt.” There is a powerful lesson here for nanotechnology.¹⁵

She’s right, and the nanotechnology industry would do well to learn from the failings she outlines. As I noted above, some companies and researchers have tended to dismiss the prospects for advanced nanotechnology in the hopes of avoiding the attention of environmental activists. That obviously isn’t working. The best defense against nano-critics is good, solid scientific information, not denial—especially given the strong promise of nanotechnology in terms of environmental improvement.

Nanotechnology legislation recently passed by Congress calls for some investigation into these issues of safety and ethics. I hope that there will be more emphasis on exploring both the scientific and the ethical issues involved in nanotechnology’s growth. That sort of exploration—done by serious people, not the charlatans and fearmongers who are sure to target the area regardless—will be important in making nanotechnology succeed.

The critics won’t shut up, of course, but some aspects of their criticism will have more weight than others, leaving the scaremongering less influential than the scaremongers hope. And if that’s not enough, the argument for nanotechnology’s role in maintaining military supremacy is likely to rear its head. Nanotechnology is likely to be as important in the twenty-first century as rocketry or nuclear physics were in the twentieth. The United States has a fairly competent nanotechnology research program, though many feel its efforts are misdirected. Europe has a substantial but comparatively muted one. Other countries seem very interested indeed.

In the United States, and especially in Europe, research into nanotechnology is facing growing resistance from the same forces that have opposed biotechnology—and, for that matter, nuclear energy and other new technologies. The claim is that concerns about the safety and morality of nanotechnology justify limitations on research and development. Even Prince Charles has weighed in against nanotechnology, although Ian Bell wonders if the real fuss is about something other than the science:

Charles is afraid that the science could, yes, run amok, with minuscule robots reproducing themselves and proceeding to turn the world into “grey goo.”

Many might suspect that the only grey goo we have to worry about is between the ears of HRH, but scientists fear that the prince could do to them what he did to the reputation of contemporary architecture. Charles, clearly, can have no way of knowing what he is talking about, but the fear he expresses is common: do any of us really know what we are doing when we follow where science leads?¹⁶

The real problem isn’t a distrust of science. It’s a distrust of people. Such fear is strongest when pessimism about humanity is at a high. Europe, perhaps understandably pessimistic about humanity’s prospects in light of recent history, leads the way in throwing some people’s only favored invention—the wet blanket—over nanotechnology research.

In the more optimistic United States, concerns exist, but they haven’t yet led to a strong interest in regulating nanotechnology. Instead, the U.S. takes an ostrich-like approach to dealing with the realities of the technology; scientific and corporate types try to shift the focus to short-term technological developments while scoffing at the prospects for genuine molecular manufacturing—the “spooky” stuff, as I’ve labeled it. Some promising developments are taking place, both at the National Nanotechnology Initiative and within the nanotechnology industry itself, but it’s still too early to tell whether this turnaround will really take hold.

MANDARINS AND MEMORIES

In the meantime, other cultures, unencumbered by the residual belief in original sin plaguing even the most secular Westerners, show far less reluctance. Perhaps they are less comfortable and more ambitious than we are, as well. Chinese interest in military nanotechnology has begun to alarm some, especially as China is already third in the world in nanotechnology patent applications.¹⁷

India’s president, Abdul Kalam, is also touting nanotechnology, and as a recent press account captured, he’s quite straightforward in saying that one reason for treating nanotechnology as important is that it will lead to revolutionary weaponry:

[Kalam] said carbon nano tubes and its composites would give rise to super strong, smart and intelligent structures in the field of material science and this in turn could lead to new production of nano robots with new types of explosives and sensors for air, land and space systems. “This would revolutionise the total concepts of future warfare,” he said.¹⁸

Yes, it would. Westerners tend to forget it, but it was a few key technologies—primarily steam navigation and repeating firearms—that made the era of Western colonialism possible. (See Daniel Headrick’s The Tools of Empire¹⁹ for more on this.)

It is, no doubt, as hard for American and European Mandarins to imagine being conquered by Chinese troops equipped with superior weaponry as it was for Chinese Mandarins to imagine the reverse two hundred years ago. Will our mandarins be smart enough to learn from that experience? That’s the question, isn’t it?

But in the long run, the growth of nanotechnology means that we won’t just be worrying about countries, but about individuals. With mature nanotechnology, individuals and small groups will possess powers once available only to nation-states. As with all powers possessed by individuals, these will sometimes be used for good, and sometimes for ill.

Of course, that’s just an extension of existing phenomena. My own neighborhood has a few dozen families in it; between them, they probably have enough guns and motorized vehicles (conveniently, mostly SUVs) to wipe out a Roman legion, or a Mongol horde—forces that, in both cases, once represented the peak of military power on the planet. Nobody worries about the military power that my neighborhood represents, because it’s (1) unlikely to be misused, and (2) negligible in a world where most anyone can afford guns and SUVs anyway.

What this suggests is that a world in which nanotechnology is ubiquitous is likely to be less threatening than one in which it’s a closely held government monopoly. A world in which nanotechnology is ubiquitous is a rich world. That doesn’t preclude bad behavior, but it helps. A world with such diffuse power makes abuse by smaller groups, or even governments, less threatening overall. The average Roman or Mongolian citizen didn’t really need guns or SUVs. Back then, the hobbyist machine shop in my neighbor’s basement would have been a tool of strategic, even world-changing, importance all by itself. Now, in a different world, it’s just a toy, even though it could, in theory, produce dangerous weaponry. It’s probably best if nano-technology works out the same way, with diffusion minimizing the risk that anyone will gain disproportionate power over the rest of us.

In his recent book, The Singularity Is Near, Ray Kurzweil notes that technology often suffices to deal with technological threats, even in the absence of governmental intervention:

When [the computer virus] first appeared, strong concerns were voiced that as they became more sophisticated, software pathogens had the potential to destroy the computer-network medium in which they live. Yet the “immune system” that has evolved in response to this challenge has been largely effective. Although destructive self-replicating software entities do cause damage from time to time, the injury is but a small fraction of the benefit we receive from the computers and communications links that harbor them.²⁰

Software viruses, of course, aren’t usually a lethal threat. But Kurzweil notes that this cuts both ways:

The fact that computer viruses are not usually deadly to humans only means that more people are willing to create and release them. The vast majority of software-virus authors would not release viruses if they thought they would kill people. It also means that our response to the danger is that much less intense. Conversely, when it comes to self-replicating entities that are potentially lethal on a large scale, our response on all levels will be vastly more serious.²¹

I think that’s right. In fact, prophetic works of science fiction—Neal Stephenson’s The Diamond Age, for instance—generally feature such defensive technologies against rogue nanotechnology. Given the greater threat potential of nanotechnologies, we may have to rely on more than Symantec and McAfee for protection—but on the other hand, given the huge benefits promised by nanotechnology, we should be willing to go ahead anyway. And I expect we will.

1. Richard P. Feynman, There’s Plenty of Room at the Bottom, ed. Horace D. Gilbert (1961), 295-96.

2. On the artificial kidneys, see "Nanotechnology Used to Help Develop Artificial Kidney," ABC News Online.

3. Information on the National Nanotechnology Initiative can be found at its website, http://www.nano.gov—but information on classified Defense Department work is, of course, classified.

4. Robert J. Freitas, Nanomedicine, Volume I: Basic Capabilities (Landes Bioscience, 1999). See also Robert J. Freitas, Nanomedicine, Volume IA: Biocompatibility (Landes Bioscience, 2003). On enhanced cognition, see Kelly Hearn, "Future Soldiers Could Get Enhanced Minds," UPI, 19 March 2001, LexisNexis Library, UPI File (describing planned use of nanotechnology to enhance soldiers’ cognition and decision-making under stress).

5. National Science and Technology Council (2004), available online at http://nano.gov/nni04_budget_supplement.pdf.

6. National Science and Technology Council, 27.

7. National Science and Technology Council.

8. National Science and Technology Council, 33.

9. For a summary of this debate, see Judith P. Swazey, et al., "Risks and Benefits, Rights and Responsibilities: A History of the Recombinant DNA Research Controversy," Volume 51, Southern California Law Review (1978), 1019.

10. Available online at http://www.dnafiles.org/PDFs/therapy.pdf.

11. See David Whitehouse, "First Synthetic Virus Created," BBC News, 11 July 2002. Available online at http://news.bbc.co.uk/2/hi/science/nature/2122619.stm.

12. Available online at http://www.greenpeace.org.uk/MultimediaFiles/Live/FullReport/5886.pdf.

13. Available online at http://www.greenpeace.org.uk/MultimediaFiles/Live/FullReport/5886.pdf.

14. Howard Lovy, Nanobot blog, http://nanobot.blogspot.com/2003_07_20_nanobot_archive.html#105905157013774164.

15. Testimony of Dr. Vicki L. Colvin, director, Center for Biological and Environmental Nanotechnology (CBEN), and associate professor of chemistry, Rice University, Houston, Texas, before the U.S. House of Representatives Committee on Science, in regard to "Nanotechnology Research and Development Act of 2003," 9 April 2003. Available online at http://www.house.gov/science/hearings/full03/apr09/colvin.htm.

16. Ian Bell, "Upgrading the Human Condition," Sunday Herald (Glasgow), 1 August 2004. Available online at http://www.sundayherald.com/43701.

17. "China’s Nanotechnology Patent Applications Rank Third in World," InvestorIdeas.com, 3 October 2003, http://www.investorideas.com/Companies/Nanotechnology/Articles/China’sNanotechnology1003,03.asp. See also Dennis Normile, "China’s R&D Power, Truth about Trade & Technology," 2 September 2005, http://www.truthabouttrade.org/article.asp?id=4364. ("Ernest Preeg, senior fellow in trade and productivity for the Manufacturers Alliance/MAPI, warns in his just released book, The Emerging Chinese Advanced Technology Superstate (jointly pub­lished by the Manufacturers Alliance/MAPI and the US Hudson Institute in June 2005) that ‘China is right up there with the US in nanotechnology and coming on strong in biotech and in genetically modified agriculture.’")

18. "Indian Scientists Should Make Breakthrough in Nano Technology: Kalam," IndiaExpress.com, 1 July 2004, http://www.indiaexpress.com/news/technology/20040701-0.html.

19. Daniel Headrick, The Tools of Empire: Technology and European Imperialism in the Nineteenth Century (Oxford University Press, 1981).

20. Ray Kurzweil, The Singularity Is Near: When Humans Transcend Biology (Viking, 2005), 415.

21. Kurzweil.

© 2006 Glenn Reynolds

In order to give you pleasant dreams tonight, let me offer a few possibilities about the days that lie ahead—changes that may occur within the next twenty or so years, roughly a single human generation. Possibilities that are taken seriously by some of today’s best minds. Potential transformations of human life on Earth and, perhaps, even what it means to be human.

For example, what if biologists and organic chemists manage to do to their laboratories the same thing that cyberneticists did to computers? Shrinking their vast biochemical labs from building-sized behemoths down to units that are utterly compact, making them smaller, cheaper, and more powerful than anyone imagined. Isn’t that what happened to those gigantic computers of yesteryear? Until, today, your pocket cell phone contains as much processing power and sophistication as NASA owned during the moon shots. People who foresaw this change were able to ride this technological wave. Some of them made a lot of money.

Biologists have come a long way already toward achieving a similar transformation. Take, for example, the Human Genome Project, which sped up the sequencing of DNA by so many orders of magnitude that much of it is now automated and miniaturized. Speed has skyrocketed, while prices plummet, promising that each of us may soon be able to have our own genetic mappings done, while-U-wait, for the same price as a simple EKG. Imagine extending this trend, by simple extrapolation, compressing a complete biochemical laboratory the size of a house down to something that fits cheaply on your desktop. A MolecuMac, if you will. The possibilities are both marvelous and frightening.

When designer drugs and therapies are swiftly modifiable by skilled medical workers, we all should benefit.

But then, won’t there also be the biochemical equivalent of “hackers”? What are we going to do when kids all over the world can analyze and synthesize any organic compound, at will? In that event, we had better hope for accompanying advances in artificial intelligence and robotics… at least to serve our fast food burgers. I’m not about to eat at any restaurant that hires resentful human adolescents, who swap fancy recipes for their home molecular synthesizers over the Internet. Would you?

Now don’t get me wrong. If we ever do have MolecuMacs on our desktops, I’ll wager that 99 percent of the products will be neutral or profoundly positive, just like most of the software creativity flowing from young innovators today. But if we’re already worried about a malicious one percent in the world of bits and bytes—hackers and cyber-saboteurs—then what happens when this kind of ‘creativity’ moves to the very stuff of life itself? Nor have we mentioned the possibility of intentional abuse by larger entities—terror cabals, scheming dictatorships, or rogue corporations.

These fears start to get even more worrisome when we ponder the next stage, beyond biotech. Deep concerns are already circulating about what will happen when nanotechnology—ultra-small machines building products atom-by-atom to precise specifications—finally hits its stride. Molecular manufacturing could result in super-efficient factories that create wealth at staggering rates of efficiency. Nano-maintenance systems may enter your bloodstream to cure disease or fine-tune bodily functions. Visionaries foresee this technology helping to save the planet from earlier human errors, for instance by catalyzing the recycling of obstinate pollutants. Those desktop units eventually may become universal fabricators that turn almost any raw material into almost any product you might desire…

… or else (some worry), nanomachines might break loose to become the ultimate pollution. A self-replicating disease, gobbling everything in sight, conceivably turning the world’s surface into gray goo.²

Others have raised this issue before, some of them in very colorful ways. Take the sensationalist novel Prey, by Michael Crichton, which portrays a secretive agency hubristically pushing an arrogant new technology, heedless of possible drawbacks or consequences. Crichton’s typical worried scenario about nanotechnology follows a pattern nearly identical to his earlier thrillers about unleashed dinosaurs, robots, and dozens of other techie perils, all of them viewed with reflexive suspicious loathing. (Of course, in every situation, the perilous excess happens to result from secrecy, a topic that we will return to, later.) A much earlier and better novel, Blood Music, by Greg Bear, presented the up and downside possibilities of nanotech with profound vividness. Especially the possibility that most worries even optimists within the nanotechnology community—that the pace of innovation may outstrip our ability to cope.

Now, at one level, this is an ancient fear. If you want to pick a single cliché that is nearly universally held, across all our surface boundaries of ideology and belief—e.g. left-versus-right, or even religious-vs-secular—the most common of all would probably be:

“Isn’t it a shame that our wisdom has not kept pace with technology?”

While this cliché is clearly true at the level of solitary human beings, and even mass-entities like corporations, agencies or political parties, I could argue that things aren’t anywhere near as clear at the higher level of human civilization. Elsewhere I have suggested that “wisdom” needs to be defined according to outcomes and processes, not the perception or sagacity of any particular individual guru or sage. Take the outcome of the Cold War… the first known example of humanity acquiring a means of massive violence, and then mostly turning away from that precipice. Yes, that means of self-destruction is still with us. But two generations of unprecedented restraint suggest that we have made a little progress in at least one kind of “wisdom.” That is, when the means of destruction are controlled by a few narrowly selected elite officials on both sides of a simple divide.

But are we ready for a new era, when the dilemmas are nowhere near as simple? In times to come, the worst dangers to civilization may not come from clearly identifiable and accountable adversaries—who want to win an explicit, set-piece competition—as much as from a general democratization of the means to do harm. New technologies, distributed by the Internet and effectuated by cheaply affordable tools, will offer increasing numbers of angry people access to modalities of destructive power–means that will be used because of justified grievance, avarice, indignant anger, or simply because they are there.

THE RETRO PRESCRIPTION—RENUNCIATION

Faced with onrushing technologies in biotech, nanotech, artificial intelligence, and so on, some bright people—like Bill Joy, former chief scientist of Sun Computers—see little hope for survival of a vigorously open society. You may have read Joy’s unhappy manifesto in Wired Magazine³, in which he quoted the Unabomber (of all people), in support of a proposal that is both ancient and new—that our sole hope for survival may be to renounce, squelch, or relinquish several classes of technological progress.

This notion of renunciation has gained credence all across the political and philosophical map, especially at the farther wings of both right and left. Take the novels and pronouncements of Margaret Atwood, whose fundamental plot premises seem almost identical to those of Michael Crichton, despite their differences over superficial politics. Both authors routinely express worry that often spills into outright loathing for the overweening arrogance of hubristic technological innovators who just cannot leave nature well enough alone.

At the other end of the left-right spectrum stands Francis Fukuyama, who is Bernard L. Schwartz Professor of International Political Economy at the Paul H. Nitze School of Advanced International Studies of Johns Hopkins University. Dr. Fukuyama’s best-known book, The End of History and the Last Man (1992) triumphally viewed the collapse of communism as likely to be the final stirring event worthy of major chronicling by historians. From that point on, we would see liberal democracy bloom as the sole path for human societies, without significant competition or incident. No more "interesting times."⁴ But this sanguine view did not last, as Fukuyama began to see potentially calamitous “history” in the disruptive effects of new technology. As a Bush Administration court intellectual and a member of the President’s Council on Bioethics, he now condemns a wide range of biological science as disruptive and even immoral. People cannot, according to Fukuyama, be trusted to make good decisions about the use of—for example—genetic therapy. Human "improvability" is so perilous a concept that it should be dismissed, almost across-the-board. In Our Posthuman Future: Consequences of the Biotechnology Revolution (2002), Fukuyama prescribes paternalistic government industry panels to control or ban whole avenues of scientific investigation, doling out those advances that are deemed suitable.

You may surmise that I am dubious. For one thing, shall we enforce this research ban worldwide? Can such tools be squelched forever? From elites, as well as the masses? If so, how?

Although some of the failure modes mentioned by Bill Joy, Ralph Peters, Francis Fukuyama, and the brightest renunciators seem plausible and worth investigating, it’s hard to grasp how we can accomplish anything by becoming neo-Luddites. Laws that seek to limit technological advancement will certainly be disobeyed by groups that simmer at the social extreme, where the worst dangers lie. Even if ferocious repression is enacted—perhaps augmented with near-omniscient and universal surveillance—this will not prevent exploration and exploitation of such technologies by social elites. (Corporate, governmental, aristocratic, criminal, foreign… choose your own favorite bogeymen of unaccountable power.) For years, I have defied renunciators to cite one example, amid all of human history, when the mighty allowed such a thing to happen. Especially when they plausibly stood to benefit from something new.

While unable to answer that challenge, some renunciators have countered that all of the new mega-technologies—including biotech and nanotechnology—may be best utilized and advanced if control is restricted to knowing elites, even in secret. With so much at stake, should not the best and brightest make decisions for the good of all? Indeed, in fairness, I should concede that the one historical example I gave earlier—that of nuclear weaponry—lends a little support to this notion. Certainly, in that case, one thing that helped to save us was the limited number of decision-makers who could launch calamitous war.

Still, weren’t the political processes constantly under public scrutiny, during that era? Weren’t those leaders supervised by the public, at least on one side? Moreover, decisions about atom bombs were not corrupted very much by matters of self-interest. (Howard Hughes did not seek to own and use a private nuclear arsenal.) But self-interest will certainly influence controlling elites when they weigh the vast benefits and potential costs of biotech and nanotechnology.

Besides, isn’t elitist secrecy precisely the error-generating mode that Crichton, Atwood and so many others portray so vividly, time and again, while preaching against technological hubris? History is rife with examples of delusional cabals of self-assured gentry, telling each other just-so stories while evading any criticism that might reveal flaws in The Plan. By prescribing a return to paternalism—control by elites who remain aloof and unaccountable—aren’t renunciators ultimately proposing the very scenario that everybody—rightfully—fears most?

Perhaps this is one reason why the renunciators—while wordy and specific about possible failure modes—are seldom very clear on which controlling entities should do the dirty work of squelching technological progress. Or how this relinquishment could be enforced, across the board. Indeed, supporters can point to no historical examples when knowledge-suppression led to anything but greater human suffering. No proposal that’s been offered so far even addresses the core issue of how to prevent some group of elites from cheating. Perhaps all elites.

In effect, only the vast pool of normal people would be excluded, eliminating their myriad eyes, ears and prefrontal lobes from civilization’s error-detecting network.

Above all, renunciation seems a rather desperate measure, completely out of character with this optimistic, pragmatic, can-do culture.

THE SELDOM-MENTIONED ALTERNATIVE—RECIPROCAL ACCOUNTABILITY

And yet, despite all this criticism, I am actually much more approving of Joy, Atwood, Fukuyama, et al, than some might expect. In The Transparent Society, I speak well of social critics who shout when they see potential danger along the road.

In a world of rapid change, we can only maximize the benefits of scientific advancement—and minimize inevitable harm—by using the great tools of openness and accountability. Above all, acknowledging that vigorous criticism is the only known antidote to error. This collective version of “wisdom” is what almost surely has saved us so far. It bears little or no resemblance to the kind of individual sagacity that we are used to associating with priests, gurus, and grandmothers… but it is also less dependent upon perfection. Less prone to catastrophe when the anointed Center of Wisdom makes some inevitable blunder.

Hence, in fact, I find fretful worry-mongers invigorating! Their very presence helps progress along by challenging the gung-ho enthusiasts. It’s a process called reciprocal accountability. Without bright grouches, eager to point at potential failure modes, we might really be in the kind of danger that they claim we are. Ironically, it is an open society—where the sourpuss Cassandras are well heard—that is unlikely to need renunciation, or the draconian styles of paternalism they prescribe.

Oh, I see the renunciators’ general point. If society remains as stupid as some people think it is—or even if it is as smart as I think it is, but gets no smarter—then nothing that folks do or plan at a thousand well-intentioned futurist conferences will achieve very much. No more than delaying the inevitable.

In that case, we’ll finally have the answer to an ongoing mystery of science—why there’s been no genuine sign of extraterrestrial civilization amid the stars.⁵ The answer will be simple. Whenever technological culture is tried, it always destroys itself. That possibility lurks, forever, in the corner of our eye, reminding us what’s at stake.

On the other hand, I see every reason to believe we have a chance to disprove that dour worry. As members of an open and questioning civilization—one that uses reciprocal accountability to find and probe every possible failure-mode—we may be uniquely equipped to handle the challenges ahead.

Anyway, believing that is a lot more fun.

THE UPSIDE SCENARIO—THE SINGULARITY

We’ve heard from the gloomy renunciators. Let’s look at another future. The scenario of those who—literally—believe the sky’s the limit. Among many of our greatest thinkers, there is a thought going around—a new ‘meme’ if you will—that says we’re poised for take-off. The idea I’m referring to is that of a coming Technological Singularity.

Science fiction author Vernor Vinge has been touted as a chief popularizer of this notion, though it has been around, in many forms, for generations. More recently, Ray Kurzweil’s book The Singularity is Near argues that our scientific competence and technologically-empowered creativity will soon skyrocket, propelling humanity into an entirely new age.

Call it a modern, high-tech version of Teilhard De Chardin’s noosphere apotheosis—an approaching time when humanity may move, dramatically and decisively, to a higher state of awareness or being. Only, instead of achieving this transcendence through meditation, good works or nobility of spirit, the idea this time is that we may use an accelerating cycle of education, creativity and computer-mediated knowledge to achieve intelligent mastery over both the environment and our own primitive drives.

In other words, first taking control over Brahma’s “wheel of life,” then learning to steer it wherever we choose.

What else would you call it…

• When we start using nanotechnology to repair bodies at the cellular level?
• When catching up on the latest research is a mere matter of desiring information, whereupon autonomous software agents deliver it to you, as quickly and easily as your arm now moves wherever you wish it to?
• When on-demand production becomes so trivial that wealth and poverty become almost meaningless terms?
• When the virtual reality experience—say visiting a faraway planet—gets hard to distinguish from the real thing?
• When each of us can have as many “servants”—either robotic or software-based—as we like, as loyal as your own right hand?
• When augmented human intelligence will soar and—trading insights with one another at light speed—helping us attain entirely new levels of thought?

Of course, it is worth pondering how this ‘singularity’ notion compares to the long tradition of contemplations about human transcendence. Indeed, the idea of rising to another plane of existence is hardly new! It makes up one of the most consistent themes in cultural history, as though arising from our basic natures.

Indeed, many opponents of science and technology clutch their own images of messianic transformation, images that—if truth be told—share many emotional currents with the tech-heavy version, even if they disagree over the means to achieve transformation. Throughout history, most of these musings dwelled upon the spiritual path, that human beings might achieve a higher state through prayer, moral behavior, mental discipline, or by reciting correct incantations. Perhaps because prayer and incantations were the only means available.

In the last century, an intellectual tradition that might be called ‘techno-transcendentalism’ added a fifth track. The notion that a new level of existence, or a more appealing state of being, might be achieved by means of knowledge and skill.

But which kinds of knowledge and skill?

Depending on the era you happen to live in, techno-transcendentalism has shifted from one fad to another, pinning fervent hopes upon the scientific flavor of the week. For example, a hundred years ago, Marxists and Freudians wove complex models of human society—or mind—predicting that rational application of these models and rules would result in far higher levels of general happiness.⁶ Subsequently, with popular news about advances in agriculture and evolutionary biology, some groups grew captivated by eugenics—the allure of improving the human animal. On occasion, this resulted in misguided and even horrendous consequences. Yet, this recurring dream has lately revived in new forms, with the promise of genetic engineering and neurotechnology.

Enthusiasts for nuclear power in the 1950s promised energy too cheap to meter. Some of the same passion was seen in a widespread enthusiasm for space colonies, in the 1970s and 80s, and in today’s ongoing cyber-transcendentalism, which promises ultimate freedom and privacy for everyone, if only we just start encrypting every Internet message, using anonymity online to perfectly mask the frail beings who are actually typing at a real keyboard. Over the long run, some hold out hope that human minds will be able to download into computers or the vast new frontier of mid-21^st Century cyberspace, freeing individuals of any remaining slavery to our crude and fallible organic bodies.

This long tradition—of bright people pouring faith and enthusiasm into transcendental dreams—tells us a lot about one aspect of our nature, a trait that crosses all cultures and all centuries. Quite often, this zealotry is accompanied by disdain for contemporary society—a belief that some kind of salvation can only be achieved outside of the normal cultural network…a network that is often unkind to bright philosophers—and nerds. Seldom is it ever discussed how much these enthusiasts have in common—at least emotionally—with believers in older, more traditional styles of apotheosis, styles that emphasize methods that are more purely mental or spiritual.

We need to keep this long history in mind, as we discuss the latest phase: a belief in the ultimately favorable effects of an exponential increase in the ability of our calculating engines. That their accelerating power of computation will offer commensurately profound magnifications of our knowledge and power. Our wisdom and happiness.

The challenge that I have repeatedly laid down is this: “Name one example, in all of history, when these beliefs actually bore fruit. In light of all the other generations who felt sure of their own transforming notion, should you not approach your newfangled variety with some caution… and maybe a little doubt?”

IT MAY BE JUST A DREAM

Are both the singularity believers and the renunciators getting a bit carried away? Let’s take that notion of doubt and give it some steam. Maybe all this talk of dramatic transformation, within our lifetimes, is just like those earlier episodes: based more on wishful (or fearful) thinking than upon anything provable or pragmatic.

Take Jonathan Huebner, a physicist who works at the Pentagon’s Naval Air Warfare Center in China Lake, California. Questioning the whole notion of accelerating technical progress, he studied the rate of “significant innovations per person.” Using as his sourcebook The History of Science and Technology, Huebner concluded that the rate of innovation peaked in 1873 and has been declining ever since. In fact, our current rate of innovation—which Huebner puts at seven important technological developments per billion people per year—is about the same as it was in 1600. By 2024, it will have slumped to the same level as it was in the Dark Ages, around 800 AD. "The number of advances wasn’t increasing exponentially, I hadn’t seen as many as I had expected."

Huebner offers two possible explanations: economics and the size of the human brain. Either it’s just not worth pursuing certain innovations since they won’t pay off—one reason why space exploration has all but ground to a halt—or we already know most of what we can know, and so discovering new things is becoming increasingly difficult.

Ben Jones, of Northwestern University in Illinois, agrees with Huebner’s overall findings, comparing the problem to that of the Red Queen in Through the Looking Glass: we have to run faster and faster just to stay in the same place. Jones differs, however, as to why this happened. His first theory is that early innovators plucked the easiest-to-reach ideas, or "low-hanging fruit," so later ones have to struggle to crack the harder problems. Or it may be that the massive accumulation of knowledge means that innovators have to stay in education longer to learn enough to invent something new and, as a result, less of their active life is spent innovating. "I’ve noticed that Nobel Prize winners are getting older," he says.

In fact, it is easy to pick away at these four arguments by Huebner and Jones.⁷ For example, it is only natural for innovations and breakthroughs to seem less obvious or apparent to the naked eye, as we have zoomed many of our research efforts down to the level of the quantum and out to the edges of the cosmos. In biology, only a few steps—like completion of the Human Genome Project—get explicit attention as “breakthroughs.” Such milestones are hard to track in a field that is fundamentally so complex and murky. But that does not mean biological advances aren’t either rapid or, overall, truly substantial. Moreover, while many researchers seem to gain their honors at an older age, is that not partly a reflection of the fact that lifespans have improved, and fewer die off before getting consideration for prizes?

Oh, there is something to be said for the singularity-doubters. Indeed, even in the 1930s, there were some famous science fiction stories that prophesied a slowdown in progress, following a simple chain of logic. Because progress would seem to be its own worst enemy. As more becomes known, specialists in each field would have to absorb more and more about less and less—or about ever narrowing fields of endeavor—in order to advance knowledge by the tiniest increments. When I was a student at Caltech, in the 1960s, we undergraduates discussed this problem at worried length. For example, every year the sheer size, on library shelves, of "Chemical Abstracts" grew dauntingly larger and more difficult for any individual to scan for relevant papers.

And yet, over subsequent decades, this trend never seemed to become the calamity we expected. In part, because Chemical Abstracts and its cousins have—in fact—vanished from library shelves, altogether! The library space problem was solved by simply putting every abstract on the Web. Certainly, literature searches—for relevant work in even distantly related fields—now take place faster and more efficiently than ever before, especially with the use of software agents and assistants that should grow even more effective in years to come.

That counter-force certainly has been impressive. Still, my own bias leans toward another trend that seems to have helped forestall a productivity collapse in science. This one (I will admit) is totally subjective. And yet, in my experience, it has seemed even more important than advances in online search technology. For it has seemed to me that the best and brightest scientists are getting smarter, even as the problems they address become more complex.

I cannot back this up with statistics or analyses. Only with my observation that many of the professors and investigators that I have known during my life now seem much livelier, more open-minded and more interested in fields outside their own—even as they advance in years—than they were when I first met them. In some cases, decades ago. Physicists seem to be more interested in biology, biologists in astronomy, engineers in cybernetics, and so on, than used to be the case. This seems in stark contrast to what you would expect, if specialties were steadily narrowing. But it is compatible with the notion that culture may heavily influence our ability to be creative. And a culture that loosens hoary old assumptions and guild boundaries may be one that’s in the process of freeing-up mental resources, rather than shutting them down.

In fact, this trend—toward overcoming standard categories of discipline—is being fostered deliberately in many places. For example, the new Sixth College of the University of California at San Diego, whose official institutional mission is to "bridge the arts and sciences," drives a nail in the coffin of C.P. Snow’s old concept that the "two cultures" can never meet. Never before have there been so many collaborative efforts between tech-savvy artists and technologists who appreciate the aesthetic and creative sides of life.⁸

What Huebner and Jones appear to miss is that complex obstacles tend best to be overcome by complex entities. Even if Einstein and others picked all the low hanging fruit within reach to individuals, that does not prevent groups—institutions and teams and entrepreneurial startups—from forming collaborative human pyramids to go after goodies that are higher in the tree. Especially when those pyramids and teams include new kinds of members, software agents and search methodologies, worldwide associative networks and even open-source participation by interested amateurs. Or when a myriad fields of endeavor see their loci of creativity get dispersed onto a multitude of inexpensive desktops, the way software has been.⁹

Dutch-American economic historian Joel Mokyr, in The Lever of Riches and The Gifts of Athena, supports this progressive view that we are indeed doing something right, something that makes our liberal-democratic civilization uniquely able to generate continuous progress. Mokyr believes that, since the 18th-century Enlightenment, a new factor has entered the human equation: the accumulation of and a free market in knowledge. As Mokyr puts it, we no longer behead people for saying the wrong thing—we listen to them. This "social knowledge" is progressive because it allows ideas to be tested and the most effective to survive. This knowledge is embodied in institutions, which, unlike individuals, can rise above our animal natures.

But Mokyr does worry that, though a society may progress, human nature does not. "Our aggressive, tribal nature is hard-wired, unreformed and unreformable. Individually we are animals and, as animals, incapable of progress." The trick is to cage these animal natures in effective institutions: education, the law, government. But these can go wrong. "The thing that scares me," he says, "is that these institutions can misfire."

While I do not use words such as "caged," I must agree that Mokyr captures the essential point of our recent, brief experiment with the Enlightenment: John Locke’s rejection of romantic oversimplification in favor of pragmatic institutions that work flexibly to maximize the effectiveness of our better efforts—the angels of our nature—enabling our creative forces to mutually reinforce. Meanwhile, those same institutions and processes would thwart our “devils”—the always-present human tendency towards self-delusion and cheating. Of course, human nature strives against these constraints. Self-deluders and cheaters are constantly trying to make up excuses to bypass the Enlightenment covenant and benefit by making these institutions less effective. Nothing is more likely to ensure the failure of any singularity than if we allow this to happen.

But then, swiveling the other way, what if it soon becomes possible not only to preserve and advance those creative enlightenment institutions, but also to do what Mokyr calls impossible? What if we actually can improve human nature?

Suppose the human components of societies and institutions can also be made better, even by a little bit? I have contended that this is already happening, on a modest scale. Imagine the effects of even a small upward-ratcheting in general human intelligence, whether inherent or just functional, by means of anything from education to “smart drugs” to technologically-assisted senses to new methods of self-conditioning.

It might not take much of an increase in effective human intelligence for markets and science and democracy, etc., to start working much better than they already do. Certainly, this is one of the factors that singularity aficionados are counting on.

What we are left with is an image that belies the simple and pure notion of a "singularity" curve… one that rises inexorably skyward, as a simple mathematical function, with knowledge and skill perpetually leveraging against itself, as if ordained by natural law. Even the most widely touted example of this kind of curve, Moore’s Law—which successfully modeled the rapid increase of computational power available at plummeting cost—has never been anything like a smooth phenomenon. Crucial and timely decisions—some of them pure happenstance—saved Moore’s Law on many occasions from collision with either technological barriers or cruel market forces.

True, we seem to have been lucky, so far. Cybernetics and education and a myriad other factors have helped to overcome the "specialization trap." But as we have seen in this section, past success is no guarantee of future behavior. Those who foresee upward curves continuing ad infinitum, almost as a matter of faith, are no better grounded than other transcendentalists, who confidently predicted other rapturist fulfillments, in their own times.

THE DAUNTING TASK OF CROSSING A MINEFIELD

Having said all of the above, let me hasten to add that I believe in the high likelihood of a coming singularity!

I believe in it because the alternatives are too awful to accept. Because, as we discussed before, the means of mass destruction, from A-bombs to germ warfare, are ‘democratizing’—spreading so rapidly among nations, groups, and individuals—that we had better see a rapid expansion in sanity and wisdom, or else we’re all doomed.

Indeed, bucking the utterly prevalent cliché of cynicism, I suggest that strong evidence does indicate some cause for tentative optimism. An upward trend is already well in place. Overall levels of education, knowledge and sagacity in Western Civilization—and its constituent citizenry—have never been higher, and these levels may continue to improve, rapidly, in the coming century. Possibly enough to rule out some of the most prevalent images of failure that we have grown up with. For example, we will not see a future that resembles Blade Runner, or any other cyberpunk dystopia. Such worlds—where massive technology is unmatched by improved wisdom or accountability—will simply not be able to sustain themselves.

The options before us appear to fall into four broad categories:

1. Self-destruction. Immolation or desolation or mass-death. Or ecological suicide. Or social collapse. Name your favorite poison. Followed by a long era when our few successors (if any) look back upon us with envy. For a wonderfully depressing and informative look at this option, see Jared Diamond’s Collapse: How Societies Choose to Fail or Succeed. (Note that Diamond restricts himself to ecological disasters that resonate with civilization-failures of the past; thus he only touches on the range of possible catastrophe modes.) We are used to imagining self-destruction happening as a result of mistakes by ruling elites. But in this article we have explored how it also could happen if society enters an age of universal democratization of the means of destruction—or, as Thomas Friedman puts it, “the super-empowerment of the angry young man”—without accompanying advances in social maturity and general wisdom.

2. Achieve some form of ‘Positive Singularity‘—or at least a phase shift to a higher and more knowledgeable society (one that may have problems of its own that we can’t imagine.) Positive singularities would, in general, offer normal human beings every opportunity to participate in spectacular advances, experiencing voluntary, dramatic self-improvement, without anything being compulsory… or too much of a betrayal to the core values of decency we share.

3. Then there is the ‘Negative Singularity‘—a version of self-destruction in which a skyrocket of technological progress does occur, but in ways that members of our generation would find unpalatable. Specific scenarios that fall into this category might include being abused by new, super-intelligent successors (as in Terminator or The Matrix), or simply being “left behind” by super entities that pat us on the head and move on to great things that we can never understand. Even the softest and most benign version of such a ‘Negative Singularity’ is perceived as loathsome by some perceptive renunciators, like Bill Joy, who take a dour view of the prospect that humans may become a less-than-pinnacle form of life on Planet Earth.¹⁰

4. Finally, there is the ultimate outcome that is implicit in every renunciation scenario: Retreat into some more traditional form of human society, like those that maintained static sameness under pyramidal hierarchies of control for at least four millennia. One that quashes the technologies that might lead to results 1 or 2 or 3. With four thousand years of experience at this process, hyper-conservative hierarchies could probably manage this agreeable task, if we give them the power. That is, they could do it for a while.

When the various paths¹¹ are laid out in this way, it seems to be a daunting future that we face. Perhaps an era when all of human destiny will be decided. Certainly not one that’s devoid of “history.” For a somewhat similar, though more detailed, examination of these paths, the reader might pick up Joel Garreau’s fine book, Radical Evolution. It takes a good look at two extreme scenarios for the future—"Heaven" and Hell"—then posits a third—"Prevail"—as the one that rings most true.

So, which of these outcomes seem plausible?

First off, despite the fact that it may look admirable and tempting to many, I have to express doubt that outcome #4 could succeed over an extended period. Yes, it resonates with the lurking tone that each of us feels inside, inherited from countless millennia of feudalism and unquestioning fealty to hierarchies, a tone that today is reflected in many popular fantasy stories and films. Even though we have been raised to hold some elites in suspicion, there is a remarkable tendency for each of us to turn a blind eye to other elites—or favorites—and to rationalize that those would rule wisely.

Certainly, the quasi-Confucian social pattern that is being pursued by the formerly Communist rulers of China seems to be an assertive, bold and innovative approach to updating authoritarian rule, incorporating many of the efficiencies of both capitalism and meritocracy.¹² This determined effort suggests that an updated and modernized version of hierarchism might succeed at suppressing whatever is worrisome, while allowing progress that’s been properly vetted. It is also, manifestly, a rejection of the Enlightenment and everything that it stands for, including John Locke’s wager that processes of regulated but mostly free human interaction can solve problems better than elite decision-making castes.

In fact, we have already seen, in just this one article, more than enough reasons to understand why retreat simply cannot work over the long run. Human nature ensures that there can never be successful rule by serene and dispassionately wise “philosopher kings.” That approach had its fair trial—at least forty centuries—and by almost any metric, it failed.

As for the other three roads, well, there is simply no way that anyone—from the most enthusiastic, “extropian” utopian-transcendentalists to the most skeptical and pessimistic doomsayers—can prove that one path is more likely than the others. (How can models, created within an earlier, cruder system, properly simulate and predict the behavior of a later and vastly more complex system?) All we can do is try to understand which processes may increase our odds of achieving better outcomes. More robust outcomes. These processes will almost certainly be as much social as technological. They will, to a large degree, depend upon improving our powers of error-avoidance.

My contention—running contrary to many prescriptions from both left and right—is that we should trust Locke a while longer. This civilization already has in place a number of unique methods for dealing with rapid change. If we pay close attention to how these methods work, they might be improved dramatically, perhaps enough to let us cope, and even thrive. Moreover, the least helpful modification would appear to be the one thing that the Professional Castes tell us we need—an increase in paternalistic control.¹³

In fact, when you look at our present culture from an historical perspective, it is already profoundly anomalous in its emphasis upon individualism, progress, and above all, suspicion of authority (SOA). These themes were actively and vigorously repressed in a vast majority of human cultures, because they threatened the stable equilibrium upon which ruling classes always depended. In Western Civilization—by way of contrast—it would seem that every mass-media work of popular culture, from movies to novels to songs, promotes SOA as a central human value.¹⁴ This may, indeed, be the most unique thing about our culture, even more than our wealth and technological prowess.

Although we are proud of the resulting society—one that encourages eccentricity, appreciation of diversity, social mobility, and scientific progress—we have no right, as yet, to claim that this new way of doing things is especially sane or obvious. Many in other parts of the world consider Westerners to be quite mad! And with some reason. Indeed, only time will tell who is right about that. For example, if we take the suspicion of authority ethos to its extreme, and start paranoically mistrusting even our best institutions—as was the case with Oklahoma City bomber Timothy McVeigh—then it is quite possible that Western Civilization may fly apart before ever achieving its vaunted aims, and lead rapidly to some of the many ways that we might achieve outcome #1.

Certainly, a positive singularity (outcome #2) cannot happen if only centrifugal forces operate and there are no compensating centripetal virtues to keep us together as a society of mutually respectful sovereign citizens.

Above all (as I point out in The Transparent Society), our greatest innovations, the accountability arenas¹⁵ wherein issues of importance get decided—science, justice, democracy and free markets—are not arbitrary, nor are they based on whim or ideology. They all depend upon adversaries competing on specially designed playing fields, with hard-learned arrangements put in place to prevent the kinds of cheating that normally prevail whenever human beings are involved. Above all, science, justice, democracy, and free markets depend on the mutual accountability that comes from open flows of information.

Secrecy is the enemy that destroys each of them, and it could easily spread like an infection to spoil our frail renaissance.

THE BEST METHODS OF ERROR-AVOIDANCE

Clearly, our urgent goal is to find (and then avoid) a wide range of quicksand pits—potential failure modes—as we charge headlong into the future. At risk of repeating an oversimplification, we do this in two ways. One method is anticipation. The other is resiliency.

The first of these uses the famous prefrontal lobes—our most recent, and most spooky, neural organs—to peer ahead, perform gedankenexperiments, forecast problems, make models and devise countermeasures in advance. Anticipation can either be a lifesaver… or one of our most colorful paths to self-deception and delusion.¹⁶

The other approach—resiliency—involves establishing robust systems, reaction sets, tools and distributed strengths that can deal with almost any problem as it arises—even surprising problems the vaunted prefrontal lobes never imagined.

Now, of course, these two methods are compatible, even complementary. We have a better computer industry, overall, because part of it is centered in Boston and part in California, where different corporate cultures reign. Companies acculturated with a “northeast mentality” try to make perfect products. Employees stay in the same company, perhaps for decades. They feel responsible. They get the bugs out before releasing and shipping. These are people you want designing a banking program, or a defense radar, because we can’t afford a lot of errors in even the beta version, let alone the nation’s ATM machines! On the other hand, people who work in Silicon Valley seem to think almost like another species. They cry, “Let’s get it out the door! Innovate first and catch the glitches later! Our customers will tell us what parts of the product to fix on the fly. They want the latest thing and to hell with perfection.” Today’s Internet arose from that kind of creative ferment, adapting quickly to emergent properties of a system that turned out to be far more complex and fertile than its original designers anticipated. Indeed, their greatest claim to fame comes from having anticipated that unknown opportunities might emerge!

Sometimes the best kind of planning involves leaving room for the unknown.

This can be hard, especially when your duty is to prepare against potential failure modes that could harm or destroy a great nation. Government and military culture have always been anticipatory, seeking to analyze potential near-term threats and coming up with detailed plans to stymie them. This resulted in incremental approaches to thinking about the future. One classic cliché holds that generals are always planning to fight a modified version of the last war. History shows that underdogs—those who lost the last campaign or who bear a bitter grudge—often turn to innovative or resilient new strategies, while those who were recently successful are in grave danger of getting mired in irrelevant solutions from the past, often with disastrous consequences.¹⁷

At the opposite extreme is the genre of science fiction, whose attempts to anticipate the future are—when done well—part of a dance of resiliency. Whenever a future seems to gather a consensus around it, as happened to “cyberpunk” in the late eighties, the brightest SF authors become bored with such a trope and start exploring alternatives. Indeed, boredom could be considered one of the driving forces of ingenious invention, not only in science fiction, but in our rambunctious civilization as a whole.

Speaking as an author of speculative novels, I can tell you that it is wrong to think that science fiction authors try to predict the future. With our emphasis more on resiliency than anticipation, we are more interested in discovering possible failure modes and quicksand pits along the road ahead, than we are in providing a detailed and prophetic travel guide for the future.

Indeed, one could argue that the most powerful kind of science fiction tale is the self-preventing prophecy—any story or novel or film that portrays a dark future so vivid, frightening and plausible that millions are stirred to act against the scenario ever coming true. Examples in this noble (if terrifying) genre—which also can encompass visionary works of non-fiction—include Fail-Safe, Brave New World, Soylent Green, Silent Spring, The China Syndrome, Das Kapital, The Hot Zone, and greatest of all, George Orwell’s Nineteen Eighty-Four, now celebrating 60 years of scaring readers half to death. Orwell showed us the pit awaiting any civilization that combines panic with technology and the dark, cynical tradition of tyranny. In so doing, he armed us against that horrible fate. By exploring the shadowy territory of the future with our minds and hearts, we can sometimes uncover failure-modes in time to evade them.

Summing up, this process of gedanken or thought experimentation is applicable to both anticipation and resiliency. But it is only most effective when it is engendered en masse, in markets and other arenas where open competition among countless well-informed minds can foster the unique synergy that has made our civilization so different from hierarchy-led cultures that came before. A synergy that withers the bad notions under criticism, while allowing good ones to combine and multiply.

I cannot guarantee that this scenario will work over the dangerous ground ahead. An open civilization filled with vastly educated, empowered, and fully-knowledgeable citizens may be able to apply the cleansing light of reciprocal accountability so thoroughly that onrushing technologies cannot be horribly abused by either secretive elites or disgruntled AYMs (angry young men).

Or else… perhaps… that solution, which brought us so far in the 20^th Century, will not suffice in the accelerating 21^st. Perhaps nothing can work. Maybe this explains the Great Silence, out there among the stars.

What I do know is this. No other prescription has even a snowball’s chance of working. Open knowledge and reciprocal accountability seem, at least, to be worth betting on. They are the tricks that got us this far, in contrast to 4,000 years of near utter failure by systems of hierarchical command.

Anyone who says that we should suddenly veer back in that direction, down discredited and failure-riven paths of secrecy and hierarchy, should bear a steep burden of proof.

VARIETIES OF SINGULARITY EXPERIENCE

All right, what if we do stay on course, and achieve something like the Positive Singularity?

There is plenty of room to argue over what type would be beneficial or even desirable. For example, might we trade in our bodies—and brains—for successively better models, while retaining a core of humanity… of soul?

If organic humans seem destined to be replaced by artificial beings who are vastly more capable than we souped-up apes, can we design those successors to at least think of themselves as human? (This unusual notion is one that I’ve explored in a few short stories.) In that case, are you so prejudiced that you would begrudge your great-grandchild a body made of silicon, so long as she visits you regularly, tells good jokes, exhibits kindness, and is good to her own kids?

Or will they simply move on, sparing a moment to help us come to terms with our genteel obsolescence?

Some people remain big fans of Teilhard de Chardin’s apotheosis—the notion that we will all combine into a single macro-entity, almost literally godlike in its knowledge and perception. Physicist Frank Tipler speaks of such a destiny in his book, The Physics of Immortality, and Isaac Asimov offered a similar prescription as mankind’s long-range goal in Foundation’s Edge. I have never found this notion particularly appealing—at least in its standard presentation, by which some macro-being simply subsumes all lesser individuals within it, and then proceeds to think deep thoughts. In Earth, I talk about a variation on this theme that might be far more palatable, in which we all remain individuals, while at the same time contributing to a new of planetary consciousness. In other words, we could possibly get to have our cake and eat it too.

At the opposite extreme, in Foundation’s Triumph, my sequel to Asimov’s famous universe, I make more explicit something that Isaac had been alluding to all along—the possibility that conservative robots might dread human transcendence, and for that reason actively work to prevent a human singularity. Fearing that it could bring us harm. Or enable us to compete with them. Or empower us to leave them behind.

In any event, the singularity is a fascinating variation on all those other transcendental notions that seem to have bubbled, naturally and spontaneously, out of human nature since before records were kept. Even more than all the others, this one can be rather frustrating at times. After all, a good parent wants the best for his or her children—for them to do and be better. And yet, it can be poignant to imagine them (or perhaps their grandchildren) living almost like gods, with nearly omniscient knowledge and perception—and near immortality—taken for granted.

It’s tempting to grumble, “Why not me? Why can’t I be a god, too?”¹⁸

But then, when has human existence been anything but poignant?

Anyway, what is more impressive? To be godlike?

Or to be natural creatures, products of grunt evolution, who are barely risen from the caves… who nevertheless manage to learn nature’s rules, revere them, and then use them to create good things, good descendants, good destinies? Even godlike ones.

All of our speculations and musings (including this one) may eventually seem amusing and naive to those dazzling descendants. But I hope they will also experience moments of respect, when they look back at us.

They may even pause and realize that we were really pretty good… for souped-up cavemen. After all, what miracle could be more impressive than for such flawed creatures as us to design and sire gods?

There may be no higher goal. Or any that better typifies arrogant hubris.

Or else… perhaps… the fulfillment of our purpose and the reason for all that pain.

To have learned the compassion and wisdom that we’ll need, more than anything else, when bright apprentices take over the Master’s workroom. Hopefully winning merit and approval, at last, as we resume the process of creation.

name="_ftn1">1. Parts of this essay were transcribed from a speech before the conference Accelerating Change 2004: "Horizons of Perception in an Era of Change" November 2004 at Stanford University. Copyright 2005, by David Brin.

title="">2. In his article, “Molecular Manufacturing: Too Dangerous to Allow?” Robert A. Freitas Jr. describes this scenario. One common argument against pursuing a molecular assembler or nanofactory design effort is that the end results are too dangerous. According to this argument, any research into molecular manufacturing (MM) should be blocked because this technology might be used to build systems that could cause extraordinary damage. The kinds of concerns that nanoweapons systems might create have been discussed elsewhere, in both the nonfictional and fictional literature. Perhaps the earliest-recognized and best-known danger of molecular nanotechnology is the risk that self replicating nanorobots capable of functioning autonomously in the natural environment could quickly convert that natural environment (e.g., "biomass") into replicas of themselves (e.g., "nanomass") on a global basis, a scenario often referred to as the "gray goo problem" but more accurately termed global ecophagy". In explaining this scenario, Freitas does not endorse it.

name="_ftn3">3. "Why the future doesn’t need us." Wired Magazine, Issue 8.04, April 2000.

name="_ftn4">4. While my description of The End of History oversimplifies a bit, one can wish that predictions in social science were as well tracked for credibility as they are in physics. Back in 1986, at the height of Reagan-era confrontations, I forecast an approaching fall of the Berlin Wall, to be followed by several decades of tense confrontation with "one or another branch of macho culture, probably Islamic."

name="_ftn5">5. For more on this quandary and its implications, see: http://www.davidbrin.com/sciencearticles.html

name="_ftn6">6. And more quasi-religious social-political mythologies followed, from the incantations of Ayn Rand to MaoZedong. All of them crafting "logical chains of cause and effect that forecast utter human transformation, by political (as opposed to spiritual or technical) means.

name="_ftn7">7. For a detailed response to Huebner’s anti-innovation argument, see Review of "A Possible Declining Trend for Worldwide Innovation" by Jonathan Huebner, published by John Smart in the September 2005 issue of Technological Forecasting and Social Change http://accelerating.org/articles/huebnerinnovation.html

name="_ftn8">8. The Exorarium Project proposes to achieve all this and more, by inviting both museum visitors and online participants to enter a unique learning environment. Combining state-of-the-art simulation and visualization systems, plus the very best ideas from astronomy, physics, chemistry, and ecology, the Exorarium will empower users to create vividly plausible extraterrestrials and then test them in realistic first contact scenarios. http://www.exorarium.com/

name="_ftn9">9. For a rather intense look at how "truth" is determined in science, democracy, courts and markets, see the lead article in the American Bar Association’s Journal on Dispute Resolution (Ohio State University), v.15, N.3, pp 597-618, Aug. 2000, "Disputation Arenas: Harnessing Conflict and Competition for Society’s Benefit" or at: http://www.davidbrin.com/disputationarticle1.html

name="_ftn10">10.In other places, I discuss various proposed ways to deal with the Problem of Loyalty, in some future age when machine intelligences might excel vastly beyond the capabilities of mere organic brains. Older proposals (e.g. Asimov’s “laws of robotics”) almost surely cannot work. It remains completely unknown whether humans can “go along for the ride” by using cyborg enhancements or “linking” with external processors. In the long run, I suggest that we might deal with this in the same way that all prior generations created new (and sometimes superior) beings without much shame or fear. By raising them to think of themselves as human beings, with our same values and goals. In other words, as our children. (See: http://www.davidbrin.com/lungfish1.html)

name="_ftn11">11. Of course, there are other possibilities, indeed many others, or I would not be worth my salt as a science fiction author or futurist. Among the more sophomorically entertaining possibilities is the one positing that we all live in a simulation, in some already post-singularity “context” such as a vast computer. The range is limitless. But these four categories seem to lay down the starkness of our challenge: to become wise, or see everything fail within a single lifespan.

name="_ftn12">12. This endeavor has been based upon earlier Asian success stories, in Japan and in Singapore, extrapolating from their mistakes. Most notable has been an apparent willingness to learn pragmatic lessons, to incorporate limited levels of criticism and democracy, accepting their value as error-correction mechanisms—while limiting their effectiveness as threats to hierarchical rule. One might imagine that this tightrope act must fail, once universal education rises beyond a certain point. But that is only a hypothesis. Certainly the neo-confucians can point to the sweep of history, supporting their wager.

name="_ftn13">13. See my essay on "Beleaguered Professionals vs. Disempowered Citizens" about a looming 21^st Century power struggle between average people and the sincere, skilled professionals who are paid to protect us: http://www.amazon.com/gp/product/B000BY2PRQ/002-1071896-8741633 In a related context, a ‘futurist essay’ points out a rather unnoticed aspect of the tragedy of 9/11/01—that citizens themselves were most effective in our civilization’s defense. The only actions that actually saved lives and thwarted terrorism on that awful day were taken amid rapid, ad hoc decisions made by private individuals, reacting with both resiliency and initiative—our finest traits—and armed with the very same new technologies that dour pundits say will enslave us. Could this point to a trend for the 21^st Century, reversing what we’ve seen throughout the 20^th… the ever-growing dependency on professionals to protect and guide and watch over us? See: http://www.futurist.com/portal/future_trends/david_brin_empowerment.htm

name="_ftn14">14. Take the essential difference between moderate members of the two major American political parties. This difference boils down to which elites you accuse of seeking to accumulate too much authority. A decent Republican fears snooty academics, ideologues, and faceless bureaucrats seeking to become paternalistic Big Brothers. A decent Democrat looks with worried eyes toward conspiratorial power grabs by conniving aristocrats, faceless corporations, and religious fanatics. (A decent Libertarian picks two from Column A and two from Column B!) I have my own opinions about which of these elites are presently most dangerous. (Hint: it is the same one that dominated most other urban cultures, for four thousand years.) But the startling irony, that is never discussed, is how much in common these fears really share. And the fact that—indeed—every one of them is right to worry. In fact, only universal SOA makes any sense. Instead of an ideologically blinkered focus on just one patch of horizon, should we not agree to watch all directions where tyranny or rationalized stupidity might arise? Again, reciprocal accountability appears to be the only possible solution.

name="_ftn15">15. For a rather intense look at how "truth" is determined in science, democracy, courts and markets, see the lead article in the American Bar Association’s Journal on Dispute Resolution (Ohio State University), v.15, N.3, pp 597-618, Aug. 2000, "Disputation Arenas: Harnessing Conflict and Competition for Society’s Benefit." or at: http://www.davidbrin.com/disputationarticle1.html

name="_ftn16">16. I say this as a prime practitioner of the art of anticipation, both in nonfiction and in fiction. Every futurist and novelist deals in creating convincing illusions of prescience… though at times these illusions can be helpful.

name="_ftn17">17. It is worth noting that the present US military Officer Corps has tried strenuously to avoid this trap, endeavoring to institute processes of re-evaluation, by which a victorious and superior force actually thinks like one that has been defeated. In other words, with a perpetual eye open to innovation. And yet, despite this new and intelligent spirit of openness, military thinking remains rife with unwarranted assumptions. Almost as many as swarm through practitioners of politics.

name="_ftn18">18. Of course, there are some Singularitarians—true believers in a looming singularity—who expect it to rush upon us so rapidly that even fellows my age (in my fifties) will get to ride the immortality wave. Yeah, right. And they call me a dreamer.

© 2006 David Brin

Nanotechnology—the precise engineering of tiny but powerful machines—is advancing quickly, leaping from the pages of science fiction into world-class research laboratories, and coming soon to a desktop near you.

Like electricity or computers before it, nanotechnology will bring greatly improved efficiency and productivity in many areas of human endeavor. In its mature form, known as molecular nanotechnology (MNT) or molecular manufacturing (MM), it will have significant impact on almost all industries and all parts of society. Personal nanofactories (PNs) may offer better built, longer lasting, cleaner, safer, and smarter products for the home, for communications, for medicine, for transportation, for agriculture, and for industry in general.

However, as a general-purpose technology, MM will be dual-use, meaning that in addition to its civilian applications, it will have military uses as well—making far more powerful weapons and tools of surveillance. Thus, it represents not only wonderful benefits for humanity, but also grave risks.

Progress toward developing the technical requirements for desktop molecular manufacturing is moving forward rapidly. By reading the collection of essays in the March 27 issue of Nanotechnology Perceptions—or reading them here at KurweilAI.net—you will learn how PNs will bring radical changes to society, and to your life.

Several factors will come together to make MM truly revolutionary.

• Cost: One PN can build another PN as easily as any other product, so nanofactories will be neither scarce nor expensive. Labor costs will also be minimal, since PNs will be automated. Small carbon-based molecules (feedstock) are quite inexpensive.

• Exponential manufacturing: One PN can be made to build two, or a small system can build one twice as big. Working in parallel, manufacturing capacity can double every few hours. Within just a few months, a single molecular manipulation device could be expanded to PN’s with a combined capacity of thousands of tons per hour. The PN architecture can even scale to individual factories of industrial size.

• Precision: Atoms of each type are identical with each other, and products made from precisely placed atoms also will be identical—more reliable and easier to manufacture.

• High performance: Small machines are more powerful than large ones—perhaps a million times more powerful, when shrunk to nano-scale—and precise materials are perhaps 100 times stronger. Also, precise surfaces can have extremely low friction and wear. Nanofactory-built products could include large numbers of small, high-performance machines.

• General-purpose manufacturing: Structures will be made by automated placement of tiny building blocks, so changing the program (blueprint) will change the product. A wide range of components and products is possible, including computers, sensors, motors, and displays, and combinations thereof.

• Rapid prototyping: Because a nanofactory will make a complete product in a few minutes from any given blueprint, new product designs could be built and tested almost immediately, and at very low cost.

OVERVIEW OF ESSAYS

In August 2005, the Center for Responsible Nanotechnology (CRN), a non-profit research and advocacy organization, announced the formation of a Task Force convened to study the societal implications of this rapidly emerging technology. Bringing together a diverse group of world-class experts from multiple disciplines, CRN is spearheading an historic, collaborative effort to develop comprehensive recommendations for the safe and responsible use of nanotechnology.

Many of the profound implications of molecular manufacturing are explored in an initial collection of 11 new essays, all written by members of the CRN Task Force and published in the March 24 issue of Nanotechnology Perceptions. From military and security issues to human enhancement, artificial intelligence, and more, we take a look under the lid of Pandora’s box to see what the future might hold. A second collection of essays exploring additional concerns will form the next issue of Nanotechnology Perceptions.

Reacting to the huge risks of MM, some advocate that all research be halted. Our first two essays, “Nanotechnology Dangers and Defenses” by inventor and author Ray Kurzweil and “Molecular Manufacturing: Too Dangerous to Allow?” by Nanomedicine author Robert A. Freitas Jr., explore these issues. They survey the dangers, discuss ways to mitigate them, and analyze the weaknesses of relinquishment.

“Nano-Guns, Nano-Germs, and Nano-Steel,” an essay by Mike Treder, explores the troubling topic of nanotech-enabled warfare. Tom Cowper, an expert in policing and criminology, offers his special perspective in “Molecular Manufacturing and 21^st Century Policing.” In “The Need For Limits,” Chris Phoenix explains that we may face unprecedented risks as MM’s revolutionary potential dissolves the barriers that keep us safe.

After Giulio Prisco explores the real-world challenge of “Globalization and Open Source Nano Economy,” Damien Broderick provides a broad historical perspective of the relationship between society and technology in “Cultural Dominants and Differential MNT Uptake.”

Advanced nanotechnology could go well beyond making better consumer goods and better weapons. In “Nanoethics and Human Enhancement,” professional ethicists Patrick Lin and Fritz Allhoff look into the controversial aspects of using MM to change our bodies and minds. Noted futurist Natasha Vita-More then lays out the problems our grey matter could face in “Strategic Sustainable Brain.”

Computers built by nanofactories may be millions of times more powerful than anything we have today. The potential for creating world-changing artificial intelligence is examined by scientist J. Storrs Hall in “Is AI Near a Takeoff Point?” Finally, if some of our worst scenarios become real, we may face truly existential dilemmas. These are surveyed in depth by best-selling author David Brin in “Singularities and Nightmares: The Range of Our Futures.”

As editors of these essays, we will be pleased if you are entertained and informed. But we will be further gratified if you are inspired to learn more. We hope you’ll want to get involved in the vital work of raising awareness and finding effective solutions to the challenges presented to the world by advanced nanotechnology.

Mike Treder, Executive Director

Chris Phoenix, Director of Research

Center for Responsible Nanotechnology (www.CRNano.org)

Note: The opinions expressed in these essays are those of the individual authors and do not necessarily represent the opinions of the Center for Responsible Nanotechnology, nor of its parent organization, World Care.

The first half of the 21st century will be characterized by three overlapping revolutions—in Genetics, Nanotechnology, and Robotics (GNR). The deeply intertwined promise and peril of these technologies has led some serious thinkers to propose that we go very cautiously, possibly even to abandon them altogether.

A few years ago, computer maven Bill Joy wrote, “We are being propelled into a new century with no plan, no control, no brakes… The only realistic alternative I see is relinquishment: to limit the development of the technologies that are too dangerous, by limiting our pursuit of certain kinds of knowledge.”¹

Joy’s deep concern about the future grew out of a conversation we had in 1998 about these emerging technologies, and an early draft of The Age of Spiritual Machines that I gave him. Although I have a reputation as a technology optimist, it turns out that at public discussions of “promise and peril,” I often spend much of my time defending Joy’s position on the feasibility of the dangers that concern him. Indeed, Joy and I agree on both promise and peril.

Technology has always been a mixed blessing, bringing us benefits such as longer and healthier lifespans, freedom from physical and mental drudgery, and many novel creative possibilities on the one hand, while introducing new dangers. Technology empowers both our creative and destructive natures.

Broad relinquishment is contrary to economic progress and is ethically unjustified given the opportunity to alleviate disease, overcome poverty, and clean up the environment. Joy and I also agree that relinquishment of major fields such as genetics (“G”), nanotechnology (“N”), or strong AI/robotics (“R”) is not the answer. There is, however, a growing movement advocating exactly that. Bill McKibben, the environmentalist who first brought global warming to our attention, argues in his book Enough that we have had “enough” technology and should not pursue more. However, regulations on safety—essentially fine-grained relinquishment—will remain an appropriate strategy. In that spirit, Joy and I recently wrote a joint op ed piece (“Recipe for Destruction”) published in the New York Times on October 17, 2005 criticizing the publication of the 1918 flu genome on the web.

Dangers to Defend Against

As technology accelerates toward the full realization of GNR, we will see interweaving potentials: a feast of creativity resulting from human intelligence expanded manyfold, combined with many grave new dangers. A quintessential concern that has received considerable attention is unrestrained nanobot replication. Early proposals for molecular manufacturing required trillions of intelligently designed devices to be useful. To scale up to such levels it would have been necessary to enable them to self-replicate, essentially the same approach used in the biological world (that’s how one fertilized egg cell becomes the trillions of cells in a human).

Although the self-replication can be hidden and blocked in a variety of ways (for example, Ralph Merkle’s proposal¹ for a “broadcast architecture” in which each replicating entity needs to get the replicating codes from a secure server), the overall system will have self-replication at some level. And in the same way that biological self-replication gone awry (that is, cancer) results in biological destruction, a defect in a mechanism curtailing nanobot self-replication—the so-called gray goo scenario—would endanger all physical entities, biological or otherwise.

Modern proposals, such as the use of large integrated manufacturing systems rather than trillions of quasi-independent nanobots, appear to prevent inadvertent release of destructive self-replication, but in general these safeguards can be worked around by a determined adversary. We see a similar situation today in biological technologies. The ethical guidelines for gene modification technologies adopted at the Asilomar Conference have worked well for over a quarter of a century, but these guidelines would not restrict a would-be bioterrorist because they don’t have to follow the guidelines (they don’t have to put their “inventions” through the FDA either).

These guidelines and strategies are likely to be effective for preventing accidental release of dangerous self-replicating nanotechnology entities. But dealing with the intentional design and release of such entities is a more complex and challenging problem. A sufficiently determined and destructive opponent could possibly defeat each of these layers of protections. Take, for example, the broadcast architecture. When properly designed, each entity is unable to replicate without first obtaining replication codes, which are not repeated from one replication generation to the next. However, a modification to such a design could bypass the destruction of the replication codes and thereby pass them on to the next generation. To counteract that possibility it has been recommended that the memory for the replication codes be limited to only a subset of the full code. However, this guideline could be defeated by expanding the size of the memory.

Another protection that has been suggested is to encrypt the codes and build in protections in the decryption systems, such as time-expiration limitations. However, we can see how easy is has been to defeat protections against unauthorized replications of intellectual property such as music files. Once replication codes and protective layers are stripped away, the information can be replicated without these restrictions.

This doesn’t mean that that protection is impossible. Rather, each level of protection will work only to a certain level of sophistication. The meta lesson here is that we will need to place twenty-first-century society’s highest priority on the continuing advance of defensive technologies, keeping them one or more steps ahead of the destructive technologies (or at least no more than a quick step behind).

Living creatures—including humans—would be the primary victims of an exponentially spreading nanobot attack. The principal designs for nanobot construction use carbon as a primary building block. Because of carbon’s unique ability to form four-way bonds, it is an ideal building block for molecular assemblies. Because biology has made the same use of carbon, pathological nanobots would find the Earth’s biomass an ideal source of this primary ingredient.

How long would it take an out-of-control replicating nanobot to destroy the Earth’s biomass? The biomass has on the order of 10⁴⁵ carbon atoms. A reasonable estimate of the number of carbon atoms in a single replicating nanobot is about 10⁶. (Note that this analysis is not very sensitive to the accuracy of these figures, only to the approximate order of magnitude.) This malevolent nanobot would need to create on the order of 10³⁴ copies of itself to replace the biomass, which could be accomplished with 113 replications (each of which would potentially double the destroyed biomass). Rob Freitas has estimated a minimum replication time of approximately 100 seconds, so 113 replication cycles would require about three hours.² However, the actual rate of destruction would be slower because biomass is not “efficiently” laid out. The limiting factor would be the actual movement of the front of destruction. Nanobots cannot travel very quickly because of their small size. It’s likely to take weeks for such a destructive process to circle the globe.

Based on this observation we can envision a more insidious possibility. In a two-phased attack, the nanobots take several weeks to spread throughout the biomass but use up an insignificant portion of the carbon atoms, say one out of every thousand trillion (10¹⁵). At this extremely low level of concentration, the nanobots would be as stealthy as possible. Then, at an “optimal” point, the second phase would begin with the seed nanobots expanding rapidly in place to destroy the biomass. For each seed nanobot to multiply itself a thousand trillionfold would require only about 50 binary replications, or about 90 minutes. With the nanobots having already spread out in position throughout the biomass, movement of the destructive wave front would no longer be a limiting factor.

The point is that without defenses, the available biomass could be destroyed by gray goo very rapidly. Clearly, we will need a nanotechnology immune system³ in place before these scenarios become a possibility. This immune system would have to be capable of contending not just with obvious destruction but any potentially dangerous (stealthy) replication, even at very low concentration.

Eric Drexler, Robert Freitas, Ralph Merkle, Mike Treder, Chris Phoenix, and others have pointed out that future nanotech manufacturing devices can be created with safeguards that would prevent the accidental creation of self-replicating nanodevices.⁴ However, this observation, although important, does not eliminate the threat of gray goo as I pointed out above. There are other reasons (beyond manufacturing) that self-replicating nanobots will need to be created. The nanotechnology immune system mentioned above, for example, will ultimately require self-replication; otherwise it would be unable to defend us against the development of increasingly sophisticated types of goo. It is also likely to find extensive military applications. Moreover, a determined adversary or terrorist can defeat safeguards against unwanted self-replication; hence, the need for defense.

Bill Joy and other observers have pointed out that such an immune system would itself be a danger because of the potential of “autoimmune” reactions (that is, the immune-system nanobots attacking the world they are supposed to defend). However, this possibility is not a compelling reason to avoid the creation of an immune system. No one would argue that humans would be better off without an immune system because of the potential of developing autoimmune diseases. Although the biological immune system can itself present a danger, humans would not last more than a few weeks (barring extraordinary efforts at isolation) without one. And even so, the development of a technological immune system for nanotechnology will happen even without explicit efforts to create one. This has effectively happened with regard to software viruses, creating an immune system not through a formal grand-design project but rather through incremental responses to each new challenge and by developing heuristic algorithms for early detection. We can expect the same thing will happen as challenges from nanotechnology-based dangers emerge. The point for public policy will be to specifically invest in these defensive technologies.

As a test case, we can take a small measure of comfort from how we have dealt with one recent technological challenge. There exists today a new fully nonbiological self-replicating entity that didn’t exist just a few decades ago: the computer virus. When this form of destructive intruder first appeared, strong concerns were voiced that as they became more sophisticated, software pathogens had the potential to destroy the computer-network medium in which they live. Yet the “immune system” that has evolved in response to this challenge has been largely effective. Although destructive self-replicating software entities do cause damage from time to time, the injury is but a small fraction of the benefit we receive from the computers and communication links that harbor them.

One might counter that computer viruses do not have the lethal potential of biological viruses or of destructive nanotechnology. This is not always the case; we rely on software to operate our 911 call centers, monitor patients in critical-care units, fly and land airplanes, guide intelligent weapons in our military campaigns, handle our financial transactions, operate our municipal utilities, and many other mission-critical tasks. To the extent that software viruses do not yet pose a lethal danger, however, this observation only strengthens my argument. The fact that computer viruses are not usually deadly to humans only means that more people are willing to create and release them. The vast majority of software virus authors would not release viruses if they thought they would kill people. It also means that our response to the danger is that much less intense. Conversely, when it comes to self-replicating entities that are potentially lethal on a large scale, our response on all levels will be vastly more serious.

Although software pathogens remain a concern, the danger exists today mostly at a nuisance level. Keep in mind that our success in combating them has taken place in an industry in which there is no regulation and minimal certification for practitioners. The largely unregulated computer industry is also enormously productive. One could argue that it has contributed more to our technological and economic progress than any other enterprise in human history.

But the battle concerning software viruses and the panoply of software pathogens will never end. We are becoming increasingly reliant on mission-critical software systems, and the sophistication and potential destructiveness of self-replicating software weapons will continue to escalate. When we have software running in our brains and bodies and controlling the world’s nanobot immune system, the stakes will be immeasurably greater.

The Right Level of Relinquishment

The only conceivable way that the accelerating pace of GNR technology advancement could be stopped would be through a worldwide totalitarian system that relinquishes the very idea of progress. Even this specter would be likely to fail in averting the dangers of GNR because the resulting underground activity would tend to favor the more destructive applications. This is because the responsible practitioners that we rely on to quickly develop defensive technologies would not have easy access to the needed tools. Fortunately, such a totalitarian outcome is unlikely because the increasing decentralization of knowledge is inherently a democratizing force.

I do think that relinquishment at the right level needs to be part of our ethical response to the dangers of 21st century technologies. One constructive example of this is the ethical guideline proposed by the Foresight Institute: namely, that nanotechnologists agree to relinquish the development of physical entities that can self-replicate in a natural environment. In my view, there are two exceptions to this guideline. First, we will ultimately need to provide a nanotechnology-based planetary immune system (nanobots embedded in the natural environment to protect against rogue self-replicating nanobots). Robert Freitas and I have discussed whether or not such an immune system would itself need to be self-replicating. Freitas writes: “A comprehensive surveillance system coupled with prepositioned resources—resources including high-capacity nonreplicating nanofactories able to churn our large numbers of nonreplicating defenders in response to specific threats—should suffice.”⁵ I agree with Freitas that a prepositioned immune system with the ability to augment the defenders will be sufficient in early stages. But once strong AI is merged with nanotechnology, and the ecology of nanoengineered entities becomes highly varied and complex, my own expectation is that we will find that the defending nanorobots need the ability to replicate in place quickly. Biological evolution essentially made the same “discovery.” The other exception is the need for self-replicating nanobot-based probes to explore planetary systems outside of our solar system.

Broad relinquishment of GNR technologies would be unwise for several reasons. However, I do think we need to take seriously the increasingly strident voices that advocate for it, even though many of these advocates are motivated by a general distrust of technology, and their proposals are not well considered. Although blanket relinquishment is not the answer, rational fear could lead to irrational solutions, and those solutions may cause severe negative consequences.

A summary of an overall strategy for defending against the downsides of emerging GNR technologies would include the following:

• We need to streamline the regulatory process for genetic and medical technologies. The regulations do not impede the malevolent use of technology but significantly delay the needed defenses. As mentioned, we need to better balance the risks of new technology (for example, new medications) against the known harm of delay.

• A global program of confidential, random serum monitoring for unknown or evolving biological pathogens should be funded. Diagnostic tools exist to rapidly identify the existence of unknown protein or nucleic acid sequences. Intelligence is key to defense, and such a program could provide invaluable early warning of an impending epidemic. Such a ‘pathogen sentinel’ program has been proposed for many years by public health authorities but has never received adequate funding.

• Well-defined and targeted temporary moratoriums, such as the one that occurred in the genetics field in 1975, may be needed from time to time. But such moratoriums are unlikely to be necessary with nanotechnology. Broad efforts at relinquishing major areas of technology serve only to continue vast human suffering by delaying the beneficial aspects of new technologies, and actually make the dangers worse.

• Efforts to define safety and ethical guidelines for nanotechnology should continue. Such guidelines will inevitably become more detailed and refined as we get closer to molecular manufacturing.

• To create the political support to fund the efforts suggested above, it is necessary to raise public awareness of these dangers. Because, of course, there exists the downside of raising alarm and generating uninformed backing for broad antitechnology mandates, we also need to create a public understanding of the profound benefits of continuing advances in technology.

• These risks cut across international boundaries—which is, of course, nothing new; biological viruses, software viruses, and missiles already cross such boundaries with impunity. International cooperation was vital to containing the SARS virus and will become increasingly vital in confronting future challenges. Worldwide organizations such as the World Health Organization, which helped coordinate the SARS response, and is now dealing with the possibility of a bird flu pandemic, need to be strengthened.

• A contentious contemporary political issue is the need for preemptive action to combat threats, such as terrorists with access to weapons of mass destruction or rogue nations that support such terrorists. Such measures will always be controversial, but the potential need for them is clear. A nuclear explosion can destroy a city in seconds. A self-replicating pathogen, whether biological or nanotechnology based, could destroy our civilization in a matter of days or weeks. We cannot always afford to wait for the massing of armies or other overt indications of ill intent before taking protective action.

• Intelligence agencies and policing authorities will have a vital role in forestalling the vast majority of potentially dangerous incidents. Their efforts need to involve the most powerful technologies available. For example, before this decade is out devices the size of dust particles will be able to carry out reconnaissance missions. When we reach the 2020s and have software running in our bodies and brains, government authorities will have a legitimate need on occasion to monitor these software streams. The potential for abuse of such powers is obvious. We will need to achieve a middle road of preventing catastrophic events while preserving our privacy and liberty.

• The above approaches will be inadequate to deal with the danger from pathological R (strong AI). Our primary strategy in this area should be to optimize the likelihood that future nonbiological intelligence will reflect our values of liberty, tolerance, and respect for knowledge and diversity. The best way to accomplish this is to foster those values in our society today and going forward. If this sounds vague, it is. But there is no purely technical strategy that is workable in this area because greater intelligence will always find a way to circumvent measures that are the product of a lesser intelligence. The nonbiological intelligence we are creating is and will be embedded in our societies and will reflect our values, as inconsistent and conflicted as these may appear to be. The transbiological phase will involve nonbiological intelligence deeply integrated with biological intelligence. This will amplify our abilities, and our application of these greater intellectual powers will be governed by the values of its creators. The transbiological era will ultimately give way to the postbiological era, but it is to be hoped that our values will remain influential. This strategy is certainly not foolproof, but it is the primary means we have today to influence the future course of strong AI.

Technology will remain a double-edged sword. It represents vast power to be used for all humankind’s purposes. GNR will provide the means to overcome age-old problems such as illness and poverty, but also will empower destructive ideologies. We have no choice but to strengthen our defenses while we apply these quickening technologies to advance our human values, despite an apparent lack of consensus on what those values should be.

1 Bill Joy, "Why the future doesn’t need us," Wired April 2000, http://www.wired.com/wired/archive/8.04/joy_pr.html

2 “Self replicating systems and low cost manufacturing” (1994) http://www.zyvex.com/nanotech/selfRepNATO.html The Gray Goo Problem

3 More fully discussed in my book, The Singularity is Near, Chapter 8

4 "Gray Goo is a Small Issue," Center for Responsible Nanotechnology, Dec. 14, 2003, http://www.crnano.org/BD-Goo.htm

5 Private correspondence

© 2006 Ray Kurzweil

One common argument against pursuing a molecular assembler or nanofactory design effort is that the end results are too dangerous. According to this argument [2, 3], any researh into molecular manufacturing (MM) should be blocked because this technology might be used to build systems that could cause extraordinary damage. The kinds of concerns that nanoweapons systems might create have been discussed elsewhere, in both the nonfictional [4-6] and fictional [7] literature. Perhaps the earliest-recognized and best-known danger of molecular manufacturing ^[.1] is the risk that self-replicating nanorobots capable of functioning autonomously in the natural environment could quickly convert that natural environment (e.g., “biomass”) into replicas of themselves (e.g., “nanomass”) on a global basis, a scenario often referred to as the “gray goo problem” but more accurately termed “global ecophagy” [4]. As Drexler first warned in Engines of Creation in 1986 [8]:

“Plants” with “leaves” no more efficient than today’s solar cells could out-compete real plants, crowding the biosphere with an inedible foliage. Tough omnivorous “bacteria” could out-compete real bacteria: They could spread like blowing pollen, replicate swiftly, and reduce the biosphere to dust in a matter of days. Dangerous replicators could easily be too tough, small, and rapidly spreading to stop—at least if we make no preparation…. We cannot afford certain kinds of accidents with replicating assemblers.

Such self-replicating systems, if not countered, could make the earth largely uninhabitable [4, 7-9]—concerns that motivated the drafting of the Foresight Guidelines for the safe development of nanotechnology [10]. But, as the Center for Responsible Nanotechnology explains [5], (reference annotations added):

Gray goo would entail five capabilities integrated into one small package. These capabilities are: Mobility—the ability to travel through the environment; Shell—a thin but effective barrier to keep out diverse chemicals and ultraviolet light; Control—a complete set of blueprints and the computers to interpret them (even working at the nanoscale, this will take significant space); Metabolism—breaking down random chemicals into simple feedstock; and Fabrication—turning feedstock into nanosystems. A nanofactory would use tiny fabricators, but these would be inert if removed or unplugged from the factory. The rest of the listed requirements would require substantial engineering and integration [4].

Although gray goo has essentially no military and no commercial value, and only limited terrorist value, it could be used as a tool for blackmail. Cleaning up a single gray goo outbreak would be quite expensive and might require severe physical disruption of the area of the outbreak (atmospheric and oceanic goos [4] deserve special concern for this reason). Another possible source of gray goo release is irresponsible hobbyists. The challenge of creating and releasing a self-replicating entity apparently is irresistible to a certain personality type, as shown by the large number of computer viruses and worms in existence. We probably cannot tolerate a community of “script kiddies” [11] releasing many modified versions of goo.

Development and use of molecular manufacturing poses absolutely no risk of creating gray goo by accident at any point. However, goo type systems do not appear to be ruled out by the laws of physics, and we cannot ignore the possibility that the five stated requirements could be combined deliberately at some point, in a device small enough that cleanup would be costly and difficult. Drexler’s 1986 statement can therefore be updated: We cannot afford criminally irresponsible misuse of powerful technologies. Having lived with the threat of nuclear weapons for half a century, we already know that.

Attempts to block or “relinquish” [3, 12] molecular manufacturing research will make the world a more, not less, dangerous place [13]. This paradoxical conclusion is founded on two premises. First, attempts to block the research will fail. Second, such attempts will preferentially block or slow the development of defensive measures by responsible groups. One of the clear conclusions reached by Freitas [4] was that effective countermeasures against self-replicating systems should be feasible, but will require significant effort to develop and deploy. (Nanotechnology critic Bill Joy, responding to this author, complained in late 2000 that any nanoshield defense to protect against global ecophagy “appears to be so outlandishly dangerous that I can’t imagine we would attempt to deploy it.” [12]) But blocking the development of defensive systems would simply insure that offensive systems, once deployed, would achieve their intended objective in the absence of effective countermeasures. James Hughes [13] concurs: “The only safe and feasible approach to the dangers of emerging technology is to build the social and scientific infrastructure to monitor, regulate and respond to their threats.”

We can reasonably conclude that blocking the development of defensive systems would be an extraordinarily bad idea. Actively encouraging rapid development of defensive systems by responsible groups while simultaneously slowing or hindering development and deployment by less responsible groups (“nations of concern”) would seem to be a more attractive strategy, and is supported by the Foresight Guidelines [10]. As even nanotechnology critic Bill Joy [14] finally admitted in late 2003: “These technologies won’t stop themselves, so we need to do whatever we can to give the good guys a head start.”

While a 100% effective ban against development might theoretically be effective at avoiding the potential adverse consequences, blocking all groups for all time does not appear to be a feasible goal. The attempt would strip us of defenses against attack, increasing rather than decreasing the risks. In addition, blocking development would insure that the substantial economic, environmental, and medical benefits [15] of this new technology would not be available.

Observes Glenn Reynolds [16]:

To the extent that such efforts [to ban all development] succeed, the cure may be worse than the disease. In 1875, Great Britain, then the world’s sole superpower, was sufficiently concerned about the dangers of the new technology of high explosives that it passed an act barring all private experimentation in explosives and rocketry. The result was that German missiles bombarded London rather than the other way around. Similarly, efforts to control nanotechnology, biotechnology or artificial intelligence are more likely to drive research underground (often under covert government sponsorship, regardless of international agreement) than they are to prevent research entirely. The research would be conducted by unaccountable scientists, often in rogue regimes, and often under inadequate safety precautions. Meanwhile, legitimate research that might cure disease or solve important environmental problems would suffer.

Finally, and as explained elsewhere [17], it is well-known [18] that self-replication activities, as distinct from the inherent capacity for self-replication, are not strictly required to achieve the anticipated broad benefits of molecular manufacturing. By restricting the capabilities of nanomanufacturing systems simultaneously along multiple design dimensions such as control autonomy (A1), nutrition (E4), mobility (E10), immutability (L3, L4), etc. [19], molecular manufacturing systems—whether microscale or macroscale—can be made inherently safe.

As Phoenix and Drexler [20] noted in a 2004 paper:

In 1959, Richard Feynman pointed out that nanometer-scale machines could be built and operated, and that the precision inherent in molecular construction would make it easy to build multiple identical copies. This raised the possibility of exponential manufacturing, in which production systems could rapidly and cheaply increase their productive capacity, which in turn suggested the possibility of destructive runaway self-replication. Early proposals for artificial nanomachinery focused on small self-replicating machines, discussing their potential productivity and their potential destructiveness if abused…. [But] nanotechnology-based fabrication can be thoroughly non-biological and inherently safe: such systems need have no ability to move about, use natural resources, or undergo incremental mutation. Moreover, self-replication is unnecessary: the development and use of highly productive systems of nanomachinery (nanofactories) need not involve the construction of autonomous self-replicating nanomachines…. Although advanced nanotechnologies could (with great difficulty and little incentive) be used to build such devices, other concerns present greater problems. Since weapon systems will be both easier to build and more likely to draw investment, the potential for dangerous systems is best considered in the context of military competition and arms control.

Of course, it must be conceded that while nanotechnology-based manufacturing systems can be made safe, they also could be made dangerous. Just because free-range self-replicators may be “undesirable, inefficient and unnecessary” [20] does not imply that they cannot be built, or that nobody will build them. How can we avoid “throwing out the baby with the bathwater”? The correct solution, first explicitly proposed by Freitas in 2000 [21] and later partially echoed by Phoenix and Drexler in 2004, [22] starts with a carefully targeted moratorium or outright legal ban on the most dangerous kinds of nanomanufacturing systems, while still allowing the safe kinds of nanomanufacturing systems to be built—subject to appropriate monitoring and regulation commensurate with the lesser risk that they pose.

Virtually every known technology comes in “safe” and “dangerous” flavors which necessarily must receive different legal treatment. For example, over-the-counter drugs are the safest and most difficult to abuse, hence are lightly regulated; prescription drugs, more easy to abuse, are very heavily regulated; and other drugs, typically addictive narcotics and other recreational substances, are legally banned from use by anyone, even for medicinal purposes. Artificial chemicals can range from lightly regulated household substances such as Clorox or ammonia; to more heavily regulated compounds such as pesticides, solvents and acids; to the most dangerous chemicals such as chemical warfare agents which are banned outright by international treaties. Another example is pyrotechnics, which range from highway flares, which are safe enough to be purchased and used by anyone; to “safe and sane” fireworks, which are lightly regulated but still available to all; to moderately-regulated firecrackers and model rocketry; to minor explosives and skyrockets, which in most states can be legally obtained and used only by licensed professionals who are heavily regulated; to high-yield plastic explosives, which are legally accessible only to military specialists; to nuclear explosives, the possession of which is strictly limited to a handful of nations via international treaties, enforced by an international inspection agency. Yet another example is aeronautics technology, which ranges from safe unregulated kites and paper airplanes; to lightly regulated powered model airplanes operated by remote control; to moderately regulated civilian aircraft, both small and large; to heavily regulated military attack aircraft such as jet fighters and bombers, which can only be purchased by approved governments; to intercontinental ballistic missiles, the possession of which is strictly limited to a handful of nations via international treaties.

Note that in all cases, the existence of a “safe” version of a technology does not preclude the existence of a “dangerous” version, and vice versa. The laws of physics permit both versions to exist. The most rational societal response has been to classify the various applications according to the risk of accident or abuse that each one poses, and then to regulate each application accordingly. The societal response to the tools and products of molecular manufacturing will be no different. Some MM-based tools and products will be deemed safe, and will be lightly regulated. Other MM-based tools and products will be deemed dangerous, and will be heavily regulated, or even legally banned in some cases.

Of course, the mere existence of legal restrictions or outright bans does not preclude the acquisition and abuse of a particular technology by a small criminal fraction of the population. For instance, in the high-risk category, drug abusers obtain and inject themselves with banned narcotics; outlaw regimes employ prohibited poison chemicals in warfare; and rogue nations seek to enter the “nuclear club” via clandestine atomic bomb development programs. Bad actors such as terrorists can also abuse less-heavily regulated products such as fully-automatic rifles or civilian airplanes (which are hijacked and flown into buildings). The most constructive response to this class of threat is to increase monitoring efforts to improve early detection and to pre-position defensive instrumentalities capable of responding rapidly to these abuses, as recommended in 2000 by this author [4] in the context of molecular manufacturing.

The risk of accident or malfunction is less problematic for new technologies than the dangers of abuse. Engineers generally try to design products that work reliably and companies generally seek to sell reliable products to maintain customer goodwill and to avoid expensive product liability lawsuits. But accidents do happen. Here again, our social system has established a set of progressive responses to deal efficiently with this problem. A good example is the ancient technology of fire. The uses of fire are widespread in society, ranging from lightly-regulated matchsticks, butane lighters, campfires, and internal combustion engines, to more heavily regulated home HVAC furnaces, municipal incinerators and industrial smelters. A range of methods are available to deal quickly and effectively with a fire that has accidentally escaped the control of its user. Home fires due to a smoldering cigarette or a blazing grease pan in the kitchen are readily doused using a common household fire extinguisher. Fires in commercial buildings (e.g., hotels) or industrial buildings (e.g., factories) are automatically quenched by overhead sprinkler systems. When these methods prove insufficient to snuff out the flames, the local fire department is called in to limit the damage to just a single building, using fire trucks, water hoses and hydrants. If many buildings are involved, more extensive fire suppression equipment and hundreds of firefighters can be brought in from all across town to hold the damage to a single city block. In the case of the largest accidental fires, like forest fires, vast quantities of heavy equipment are deployed including thousands of firefighters wielding specialized tools, bulldozers to dig firebreaks, helicopters with pendulous water buckets, and great fleets of air tankers dropping tons of fire retardants. (These progressive measures also protect the public in cases of deliberate arson.) The future emergency response hierarchy for dealing with MM-based accidents will be no less exhaustive and may be equally effective in preserving human life and property, while allowing us to enjoy the innumerable benefits of this new technology. Notes Steen Rasmussen of Los Alamos National Laboratory in New Mexico: “The more powerful technology you unleash, the more careful you have to be.” [23]

The study of the ethical [24], socioeconomic [25-28] and legal [29] impact of replication-capable machines such as molecular assemblers and machines such as nanofactories that could build replicators is still in its earliest stages, and there is additional discussion of safety issues elsewhere [30]. However, two important general observations about replicators and self-replication should be noted here.

First, replication is nothing new. Humanity has thousands, arguably even millions, of years of experience living with entities that are capable of kinematic self-replication. These replicators range from the macroscale (e.g., insects, birds, horses, other humans) on down to the microscale (e.g. bacteria, protozoa) and even the nanoscale (e.g., prions, viruses). As a species, we have successfully managed the eternal tradeoff between risk and reward, and have successfully negotiated the antipodes of danger and progress. There is every reason to expect this success to continue. (As shown by the problem of invasive species, the biosphere requires time to adapt to new replicators, so human intervention may be required to prevent severe damage.)

The technologies of engineered self-replication, even at the microscale, are already in wide commercial use throughout the world. Indeed, human civilization is utterly dependent on self-replication technologies. Many important foods including beer, wine, cheese, yogurt, and kefir (a fermented milk), along with various flavors, nutrients, vitamins and other food ingredients, are produced by specially cultured microscopic replicators such as algae, fungi (yeasts) and bacteria. Virtually all of the rest of our food is made by macroscale replicators such as agricultural crop plants, trees, and farm animals. Many of our most important drugs are produced using microscopic self-replicators—from penicillin produced by natural replicating molds starting in the 1940s [15] to the first use of artificial (engineered) self-replicating bacteria to manufacture human insulin by Eli Lilly in 1982 [31]. These uses continue today in the manufacture of many other important drug products such as: (a) human growth hormone (HGH) and erythropoietin (EPO), (b) precursors for antibiotics such as erythromycin [32], and (c) therapeutic proteins such as Factor VIII. A few species of self-replicating bacteria are even used directly as therapeutic medicines, such as the widely available swallowable pills containing bacteria (i.e., natural biological nanomachines) for gastrointestinal refloration, as for example Salivarex^TM which “contains a minimum of 2.9 billion beneficial bacteria per capsule” [33], and Alkadophilus^TM which “contains 1.5 billion organisms per capsule” [34], both at a 2005 price of ~$(0.1-0.2) x 10^-9 per microscale replicator (i.e., per bacterium). Some replicating viruses, notably bacteriophages, are used as therapeutic agents to combat and destroy unhealthful infectious bacterial replicators [35], and for decades viruses have served as transfer vectors to attempt gene therapies [36]. In industry, bacteria are already employed as “self-replicating factories” [37] for various useful products, and microorganisms are also used as workhorses for environmental bioremediation [38, 39], biomining of heavy metals [40], and other applications. In due course, we will learn to safely harness the abilities of nonbiological replication-capable machines for human benefit as well.

Second, replicators can be made inherently safe. An “inherently safe” kinematic replicator is a replicating system that, by its very design, is inherently incapable of surviving mutation or of undergoing evolution (and thus evolving out of our control or developing an independent agenda), and that, equally importantly, does not compete with biology for resources (or worse, use biology as a raw materials resource [4]). One primary route for ensuring inherent safety is to combine the broadcast architecture for control [41] and the vitamin architecture for materials [42], which together eliminate the likelihood that the system can replicate outside of a very controlled and highly artificial setting. There are numerous other routes to this end [10, 19]. Many dozens of additional safeguards may be incorporated into replicator designs to provide redundant embedded controls and thus an arbitrarily low probability of replicator malfunctions of various kinds, simply by selecting the appropriate design parameters [19].

Artificial kinematic replication-capable systems which are not inherently safe should not be designed or constructed, and indeed should be legally prohibited by appropriate juridical and economic sanctions, with these sanctions to be enforced in both national and international regimes. In the case of individual lawbreakers or rogue states that might build and deploy unsafe artificial mechanical replicators, the defenses we have already developed against harmful biological replicators all have analogs in the mechanical world that should provide equally effective, or even superior, defenses. Molecular manufacturing will make possible ever more sophisticated methods of environmental monitoring, prophylaxis and safety. However, advance planning and strategic foresight will be essential in maintaining this advantage.

References and Footnotes

1. An earlier version of this essay appeared as portions of Sections 5.11 and 6.3.1 in: Robert A. Freitas Jr., Ralph C. Merkle, Kinematic Self-Replicating Machines, Landes Bioscience, Georgetown TX, 2004, p. 199 and pp. 204-206; http://www.MolecularAssembler.com/KSRM/5.11.htm#p44 and http://www.MolecularAssembler.com/KSRM/6.3.1.htm.Copyright 2006 Robert A Freitas Jr.

2. Sean Howard, “Nanotechnology and mass destruction: The need for an inner space treaty,” Disarmament Diplomacy 65 (2002); http://www.acronym.org.uk/dd/dd65/65op1.htm; Lee-Anne Broadhead, Sean Howard, “The Heart of Darkness,” Resurgence #221, November/December 2003; http://resurgence.gn.apc.org/issues/broadhead221.htm.

3. Bill Joy, “Why the future doesn’t need us,” Wired 8(April 2000); http://www.wired.com/wired/archive/8.04/joy.html. Response by Ralph Merkle, “Text of prepared comments by Ralph C. Merkle at the April 1, 2000 Stanford Symposium organized by Douglas Hofstadter,” at: http://www.zyvex.com/nanotech/talks/stanford000401.html.

4. Robert A. Freitas Jr., “Some Limits to Global Ecophagy by Biovorous Nanoreplicators, with Public Policy Recommendations,” Zyvex preprint, April 2000; http://www.rfreitas.com/Nano/Ecophagy.htm.

5. “Dangers of Molecular Manufacturing,” Center for Responsible Nanotechnology, 2004; http://crnano.org/dangers.htm.

6. K. Eric Drexler, “Chapter 11. Engines of Destruction,” Engines of Creation: The Coming Era of Nanotechnology, Anchor Press/Doubleday, New York, 1986; http://www.foresight.org/EOC/EOC_Chapter_11.html. Mark Avrum Gubrud, “Nanotechnology and international security,” paper presented at the 5th Foresight Conference, November 1997; http://www.foresight.org/Conferences/MNT05/Papers/Gubrud/. Lev Navrozov, “Molecular nano weapons: Research in China and talk in the West,” NewsMax.com, 27 February 2004; http://www.newsmax.com/archives/articles/2004/2/27/101732.shtml. Jurgen Altmann, “Military uses of nanotechnology: Perspectives and concerns,” Security Dialogue 35(March 2004):61-79. Ray Kurzweil, The Singularity is Near: When Humans Transcend Biology, Penguin Books, New York, 2005.

7. Michael Crichton, Prey, HarperCollins Publishers, New York, 2002. Britt D. Gillette, Conquest of Paradise: An End-times Nano-Thriller, Writers Club Press, New York, 2003. John Robert Marlow, Nano, St. Martin’s Press, New York, 2004.

8. K. Eric Drexler, Engines of Creation: The Coming Era of Nanotechnology, Anchor Press/Doubleday, New York, 1986; http://www.foresight.org/EOC/

9. Philip K. Dick, “Second Variety,” Space Science Fiction, May 1953; also available in: Philip K. Dick, Second Variety and Other Classic Stories by Philip K. Dick, Citadel Press, 1991. Greg Bear, The Forge of God, Gollancz, New York, 1987; http://www.wikipedia.org/wiki/The_Forge_of_God (brief summary). Greg Bear, Anvil of Stars, Century, London, U.K., 1992; http://postviews.editthispage.com/books/byTitle/AnvilOfStars (review).

10. Foresight Institute, “Molecular Nanotechnology Guidelines: Draft Version 3.7,” 4 June 2000; http://www.foresight.org/guidelines/. Extensive excerpt at: http://www.MolecularAssembler.com/KSRM/5.11.htm#p8.

11. According to cyberjournalist Clive Thompson [43], elite writers of software viruses openly publish their code on Web sites, often with detailed descriptions of how the program works, but don’t actually release them. The people who do release the viruses are often anonymous mischief-makers, or "script kiddies"—a derisive term for aspiring young hackers, "usually teenagers or curious college students, who don’t yet have the skill to program computers but like to pretend they do. They download the viruses, claim to have written them themselves and then set them free in an attempt to assume the role of a fearsome digital menace. Script kiddies often have only a dim idea of how the code works and little concern for how a digital plague can rage out of control. Our modern virus epidemic is thus born of a symbiotic relationship between the people smart enough to write a virus and the people dumb enough—or malicious enough—to spread it."

Thompson goes on to describe his early 2004 visit to an Austrian programmer named Mario, who cheerfully announced that in 2003 he had created, and placed online at his website, freely available, a program called "Batch Trojan Generator" that autogenerates malicious viruses. Thompson described a demonstration of this program: "A little box appears on his laptop screen, politely asking me to name my Trojan. I call it the ‘Clive’ virus. Then it asks me what I’d like the virus to do. Shall the Trojan Horse format drive C:? Yes, I click. Shall the Trojan Horse overwrite every file? Yes. It asks me if I’d like to have the virus activate the next time the computer is restarted, and I say yes again. Then it’s done. The generator spits out the virus onto Mario’s hard drive, a tiny 3KB file. Mario’s generator also displays a stern notice warning that spreading your creation is illegal. The generator, he says, is just for educational purposes, a way to help curious programmers learn how Trojans work. But of course I could ignore that advice."

Apparently top "malware" writers do take some responsible precautions, notes Thompson. For example, one hacker’s "main virus-writing computer at home has no Internet connection at all; he has walled it off like an airlocked biological-weapons lab, so that nothing can escape, even by accident." Some writers, after finishing a new virus, "immediately e-mail a copy of it to antivirus companies so the companies can program their software to recognize and delete the virus should some script kiddie ever release it into the wild."

12. Bill Joy, “Act now to keep new technologies out of destructive hands,” New Perspectives Quarterly 17(Summer 2000); http://www.pugwash.org/reports/pim/pim18.htm.

13. James R. Hughes, “Relinquishment or Regulation: Dealing with Apocalyptic Technological Threats,” Trinity College, Fall 2001; http://www.changesurfer.com/Acad/RelReg.pdf.

14. Spencer Reiss, “Hope Is a Lousy Defense,” Wired, December 2003; http://www.wired.com/wired/archive/11.12/billjoy_pr.html.

15. Robert A. Freitas Jr., Nanomedicine, Volume I: Basic Capabilities, Landes Bioscience, Georgetown, TX, 1999; http://www.nanomedicine.com/NMI.htm. Robert A. Freitas Jr., Nanomedicine, Volume IIA: Biocompatibility, Landes Bioscience, Georgetown, TX, 2003; http://www.nanomedicine.com/NMIIA.htm. Robert A. Freitas Jr., “Current Status of Nanomedicine and Medical Nanorobotics (Invited Survey),” J. Comput. Theor. Nanosci. 2(March 2005):1-25; http://www.nanomedicine.com/Papers/NMRevMar05.pdf.

16. Glenn Harlan Reynolds “Techno Worries Miss the Target,” SpeakOut.com, 8 June 2000; http://speakout.com/activism/opinions/5298-1.html.

17. Robert A. Freitas Jr., Ralph C. Merkle, Kinematic Self-Replicating Machines, Landes Bioscience, Georgetown TX, 2004; Sections 3.13.2.2, 4.9.3, 4.14, 4.17, 4.19, 5.7, 5.9.4; http://www.MolecularAssembler.com/KSRM.htm.

18. K. Eric Drexler, Nanosystems: Molecular Machinery, Manufacturing, and Computation, John Wiley & Sons, New York, 1992; http://www.zyvex.com/nanotech/nanosystems.html.

19. Robert A. Freitas Jr., Ralph C. Merkle, Kinematic Self-Replicating Machines, Landes Bioscience, Georgetown TX, 2004, Section 5.1.9; http://www.MolecularAssembler.com/KSRM/5.1.9.htm. The notations (A1, etc.) refer to specific sections in the cited literature.

20. Chris Phoenix, Eric Drexler, “Safe exponential manufacturing,” Nanotechnology 15(2004):869-872; http://www.iop.org/EJ/news/-topic=763/journal/0957-4484. See also: Paul Rincon, “Nanotech guru turns back on ‘goo’,” BBC News Online UK Edition, 9 June 2004; http://news.bbc.co.uk/1/hi/sci/tech/3788673.stm; and Liz Kalaugher, “Drexler dubs ‘grey goo’ fears obsolete,” Nanotechweb.org, 9 June 2004; http://www.nanotechweb.org/articles/society/3/6/1/1.

21. From Freitas (2000) [4]: “Specific public policy recommendations suggested by the results of the present analysis include: (1) an immediate international moratorium on all artificial life experiments implemented as nonbiological hardware. In this context, ‘artificial life’ is defined as autonomous foraging replicators, excluding purely biological implementations (already covered by NIH guidelines tacitly accepted worldwide) and also excluding software simulations which are essential preparatory work and should continue. Alternative ‘inherently safe’ replication strategies such as the broadcast architecture are already well-known….”

22. From Phoenix and Drexler (2004) [20]: “The construction of anything resembling a dangerous self-replicating nanomachine can and should be prohibited.”

23. Ronald Kotulak, “Science on verge of new ‘Creation’: Labs say they have nearly all the tools to make artificial life,” Sun-Sentinel Tribune, 28 March 2004; http://www.sun-sentinel.com/news/local/southflorida/chi-0403280359mar28,0,4395528.story?coll=sfla-home-headlines.

24. David S. Goodsell, Bionanotechnology: Lessons from Nature, John Wiley & Sons, New York, 2004.

25. Robert A. Freitas Jr., William P. Gilbreath, eds., Advanced Automation for Space Missions, NASA Conference Publication CP-2255 (N83-15348), 1982; http://www.islandone.org/MMSG/aasm and Robert A. Freitas Jr., “Noninflationary Nanofactories,” Nanotechnology Perceptions 2 (May 2006), http://www.rfreitas.com/Nano/NoninflationaryPN.pdf.

26. Murray Leinster, The Duplicators, Ace Books, New York, 1964; originally published as “The Lost Race,” Thrilling Wonder Stories, April 1949. Gerald D. Nordley, “On the socioeconomic impact of smart self-replicating machines,” CONTACT 2000, NASA/Ames Research Center; http://www.contact-conference.com/archive/00.html.

27. V. Weil, “Ethical Issues in Nanotechnology,” in M.C. Roco, W.S. Bainbridge, eds., Societal Implications of Nanoscience and Nanotechnology, Kluwer, Dordrecht, 2001, pp. 193-198. R.H. Smith, “Social, Ethical, and Legal Implications of Nanotechnology,” in M.C. Roco, W.S. Bainbridge, eds., Societal Implications of Nanoscience and Nanotechnology, Kluwer, Dordrecht, 2001, pp. 203-211. See also http://itri.loyola.edu/nano/societalimpact/nanosi.pdf.

28. “Task Area 3: Problems of Self-replication, Risk, and Cascading Effects in Nanotechnology: Analogies between Biological Systems and Nanoengineering,” in Philosophical and Social Dimensions of Nanoscale Research—From Laboratory to Society: Developing an Informed Approach to Nanoscale Science and Technology, Working Group for the Study of the Philosophy and Ethics of Complexity and Scale [SPECS], University of South Carolina NanoCenter, 17 March 2003; http://www.cla.sc.edu/cpecs/nirt/nirt.html.

29. Frederick A. Fiedler, Glenn H. Reynolds, “Legal Problems of Nanotechnology: An Overview,” Southern California Interdisciplinary Law Journal 3(1994):593-629. Ty S. Wahab Twibell, “Nano law: The legal implications of self-replicating nanotechnology,” Nanotechnology Magazine, 2000; http://www.irannano.org/English/publication/Articles/Nano-law.htm. John Miller, “Beyond Biotechnology: FDA Regulation Of Nanomedicine,” Columbia Science and Technology Law Review, Vol. IV, 2002-2003; http://www.stlr.org/html/volume4/miller.pdf. Glenn Harlan Reynolds, “Nanotechnology and regulatory policy: three futures,” Harv. J. Law & Technol. 17 (Fall 2003); http://instapundit.com/lawrev/HJOLTnano.pdf.

30. Robert A. Freitas Jr., Ralph C. Merkle, Kinematic Self-Replicating Machines, Landes Bioscience, Georgetown TX, 2004; Sections 2.1.5, 2.3.6, 5.1.9(L), 6.3.1, 6.4.4; http://www.MolecularAssembler.com/KSRM.htm.

31. “Milestones in Medical Research,” Eli Lilly; http://www.lilly.com/about/milestones.html.

32. B.A. Pfeifer, S.J. Admiraal, H. Gramajo, D.E. Cane, Chaitan Khosla, “Biosynthesis of complex polyketides in a metabolically engineered strain of E. coli,” Science 291(2 March 2001):1790-1792, 1683 (comment).

33. “L-Salivarius Plus Other Beneficial Microflora,” Product Information Sheet No. 8058, Life Plus, 1996, at http://www.lightplus.com/lifeplus/8058.html; “Life Plus Vitamin/Herbal Answer For a Healthy Digestive Tract,” at http://members.aol.com/probb0254/salivrex.html; “Support Digestion Naturally: Salivarex,” at http://www.healthyway.net/products/digestion.htm.

34. “Alkadophilus: The Non-Refrigerated Acidophilus,” at: http://www.morter.com/HTML-FILES/ALKAdophilus.HTM, http://www.backcare-center.com/NC-AlkaLine.htm, and http://www.nutritionforhealth.com/herbalformulas.htm.

35. R.J. Payne, D. Phil, V.A. Jansen, “Phage therapy: the peculiar kinetics of self-replicating pharmaceuticals,” Clin. Pharmacol. Ther. 68(September 2000):225-230.

36. Michael G. Kaplitt, Arthur D. Loewy, eds., Viral Vectors: Gene Therapy and Neuroscience Applications, Academic Press, New York, 1995. Angel Cid-Arregui, Alejandro Garcia-Carranca, eds., Viral Vectors: Basic Science and Gene Therapy, Eaton Publishing Co., 2000. David Latchman, Viral Vectors for Treating Diseases of the Nervous System, Academic Press, New York, 2003. Curtis A. MacHida, Jules G. Constant, eds., Viral Vectors for Gene Therapy: Methods and Protocols, Humana Press, 2003.

37. Jonathan King, “Chapter 9. The biotechnology revolution: self-replicating factories and the ownership of life forms,” in Jim Davis, Thomas A. Hirschl, Michael Stack, eds., Cutting Edge: Technology, Information Capitalism and Social Revolution, Verso Books, 1997. M. Kleerebezemab, P. Hols, J. Hugenholtz, “Lactic acid bacteria as a cell factory: rerouting of carbon metabolism in Lactococcus lactis by metabolic engineering,” Enzyme Microb. Technol. 26(1 June 2000):840-848. J. Hugenholtz, M. Kleerebezem, M. Starrenburg, J. Delcour, W. de Vos, P. Hols, “Lactococcus lactis as a cell factory for high-level diacetyl production,” Appl. Environ. Microbiol. 66 (September 2000):4112-4114; http://aem.asm.org/cgi/content/full/66/9/4112. Bernard R. Glick, Jack J. Pasternak, Molecular Biotechnology: Principles and Applications of Recombinant DNA, American Society for Microbiology, Washington, DC, 2003.

38. According to Press [44]: “The first patented form of life produced by genetic engineering was a greatly enhanced oil-eating microbe. The patent [45] was registered to Dr. Ananda Chakrabarty of the General Electric Company in 1981 and was initially welcomed as an answer to the world’s petroleum pollution problem. But anxieties about releasing ‘mutant bacteria’ soon led the U.S. Congress and the Environmental Protection Agency (EPA) to prohibit the use of genetically engineered microbes outside of sealed laboratories. The prohibition set back bioremediation for a few years, until scientists developed improved forms of oil-eating bacteria without using genetic engineering. After large-scale field tests in 1988, the EPA reported that bioremediation eliminated both soil and water-borne oil contamination at about one-fifth the cost of previous methods. Since then, bioremediation has been increasingly used to clean up oil pollution on government sites across the United States.”

39. P. Kotrba, L. Doleckova, V. de Lorenzo, T. Ruml, “Enhanced bioaccumulation of heavy metal ions by bacterial cells due to surface display of short metal binding peptides,” Appl. Environ. Microbiol. 65(March 1999):1092-1098; http://aem.asm.org/cgi/content/full/65/3/1092?view=full&pmid=10049868. W. Bae, R.K. Mehra, A. Mulchandani, W. Chen, “Genetic engineering of Escherichia coli for enhanced uptake and bioaccumulation of mercury,” Appl. Environ. Microbiol. 67(November 2001):5335-5338; http://aem.asm.org/cgi/content/full/67/11/5335?view=full&pmid=11679366. X. Deng, Q.B. Li, Y.H. Lu, D.H. Sun, Y.L. Huang, X.R. Chen, “Bioaccumulation of nickel from aqueous solutions by genetically engineered Escherichia coli,” Water Res. 37(May 2003):2505-2511.

40. I. Suzuki, “Microbial leaching of metals from sulfide minerals,” Biotechnol. Adv. 19(1 April 2001):119-132. D.V. Rao, C.T. Shivannavar, S.M. Gaddad, “Bioleaching of copper from chalcopyrite ore by fungi,” Indian J. Exp. Biol. 40(March 2002):319-324. D.E. Rawlings, D. Dew, C. du Plessis, “Biomineralization of metal-containing ores and concentrates,” Trends Biotechnol. 21(January 2003):38-44. G.J. Olson, J.A. Brierley, C.L. Brierley, “Bioleaching review part B: progress in bioleaching: applications of microbial processes by the minerals industries,” Appl. Microbiol. Biotechnol. 63(December 2003):249-257.

41. Robert A. Freitas Jr., Ralph C. Merkle, Kinematic Self-Replicating Machines, Landes Bioscience, Georgetown TX, 2004, Section 4.11.3.3; http://www.MolecularAssembler.com/KSRM/4.11.3.3.htm.

42. Robert A. Freitas Jr., Ralph C. Merkle, Kinematic Self-Replicating Machines, Landes Bioscience, Georgetown TX, 2004, Section 4.3.7; http://www.MolecularAssembler.com/KSRM/4.3.7.htm.

43. Clive Thompson, “The Virus Underground,” The New York Times, 8 February 2004; http://www.nytimes.com/2004/02/08/magazine/08WORMS.html.

44. Joseph Henry Press, “Chapter 5. Biotechnology and the Environment,” Biotechnology Unzipped: Promises and Realities, National Academy of Sciences, Washington, DC, 2003, pp. 134-160; http://books.nap.edu/books/0309057779/html/134.html.

45. Ananda M. Chakrabarty, “Microorganisms having multiple compatible degradative energy-generating plasmids and preparation thereof,” United States Patent No. 4,259,444, 31 March 1981; Ananda M. Chakrabarty, Scott T. Kellogg, “Bacteria capable of dissimilation of environmentally persistent chemical compounds,” United States Patent No. 4,535,061, 13 August 1985.

© 2006 Robert A. Freitas, Jr.

It turns out that information technology is increasingly encompassing everything of value. It’s not just computers, it’s not just electronic gadgets. It now includes the field of biology. We’re beginning to understand how life processes, disease, aging, are manifested as information processes and gaining the tools to actually manipulate those processes. It’s true of all of our creations of intellectual and cultural endeavors, our music, movies are all facilitated by information technology, and are distributed, and represented as information.

Evolutionary processes work through indirection. Evolution creates a capability, and then it uses that capability to evolve the next stage. That’s why the next stage goes more quickly, and that’s why the fruits of an evolutionary process grow exponentially.

The first paradigm shift in biological evolution, the evolution of cells, and in particular DNA (actually, RNA came first)—the evolution of essentially a computer system or an information processing backbone that would allow evolution to record the results of its experiments—took billions of years. Once DNA and RNA were in place, the next stage, the Cambrian explosion, when all the body plans of the animals were evolved, went a hundred times faster. Then those body plans were used by evolution to concentrate on higher cognitive functions. Biological evolution kept accelerating in this manner. Homo sapiens, our species, evolved in only a few hundred thousand years, the blink of an eye in evolutionary terms.

Then again working through indirection, biological evolution used one of its creations, the first technology-creating species to usher in the next stage of evolution, which was technology. The enabling factors for technology were a higher cognitive function with an opposable appendage, so we could manipulate and change the environment to reflect our models of what could be. The first stages of technology evolution—fire, the wheel, stone tools—only took a few tens of thousands of years.

Technological evolution also accelerated. Half a millennium ago the printing press took a century to be adopted, half a century ago the first computers were designed pen on paper. Now computers are designed in only a few weeks’ time by computer designers sitting at computers, using advanced computer assisted design software. When I was at MIT [in the mid-1960s] a computer that took about the size of this room cost millions of dollars yet was less powerful than the computer in your cell phone today.

One of the profound implications is that we are understanding our biology as information processes. We have 23,000 little software programs inside us called genes. These evolved in a different era. One of those programs, called the fat insulin receptor gene, says, basically, hold onto every calorie because the next hunting season might not work out so well. We’d like to change that program now. We have a new technology that has just emerged in the last couple years called RNA interference, in which we put fragments of RNA inside the cell, as a drug, to inhibit selected genes. It can actually turn genes off by blocking the messenger RNA expressing that gene. When the fat insulin receptor was turned off in mice, the mice ate ravenously and remained slim. They didn’t get diabetes, didn’t get heart disease, lived 20% longer: they got the benefit of caloric restriction without the restriction.

Every major disease, and every major aging process has different genes that are used in the expression of these disease and aging processes. Being able to actually select when we turn them off is one powerful methodology. We also have the ability to turn enzymes off. Torcetrapib, a drug that’s now in FDA Phase 3 trials, turns off a key enzyme that destroys the good cholesterol, HDL, in the blood. If you inhibit that enzyme, HDL levels soar and atherosclerosis slows down or stops.

There are thousands of these developments in the pipeline. The new paradigm of rational drug design involves actually understanding the information processes underlying biology, the exact sequence of steps that leads up to a process like atherosclerosis, which causes heart attacks, or cancer, or insulin resistance, and providing very precise tools to intervene. Our ability to do this is also growing at an escalating rate.

Another exponential process is miniaturization. We’re showing the feasibility of actually constructing things at the molecular level that can perform useful functions. One of the biggest applications of this, again, will be in biology, where we will be able to go inside the human body and go beyond the limitations of biology.

Rob Freitas has designed a nanorobotic red blood cell, which is a relatively simple device, it just stores oxygen and lets it out. A conservative analysis of these robotic respirocytes shows that if you were to replace ten percent of your red blood cells with these robotic versions you could do an Olympic sprint for 15 minutes without taking a breath, or sit at the bottom of your pool for four hours. It will be interesting to see what we do with these in our Olympic contests. Presumably we’ll ban them, but then we’ll have the specter of high school students routinely outperforming the Olympic athletes.

A robotic white blood cell is also being designed. A little more complicated, it downloads software from the Internet to combat specific pathogens. If it sounds very futuristic to download information to a device inside your body to perform a health function, I’ll point out that we’re already doing that. There are about a dozen neural implants either FDA-approved or approved for human testing. One implant that is FDA-approved for actual clinical use replaces the biological neurons destroyed by Parkinson’s disease. The neurons in the vicinity of this implant then receive signals from the computer that’s inside the patient’s brain. This hybrid of biological and non-biological intelligence works perfectly well. The latest version of this device allows the patient to download new software to the neural implant in his brain from outside his body.

These are devices that today require surgery to be implanted, but when we get to the 2020s, we will ultimately have the “killer app” of nanotechnology, nanobots, which are blood cell-sized devices that can go inside the body and brain to perform therapeutic functions, as well as advance the capabilities of our bodies and brains.

If that sounds futuristic, I’ll point out that we already have blood cell-size devices that are nano-engineered, working to perform therapeutic functions in animals. For example, one scientist cured type I diabetes in rats with this type of nanoengineered device. And some of these are now approaching human trials. The 2020s really will be the “golden era” of nanotechnology.

It is a mainstream view now among informed observers that by the 2020s we will have sufficient computer processing to emulate the human brain. The current controversy, or I would say, the more interesting question is, will we have the software or methods of human intelligence? To achieve the methods, the algorithms of human intelligence, there is underway a grand project to reverse-engineer the brain. And there, not surprisingly, we are also making exponential progress. If you follow the trends in reverse brain engineering it’s a reasonable conclusion that we will have reverse-engineered the several hundred regions of the brain by the 2020s.

By early in the next decade, computers won’t look like today’s notebooks and PDAs, they will disappear, integrated into our clothing and environment. Images will be written to our retinas for our eyeglasses and contact lenses, we’ll have full-immersion virtual reality. We’ll be interacting with virtual personalities; we can see early harbingers of this already. We’ll have effective language translation.

If we go out to 2029, there will be many turns of the screw in terms of this exponential progression of information technology. There will be about thirty doublings in the next 25 years. That’s a factor of a billion in capacity and price performance over today’s technology, which is already quite formidable.

By 2029, we will have completed reverse engineering of the brain, we will understand how human intelligence works, and that will give us new insight into ourselves. Non-biological intelligence will combine the suppleness and subtlety of our pattern-recognition capabilities with ways computers have already demonstrated their superiority. Every time you use Google you can see the power of non-biological intelligence. Machines can remember things very accurately. They can share their knowledge instantly. We can share our knowledge, too, but at the slow bandwidth of language.

This will not be an alien invasion of intelligent machines coming from over the horizon to compete with us, it’s emerging from within our civilization, it’s extending the power of our civilization. Even today we routinely do intellectual feats that would be impossible without our technology. In fact our whole economic infrastructure couldn’t manage without the intelligent software that’s underlying it.

The most interesting application of computerized nanobots will be to interact with our biological neurons. We’ve already shown the feasibility of using electronics and biological neurons to interact non-invasively. We could have billions of nanobots inside the capillaries of our brains, non-invasively, widely distributed, expanding human intelligence, or providing full immersion virtual reality encompassing all of the senses from within the nervous system. Right now we have a hundred trillion connections. Although there’s a certain amount of plasticity, biological intelligence is essentially fixed. Non-biological intelligence is growing exponentially; the crossover point will be in the 2020s. When we get to the 2030s and 2040s, it will be the non-biological portion of our civilization that will be predominant. But it will still be an expression of human civilization.

Every time we have technological gains we make gains in life expectancy. Sanitation was a big one, antibiotics was another. We’re now in the beginning phases of this biotechnology revolution. We’re exploring, understanding and graining the tools to reprogram the information processes underlying biology; and that will result in another big gain in life expectancy. So, if you watch your health today, the old-fashioned way, you can actually live to see the remarkable 21^st century.

© 2006 Ray Kurzweil. Reprinted with permission.

Abstract

Goal 7 of the NASA Astrobiology Roadmap states: “Determine how to recognize signatures of life on other worlds and on early Earth. Identify biosignatures that can reveal and characterize past or present life in ancient samples from Earth, extraterrestrial samples measured in situ, samples returned to Earth, remotely measured planetary atmospheres and surfaces, and other cosmic phenomena.” The cryptic reference to “other cosmic phenomena” would appear to be broad enough to include the possible identification of biosignatures embedded in the dimensionless constants of physics. The existence of such a set of biosignatures—a life-friendly suite of physical constants—is a retrodiction of the Selfish Biocosm (SB) hypothesis. This hypothesis offers an alternative to the weak anthropic explanation of our indisputably life-friendly cosmos favored by (1) an emerging alliance of M-theory-inspired cosmologists and advocates of eternal inflation like Linde and Weinberg, and (2) supporters of the quantum theory-inspired sum-over-histories cosmological model offered by Hartle and Hawking. According to the SB hypothesis, the laws and constants of physics function as the cosmic equivalent of DNA, guiding a cosmologically extended evolutionary process and providing a blueprint for the replication of new life-friendly progeny universes.

Introduction

The notion that we inhabit a universe whose laws and physical constants are fine-tuned in such a way as to make it hospitable to carbon-based life is an old idea (Gardner, 2003). The so-called “anthropic” principle comes in at least four principal versions (Barrow and Tipler, 1988) that represent fundamentally different ontological perspectives. For instance, the “weak anthropic principle” is merely a tautological statement that since we happen to inhabit this particular cosmos it must perforce by life-friendly or else we would not be here to observe it. As Vilenkin put it recently (Vilenkin, 2004), “the ‘anthropic’ principle, as stated above, hardly deserves to be called a principle: it is trivially true.” By contrast, the “participatory anthropic principle” articulated by Wheeler and dubbed “it from bit” (Wheeler, 1996) is a radical extrapolation from the Copenhagen interpretation of quantum physics and a profoundly counterintuitive assertion that the very act of observing the universe summons it into existence.

All anthropic cosmological interpretations share a common theme: a recognition that key constants of physics (as well as other physical aspects of our cosmos such as its dimensionality) appear to exhibit a mysterious fine-tuning that optimizes their collective bio-friendliness. Rees noted (Rees, 2000) that virtually every aspect of the evolution of the universe—from the birth of galaxies to the origin of life on Earth—is sensitively dependent on the precise values of seemingly arbitrary constants of nature like the strength of gravity, the number of extended spatial dimensions in our universe (three of the ten posited by M-theory), and the initial expansion speed of the cosmos following the Big Bang. If any of these physical constants had been even slightly different, life as we know it would have been impossible:

The [cosmological] picture that emerges—a map in time as well as in space—is not what most of us expected. It offers a new perspective on a how a single “genesis event” created billions of galaxies, black holes, stars and planets, and how atoms have been assembled—here on Earth, and perhaps on other worlds—into living beings intricate enough to ponder their origins. There are deep connections between stars and atoms, between the cosmos and the microworld…. Our emergence and survival depend on very special “tuning” of the cosmos—a cosmos that may be even vaster than the universe that we can actually see.

As stated recently by Smolin (Smolin, 2004), the challenge is to provide a genuinely scientific explanation for what he terms the “anthropic observation”:

The anthropic observation: Our universe is much more complex than most universes with the same laws but different values of the parameters of those laws. In particular, it has a complex astrophysics, including galaxies and long lived stars, and a complex chemistry, including carbon chemistry. These necessary conditions for life are present in our universe as a consequence of the complexity which is made possible by the special values of the parameters.

There is good evidence that the anthropic observation is true. Why it is true is a puzzle that science must solve.

It is a daunting puzzle indeed. The strangely (and apparently arbitrarily) biophilic quality of the physical laws and constants poses, in Greene’s view, the deepest question in all of science (Greene, 2004). In the words of Davies (Gardner, 2003), it represents “the biggest of the Big Questions: why is the universe bio-friendly?”

Modern History of Anthropic Reasoning

Modern statements of the cosmological anthropic principle date from the publication of a landmark book by Henderson in 1913 entitled The Fitness of the Environment (Henderson, 1913). Henderson’s book was an extended reflection on the curious fact that there are particular substances present in the environment—preeminently water—whose peculiar qualities rendered the environment almost preternaturally suitable for the origin, maintenance, and evolution of organic life. Indeed, the strangely life-friendly qualities of these materials led Henderson to the view that “we were obliged to regard this collocation of properties in some intelligible sense a preparation for the process of planetary evolution…. Therefore the properties of the elements must for the present be regarded as possessing a teleological character.”

Thoroughly modern in outlook, Henderson dismissed this apparent evidence that inanimate nature exhibited a teleological character as indicative of divine design or purpose. Indeed, he rejected the notion that nature’s seemingly teleological quality was in any way inconsistent with Darwin’s theory of evolution through natural selection. On the contrary, he viewed the bio-friendly character of the inanimate natural environment as essential to the optimal operation of the evolutionary forces in the biosphere. Absent the substrate of a superbly “fit” inanimate environment, Henderson contended, Darwinian evolution could never have achieved what it has in terms of species multiplication and diversification.

The mystery of why the physical qualities of the inanimate universe happened to be so oddly conducive to life and biological evolution remained just that for Henderson—an impenetrable mystery. The best he could do to solve the puzzle was to speculate that the laws of chemistry were somehow fine-tuned in advance by some unknown cosmic evolutionary mechanism to meet the future needs of a living biosphere:

The properties of matter and the course of cosmic evolution are now seen to be intimately related to the structure of the living being and to its activities; they become, therefore, far more important in biology than has previously been suspected. For the whole evolutionary process, both cosmic and organic, is one, and the biologist may now rightly regard the Universe in its very essence as biocentric.

Henderson’s iconoclastic vision was far ahead of its time. His potentially revolutionary book was largely ignored by his contemporaries or dismissed as a mere tautology. Of course there should be a close match-up between the physical requirements of life and the physical world that life inhabits, contemporary skeptics pointed out, since life evolved to survive the very challenges presented by that pre-organic world and to take advantage of the biochemical opportunities it offered.

While lacking broad influence at the time, Henderson’s pioneering vision proved to be the precursor to modern formulations of the cosmological anthropic principle. One of the first such formulations was offered by British astronomer Fred Hoyle. A storied chapter in the history of the principle is the oft-told tale of Hoyle’s prediction of the details of the triple-alpha process (Mitton 2005). This prediction, which seems to qualify as the first falsifiable implication to flow from an anthropic hypothesis, involves the details of the process by which the element carbon (widely viewed as the essential element of abiotic precursor polymers capable of autocatalyzing the emergence of living entities) emerges through stellar nucleosynthesis. As noted by Livio (Livio, 2003):

Carbon features in most anthropic arguments. In particular, it is often argued that the existence of an excited state of the carbon nucleus is a manifestation of fine-tuning of the constants of nature that allowed for the appearance of carbon-based life. Carbon is formed through the triple-alpha process in two steps. In the first, two alpha particles form an unstable (lifetime ~10^-16s)⁸Be. In the second, a third alpha particle is captured, via ⁸Be(α,γ)¹²C. Hoyle argued than in order for the 3α reaction to proceed at a rate sufficient to produce the observed cosmic carbon, a resonant level must exist in ¹²C, a few hundred keV about the ⁸Be+⁴He threshold. Such a level was indeed found experimentally.

Other chapters in the modern history of the anthropic principle are treated comprehensively by Barrow and Tipler (Barrow and Tipler, 1988) and will not be revisited here.

The New Urgency of Anthropic Investigation

Two recent developments have imparted a renewed sense of urgency to investigations of the anthropic qualities of our cosmos. The first is the discovery that the value of dark energy density is exceedingly small but not quite zero—an apparent happenstance, unpredictable from first principles, with profound implications for the bio-friendly quality of our universe. As noted recently by Goldsmith (Goldsmith, 2004):

A relatively straightforward calculation [based on established principles of theoretical physics] does yield a theoretical value for the cosmological constant, but that value is greater than the measured one by a factor of about 10¹²⁰—probably the largest discrepancy between theory and observation science has ever had to bear.

If the cosmological constant had a smaller value than that suggested by recent observations, it would cause no trouble (just as one would expect, remembering the happy days when the constant was thought to be zero). But if the constant were a few times larger than it is now, the universe would have expanded so rapidly that galaxies could not have endured for the billions of years necessary to bring forth complex forms of life.

The second development is the realization that M-theory—arguably the most promising contemporary candidate for a theory capable of yielding a deep synthesis of relativity and quantum physics—permits, in Bjorken’s phrase (Bjorken, 2004), “a variety of string vacuua, with different standard-model properties.”

M-theorists had initially hoped that their new paradigm would be “brittle” in the sense of yielding a single mathematically unavoidable solution that uniquely explained the seemingly arbitrary parameters of the Standard Model. As Susskind has put it (Susskind, 2003):

The world-view shared by most physicists is that the laws of nature are uniquely described by some special action principle that completely determines the vacuum, the spectrum of elementary particles, the forces and the symmetries. Experience with quantum electrodynamics and quantum chromodynamics suggests a world with a small number of parameters and a unique ground state. For the most part, string theorists bought into this paradigm. At first it was hoped that string theory would be unique and explain the various parameters that quantum field theory left unexplained.

This hope has been dashed by the recent discovery that the number of different solutions permitted by M-theory (which correspond to different values of Standard Model parameters) is, in Susskind’s words, “astronomical, measured not in millions or billions but in googles or googleplexes.” This development seems to deprive our most promising new theory of fundamental physics of the power to uniquely predict the emergence of anything remotely resembling our universe. As Susskind puts it, the picture of the universe that is emerging from the deep mathematical recesses of M-theory is not an “elegant universe” but rather a Rube Goldberg device, cobbled together by some unknown process in a supremely improbable manner that just happens to render the whole ensemble fit for life. In the words of University of California theoretical physicist Steve Giddings, “No longer can we follow the dream of discovering the unique equations that predict everything we see, and writing them on a single page. Predicting the constants of nature becomes a messy environmental problem. It has the complications of biology.”[1]

Two Contemporary Restatements of the Weak Anthropic Principle: Eternal Inflation Plus M-Theory and Many-Worlds Quantum Cosmology

There have been two principal approaches to the task of enlisting the weak anthropic principle to explain the mysteriously small (and thus bio-friendly) value of the density of dark energy and the apparent happenstance by which our bio-friendly universe was selected from the enormously large “landscape” of possible solutions permitted by M-theory, only a tiny fraction of which correspond to anything resembling the Standard Model prevalent in our cosmos.

Eternal Inflation Meets M-Theory

The first approach, favored by Susskind (Susskind, 2003). Linde (Linde, 2002), Weinberg (Weinberg, 1999), and Vilenkin (Vilenkin, 2004) among others, overlays the model of eternal inflation with the key assumption that M-theory-permitted solutions (corresponding to different values of Standard Model parameters) and dark energy density values will vary randomly from bubble universe to bubble universe within an eternally expanding ensemble variously termed a multiverse or a meta-univers. Generating a life-friendly cosmos is simply a matter of randomly reshuffling the set of permissible parameters and values a sufficient number of times until a particular Big Bang yields, against odds of perhaps a googleplex-to-one, a permutation that just happens to possess the right mix of Standard Model parameters to be bio-friendly.

Sum-Over-Histories Quantum Cosmological Model

The second approach invokes a quantum theory-derived sum-over-histories cosmological model inspired by Everett’s “many worlds” interpretation of quantum physics. This approach, which has been prominently embraced by Hawking (Hawking and Hertog, 2002), was summarized as follows by Hogan (Hogan, 2004):

In the original formulation of quantum mechanics, it was said that an observation collapsed a wavefunction to one of the eignestates of the observed quantity. The modern view is that the cosmic wavefunction never collapses, but only appears to collapse from the point of view of observers who are part of the wavefunction. When Schrödinger’s cat lives or dies, the branch of the wavefunction with the dead cat also contains observers who are dealing with a dead cat, and the branch with the live cat also contains observers who are petting a live one.

Although this is sometimes called the “Many Worlds” interpretation of quantum mechanics, it is really about having just one world, one wavefunction, obeying the Schrödinger equation: the wavefunction evolves linearly from one time to the next based on its previous state.

Anthropic selection in this sense is built into physics at the most basic level of quantum mechanics. Selection of a wavefunction branch is what drives us into circumstances in which we thrive. Viewed from a disinterested perspective outside the universe, it looks like living beings swim like salmon up their favorite branches of the wavefunction, chasing their favorite places.

Hawking and Hertog (Hawking and Hertog, 2002) have explicitly characterized this “top down” cosmological model as a restatement of the weak anthropic principle:

We have argued that because our universe has a quantum origin, one must adopt a top down approach to the problem of initial conditions in cosmology, in which histories that contribute to the path integral, depend on the observable being measured. There is an amplitude for empty flat space, but it is not of much significance. Similarly, the other bubbles in an eternally inflating spacetime are irrelevant. They are to the future of our past light cone, so they don’t contribute to the action for observables and should be excised by Ockham’s razor. Therefore, the top down approach is a mathematical formulation of the weak anthropic principle. Instead of starting with a universe and asking what a typical observer would see, one specifies the amplitude of interest.

Critique of Contemporary Restatements of the Weak Anthropic Principle

Apart from the objections on the part of those who oppose in principle any use of the anthropic principle in cosmology, there are at least three reasons why both the Hawking/Hogan and the Susskind/Linde/Weinberg restatements of the weak anthropic principle are objectionable.

First, both approaches appear to be resistant (at the very least) to experimental testing. Universes spawned by Big Bangs other than our own are inaccessible from our own universe, at least with the experimental techniques currently available to science. So too are quantum wavefunction branches that we cannot, in principle, observe. Accordingly, both approaches appear to be untestable—perhaps untestable in principle. For this reason, Smolin recently argued (Smolin, 2004) “not only is the Anthropic Principle not science, its role may be negative. To the extent that the Anthropic Principle is espoused to justify continued interest in unfalsifiable theories, it may play a destructive role in the progress of science.”

Second, both approaches violate the mediocrity principle. The mediocrity principle, a mainstay of scientific theorizing since Copernicus, is a statistically based rule of thumb that, absent contrary evidence, a particular sample (Earth, for instance, or our particular universe) should be assumed to be a typical example of the ensemble of which it is a part. The Susskind/Linde/Weinberg approach, in particular, flouts this principle. Their approach simply takes refuge in a brute, unfathomable mystery—the conjectured lucky roll of the dice in a crap game of eternal inflation—and declines to probe seriously into the possibility of a naturalistic cosmic evolutionary process that has the capacity to yield a life-friendly set of physical laws and constants on a nonrandom basis.

Third, both approaches extravagantly inflate the probabilistic resources required to explain the phenomenon of a life-friendly cosmos. (Think of a googleplex of monkeys typing away randomly until one of them, by pure chance, accidentally composes a set of equations that correspond to the Standard Model.) This should be a hint that something fundamental is being overlooked and that there may exist an unknown natural process, perhaps functionally akin in some manner to terrestrial evolution, capable of effecting the emergence and prolongation of physical states of nature that are, in the abstract, vanishingly improbable.

The Darwinian Precedent

Hogan (Hogan, 2004) has analogized the quantum theory-inspired sum-over-histories version of the weak anthropic principle to Darwinian theory:

This blending of empirical cosmology and fundamental physics is reminiscent of our Darwinian understanding of the tree of life. The double helix, the four-base codon alphabet and the triplet genetic code for amino acids, any particular gene for a protein in a particular organism—all are frozen accidents of evolutionary history. It is futile to try to understand or explain these aspects of life, or indeed any relationships in biology, without referring to the way the history of life unfolded. In the same way that (in Dobzhansky’s phrase), “nothing in biology makes sense except in the light of evolution,” physics in these models only makes sense in the light of cosmology.

Ironically, Hogan misses the key point that neither the branching wavefunction nor the eternal inflation-plus-M-theory versions of the weak anthropic principle hypothesize the existence of anything corresponding to the main action principle of Darwin’s theory: natural selection. Both restatements of the weak anthropic principle are analogous, not to Darwin’s approach, but rather to a mythical alternative history in which Darwin, contemplating the storied tangled bank (the arresting visual image with which he concludes The Origin of Species), had confessed not a magnificent obsession with gaining an understanding of the mysterious natural processes that had yielded “endless forms most beautiful and most wonderful,” but rather a smug satisfaction that of course the earthly biosphere must have somehow evolved in a just-so manner mysteriously friendly to humans and other currently living species, or else Darwin and other humans would not be around to contemplate it.

Indeed, the situation that confronts cosmologists today is reminiscent of that which faced biologists before Darwin propounded his revolutionary theory of evolution through natural selection. Darwin confronted the seemingly miraculous phenomenon of a fine-tuned natural order in which every creature and plant appeared to occupy a unique and well-designed niche. Refusing to surrender to the brute mystery posed by the appearance of nature’s design, Darwin masterfully deployed the art of metaphor[2] to elucidate a radical hypothesis—the origin of species through natural selection—that explained the apparent miracle as a natural phenomenon.

A significant lesson drawn from Darwin’s experience is important to note at this point. Answering the question of why the most eminent geologists and naturalists had, until shortly before publication of The Origin of Species, disbelieved in the mutability of species, Darwin responded that this false conclusion was “almost inevitable as long as the history of the world was thought to be of short duration.” It was geologist Charles Lyell’s speculations on the immense age of Earth that provided the essential conceptual framework for Darwin’s new theory. Lyell’s vastly expanded stretch of geological time provided an ample temporal arena in which the forces of natural selection could sculpt and reshape the species of Earth and achieve nearly limitless variation.

The central point for purposes of this paper is that collateral advances in sciences seemingly far removed from cosmology (complexity theory and evolutionary theory among them) can help dissipate the intellectual limitations imposed by common sense and naïve human intuition. And, in an uncanny reprise of the Lyell/Darwin intellectual synergy, it is a realization of the vastness of time and history that gives rise to the novel theoretical possibility to be discussed subsequently. Only in this instance, it is the vastness of future time and future history that is of crucial importance. In particular, sharp attention must be paid to the key conclusion of Wheeler: most of the time available for life and intelligence to achieve their ultimate capabilities lie in the distant cosmic future, not in the cosmic past. As Tipler (Tipler, 1994) has stated, “Almost all of space and time lies in the future. By focusing attention only on the past and present, science has ignored almost all of reality. Since the domain of scientific study is the whole of reality, it is about time science decided to study the future evolution of the universe.” The next section of this paper describes an attempt to heed these admonitions.

The Selfish Biocosm Hypothesis

In a paper published in Complexity (Gardner, 2000), I first advanced the hypothesis that the anthropic qualities which our universe exhibits might be explained as incidental consequences of a cosmic replication cycle in which the emergence of a cosmologically extended biosphere could conceivably supply two of the logically essential elements of self-replication identified by von Neumann (von Neumann, 1948): a controller and a duplicating device. The hypothesis proposed in that paper was an attempt to extend and refine Smolin’s conjecture (Smolin, 1997) that the majority of the anthropic qualities of the universe can be explained as incidental consequences of a process of cosmological replication and natural selection (CNS) whose utility function is black hole maximization. Smolin’s conjecture differs crucially from the concept of eternal inflation advanced by Linde (Linde, 1998) in that it proposes a cosmological evolutionary process with a specific and discernible utility function—black hole maximization. It is this aspect of Smolin’s conjecture rather than the specific utility function he advocates that renders his theoretical approach genuinely novel.

As demonstrated previously (Rees, 1997; Baez, 1998), Smolin’s conjecture suffers from two evident defects: (1) the fundamental physical laws and constants do not, in fact, appear to be fine-tuned to favor black hole maximization and (2) no mechanism is proposed corresponding to two logically required elements of any von Neumann self-replicating automaton: a controller and a duplicator.[3] The latter are essential elements of any replicator system capable of Darwinian evolution, as noted by Dawkins (Gardner, 2000) in a critique of Smolin’s conjecture:

Note that any Darwinian theory depends on the prior existence of the strong phenomenon of heredity. There have to be self-replicating entities (in a population of such entities) that spawn daughter entities more like themselves than the general population.

Theories of cosmological eschatology previously articulated (Kurzweil, 1999; Wheeler, 1996; Dyson, 1988) predict that the ongoing process of biological and technological evolution is sufficiently robust and unbounded that, in the far distant future, a cosmologically extended biosphere could conceivably exert a global influence on the physical state of the cosmos. A related set of insights from complexity theory (Gardner, 2000) indicates that the process of emergence resulting from such evolution is essentially unbounded.

A synthesis of these two sets of insights yielded the two key elements of the Selfish Biocosm (SB) hypothesis. The essence of that synthesis is that the ongoing process of biological and technological evolution and emergence could conceivably function as a von Neumann controller and that a cosmologically extended biosphere could, in the very distant future, function as a von Neumann duplicator in a hypothesized process of cosmological replication.

In a paper published in Acta Astronautica (Gardner, 2001) I suggested that a falsifiable implication of the SB hypothesis is that the process of the progression of the cosmos through critical epigenetic thresholds in its life cycle, while perhaps not strictly inevitable, is relatively robust. One such critical threshold is the emergence of human-level and higher intelligence, which is essential to the eventual scaling up of biological and technological processes to the stage at which those processes could conceivably exert a global influence on the state of the cosmos. Four specific tests of the robustness of the emergence of human-level and higher intelligence were proposed.

In a subsequent paper published in the Journal of the British Interplanetary Society (Gardner, 2002) I proposed that an additional falsifiable implication of the SB hypothesis is that there exists a plausible final state of the cosmos that exhibits maximal computational potential. This predicted final state appeared to be consistent with both the modified ekpyrotic cyclic universe scenario (Khoury, Ovrut, Seiberg, Steinhardt, and Turok, 2001; Steinhardt and Turok, 2001) and with Lloyd’s description (Lloyd, 2000) of the physical attributes of the ultimate computational device: a computer as powerful as the laws of physics will allow.

Key Retrodiction of the SB Hypothesis: A Life-Friendly Cosmos

The central assertions of the SB hypothesis are: (1) that highly evolved life and intelligence play an essential role in a hypothesized process of cosmic replication and (2) that the peculiarly life-friendly laws and physical constants that prevail in our universe—an extraordinarily improbable ensemble that Pagels dubbed the cosmic code (Pagels, 1983)—play a cosmological role functionally equivalent to that of DNA in an earthly organism: they provide a recipe for cosmic ontogeny and a blueprint for cosmic reproduction. Thus, a key retrodiction of the SB hypothesis is that the suite of physical laws and constants that prevail in our cosmos will, in fact, be life-friendly. Moreover—and alone among the various cosmological scenarios offered to explain the phenomenon of a bio-friendly universe—the SB hypothesis implies that this suite of laws and constants comprise a robust program that will reliably generate life and advanced intelligence just as the DNA of a particular species constitutes a robust program that will reliably generate individual organisms that are members of that particular species. Indeed, because the hypothesis asserts that sufficiently evolved intelligent life serves as a von Neumann duplicator in a putative process of cosmological replication, the biophilic quality of the suite emerges as a retrodicted biosignature of the putative duplicator and duplication process within the meaning of Goal 7 of the NASA Astrobiology Roadmap, which provides in pertinent part:

Determine how to recognize signatures of life on other worlds and on early Earth. Identify biosignatures that can reveal and characterize past or present life in ancient samples from Earth, extraterrestrial samples measured in situ, samples returned to Earth, remotely measured planetary atmospheres and surfaces, and other cosmic phenomena.

Does this retrodiction qualify as a valid scientific test of the validity of the SB hypothesis? I propose that it may, provided two additional qualifying criteria are satisfied:

• The underlying hypothesis must enjoy consilience[4] with mainstream scientific paradigms and conjectural frameworks (in particular, complexity theory, evolutionary theory, M-theory, and theoretically acceptable conjectures by mainstream cosmologists concerning the feasibility, at least in principle, of “baby universe” fabrication); and

• The retrodiction must be augmented by falsifiable predictions of phenomena implied by the SB hypothesis but not yet observed.

Retrodiction as a Tool for Testing Scientific Hypotheses

There is a lively literature debating the propriety of employing retrodiction as a tool for testing scientific hypotheses (Cleland, 2002; Cleland, 2001; Gee, 1999; Oldershaw, 1988). Oldershaw (Oldershaw, 1988) has discussed the use of falsifiable retrodiction (as opposed to falsifiable prediction) as a tool of scientific investigation:

A second type of prediction is actually not a prediction at all, but rather a “retrodiction.” For example, the anomalous advance of the perihelion of Mercury had been a tiny thorn in the side of Newtonian gravitation long before general relativity came upon the scene. Einstein found that his theory correctly “predicted,” actually retrodicted, the numerical value of the perihelion advance. The explanation of the unexpected result of the Michelson-Morley experiment (constancy of the velocity of light) in terms of special relativity is another example.

As he went on to note, “Retrodictions usually represent falsification tests; the theory is probably wrong if it fails the test, but should not necessarily be considered right if it passes the test since it does not involve a definitive prediction.” Despite their legitimacy as falsification tests of hypotheses, falsifiable retrodictions are qualitatively inferior to falsifiable predictions, in Oldershaw’s view:

But, in the final analysis, only true definitive predictions can justify the promotion of a theory from being viewed as one of many plausible hypotheses to being recognized as the best available approximation of how nature actually works. A theory that cannot generate definitive predictions, or whose definitive predictions are impossible to test, can be regarded as inherently untestable.”

A less sympathetic view concerning the validity of retrodiction as a scientific tool was offered by Gee (Gee, 1999), who dismissed the legitimacy of all historical hypotheses on the ground that “they can never be tested by experiment, and so they are unscientific…. No science can ever be historical.” This viewpoint, in turn, has been challenged by Cleland (Cleland, 2001) who contends that “when it comes to testing hypotheses, historical science is not inferior to classical experimental science” but simply exploits the available evidence in a different way:

There [are] fundamental differences in the methodology used by historical and experimental scientists. Experimental scientists focus on a single (sometimes complex) hypothesis, and the main research activity consists in repeatedly bringing about the test conditions specified by the hypothesis, and controlling for extraneous factors that might produce false positives and false negatives. Historical scientists, in contrast, usually concentrate on formulating multiple competing hypotheses about particular past events. Their main research efforts are directed at searching for a smoking gun, a trace that sets apart one hypothesis as providing a better causal explanation (for the observed traces) than do the others. These differences in methodology do not, however, support the claim that historical science is methodologically inferior, because they reflect an objective difference in the evidential relations at the disposal of historical and experimental researchers for evaluating their hypotheses.

Cleland’s approach has the merit of preserving as “scientific” some of the most important hypotheses advanced in such historical fields of inquiry as geology, evolutionary biology, cosmology, paleontology, and archaeology. As Cleland has noted (Cleland, 2002):

Experimental research is commonly held up at the paradigm of successful (a.k.a.good) science. The role classically attributed to experiment is that of testing hypotheses in controlled laboratory settings. Not all scientific hypotheses can be tested in this manner, however. Historical hypotheses about the remote past provide good examples. Although fields such as paleontology and archaeology provide the familiar examples, historical hypotheses are also common in geology, biology, planetary science, astronomy, and astrophysics. The focus of historical research is on explaining existing natural phenomena in terms of long past causes. Two salient examples are the asteroid-impact hypothesis for the extinction of the dinosaurs, which explains the fossil record of the dinosaurs in terms of the impact of a large asteroid, and the “big-bang” theory of the origin of the universe, which explains the puzzling isotropic three-degree background radiation in terms of a primordial explosion. Such work is significantly different from making a prediction and then artificially creating a phenomenon in a laboratory.

In a paper presented to the 2004 Astrobiology Science Conference (Cleland, 2004), Cleland extended this analytic framework to the consideration of putative biosignatures as evidence of the past or present existence of extraterrestrial life. Acknowledging that “because biosignatures represent indirect traces (effects) of life, much of the research will be historical (vs. experimental) in character even in cases where the traces represent recent effects of putative extant organisms,” Cleland concluded that it was appropriate to employ the methodology that characterizes successful historical research:

Successful historical research is characterized by (1) the proliferation of alternative competing hypotheses in the face of puzzling evidence and (2) the search for more evidence (a “smoking gun”) to discriminate among them.

From the perspective of the evidentiary standards applicable to historical science in general and astrobiology in particular, the key retrodiction of the SB hypothesis—that the fundamental constants of nature that comprise the Standard Model as well as other physical features of our cosmos (included the number of extended physical dimensions and the extremely low value of dark energy) will be collectively bio-friendly—appears to constitute a legitimate scientific test of the hypothesis. Moreover, within the framework of Goal 7 of the NASA Astrobiology Roadmap, the retrodicted biophilic quality of our universe appears, under the SB hypothesis, to constitute a possible biosignature.

Caution Regarding the Use of Retrodiction to Test the SB Hypothesis

Because the SB hypothesis is radically novel and because the use of falsifiable retrodiction as a tool to test such an hypothesis creates at least the appearance of a “confirmatory argument resemble[ing] just-so stories (Rudyard Kipling’s fanciful stories, e.g., how leopards got their spots)” (Cleland, 2001), it is important (as noted previously) that two additional criteria be satisfied before this retrodiction can be considered a legitimate test of the hypothesis:

• The SB hypothesis must generate falsifiable predictions as well as falsifiable retrodictions; and

• The SB hypothesis must be consilient with key theoretical constructs in such “adjoining” area of scientific investigation as M-theory, cosmogenesis, complexity theory, and evolutionary theory.

As argued at length elsewhere (Gardner, 2003), the SB hypothesis is both consilient with central concepts in these “adjoining” fields and fully capable of generating falsifiable predictions.

Concluding Remarks

In his book The Fifth Miracle (Davies, 1999) Davies offered this interpretation of NASA’s view that the presence of liquid water on an alien world was a reliable marker of a life-friendly environment:

In claiming that water means life, NASA scientists are… making—tacitly—a huge and profound assumption about the nature of nature. They are saying, in effect, that the laws of the universe are cunningly contrived to coax life into being against the raw odds; that the mathematical principles of physics, in their elegant simplicity, somehow know in advance about life and its vast complexity. If life follows from [primordial] soup with causal dependability, the laws of nature encode a hidden subtext, a cosmic imperative, which tells them: “Make life!” And, through life, its by-products: mind, knowledge, understanding. It means that the laws of the universe have engineered their own comprehension. This is a breathtaking vision of nature, magnificent and uplifting in its majestic sweep. I hope it is correct. It would be wonderful if it were correct. But if it is, it represents a shift in the scientific world-view as profound as that initiated by Copernicus and Darwin put together.

An emerging consensus among mainstream physicists and cosmologists is that the particular universe we inhabit appears to confirm what Smolin calls the “anthropic observation”: the laws and constants of nature seem to be fine-tuned, with extraordinary precision and against enormous odds, to favor the emergence of life and its byproduct, intelligence. As Dyson put it eloquently more than two decades ago (Dyson, 1979):

The more I examine the universe and study the details of its architecture, the more evidence I find that the universe in some sense must have known that we were coming. There are some striking examples in the laws of nuclear physics of numerical accidents that seem to conspire to make the universe habitable.

Why this should be so remains a profound mystery. Indeed, the mystery has deepened considerably with the recent discovery of the inexplicably tiny value of dark energy density and the realization that M-theory encompasses an unfathomably vast landscape of possible solutions, only a minute fraction of which correspond to anything resembling the universe that we inhabit.

Confronted with such a deep mystery, the scientific community ought to be willing to entertain plausible explanatory hypotheses that may appear to be unconventional or even radical. However, such hypotheses, to be taken seriously, must:

• be consilient with the key paradigms of “adjoining” scientific fields,

• generate falsifiable predictions, and

• generate falsifiable retrodictions.

The SB hypothesis satisfies these criteria. In particular, it generates a falsifiable retrodiction that the physical laws and constants that prevail in our cosmos will be biophilic—which they are.

References

Baez, J. 1998 on-line commentary on The Life of the Cosmos (available at http://www.aleph.se/Trans/Global/Omega/smolin.txt).

Barrow, J. and Tipler, F. 1988 The Anthropic Cosmological Principle, Oxford University Press.

Bjorken, J. 2004 “The Classification of Universes,” astro-ph/0404233.

Cleland, C. 2001 “Historical science, experimental science, and the scientific method,” Geology, 29, pp. 978-990.

Cleland, C. 2002 “Methodological and Epistemic Differences Between Historical Science and Experimental Science,” Philosophy of Science, 69, pp. 474-496.

Cleland, C. 2004 “Historical Science and the Use of Biosignatures,” unpublished summary of presentation abstracted in International Journal of Astrobiology, Supplement 2004, p. 119.

Davies, P. 1999 The Fifth Miracle, Simon & Schuster.

Dyson, F. 1979 Disturbing the Universe, Harper & Row.

Dyson, F. 1988 Infinite in All Directions, Harper & Row.

Gardner, J. 2000 “The Selfish Biocosm: Complexity as Cosmology,” Complexity, 5, no. 3, pp. 34-45..

Gardner, J. 2001 “Assessing the Robustness of the Emergence of Intelligence: Testing the Selfish Biocosm Hypothesis,” Acta Astronautica, 48, no. 5-12, pp. 951-955.

Gardner, J. 2002 “Assessing the Computational Potential of the Eschaton: Testing the Selfish Biocosm Hypothesis,” Journal of the British Interplanetary Society 55, no. 7/8, pp. 285-288.

Gardner, J. 2003 Biocosm, Inner Ocean Publishing.

Gee, H. 1999 In Search of Deep Time, The Free Press.

Goldsmith, D. 2004 “The Best of All Possible Worlds,” Natural History, 5, no. 6, pp. 44-49.

Greene, B. 2004 The Fabric of the Cosmos, Knopf.

Hawking, S. and Hertog, T. 2002 “Why Does Inflation Start at the Top of the Hill?” hep-th/0204212.

Henderson, L. 1913 The Fitness of the Environment, Harvard University Press.

Hogan, C. 2004 “Quarks, Electrons, and Atoms in Closely Related Universes,” astro-ph/0407086.

Khoury, J., Ovrut, B. A., Seiberg, N., Steinhardt, P., and Turok, N. 2001 “From Big Crunch to Big Bang,” hep-th/0108187.

Kurzweil, R. 1999 The Age of Spiritual Machines, Viking.

Linde, A. 2002 “Inflation, Quantum Cosmology and the Anthropic Principle,” hep-th/0211048.

Linde, A.1998 “The Self-Reproducing Inflationary Universe,” Scientific American, 9(20), pp. 98-104.

Livio, M. 2003 “Cosmology and Life,” astro-ph/0301615.

Lloyd, S. 2000 “Ultimate Physical Limits to Computation,” Nature, 406, pp. 1047-1054.

Mitton, S. 2005 Conflict in the Cosmos: Fred Hoyle’s Life in Science, Joseph Henry Press.

Oldershaw, R. 1988 “The new physics: physical or mathematical science?” American Journal of Physics, 56(12).

Pagels, H. 1983 The Cosmic Code, Bantam.

Rees, M. 1997 Before the Beginning, Addison Wesley.

Rees, M. 2000 Just Six Numbers, Basic Books.

Smolin, L. 1997 The Life of the Cosmos, Oxford University Press.

Smolin, L. 2004 “Scientific Alternatives to the Anthropic Principle,” hep-th/0407213.

Steinhardt, P. and Turok, N. 2001 “Cosmic Evolution in a Cyclic Universe,” hep-th/0111098.

Susskind, L. 2003 “The Anthropic Landscape of String Theory,” hep-th/0302219.

Tipler, F. 1994 The Physics of Immortality, Doubleday.

Vilenkin, A. 2004 “Anthropic predictions: The Case of the Cosmological Constant,” astro-ph/0407586.

von Neumann, J. 1948 “On the General and Logical Theory of Automata.”

Weinberg, S. 21 October 1999 “A Designer Universe?” New York Review of Books.

Wheeler, J. 1996 At Home in the Universe, AIP Press.

Wilson, E. O. 1998 “Scientists, Scholars, Knaves and Fools,” American Scientist, 86, pp. 6-7.

[1] http://www.edge.org/discourse/landscape.html.

[2] The metaphor furnished by the familiar process of artificial selection was Darwin’s crucial stepping stone. Indeed, the practice of artificial selection through plant and animal breeding was the primary intellectual model that guided Darwin in his quest to solve the mystery of the origin of species and to demonstrate in principle the plausibility of his theory that variation and natural selection were the prime movers responsible for the phenomenon of speciation.

[3] Both defects were emphasized by Susskind in a recent on-line exchange with Smolin which appears at www.edge.org. Smolin has argued that his CNS hypothesis has not been falsified on the first ground (Smolin, 2004) but conceded that his conjecture lacks any hypothesized mechanism that would endow the putative process of proliferation of black-hole-prone universes with a heredity function:

The hypothesis that the parameters p change, on average, by small random amounts, should be ultimately grounded in fundamental physics. We note that this is compatible with string theory, in the sense that there are a great many string vacua, which likely populate the space of low energy parameters well. It is plausible that when a region of the universe is squeezed to Planck densities and heated to Planck temperatures, phase transitions may occur leading to a transition from one string vacua to another. But there have so far been no detailed studies of these processes which would check the hypothesis that the change in each generation is small.

As Smolin noted in the same paper, it is crucial that such a mechanism exist in order to avoid the conclusion that each new universe’s set of physical laws and constants would constitute a merely random sample of the vast parameter space permitted by the extraordinarily large “landscape” of M-theory-allowed solutions:

It is important to emphasize that the process of natural selection is very different from a random sprinkling of universes on the parameter space P. This would produce only a uniform distribution p_random(p). To achieve a distribution peaked around the local maxima of a fitness function requires the two conditions specified. The change in each generation must be small so that the distribution can “climb the hills” in F(p) rather than jump around randomly, and so it can stay in the small volume of P where F(p) is large, and not diffuse away. This requires many steps to reach local maxima from random starts, which implies that long chains of descendants are needed.

[4] Wilson has identified consilience as one of the “diagnostic features of science that distinguishes it from pseudoscience” (Wilson, 1998):

The explanations of different phenomena most likely to survive are those that can be connected and proved consistent with one another.

© 2005 James N. Gardner. Reprinted with permission.

An exciting revolution in health care and medical technology looms large on the horizon. Yet the agents of change will be microscopically small, future products of a new discipline known as nanotechnology. Nanotechnology is the engineering of molecularly precise structures—typically 0.1 microns or smaller—and, ultimately, molecular machines. Nanomedicine^1-4 is the application of nanotechnology to medicine. It is the preservation and improvement of human health, using molecular tools and molecular knowledge of the human body. Present-day nanomedicine exploits carefully structured nanoparticles such as dendrimers,⁵ carbon fullerenes (buckyballs)⁶ and nanoshells⁷ to target specific tissues and organs. These nanoparticles may serve as diagnostic and therapeutic antiviral, antitumor or anticancer agents. But as this technology matures in the years ahead, complex nanodevices and even nanorobots will be fabricated, first of biological materials but later using more durable materials such as diamond to achieve the most powerful results.

Early Vision

Can it be that someday nanorobots will be able to travel through the body searching out and clearing up diseases, such as an arterial atheromatous plaque?⁸ The first and most famous scientist to voice this possibility was the late Nobel physicist Richard P. Feynman. In his remarkably prescient 1959 talk “There’s Plenty of Room at the Bottom,” Feynman proposed employing machine tools to make smaller machine tools, these to be used in turn to make still smaller machine tools, and so on all the way down to the atomic level, noting that this is “a development which I think cannot be avoided.”⁹

Feynman was clearly aware of the potential medical applications of this new technology. He offered the first known proposal for a nanorobotic surgical procedure to cure heart disease: “A friend of mine (Albert R. Hibbs) suggests a very interesting possibility for relatively small machines. He says that, although it is a very wild idea, it would be interesting in surgery if you could swallow the surgeon. You put the mechanical surgeon inside the blood vessel and it goes into the heart and looks around. (Of course the information has to be fed out.) It finds out which valve is the faulty one and takes a little knife and slices it out. …[Imagine] that we can manufacture an object that maneuvers at that level!… Other small machines might be permanently incorporated in the body to assist some inadequately functioning organ.”⁹

Medical Microrobotics

There are ongoing attempts to build microrobots for in vivo medical use. In 2002, Ishiyama et al at Tohoku University developed tiny magnetically-driven spinning screws intended to swim along veins and carry drugs to infected tissues or even to burrow into tumors and kill them with heat.¹⁰ In 2003, the “MR-Sub” project of Martel’s group at the NanoRobotics Laboratory of Ecole Polytechnique in Montreal tested using variable MRI magnetic fields to generate forces on an untethered microrobot containing ferromagnetic particles, developing sufficient propulsive power to direct the small device through the human body.¹¹ Brad Nelson’s team at the Swiss Federal Institute of Technology in Zurich continued this approach. In 2005 they reported the fabrication of a microscopic robot small enough (~200 microns) to be injected into the body through a syringe. They hope this device or its descendants might someday be used to deliver drugs or perform minimally invasive eye surgery.¹² Nelson’s simple microrobot has successfully maneuvered through a watery maze using external energy from magnetic fields, with different frequencies able to vibrate different mechanical parts on the device to maintain selective control of different functions. Gordon’s group at the University of Manitoba has also proposed magnetically-controlled “cytobots” and “karyobots” for performing wireless intracellular and intranuclear surgery.¹³

 Manufacturing Medical Nanorobots

 The greatest power of nanomedicine will emerge, perhaps in the 2020s, when we can design and construct complete artificial nanorobots using rigid diamondoid nanometer-scale parts like molecular gears and bearings.¹⁴ These nanorobots will possess a full panoply of autonomous subsystems including onboard sensors, motors, manipulators, power supplies, and molecular computers. But getting all these nanoscale components to spontaneously self-assemble in the right sequence will prove increasingly difficult as machine structures become more complex. Making complex nanorobotic systems requires manufacturing techniques that can build a molecular structure by what is called positional assembly. This will involve picking and placing molecular parts one by one, moving them along controlled trajectories much like the robot arms that manufacture cars on automobile assembly lines. The procedure is then repeated over and over with all the different parts until the final product, such as a medical nanorobot, is fully assembled.

The positional assembly of diamondoid structures, some almost atom by atom, using molecular feedstock has been examined theoretically^14,15 via computational models of diamond mechanosynthesis (DMS). DMS is the controlled addition of carbon atoms to the growth surface of a diamond crystal lattice in a vacuum manufacturing environment. Covalent chemical bonds are formed one by one as the result of positionally constrained mechanical forces applied at the tip of a scanning probe microscope apparatus, following a programmed sequence. Mechanosynthesis using silicon atoms was first achieved experimentally in 2003.¹⁶ Carbon atoms should not be far behind.¹⁷

To be practical, molecular manufacturing must also be able to assemble very large numbers of medical nanorobots very quickly. Approaches under consideration include using replicative manufacturing systems or massively parallel fabrication, employing large arrays of scanning probe tips all building similar diamondoid product structures in unison.¹⁸

For example, simple mechanical ciliary arrays consisting of 10,000 independent microactuators on a 1 cm² chip have been made at the Cornell National Nanofabrication Laboratory for microscale parts transport applications, and similarly at IBM for mechanical data storage applications.¹⁹ Active probe arrays of 10,000 independently-actuated microscope tips have been developed by Mirkin’s group at Northwestern University for dip-pen nanolithography²⁰ using DNA-based “ink”. Almost any desired 2D shape can be drawn using 10 tips in concert. Another microcantilever array manufactured by Protiveris Corp. has millions of interdigitated cantilevers on a single chip. Martel’s group has investigated using fleets of independently mobile wireless instrumented microrobot manipulators called NanoWalkers to collectively form a nanofactory system that might be used for positional manufacturing operations.²¹ Zyvex Corp of Richardson TX has a $25 million, five-year, National Institute of Standards and Technology (NIST) contract to develop prototype microscale assemblers using microelectromechanical systems. This research may eventually lead to prototype nanoscale assemblers using nanoelectromechanical systems.

Respirocytes and Microbivores 

The ability to build complex diamondoid medical nanorobots to molecular precision, and then to build them cheaply enough in sufficiently large numbers to be useful therapeutically, will revolutionize the practice of medicine and surgery.¹ The first theoretical design study of a complete medical nanorobot ever published in a peer-reviewed journal (in 1998) described a hypothetical artificial mechanical red blood cell or “respirocyte” made of 18 billion precisely arranged structural atoms.²² The respirocyte is a bloodborne spherical 1-micron diamondoid 1000-atmosphere pressure vessel with reversible molecule-selective surface pumps powered by endogenous serum glucose. This nanorobot would deliver 236 times more oxygen to body tissues per unit volume than natural red cells and would manage carbonic acidity, controlled by gas concentration sensors and an onboard nanocomputer. A 5 cc therapeutic dose of 50% respirocyte saline suspension containing 5 trillion nanorobots could exactly replace the gas carrying capacity of the patient’s entire 5.4 liters of blood.  Nanorobotic artificial phagocytes called “microbivores” could patrol the bloodstream, seeking out and digesting unwanted pathogens including bacteria, viruses, or fungi.²³ Microbivores would achieve complete clearance of even the most severe septicemic infections in hours or less. This is far better than the weeks or months needed for antibiotic-assisted natural phagocytic defenses. The nanorobots don’t increase the risk of sepsis or septic shock because the pathogens are completely digested into harmless sugars, amino acids and the like, which are the only effluents from the nanorobot.

Surgical Nanorobotics

Surgical nanorobots could be introduced into the body through the vascular system or at the ends of catheters into various vessels and other cavities in the human body. A surgical nanorobot, programmed or guided by a human surgeon, could act as an semi-autonomous on-site surgeon inside the human body. Such a device could perform various functions such as searching for pathology and then diagnosing and correcting lesions by nanomanipulation, coordinated by an on-board computer while maintaining contact with the supervising surgeon via coded ultrasound signals. The earliest forms of cellular nanosurgery are already being explored today. For example, a rapidly vibrating (100 Hz) micropipette with a <1 micron tip diameter has been used to completely cut dendrites from single neurons without damaging cell viability.²⁴ Axotomy of roundworm neurons was performed by femtosecond laser surgery, after which the axons functionally regenerated.²⁵ A femtolaser acts like a pair of “nano-scissors” by vaporizing tissue locally while leaving adjacent tissue unharmed. Femtolaser surgery has performed: (1) localized nanosurgical ablation of focal adhesions adjoining live mammalian epithelial cells,²⁶ (2) microtubule dissection inside yeast cells,²⁷ (3) noninvasive intratissue nanodissection of plant cell walls and selective destruction of intracellular single plastids or selected parts of them,²⁸ and even (4) the nanosurgery of individual chromosomes (selectively knocking out genomic nanometer-sized regions within the nucleus of living Chinese hamster ovary cells²⁹). These procedures don’t kill the cells upon which the nanosurgery was performed. Atomic force microscopes have also been used for bacterium cell wall dissection in situ in aqueous solution, with 26 nm thick twisted strands revealed inside the cell wall after mechanically peeling back large patches of the outer cell wall.³⁰  Future nanorobots equipped with operating instruments and mobility will be able to perform precise and refined intracellular surgeries which are beyond the capabilities of direct manipulation by the human hand. We envision biocompatible³¹ surgical nanorobots that can find and eliminate isolated cancerous cells, remove microvascular obstructions and recondition vascular endothelial cells, perform “noninvasive” tissue and organ transplants, conduct molecular repairs on traumatized extracellular and intracellular structures, and even exchange new whole chromosomes for old ones inside individual living human cells.

References

1. Freitas RA Jr. Nanomedicine, Vol. I: Basic Capabilities. Georgetown (TX): Landes Bioscience; 1999. Also available at: http://www.nanomedicine.com/NMI.htm.

2. Robert A. Freitas Jr. Nanodentistry. J Amer Dent Assoc 2000; 131:1559-66.

3. Freitas RA Jr. Current status of nanomedicine and medical nanorobotics (invited survey). J Comput Theor Nanosci 2005; 2:1-25. Also available at: http://www.nanomedicine.com/Papers/NMRevMar05.pdf.

4. Freitas RA Jr. What is nanomedicine? Nanomedicine: Nanotech Biol Med 2005; 1:2-9. Also available at: http://www.nanomedicine.com/Papers/WhatIsNMMar05.pdf.

5. Borges AR, Schengrund CL. Dendrimers and antivirals: a review. Curr Drug Targets Infect Disord 2005; 5:247-54.  

6. Mashino T, Shimotohno K, Ikegami N, Nishikawa D, Okuda K, Takahashi
K, Nakamura S, Mochizuki M. Human immunodeficiency virus-reverse transcriptase
inhibition and hepatitis C virus RNA-dependent RNA polymerase inhibition
activities of fullerene derivatives. Bioorg Med Chem Lett 2005;
15:1107-9.  

7. O’Neal DP, Hirsch LR, Halas NJ, Payne JD, West JL. Photo-thermal tumor ablation in mice using near infrared-absorbing nanoparticles. Cancer Lett 2004; 209:171-6.

8. Dewdney AK. Nanotechnology—wherein molecular computers control tiny circulatory submarines. Sci Am 1988 Jan; 258:100-3.

9. Feynman RP. There’s plenty of room at the bottom. Eng Sci 1960 Feb; 23:22-36. Also available at: http://www.zyvex.com/nanotech/feynman.html.

10. Ishiyama K, Sendoh M, Arai KI. Magnetic micromachines for medical applications. J Magnetism Magnetic Mater 2002; 242-245:1163-5.

11. Mathieu JB, Martel S, Yahia L, Soulez G, Beaudoin G. MRI systems as a mean of propulsion for a microdevice in blood vessels. Proc. 25th Ann. Intl. Conf., IEEE Engineering in Medicine and Biology; 2003 Sep 17-21; Cancun, Mexico; 2003. Also available at: http://www.nano.polymtl.ca/Articles/2003/MRI%20Syst%20Mean%20Prop%20Microdev%20Blood%20Vess%20proceedings%20P3419.pdf

12. Nelson B, Rajamani R. Biomedical micro-robotic system. 8th Intl. Conf. on Medical Image Computing and Computer Assisted Intervention (MICCAI 2005/ www.miccai2005.org), Palm Springs CA, 26-29 October 2005.

13. Chrusch DD, Podaima BW, Gordon R. Cytobots: intracellular robotic micromanipulators. In: Kinsner W, Sebak A, eds. Conf. Proceedings, 2002 IEEE Canadian Conference on Electrical and Computer Engineering; 2002 May 12-15; Winnipeg, Canada. Winnipeg: IEEE; 2002.

14. Drexler KE. Nanosystems: Molecular Machinery, Manufacturing, and Computation. New York: John Wiley & Sons; 1992.

15. Merkle RC, Freitas RA Jr. Theoretical analysis of a carbon-carbon dimer placement tool for diamond mechanosynthesis. J Nanosci Nanotechnol 2003; 3:319-24. Also available at: http://www.rfreitas.com/Nano/JNNDimerTool.pdf.

16. Oyabu N, Custance O, Yi I, Sugawara Y, Morita S. Mechanical vertical manipulation of selected single atoms by soft nanoindentation using near contact atomic force microscopy. Phys Rev Lett 2003; 90:176102.

17. Freitas RA Jr. A Simple Tool for Positional Diamond Mechanosynthesis, and its Method of Manufacture. U.S. Provisional Patent Application No. 60/543,802, filed 11 February 2004; U.S. Patent Pending, 11 February 2005. Also available at: http://www.MolecularAssembler.com/Papers/DMSToolbuildProvPat.htm.

18. Freitas RA Jr., Merkle RC. Kinematic Self-Replicating Machines.Georgetown (TX): Landes Bioscience; 2004. Also available at: http://www.molecularassembler.com/KSRM.htm.

19. Vettiger P, Cross G, Despont M, Drechsler U, Duerig U, Gotsmann B, Haeberle W, Lantz M, Rothuizen H, Stutz R, Binnig G. The Millipede—nanotechnology entering data storage. IEEE Trans Nanotechnol 2002 Mar; 1:39-55.

20. Bullen D, Chung S, Wang X, Zou J, Liu C, Mirkin C. Development of parallel dip pen nanolithography probe arrays for high throughput nanolithography. (Invited) Symposium LL: Rapid Prototyping Technologies, Materials Research Society Fall Meeting; 2-6 Dec 2002; Boston, MA. Proc. MRS, Vol. 758, 2002. Also available at: http://mass.micro.uiuc.edu/publications/papers/84.pdf.

21. Martel S, Hunter I. Nanofactories based on a fleet of scientific instruments configured as miniature autonomous robots. Proc. of the 3rd Intl. Workshop on Microfactories; 16-18 Sep 2002; Minneapolis, MN, USA; 2002, pp. 97-100.

22. Freitas RA Jr. Exploratory design in medical nanotechnology: a mechanical artificial red cell. Artif Cells Blood Subst Immobil Biotech 1998; 26:411-30. Also available at: http://www.foresight.org/Nanomedicine/Respirocytes.html.

23. Freitas RA Jr. Microbivores: artificial mechanical phagocytes using digest and discharge protocol. J Evol Technol 2005 Apr; 14:1-52. http://jetpress.org/volume14/Microbivores.pdf.

24. Kirson ED, Yaari Y. A novel technique for micro-dissection of neuronal processes. J Neurosci Methods 2000; 98:119-22.

25. Yanik MF, Cinar H, Cinar HN, Chisholm AD, Jin Y, Ben-Yakar A.Neurosurgery: functional regeneration after laser axotomy. Nature 2004; 432:822.

26. Kohli V, Elezzabi AY, Acker JP. Cell nanosurgery using ultrashort (femtosecond) laser pulses: Applications to membrane surgery and cell isolation. Lasers Surg Med 2005; 37:227-30.

27. Sacconi L, Tolic-Norrelykke IM, Antolini R, Pavone FS. Combined intracellular three-dimensional imaging and selective nanosurgery by a nonlinear microscope. J Biomed Opt 2005; 10:14002.

28. Tirlapur UK, Konig K. Femtosecond near-infrared laser pulses as a versatile non-invasive tool for intra-tissue nanoprocessing in plants without compromising viability. Plant J 2002; 31:365-74.

29. Konig K, Riemann I, Fischer P, Halbhuber KJ. Intracellular nanosurgery with near infrared femtosecond laser pulses. Cell Mol Biol 1999; 45:195-201.

30. Firtel M, Henderson G, Sokolov I. Nanosurgery: observation of peptidoglycan strands in Lactobacillus helveticus cell walls. Ultramicroscopy 2004 Nov;101:105-9.

31. Freitas RA Jr. Nanomedicine, Vol. IIA: Biocompatibility. Georgetown (TX): Landes Bioscience; 2003. Also available at: http://www.nanomedicine.com/NMIIA.htm.

Robert A. Freitas Jr. has written pioneering books on nanomedicine,
nanorobots, and molecular manufacturing. What’s next? The last two books in the Nanomedicine series and a book on fundamentals of nanomechanical engineering, extending Eric Drexler’s classic Nanosystems, he reveals in this interview.

Originally published on Nanotech.biz November 4, 2005. Reprinted on KurzweilAI.net February 2, 2006.

Robert A. Freitas Jr., J.D., published the first detailed technical design study of a mechanical nanorobot ever published in a peer-reviewed mainstream biomedical journal and is the author of nanomedicine, the first book-length technical discussion of the medical applications of nanotechnology and medical nanorobotics.

Question 1: Tell us about yourself. What is your background, and what is your current affiliation?

I received an undergraduate B.S. degree from Harvey Mudd College (dual major, physics and psychology) in 1974 and a Juris Doctor (J.D.) graduate degree from University of Santa Clara School of Law in 1978. In the late 1970s and early 1980s I published numerous editions of Lobbying for Space, the first space program political advocacy handbook ever published, and conducted three separate observational SETA/SETI programs with a colleague, using both optical and radio telescopes. I co-edited the 1980 NASA feasibility analysis of self-replicating space factories and in 1996 authored the first detailed technical design study of a medical nanorobot ever published in a peer-reviewed mainstream biomedical journal. After a stint as Research Scientist at Zyvex Corp. from 2000-2004, I’m now back with the Institute for Molecular Manufacturing, my previous and current primary affiliation, as their Senior Research Fellow.

Question 2: When and how did you first hear about molecular nanotechnology? At what point did you decide to devote your career to this field?

The first time I ever thought about atomic-scale engineered objects was probably in 1977-78, when I was working on my first treatise-length book project (Xenology). In Chapter 16 of that book, I hypothesized that “using molecular electronics with components on the order of 10 Å in size, 10¹⁰ microneurons could be packed into a space of a few microns” which would be “small enough to hide inside a bacterium.” During my NASA work on self-replicating machines in the summer of 1980, I wondered how small machine replicators might be made. I briefly studied the emerging micromachine technology, but by the time Engines of Creation came out in 1986 I had temporarily left the field in pursuit of more pragmatic opportunities. In early 1994 I happened to pick up and read a copy of Unbounding the Future. This was my first exposure to what has come to be known as molecular nanotechnology (MNT). I studied the detailed technical arguments presented in Nanosystems, which confirmed what I had already suspected based on my own knowledge—namely, that the technical case for molecular nanotechnology was very solid.

Having fully absorbed the MNT paradigm, I immediately realized that medicine would be the single most important application area of this new technology. In particular, nanomedicine offered a chance for significant healthspan (healthy lifespan) extension. It also appeared that this objective could possibly be achieved within the several decades of life actuarially remaining to me and others of my generation. But was anyone pushing it forward? I contacted the Foresight Institute and learned that nobody had yet written any systematic treatment of this area, nor was anyone planning to do so in the near future. So I took up the challenge of writing Nanomedicine, the first book-length technical discussion of the potential medical applications of molecular nanotechnology and medical nanorobotics.

I’ve been writing the Nanomedicine book series since 1994.  This technical book is my attempt to rationally assess various possible nanorobotic capabilities and medical systems to determine which ones might be plausible (and which ones not) if we could build nanorobots at some point in the future. The first volume (I) was published by Landes Bioscience in 1999 and is freely available online at http://www.nanomedicine.com/NMI.htm.  The second volume (IIA) was also published by Landes Bioscience, in 2003, and is also freely available online at http://www.nanomedicine.com/NMIIA.htm.  I’m still writing the last 2 volumes (IIB, III) of this book series, an ongoing effort that will continue during 2005-2010.

Question 3: Recently, the Foresight community has de-emphasized molecular assemblers in favor of a desktop manufacturing paradigm. What brought about this shift? How will this change affect the field?

While I’m not involved in the decisions of the Foresight Institute, I believe the shift occurred primarily as an attempt to redirect the often rancorous scientific debate away from the growing fears of runaway motile free-range replicators, and away from the seeming impossibility of building self-replicating machines (a prejudice common to many ill-informed scientists), and towards a more rational consideration of the underlying technologies and their benefits. The civility of the public discourse may improve as a result, and to the extent that the mainstream scientific community begins to pay attention, it is possible that the research funding situation might also improve. I’m all for it.

However, the change probably won’t much affect the actual research in the field, nor the achievement of useful results per dollar spent, because in truth the distinction between “molecular assemblers” and “nanofactories” is largely cosmetic. That is, if you possess either one, you can use it either to replicate itself or to build the other in very short order. Either one can be used equally well to build life-saving medical nanorobots or life-denying nanoweapons, including everyone’s favorite bugaboo, the marauding ecophages. Both assemblers and nanofactories are examples of molecular manufacturing, which depends at its core on some form of replicative or massively-parallel fabrication and assembly capability in order to be able to economically generate macroscale quantities of useful end products. The two approaches differ mainly in their technical design/performance tradeoffs. Each approach has different strengths and weaknesses (as manufacturing systems) that can be readily enumerated. I’ve been writing about both approaches since the 1980 NASA replicating factory study – wherein I was actually the main proponent for the factory approach. The key thing is that molecular assemblers and nanofactories are both molecular manufacturing systems. Each requires almost exactly the same set of enabling technologies. Developing those enabling technologies as soon as possible should be our primary focus right now.

Question 4: Do you agree with those who claim that the desktop manufacturing paradigm could become a reality by 2020?

I would not be surprised if the fabrication of medical nanorobots (and other useful nanorobotic systems) via molecular manufacturing – whether via molecular assemblers or nanofactories – arrives during the decade of the 2020s.

As noted earlier, I undertook the Nanomedicine book series in an attempt to establish a solid foundation for the single most important future application of MNT. The book introduces a long-term vision for nanorobotic medicine and articulates the technical underpinnings of that vision, so that when the day arrives that we have the technology to build such devices, we’ll have a clearer idea what can be done with them, and how.

More recently, to answer those who remain skeptical of the entire MNT enterprise, including the possibility of medical nanorobotics, I’ve turned my attention to figuring out how to build the nanorobots – the issue of implementation of the long-term vision.  My early work on diamond mechanosynthesis is described in a lecture I gave at the 2004 Foresight Conference in Washington DC, the text of which (plus many images) is available online.  I’m now involved in 6 research collaborations with various university and corporate groups in the U.S, U.K. and Russia in an effort to push forward the technology in this area as fast as possible. These collaborations include a variety of computational chemistry simulations of plausible mechanosynthetic tooltips and reaction sequences, coupled with a nascent experimental effort that is just starting up. I have several new papers on diamond mechanosynthesis nearing completion for journal submission, for publication in 2006. Earlier this year I also filed the first-ever U.S. patent on diamond mechanosynthesis that describes a specific process for achieving molecularly precise diamond structures in a practical way.

Ralph Merkle and I are also writing an entire book-length discussion of diamond mechanosynthesis, entitled Diamond Surfaces and Diamond Mechanosynthesis (DSDM), to be published in 2006 or 2007. The first half of this book is an extensive review of all that is presently known about diamond surfaces, and has been mostly written for several years. The second half describes specific tools and reaction pathways for building those surfaces using positionally controlled mechanosynthetic tools, and methods for building those tools. This part has been about 50% written for several years. But finishing this part has been put on hold because our many current research collaborations involving ab initio and DFT-based quantum chemistry simulations are providing so much new information that we think it’s better to wait and incorporate this new material into the book. (Otherwise we could’ve published DSDM in 2005.) Until then, I’ve put together a brief technical bibliography of research on positional mechanosynthesis (including diamond). Watch the Molecular Assembler website for updates and further news about DSDM.

Question 5: Tell us about your latest book, Kinematic Self-Replicating Machines. Was this book written with the specific aim of convincing skeptics of the feasibility of molecular manufacturing?

Yes it was, though of course the subject of self-replicating machines has been a long-standing professional interest of mine, across 3 decades. For instance, I published the first quantitative closure analysis for a self-replicating machine system in 1979-1980 and participated in (and edited) the first comprehensive technical analysis of a self-replicating lunar factory for NASA in 1980.

But focusing again on molecular manufacturing: Once diamond mechanosynthesis and the fabrication of nanoparts becomes feasible, we will also need a massively parallel molecular manufacturing capability in order to assemble nanorobots cheaply, precisely, and in vast quantities.  Kinematic Self-Replicating Machines (KSRM) (Landes Bioscience, 2004, and freely available online), co-authored with Ralph Merkle, surveys all known current work in the field of self-replication and replicative manufacturing, including all known concepts of molecular assemblers and nanofactories. It is intended as a general introduction to the systems-level analysis of self-replicative manufacturing machinery. With 200+ illustrations and 3200+ literature references, KSRM describes all proposed and experimentally realized self-replicating systems that were publicly disclosed as of 2004, ranging from nanoscale to macroscale systems.  The book extensively describes the historical development of the field.  It presents for the first time a detailed 137-dimensional map of the entire kinematic replicator design space to assist future engineering efforts.  It includes a primer on the mathematics of self-replication, and has an extensive discussion of safety issues and implementation issues related to molecular assemblers and nanofactories. KSRM has been cited in two articles appearing in Nature this year (Zykov et al, Nature 435, 163 (12 May 2005) and Griffith et al, Nature 437, 636 (29 September 2005)) and appears well on its way to becoming the classic reference in this field.

Perhaps the most salutary effect of KSRM is that it provides a number of physical examples of self-replicating systems (beyond the relatively simple autocatalytic-type replicators from the 1950s by Penrose, Jacobson and Morowitz and the more recent related examples by Lohn and Griffith) that have already been built and operated in a laboratory environment. This provides a ready answer to the tedious and recurring objection by the ill-informed that such things are “impossible”: The machines have actually been built. Interestingly, one of these experimental replicators is a fully autonomous machine that runs around on a table and procures its own parts, which it then assembles into a working copy of itself, a crude analog of the molecular assembler approach. Another of these experimental replicators is a computer-controlled manipulator arm anchored to a surface, that grabs its parts from a “warehouse” area and assembles these parts into a working copy of itself, a crude analog of the nanofactory approach (where the nanofactories are being used to make more nanofactories, rather than nonfactory product).

Question 6: You and Dr. Hall are working on Fundamentals of Nanomechanical Engineering. Is this book intended to serve as a guide book for college students hoping to enter the nanotechnology field?

Fundamentals will be sharply focused on nanomechanical design, with a concentration on diamondoid molecular machine systems, intended for use in a mechanical engineering curriculum at the 2nd- or 3rd-year undergraduate level. We hope the book will be widely used in nanotechnology courses and will help to train the first generation of nanomechanical design engineers.

This is the primary purpose of the book. A second purpose is to extend and complement the analyses already published in Drexler’s Nanosystems, providing more design details and engineering analysis of how to build structures with molecular precision, and understanding how molecular machines might function (and the limitations on them).  Once an experimental ability to build diamondoid gears, bearings, rods, and the like has been demonstrated in the laboratory, I think the development of nanorobotics will move very rapidly from that point, because the potential payoff to human welfare is so large and the design space will have become accessible to active experimentation.

Question 7: MIT granted Eric Drexler a PhD in molecular nanotechnology. How long do you estimate before other Universities offer undergraduate and graduate degrees in molecular nanotechnology?

If you mean college degrees in MNT, in the sense of diamondoid molecular machine systems, I think this will begin to occur as soon as the field gains scientific credibility – i.e., as soon as it becomes clear that there is a “there,” there. The key here, in my opinion, will be the first experimental demonstration of positionally controlled diamond mechanosynthesis in the laboratory. Once this has been done, it will no longer be possible for critics to deny that such a thing is possible, though they still may claim that perhaps such a thing is not very useful for anything important. But with the newfound ability to tinker with real atoms, I expect legions of graduate students to rise up and prove the critics wrong on that score too, and the results of this revolution will rapidly trickle down to the undergraduate level as well.

How long until the first simple experimental demonstration of positionally controlled diamond mechanosynthesis in the laboratory? Perhaps not as far off as you might think. I’d be shocked if it was longer than 10 years, and 5 years would not surprise me. It depends on how fast we’re able to push it forward.

Question 8: You are a scientific advisor to the startup nanotechnology company Nanorex. What role do you anticipate that Nanorex will play in the development of nanotechnology? What is your contribution to that company?

Nanorex is creating an incredibly cool piece of software called NanoEngineer that allows the user to quickly and easily design molecular machine systems of up to perhaps 100,000 atoms in size, then perform various computational simulations on the system such as energy minimization (geometry optimization) or a quantitative analysis of applied forces and torques. It’s a CAD system for molecules, with a special competence in the area of diamondoid structures. Once this software is released, users anywhere in the world will be able to begin creating designs for relatively complex nanomachine components. We’d expect the library of designed machine systems to rapidly expand from the current 1-2 dozen items (including mostly just a few bearings, gears, and joints) into the hundreds or thousands in just a few years. The existence of this expanded library of nanoparts will then make it easier to begin thinking about designs for more complex systems that may be built from thousands or more of these parts, containing millions or even billions of atoms. It’s a big step along the molecular machine design and development pathway.

I’m a member of the Scientific Advisory Board of Nanorex. The Board provides feedback in the development of the NanoEngineer software, especially including our respective “wish lists” of what the ideal molecular machine design package should include – most of which features were then incorporated into the software. Thus the new software reflects the collective experience of those few of us in the world who have ever actually designed a molecular machine component the hard way – laboriously, atom by atom, using some previously existing (inadequate) software package.

Nanorex is also directly supporting the writing of Fundamentals of Nanomechanical Engineering, which we hope will be used to train the first generation of serious nanomechanical design engineers.

Question 9: Is the mainstream scientific establishment’s assessment of the feasibility of molecular nanotechnology changing? It appears that the National Nanotechnology Initiative’s long term forecasts have become bolder, yet the standard response by Smalley and others is that molecular manufacturing is inherently unworkable.

I think this feasibility assessment may be slowly changing, but this change is probably being driven mainly by published experimental results, especially in the field of STM/AFM(scanning probe)-moderated single-atom chemistry. For example, the first experimental demonstration of pure mechanosynthesis of any kind was reported in 2003 and is just now becoming more widely known. Publication of credible high-accuracy theoretical results, demonstrating the feasibility of diamond mechanosynthesis, will also help change this assessment. (For instance, in 2006 Merkle and I will be publishing a key theoretical paper in diamond mechanosynthesis representing ~10,000 CPU-hours of quantum chemistry simulations on 2600+ molecular structures to elucidate possible reaction pathways for building diamond. Watch for it.)

Smalley has marginalized himself in this area by taking such an extreme and indefensible position, using fallacious arguments. One by one, his arguments are slowly melting away under the hot glare of brilliant (but hard-won) experimental results.

Question 10: The Foresight Institute is collaborating with Battelle to create a nanotechnology roadmap. What do you know about this roadmap? Will it affect the mainstream scientific community’s assessment of molecular manufacturing?

Actually, I don’t know anything about their roadmap. They haven’t consulted me at all on this, and I have no idea what they’re up to except what I “read in the newspaper”. I believe it is an attempt to involve mainstream players in an assessment of possible development pathways leading toward some flavor(s) of molecular manufacturing. Whether these flavor(s) will include some or all of biological systems, protein systems, polymer systems, MEMS systems, metal systems, diamondoid systems, or something else, I cannot say.

Meanwhile, over the last two years Ralph Merkle and I have worked hard to establish a small independent network of research (both theoretical and experimental) collaborators with a sharp focus on the implementation of diamondoid molecular machine systems. Last June we put together a simple draft implementation flowchart that has about 100 boxes and lots of arrows, that starts from where we are today and ends with the manufacturing of complex molecular machine systems, including simple nanorobots. The plan includes specific theoretical and experimental milestones in a particular sequence. So we’re in the process of simplifying (and implementing) our own “roadmap”. Perhaps this (or something like it) will eventually be incorporated in the broader Foresight nanotechnology roadmap. Watch the Molecular Assembler website for more details in the months ahead.

Continued in Interview with Robert Freitas: Part 2.

©2006 Sander Olson. Reprinted with permission.

This is the complete original document describing the “Freitas process” to the level of detail that was known on 12 January 2004, following its initial conception on 1 November 2003. The actual Provisional Patent Application, prepared subsequently with the assistance of legal counsel, was abstracted from (and thus differs in some particulars from) this complete original document. A full utility patent on this process (containing numerous claims and some additional material, running a total of 133 pages in length) was subsequently filed on 11 February 2005. This patent is now pending before the USPTO. It is the first known patent ever filed on positional mechanosynthesis, and the first known patent ever filed on positional diamond mechanosynthesis.

Note: Philip Moriarty at the University of Nottingham (U.K.) has posted online several technical objections to one of the two proposed toolbuilding pathways, which Freitas says he is currently working through, point by point, with Moriarty via private correspondence in the manner of a friendly collaboration.

Abstract. A method is described for building a mechanosynthesis tool intended to be used for the molecularly precise fabrication of physical structures–as for example, diamond structures. The exemplar tool consists of a bulk-synthesized dimer-capped triadamantane tooltip molecule which is initially attached to a deposition surface in tip-down orientation, whereupon CVD or equivalent bulk diamond deposition processes are used to grow a large crystalline handle structure around the tooltip molecule. The large handle with its attached tooltip can then be mechanically separated from the deposition surface, yielding an integral finished tool that can subsequently be used to perform diamond mechanosynthesis in vacuo. The present disclosure is the first description of a complete tool for positional diamond mechanosynthesis, along with its method of manufacture. The same toolbuilding process may be extended to other classes of tooltip molecules, other handle materials, and to mechanosynthetic processes and structures other than those involving diamond.

OUTLINE

Abstract

1.Background of the Invention

1.1 Conventional Diamond Manufacturing     

1.2 Diamond Manufacturing via Positional Diamond Mechanosynthesis

2. Description of the Invention     

2.1 STEP 1: Synthesis of Capped Tooltip Molecule     

2.2 STEP 2: Attach Tooltip Molecule to Deposition Surface in Preferred Orientation          

2.2.1 Surface Nucleation and Choice of Deposition Substrate          

2.2.2 Tooltip Attachment Method A: Ion Bombardment in Vacuo          

2.2.3 Tooltip Attachment Method B: Surface Decapping in Vacuo          

2.2.4 Tooltip Attachment Method C: Solution Chemistry     

2.3 STEP 3: Attach Handle Structure to Tooltip Molecule          

2.3.1 Handle Attachment Method A: Nanocrystal Growth          

2.3.2 Handle Attachment Method B: Direct Handle Bonding     

2.4 STEP 4: Separate Finished Tool from Deposition Surface

References

1. Background of the Invention

The properties of diamond, such as its extraordinary hardness, coefficient of friction, tensile strength and low compressibility, electrical resistivity, electrical carrier (electron and hole) mobility, high energy bandgap and saturation velocity, dielectric breakdown strength, low neutron cross-section (radiation-hardness), thermal conductivity, thermal expansion resistance, optical transmittance and refractive index, and chemical inertness allow this material to serve a vital role in a wide variety of industrial and technical applications.

The present invention relates generally to methods for the manufacture of synthetic diamond. More particularly, the invention is concerned with the physical structure and method of manufacture of a tool, which can itself subsequently be employed in the mechanosynthetic manufacture of other molecularly precise diamond structures. However, the same toolbuilding process is readily extended to other classes of tooltip molecules, handle materials, and mechanosynthetic processes and structures other than diamond.

1.1 Conventional Diamond Manufacturing

All prior art methods of manufacturing diamond are bulk processes in which the diamond crystal structure is manufactured by statistical processes. In such processes, new atoms of carbon arrive at the growing diamond crystal structure having random positions, energies, and timing. Growth extends outward from initial nucleation centers having uncontrolled size, shape, orientation and location. Existing bulk processes can be divided into three principal methods – high pressure, low pressure hydrogenic, and low pressure nonhydrogenic.

(A) In the first or high pressure bulk method of producing diamond artificially, powders of graphite, diamond, or other carbon-containing substances are subjected to high temperature and high pressure to form crystalline diamond. High pressure processes are of several types [1]:

     (1) Impact Process. The starting powder is instantaneously brought under high pressure by applying impact generated by, for example, the explosion of explosives and the collision of a body accelerated to a high speed. This produces granular diamond by directly converting the starting powder material having a graphite structure into a powder composed of grains having a diamond structure. This process has the advantage that no press as is required, as in the two other processes, but there is difficulty in controlling the size of the resulting diamond products. Nongraphite organic compounds can also be shock-compressed to produce diamond [2].

     (2) Direct Conversion Process. The starting powder is held under a high static pressure of 13-16 GPa and a high temperature of 3,000-4,000 oC in a sealed high pressure vessel. This establishes stability conditions for diamond, so the powder material undergoes direct phase transition from graphite into diamond, through graphite decomposition and structural reorganization into diamond. In both direct conversion and flux processes, a press is widely used and enables single crystal diamonds to be grown as large as several millimeters in size.

     (3) Flux Process. As in direct conversion, a static pressure and high temperature are applied to the starting material, but here fluxes such as Ni and Fe are added to allow the reaction to occur under lower pressure and temperature conditions, accelerating the atomic rearrangement which occurs during the conversion process. For example, high-purity graphite powder is heated to 1500-2000 oC under 4-6 GPa of pressure in the presence of iron catalyst, and under this extreme, but equilibrium, condition of pressure and temperature, graphite is converted to diamond: The flux becomes a saturated solution of solvated graphite, and because the pressure inside the high pressure vessel is maintained in the stability range for diamond, the solubility for graphite far exceeds that for diamond, leading to diamond precipitation and dissolution of graphite into the flux. Every year about 75 tons of diamond are produced industrially this way [14].

(B) In the second or low pressure hydrogenic bulk method of producing diamond artificially, widely known as CVD or Chemical Vapor Deposition, hydrogen (H2) gas mixed with a few percent of methane (CH4) is passed over a hot filament or through a microwave discharge, dissociating the methane molecule to form the methyl radical (CH3) and dissociating the hydrogen molecule into atomic hydrogens (H). Acetylene (C2H2) can also be used in a similar manner as a carbon source in CVD. Diamond or diamond-like carbon films can be grown by CVD epitaxially on diamond nuclei, but such films invariably contain small contaminating amounts (0.1-1%) of hydrogen which gives rise to a variety of structural, electronic and chemical defects relative to pure bulk diamond. Currently, diamond synthesis from CVD is routinely achieved by more than 10 different methods [163].

As noted by McCune and Baird [3], a diamond particle is a special cubic lattice grown from a single nucleus of four-coordinated carbon atoms. The diamond-cubic lattice consists of two interpenetrating face-centered cubic lattices, displaced by one quarter of the cube diagonal. Each carbon atom is tetrahedrally coordinated, making strong, directed sp3 bonds to its neighbors using hybrid atomic orbitals. The lattice can also be visualized as planes of six-membered saturated carbon rings stacked in an ABC ABC ABC sequence along <111> directions. Each ring is in the “chair” conformation and all carbon-carbon bonds are staggered. A lattice with hexagonal symmetry, lonsdaleite, can be constructed with the same tetrahedral nearest neighbor configuration. In lonsdaleite, however, the planes of chairs are stacked in an AB AB AB sequence, and the carbon-carbon bonds normal to these planes are eclipsed. In simple organic molecules, the eclipsed conformation is usually less stable than the staggered because steric interactions are greater. Thermodynamically, diamond is slightly unstable with respect to crystalline graphite. At 298 K and 1 atm the free energy difference is 0.026 eV per atom, only slightly greater than kBT, where kB is the Boltzmann constant and T is the absolute temperature in degrees Kelvin.

The basic obstacle to crystallization of diamond at low pressures is the difficulty in avoiding co-deposition of graphite and/or amorphous carbon when operating in the thermodynamically stable region of graphite [3]. In general, the possibility of forming different bonding networks of carbon atoms is understandable from their ability to form different electronic configurations of the valence electrons. These bond types are classified as sp3 (tetrahedral), sp2 (planar), and sp1 (linear), and are related to the various carbon allotropes including cubic diamond and hexagonal diamond or lonsdaleite (sp3), graphite (sp2), and carbenes (sp1), respectively.

Hydrogen is generally regarded as an essential part of the reaction steps in forming diamond film during CVD, and atomic hydrogen must be present during low pressure diamond growth to: (1) stabilize the diamond surface, (2) reduce the size of the critical nucleus, (3) “dissolve” the carbon in the feedstock gas, (4) produce carbon solubility minimum, (5) generate condensable carbon radicals in the feedstock gas, (6) abstract hydrogen from hydrocarbons attached to the surface, (7) produce vacant surface sites, (8) etch (regasify) graphite, hence suppressing unwanted graphite formation, and (9) terminate carbon dangling bonds [4, 6]. Both diamond and graphite are etched by atomic hydrogen, but for diamond, the deposition rate exceeds the etch rate during CVD, leading to diamond (tetrahedral sp3 bonding) growth and the suppression of graphite (planar sp2 bonding) formation. (Note that most potential atomic hydrogen substitutes such as atomic halogens etch graphite at much higher rates than atomic hydrogen [4].)

Low pressure or CVD hydrogenic metastable diamond growth processes are of several types [3-5]:

     (1) Hot Filament Chemical Vapor Deposition (HFCVD). Filament deposition involves the use of a dilute (0.1-2.5%) mixture of hydrocarbon gas (typically methane) and hydrogen gas (H2) at 50-1000 torr which is introduced via a quartz tube located just above a hot tungsten filament or foil which is electrically heated to a temperature ranging from 1750-2800 oC. The gas mixture dissociates at the filament surface, yielding dissociation products consisting mainly of radicals including CH3, CH2, C2H, and CH, acetylene, and atomic hydrogen, as well as unreacted CH4 and H2. A heated deposition substrate placed just below the hot tungsten filament is held in a resistance heated boat (often molybdenum) and maintained at a temperature of 500-1100 oC, whereupon diamonds are condensed onto the heated substrate. Filaments of W, Ta, and Mo have been used to produce diamond. The filament is typically placed within 1 cm of the substrate surface to minimize thermalization and radical recombination, but radiation heating can produce excessive substrate temperatures leading to nonuniformity and even graphitic deposits. Withdrawing the filament slightly and biasing it negatively to pass an electron current to the substrate assists in preventing excessive radiation heating.

     (2) High Frequency Plasma-Assisted Chemical Vapor Deposition (PACVD). Plasma deposition involves the addition of a plasma discharge to the foregoing filament process. The plasma discharge increases the nucleation density and growth rate, and is believed to enhance diamond film formation as opposed to discrete diamond particles. There are three basic plasma systems in common use: a microwave plasma system, a radio frequency or RF (inductively or capacitively coupled) plasma system, and a direct current or DC plasma system. The RF and microwave plasma systems use relatively complex and expensive equipment which usually requires complex tuning or matching networks to electrically couple electrical energy to the generated plasma. The diamond growth rate offered by these two systems can be quite modest, on the order of ~1 micron/hour. Diamonds can also be grown in microwave discharges in a magnetic field, under conditions where electron cyclotron resonance is considerably modified by collisions. These “magneto-microwave” plasmas can have significantly higher densities and electron energies than isotropic plasmas and can be used to deposit diamond over large areas.

     (3) Oxyacetylene Flame-Assisted Chemical Vapor Deposition. Flame deposition of diamond occurs via direct deposit from acetylene as a hydrocarbon-rich oxyacetylene flame. In this technique, conducted at atmospheric pressure, a specific part of the flame (in which both atomic hydrogen (H) and carbon dimers (C2) are present [19]) is played on a substrate on which diamond grows at rates as high as >100 microns/hour [7].

(C) In the third or low pressure nonhydrogenic bulk method of producing diamond artificially [8-17], a nonhydrogenic fullerene (e.g., C60) vapor suspended in a noble gas stream or a vapor of mixed fullerenes (e.g., C60, C70) is passed into a microwave chamber, forming a plasma in the chamber and breaking down the fullerenes into smaller fragments including isolated carbon dimer radicals (C2) [6]. (Often a small amount of H2, e.g., ~1%, is added to the feedstock gas.) These fragments deposit onto a single-crystal silicon wafer substrate, forming a thickness of good-quality smooth nanocrystalline diamond (15 nm average grain size, range 10-30 nm crystallites [8-10]) or ultrananocrystalline diamond (UNCD) diamond films with intergranular boundaries free from graphitic contamination [9], even when examined by high resolution TEM [16] at atomic resolution [10]. Fullerenes are allotropes of carbon, containing no hydrogen, so diamonds produced from fullerene precursors are hydrogen-defect free [11] – indeed, the Ar/C60 film is close in both smoothness and hardness to a cleaved single crystal diamond sample [10]. The growth rate of diamond film is ~1.2 microns/hour, comparable to the deposition rate observed using 1% methane in hydrogen under similar system deposition conditions [9, 10]. Diamond films can, using this process, be grown at relatively low temperatures (<500 oC) [10] as opposed to conventional diamond growth processes which require substrate temperatures of 800-1000 oC.

Ab initio calculations indicate that C2 insertion into carbon-hydrogen bonds is energetically favorable with small activation barriers, and that C2 insertion into carbon-carbon bonds is also energetically favorable with low activation barriers [15]. A mechanism for growth on the diamond C(100) (2×1):H reconstructed surface with C2 has been proposed [16]. A C2 molecule impinges on the surface and inserts into a surface carbon-carbon dimer bond, after which the C2 then inserts into an adjacent carbon-carbon bond to form a new surface carbon dimer. By the same process, a second C2 molecule forms a new surface dimer on an adjacent row. Then a third C2 molecule inserts into the trough between the two new surface dimers, so that the three C2 molecules incorporated into the diamond surface form a new surface dimer row running perpendicular to the previous dimer row. This C2 growth mechanism requires no hydrogen abstraction reactions from the surface and in principle should proceed in the absence of gas phase atomic hydrogen.

The UNCD films were grown on silicon (Si) substrates polished with 100 nm diamond grit particles to enhance nucleation [16]. Deposition of UNCD on a sacrificial release layer of SiO2 substrate is very difficult because the nucleation density is 6 orders of magnitude smaller on SiO2 than on Si [18]. However, the carbon dimer growth species in the UNCD process can insert directly into either the Si or SiO2 surface, and the lack of atomic hydrogen in the UNCD fabrication process permits both a higher nucleation density and a higher renucleation rate than the conventional H2/CH4 plasma chemistry [18], so it is therefore possible to grow UNCD directly on SiO2.

Besides fullerenes, it has been proposed that “diamondoids” or polymantanes, small hydrocarbons made of one or more fused cages of adamantane (C10H16, the smallest unit cell of hydrogen-terminated crystalline diamond) could be used as the carbon source in nonhydrogenic diamond CVD [20-22]. Dahl, Carlson and Liu [22] suggest that the injection of diamondoids could facilitate growth of CVD-grown diamond film by allowing carbon atoms to be deposited at a rate of about 10-100 or more at a time, unlike conventional plasma CVD in which carbons are added to the growing film one atom at a time, possibly increasing diamond growth rates by an order of magnitude or better. However, Plaisted and Sinnott [23] used atomistic simulations to study thin-film growth via the deposition of very hot (119-204 eV/molecule; 13-17 km/sec) beams of adamantane molecules on hydrogen-terminated diamond (111) surfaces, with forces on the atoms in the simulations calculated using a many-body reactive empirical potential for hydrocarbons. During the deposition process the adamantane molecules react with one another and the surface to form hydrocarbon thin films that are primarily polymeric with the amount of adhesion depending strongly on incident energy. Despite the fact that the carbon atoms in the adamantane molecules are fully sp3 hybridized, the films contain primarily sp2 hybridized carbon with the percentage of sp2 hybridization increasing as the incident velocity goes up. However, cooler beams might allow more consistent sp3 diamond deposition, and other techniques [24] have deposited diamond-like carbon (DLC) films with a higher percentage of sp3 hybridization from adamantane.

1.2 Diamond Manufacturing via Positional Diamond Mechanosynthesis

A new non-bulk non-statistical method of manufacturing diamond, called positional diamond mechanosynthesis, was proposed theoretically by Drexler in 1992 [32]. In this method, positionally controlled carbon deposition tools are manipulated to sub-Angstrom tolerances via SPM (Scanning Probe Microscopy) or similar atomic-resolution manipulator mechanisms to build diamond in vacuo. Each carbon deposition tool includes a tooltip molecule attached to a larger handle structure which is grasped by the atomic-resolution manipulator mechanism. One or more carbon atoms having one or more dangling bonds are relatively loosely bound to the tip of the tooltip molecule. When the tip is brought into contact with the substrate surface at a specific location and sufficient mechanical forces (compression, torsion, etc.) are applied, a stronger covalent bond is formed between the tip-bound carbon atom(s) and the surface, via the dangling bonds, than previously existed between the tip-bound carbon atom(s) and the tooltip structure. As a result, the tool may subsequently be retracted from the substrate and the tip-bound carbon atom(s) will be left behind on the substrate surface at the specific location and orientation desired. By repeating this process of positionally-constrained chemistry or mechanosynthesis, using a succession of similar tools, a large variety of molecularly precise diamond structures can be fabricated, placing one or a few atoms at a time on the growing workpiece.

Several analyses using the increasingly accurate methods of computational chemistry have confirmed the theoretical validity of the proposed process of positional diamond mechanosynthesis for hydrogen abstraction [25-33] and hydrogen donation [32, 33], in respect to the surface passivating hydrogen atoms, and carbon deposition [32-38], in respect to diamond surfaces and the body of diamond nanostructures. While positional diamond mechanosynthesis has not yet been demonstrated experimentally, early experiments [39] have demonstrated single-molecule positional covalent bond formation on surfaces via SPM, though in these cases bond formation was not purely mechanochemical but included electrochemical or other means. Mechanosynthesis of the Si(111) lattice has been studied theoretically [40, 41] and the first laboratory demonstration of nonelectrical, purely mechanical positional covalent bond formation on a silicon surface using a simple SPM tip was reported in 2003 [42]. In this demonstration, Osaka University researchers lowered a silicon AFM tip toward the silicon Si(111)-(7×7) surface and pushed down on a single atom. The focused pressure forced the atom free of its bonds to neighboring atoms, which allowed it to bind to the AFM tip. After lifting the tip and imaging the material, there was a hole where the atom had been (Figure 1). Pressing the tip back into the vacancy redeposited the tip-bound selected single atom, this time using the pressure to break the bond with the tip. These manipulation processes were purely mechanical since neither bias voltage nor voltage pulse was applied between probe and sample [42].

Figure 1. Mechanosynthesis of a single silicon atom on the silicon Si(111)-(7×7) surface

Phys. Rev. Lett. 90, 176102 (2003)

Existing mechanosynthetic tools can only be used at ultralow temperatures near absolute zero, and hold the atom or molecule to be deposited only very weakly, and can be employed only very slowly (minutes or hours per mechanosynthetic operation). These tools include the simple diamond stylus [43] and other crude tools such as nanocrystalline diamond grown (a) on standard silicon [44, 48] AFM tips with a 30 nm radius [48], (b) on silicon cantilever tips [46, 47], (c) on tungsten STM tips [45], or (d) on 12 nm radius doped-diamond STM tips [49], using CVD [44-49] including HFCVD [44, 46] or PACVD [45] diamond deposition processes. There is a need for improved mechanosynthetic tools with a molecularly precise <0.3 nm tip radius that can operate at liquid nitrogen or even room temperatures, and can perform mechanosynthetic operations in seconds or even faster cycle times, and can conveniently be precisely manipulated to sub-Angstrom positional accuracy using conventional SPM instruments.

In 2002, Merkle and Freitas [36] proposed the first design for a class of precision tooltip molecules intended to positionally deposit individual carbon dimers on a growing diamond substrate via diamond mechanosynthesis (Figure 2), and subsequent theoretical analysis [37, 38, 235] has verified that this class of tooltip molecules should be useful for depositing carbon dimers on a dehydrogenated diamond C(110) crystal surface, for the purpose of building additional C(110) surface or other molecularly precise structures at liquid nitrogen or room temperatures.

Figure 2. DCB6-Si dimer placement tooltip molecule [36]

(A) Wire frame view of tooltip molecule

(B) Overlapping spheres view of (A)

(C) Iceane

No specific proposals for attaching tooltip molecules such as the one illustrated in Figure 2 A/B to larger tool handles, or complete tools for positional diamond mechanosynthesis, have previously been reported in the scientific, engineering or patent literature. While others have previously noted the need for a handle structure to manipulate the active mechanosynthetic tooltip [32, 33, 36, 38], this invention is the first practical description of how to manufacture and to attach tooltips to such a handle structure, and thus to manufacture a complete mechanosynthetic tool.

The present invention is not limited to a method for the manufacture of a complete tool which can be used for diamond mechanosynthesis. The same toolbuilding process is readily extended to other classes of tooltip molecules, handle materials, and mechanosynthetic processes and structures other than diamond. As examples, which in no way limit or exhaust the possible applications of this invention, the same method as described herein can be used to build complete mechanosynthetic tools and attach handles to: (1) other possible C2 dimer deposition tooltips proposed by Drexler [32] and Merkle [33, 34] for the building of molecularly precise diamond structures; (2) other possible carbon deposition tooltips, including but not limited to carbene tooltips as proposed by Drexler [32] and Merkle [33, 34] and monoradical methylene tooltips as proposed by Freitas [234], for the deposition of carbon or hydrocarbon moieties during the building of molecularly precise diamond structures, or other tooltips that may be used for the removal of individual carbon atoms, C2 dimers [38], or other hydrocarbon moieties from a growing diamond surface; (3) tooltips for the abstraction [25-33] and donation [32, 33] of hydrogen atoms, for the purpose of positional surface passivation or depassivation during the building of molecularly precise diamond structures, or during the building of molecularly precise structures other than diamond, or of other atoms similarly employed for passivation purposes; or (4) tooltips for the deposition or abstraction of atoms, dimers, or other moieties, to or from materials including, but not limited to, covalent solids other than diamond, silicon, germanium or other semiconductors, intermetallics, ceramics, and metals.

2. Description of the Invention

The present invention is concerned with the physical structure and method of manufacture of a complete tool for positional diamond mechanosynthesis, which can subsequently be employed in the mechanosynthetic manufacture of other molecularly precise diamond structures, including other tools for positional diamond mechanosynthesis.

The present invention is the first description of a complete tool for positional diamond mechanosynthesis, along with its method of manufacture. The subject mechanosynthetic tool is constructed using only bulk chemical and mechanical processes, and yet, once fabricated, is capable of molecularly precise carbon dimer deposition to produce molecularly precise diamond structures. The present invention provides a tool by which the trajectory and timing of each new carbon atom added to a growing diamond nanostructure can be precisely controlled, thus allowing the manufacture of molecularly precise three-dimensional diamond structures of specified size, shape, orientation, location, and chemical composition, a significant improvement over all known bulk methods for fabricating synthetic diamond and a significant improvement over all existing mechanosynthetic SPM tips or styluses.

The positional diamond mechanosynthesis tool described herein enables the convenient manufacture of large numbers and varieties of diamond mechanosynthesis tools of similar or improved types, and also enables the convenient manufacture of a wide variety of molecularly precise nanoscale, microscale, and other diamond structures that cannot be fabricated by any known bulk process, including, but not limited to, molecularly-sharp scanning probe tips, shaped nanopores and custom binding sites, complex nanosensors, interleaved nanomechanical structures, compact mechanical nanocomputer components, nanoelectronic and quantum computational devices, aperiodically nanostructured optical materials, and many other complex nanodevices, nanomachines, and nanorobots. The tool can also be used in the fabrication of additional tools for the positional mechanosynthetic manufacture of molecularly precise structures made of materials other than diamond, employing either carbon (e.g., nanotubes and other graphene sheet structures) or carbon together with elements other than carbon, such as nanostructured nondiamond hydrocarbons, nanostructured fluorocarbons, nanostructured sapphire/alumina, and even DNA and other organic polymeric materials.

The positional diamond mechanosynthesis tool consists of two distinct parts which are covalently joined.

The first part of the positional diamond mechanosynthesis tool is the tooltip molecule (Figure 2). In the preferred embodiment the tooltip molecule consists of one or more adamantane molecules arranged in a polymantane or lonsdaleite (iceane; Figure 2C) configuration making a triadamantane base molecule. One or more dimerholder atoms (most preferably the Group IV elements Si, Ge, Sn, and Pb with three bonds into the base, but Group V elements N, P, As, Sb and Bi and Group III elements B, Al, Ga, In, and Tl with two bonds into the base may also be used [36]) are substituted into each of the adamantane molecules composing the triadamantane base molecule. A single carbon dimer (C2) molecule is bonded to two dimerholder atoms integral to the triadamantane base molecule; the carbon dimer is held by the tooltip but is later mechanically released during a mechanosynthetic dimer placement operation. Finally, a capping group is temporarily bonded to the two dangling bonds of the carbon dimer, passivating the dangling bonds and chemically stabilizing the tooltip molecule for a solution-phase environment. The capping group must be removed from the tooltip, exposing the dimer dangling bonds and activating the tooltip molecule, prior to use in a diamond mechanosynthesis operation.

The second part of the positional diamond mechanosynthesis tool is the handle structure (e.g., Figure 17). The handle structure may be a large rigid molecule, consisting in the preferred embodiment of a regular crystal, or a rod, or a cone, of pure hydrogen-terminated diamond, thus providing the greatest possible mechanical rigidity and thermal stability. At the base of the handle, the handle structure is sufficiently wide (0.1-10 microns in diameter) to be securely grasped by, or bonded to, a conventional SPM tip, a MEMS robotic end-effector, or other similarly rigid and well-controlled microscale manipulator device. Near the apex of the handle structure, the tooltip molecule is covalently bonded to the handle structure, forming an intimate and permanent connection thereto. The tooltip molecule is oriented coaxially with the handle structure, with the carbon dimer (whether capped or uncapped) of the tooltip molecule occupying the location most distal from the base of the handle structure, just as the writing tip of a sharpened pencil is most distal from the pencil eraser end.

The manufacture of the complete positional diamond mechanosynthesis tool requires four distinct steps, including (1) synthesis of capped tooltip molecule (Section 2.1), (2) attachment of tooltip molecule to deposition surface in a preferred orientation (Section 2.2), (3) attaching handle structures onto the tooltip molecules (Section 2.3), and finally (4) separating the finished tools from the deposition surface (Section 2.4). The concept of seeded growth of a useful nanoscale tool has previously been employed in the CVD growth of carbon nanotube tips for AFM [50-52].

2.1 STEP 1: Synthesis of Capped Tooltip Molecule

STEP 1. Synthesize the triadamantane tooltip molecule, with its active C2 dimer tip appropriately capped, using methods of bulk chemical synthesis derived from known synthesis pathways for functionalized polyadamantanes as found in the existing chemical literature.

While an explicit synthesis of the exact DCB6-X (X = Si, Ge, Sn, Pb) capped tooltip molecule has not yet been located in the chemical literature, the sila-adamantanes have been investigated since at least the early 1970s [53-55] and multiply-substituted adamantanes such as 1,3,5,7-tetramethyl-tetrasilaadamantane [53, 56] and other 1,3,5,7-tetrasilaadamantanes [57] have been synthesized. Adamantanes are readily functionalized with alkene C=C bonds, e.g., 2,2-divinyladamantane, a colorless liquid at room temperature [161]. Polymantanes as a class of molecules can be functionalized [58, 60] and assembled to a limited extent, including biadamantanes [63], diadamantanes [64-66] and diamantanes [67], triamantanes [68, 69], and tetramantanes [70, 71]. The Beilstein database lists over 20,000 adamantane variants and there are several excellent literature reviews of adamantane chemistry [59-63]. The molecular geometries of diamantane, triamantane, and isotetramantane have been investigated theoretically using molecular mechanics, semiempirical and ab initio approaches [72]. The core of the DCB6-X (X = Si, Ge, Sn, Pb) class of adamantane-based tooltip molecules is a single iceane molecule (Figure 2C), the smallest unit cage of lonsdaleite or hexagonal diamond (the counterpart to adamantane which is the unit cage for the more common cubic diamond lattice). The iceane molecule was first synthesized experimentally in 1974 [73-75] and more recently has been studied using the customary methods of computational chemistry [77-80]; commercial sources for hexagonal diamond (lonsdaleite) powder already exist [76].

A crucial decision to be made in a particular application of this invention is the choice of capping group to be used to passivate the two dangling bonds of the C2 dimer that is held by the tooltip molecule. The presence of the capping group converts the otherwise highly reactive C2 dimer radical into a chemically stable moiety in solution phase for the duration of the synthesis process. Only when the capping group is later removed (Section 2.2), in vacuo, does the C2 dimer resume its status as a chemically active radical. Note that for some choices of capping group it may be simpler to synthesize the capped tooltip molecule in the configuration of a double-capped single-bonded C-C dimer, then employ a subsequent process to alkenate the dimer bond to C=C which would include removing half of the capping groups.

Many possible capping groups could in principle provide electronic closed-shell termination of the C2 dangling bonds, thus maximizing tooltip molecule chemical stability during conventional solution synthesis in Step 1 and during tooltip molecule attachment in Step 2 (Section 2.2). In some procedures, attachment is facilitated if the chemical structure of the capping group is highly dissimilar to the adamantane structure of the tooltip molecule, so that the capping group may be conveniently removed, e.g., by selective bond resonance excitation, during the tooltip attachment process. (Thus purely hydrocarbon-based and some other organic radicals may be problematic as capping groups.) For simplicity of analysis, ease of tooltip molecule synthesis, and ease of capping group removal, the capping group should have as few atoms as possible, all else equal. An enumeration of 400 potentially useful capping groups fulfilling the above requirements is given in Table 1, though the present invention is not limited to this partial list of illustrative exemplar moieties. As the number of atoms in the capping group increases, the combinatoric possibilities expand enormously. Some of the groups listed in Table 1 may yield tooltip molecules that are stable only at very low temperatures or only in particular chemical environments, and a few may not yet have been verified as experimentally available or even chemically stable.

The precise choice of capping group is determined by the desired interactions of tooltip molecules with the selected deposition surface (as described in Step 2 (Section 2.2) and Step 4 (Section 2.4)), but also by the desired interactions of tooltip molecules with themselves, e.g., during synthesis. There are at least four relevant factors which must be considered.

First, from the standpoint of basic utility the ideal capping group: (1) should be loosely bound to the dimer, thus easily released in order to uncap (and activate) the tooltip; (2) should form only a single bond with carbon; and (3) should be very simple, hence relatively easy to synthesize in a polymantane system. A few capping atoms that meet these criteria are given in Table 2.

For ease of release alone, Table 2 implies that a preferred embodiment is to use two iodine atoms as the C2 dimer capping group of the tooltip molecule, as shown in Figure 3 below, right, though other capping groups may also serve in this capacity.

Figure 3. DCB6-Ge tooltip molecule, uncapped (left), and capped (right) with iodine atoms

(A) uncapped

(B)) capped with iodine atoms

Second, during bulk chemical synthesis using conventional techniques in solution phase, the capped tooltip molecule should not spontaneously dimerize across the C2 working tips. Dimerization can occur between two tooltip molecules across one bond or two bonds, as shown in Figure 4. Table 3 shows the results of geometry optimization energy minimization calculations using semi-empirical AM1 for the DCB6-Ge capped tooltip molecule [235] in various stages of “tip-on-tip” dimerization, for a variety of capping groups, in vacuo.

With no protective capping group in place, tip-to-tip dimerization is very energetically favorable. Tooltip molecule dimerization is energetically unfavorable to varying degrees for 1-atom capping groups consisting of, for example, -I, -Cl, -F, -Na, and -Li, and also for several 2-atom capping groups including hydroxyl (-OH), amine (-NH2), oxylithyl (-OLi), oxyiodinyl (-OI), and sulfiodinyl (-SI). In the case of some 2-atom oxyl (-OF), sulfyl (-SS-, -SH, -SF), and selenyl (-SeH) capping groups, dimerization is energetically unfavorable for direct =C-C= bonds linking the two tooltip molecules but appears likely to occur if dimerization occurs through an oxygen, sulfur (e.g., =C-S-C= or =C-S-S-C=) or selenium atom in the dimerization bond(s) linking the two tooltip molecules. Single-bond dimerization of an H-capped tooltip molecule with release of H2 is also energetically favorable, though double-bond dimerization for H-capped tooltips with the release of 2H2 appears unfavorable.

These analyses should be repeated using ab initio techniques, and should be extended to include a calculation of activation energy barriers (which could be substantial), weak ionic forces that could lead to crystallization (in the case of capping groups containing metal or semi-metal atoms), and solvent effects, all of which could affect the results. As a limited example of one such study, Mann et al [38] found that the dimerization reaction enthalpies of uncapped DCB6-Si and DCB6-Ge tooltip molecules are -1.64 eV and -1.84 eV, but that the energy barriers to the dimerization reaction were 1.93 eV and 1.86 eV, respectively. Therefore the dimerization of uncapped DCB6-Si and DCB6-Ge tooltip molecules “is thermodynamically favored but not kinetically favored. Due to the electron correlation errors in DFT these barrier heights may be considerably overestimated, therefore both reactions may be kinetically accessible at room temperature.” Subsequent work [235] appears to have confirmed that both tooltips work well as expected on the diamond C(110) surface, with the DCB6-Ge structure emerging as the preferred dimer placement tooltip molecule [235].

Figure 4. Progressive stages of possible “tip-on-tip” dimerization of capped tooltip molecules

(A) undimerized

(B) dimerized (1-bond)

(C) dimerized (2-bond)

In the case of bromine, and to a lesser extent in several other cases, the undimerized and 1-bond dimerized forms appear energetically almost equivalent, although 2-bond dimerization is energetically unlikely. Application of the process described in Step 2 using a capping group having this characteristic could result in a mixture of undimerized and 1-bond dimerized tooltips attached to the deposition surface. In the event that some 1-bond dimerizations occur and that a few dimerized tooltip molecules are subsequently inserted into the deposition surface during Step 2, the distinctive two-lobed geometric signature of these dimerized nucleation seeds can be detected and mapped via SPM scan prior to Step 3, and subsequently avoided during tool detachment in Step 4. Surface editing is another approach. Due to the low surface nucleation density (Section 2.2.1), after the aforementioned mapping procedure it may be possible to selectively detach and remove from the surface all attached dimerized tooltip molecules that are detected, e.g., using focused ion beam, electron beam, or NSOM photoionization, subtractively editing the deposition surface prior to commencing CVD in Step 3. An alternative to subtractive editing is additive editing, wherein FIB deposition of new substrate atoms on and around the dimerized tooltip molecule can effectively bury it under a smooth mound of fresh substrate, again preventing nucleation of diamond at that site during Step 3.

Third, the capped-C2 tip of the capped tooltip molecule should not spontaneously recombine into the side or the bottom of the adamantane base of neighboring tooltip molecules, during synthesis or storage, as illustrated in Figure 5 for a side-bonding event. Recombination can occur between two tooltip molecules across one bond or two bonds. Table 4 shows the results of semi-empirical energy calculations using AM1 for the DCB6-Ge capped tooltip molecule in two particular cases of “tip-on-base” side-bonding recombination, for a variety of capping groups, in vacuo.

With no protective capping group, tip-on-base recombination is very energetically preferred, with 1-bond recombination preferred over 2-bond when the H atom released from the adamantane base during formation of the 1-bond link becomes bonded with the remaining dangling bond of the tip-held C2 dimer. Mann et al [38] showed that intermolecular dehydrogenation from the bottom of the adamantane base by a neighboring uncapped tooltip molecule is exothermic and kinetically accessible (against a 0.48 eV reaction energy barrier) at room temperature. However, with an appropriate cap in place, tooltip molecule recombination is energetically unfavorable to varying degrees, e.g., for 1-atom capping groups consisting of -I, -Br, -Na, and -Li, and also for several 2-atom capping groups including hydroxyl (-OH), amine (-NH2), oxylithyl (-OLi), seleniodinyl (-SeI), several sulfyl groups including sulfhydryl (-SH), sulfiodinyl (-SI), and sulfalithyl (-SLi), and dimagnesyl (-MgMg-). There may be some tip-to-tip ionic bonding for beryllium (-Be-), lithium, oxylithyl, seleniodinyl, selenobromyl (-SeBr), berylfluoryl (-BeF) and berylchloryl (-BeCl) capping groups, and the imide (-NH-) cap appears to twist the tooltip dimer out of horizontal alignment. In the case of some 2-atom sulfyl (-SF, -SBr), and selenyl (-SeH) capping groups, recombination is energetically unfavorable for direct =C-C= bonds linking the two tooltip molecules but appears likely to occur if recombination occurs through a sulfur (e.g., =C-S-C= or =C-S-S-C=) or selenium atom in the recombination bond(s) linking the two tooltip molecules. Single-bond recombination of an H-capped tooltip molecule with release of H2 is slightly energetically favorable, though double-bond dimerization for H-capped tooltips with release of 2H2 appears very unfavorable energetically. These analyses should be repeated using ab initio techniques, and should be extended to include a calculation of activation energy barriers (which could be substantial), weak ionic forces that could lead to crystallization (in the case of capping groups containing metal atoms), and solvent effects, all of which could affect the results.

Figure 5. Progressive stages of possible “tip-on-base” recombination of capped tooltip molecules

(A) unrecombined

(B) 1-bond recombination

(C) 2-bond recombination

In the case of chlorine, and to a lesser extent in several other cases, the unrecombined and 1-bond recombined forms appear energetically almost equivalent, although 2-bond recombination is energetically unlikely. Application of the process described in Step 2 using a capping group having this characteristic could result in a mixture of unrecombined and 1-bond recombined tooltips attached to the deposition surface. In the event that some 1-bond recombinations occur and that a few recombined tooltip molecules are subsequently inserted into the deposition surface during Step 2, the distinctive two-lobed geometric signature of these recombined nucleation seeds can be detected and mapped via SPM scan prior to Step 3, and subsequently avoided during tool detachment in Step 4. Surface editing is another approach. Due to the low surface nucleation density (Section 2.2.1), after the aforementioned mapping procedure it may be possible to selectively detach and remove from the surface all attached recombined tooltip molecules that are detected, e.g., using focused ion beam, electron beam, or NSOM photoionization, subtractively editing the deposition surface prior to commencing CVD in Step 3. An alternative to subtractive editing is additive editing, wherein FIB deposition of new substrate atoms on and around the recombined tooltip molecule can effectively bury it under a smooth mound of fresh substrate, again preventing nucleation of diamond at that site during Step 3.

Fourth, the capped-C2 tip of the capped tooltip molecule should not spontaneously react with solvent, feedstock, or catalyst molecules that are employed during conventional techniques for the bulk chemical synthesis of functionalized adamantanes in solution phase. A definitive result regarding this capping-group selection factor depends critically upon the exact synthesis pathways required.

As a proxy for these many pathways, it has been shown that even straight-chain hydrocarbons, upon exposure to the customary aluminum halide catalysts at high temperature, readily produce mixtures of various polymethyladamantanes [81]. The simplest-case recombination event illustrated in Figure 6 was analyzed via semi-empirical energy calculations using AM1 for the DCB6-Ge iodine-capped tooltip molecule in the specific instances of 1-bond and 2-bond side-bonding recombination with a simple straight-chain hydrocarbon molecule (n-octane). The 2-bond analysis includes one event in which the second bond occurs adjacent to the first, producing a 4-carbon ring with the octane molecule, and a second alternative event in which the second bond occurs with an octane chain carbon atom three positions down the chain, producing a more stable 6-carbon ring with the octane molecule. Since solvent effects, temperature, reverse reaction rates, and so forth will determine whether the reaction can occur, and will also determine the relative yields of various products and reactants, the thermodynamics results indicate primarily the relative ease or difficulty of maintaining the given capped tooltip molecule stably in solution with liquid n-octane. The data in Table 5 show that iodine (-I), hydrogen (-H), amine (-NH2), and perhaps bromine (-Br) capped tooltip molecules should be the most stable in hydrocarbon media, as should seleniodinyl (-SeI) and several sulfyl-capped molecules including sulfhydryl (-SH), sulfiodinyl (-SI), and sulfobromyl (-SBr). Fluorine- and oxygen-containing capping groups may be (relatively) less stable.

Figure 6. Progressive stages of possible side-bonding recombination reaction between an iodine-capped DCB6-Ge tooltip molecule (above) and a molecule of n-octane (below)

	->
(A) unrecombined		(B) 1-bond recombination
->
	(C) 2-bond recombination (4-carbon ring)	(D) 2-bond recombination (6-carbon ring)

2.2 STEP 2: Attach Tooltip Molecule to Deposition Surface in Preferred Orientation

STEP 2. Attach a small number of tooltip molecules to an appropriate deposition surface in tip-down orientation, so that the tooltip-bound dimer is bonded to the deposition surface.

The appropriate deposition surface material (Section 2.2.1) is determined by choosing a surface which is not readily amenable to bulk diamond deposition, under the thermal and chemical conditions that will prevail during the diamond deposition processes described in Step 3. In Attachment Method A (Section 2.2.2), tooltip molecules may be bonded to the deposition surface in the desired orientation via low-energy ion bombardment of the deposition surface in vacuo, creating a low density of preferred diamond nucleation sites. In Attachment Method B (Section 2.2.3), tooltip molecules may be bonded to the deposition surface in the desired orientation by non-impact dispersal and weak physisorption on the deposition surface, followed by tooltip molecule decapping via targeted energy input producing dangling bonds at the C2 dimer which can then bond into the deposition surface in vacuo, also creating a low density of preferred diamond nucleation sites. In Attachment Method C (Section 2.2.4), the techniques of conventional solution-phase chemical synthesis are used to attach tooltip molecules to a deposition surface in the preferred orientation, again creating diamond nucleation sites.

2.2.1 Surface Nucleation and Choice of Deposition Substrate

The intention of this invention is to grow a handle molecule as a single crystal of bulk diamond large enough to permit convenient physical manipulation of the attached C2 dimer-bearing tooltip. Since this single crystal will be in the size range of 0.1-10 microns, and since sufficient room must be allowed around each single crystal to afford access to a MEMS-scale gripping mechanism, the maximum surface nucleation density appropriate for this process in the preferred embodiment will be ~105 cm-2, giving a mean separation between handle molecule crystals of ~32 microns on the deposition surface. In other embodiments in which much smaller 100 nm handle molecule crystals can be employed with narrower attachment clearances for the external gripping mechanism, the maximum surface nucleation density could be as high as ~109 cm-2, giving a mean separation between surface-grown handle molecule crystals of ~320 nm.

Conventional diamond films grown by CVD on smooth nondiamond substrates are characterized by very low nucleation densities, typically <104 cm-2 when diamond is deposited on a polished silicon wafer surface, which is many orders of magnitude less than that exhibited by most materials [127]. (Interestingly, the CVD nucleation density of diamond nanocrystals on an SiO2 substrate is 6 orders of magnitude smaller than on pure silicon [18].) The commercial preparation of continuous diamond films requires separately nucleated diamond crystals eventually to grow together to form a single sheet, hence is maximally efficient under conditions of high nucleation density. Therefore diamond film growth procedures often include preliminary substrate preparation techniques which attempt to increase the nucleation density to a practicable level. Such techniques typically involve introduction of surface discontinuities by scratching or abrading the substrate surface with a fine diamond grit powder or paste. Such surface discontinuities either create preferential geometrical sites for diamond crystal nucleation, or more probably embedded residues from the diamond abrading powder may serve as nucleation sites from which diamond growth can occur by accumulation. The presence of carbon particles on the surface of a substrate can provide a high density of nucleation sites for subsequent diamond growth [82]. As shown in Table 6, despite abrasive surface preparation the nucleation densities for diamond films prepared by such techniques remain relatively low, on the order of ~108 cm-2 (~1 µm-2) (vs. ~1015 cm-2 available atomic sites), and the surface structure of such films is unpredictable and typically exhibits very disordered surface patterns [127]. Nucleation has also been enhanced by coating substrate surfaces with a thin (10-20 nm) layer of hydrocarbon oil [83].

Since the purpose of this invention is to grow isolated micron-scale diamond single crystals over tooltip molecule nucleation sites, rather than a continuous diamond film, the deposition surface ideally is chosen so as to minimize the number of natural (non-tooltip molecule) nucleation sites. If tooltip molecules are attached at a number density of ~105 cm-2 to a surface of polished silicon otherwise having no pretreatment, the number density of naturally occurring nucleation sites can be held to at most 103-105 cm-2. This implies that from 50% to 99% of the isolated micron-scale diamond single crystals that are grown during Step 3 (Section 2.3) will be correctly nucleated by surface-bound undimerized tooltip molecules. An SPM scan of the deposition surface, following the completion of Step 2 but prior to the commencement of Step 3, can identify and map the positions of all of the undimerized surface-bound tooltip molecules, so that the isolated micron-scale diamond single crystals that are later grown and properly nucleated by surface-bound tooltip molecules can be identified prior to selection and detachment in Step 4 (Section 2.4).

As noted by May [85], most of the CVD diamond films reported to date have been grown on single-crystal Si wafers, mainly due to the availability, low cost, and favorable properties of Si wafers. But this is not the only possible substrate material. Candidate substrates for diamond handle molecule crystal growth must satisfy five important basic criteria [85], the first four of which are summarized quantitatively in Table 7.

First, the substrate must have a melting point (at the process pressure) higher than the temperature required for diamond growth (at least 300-500 oC, but normally greater than 700 oC). This precludes the use of low-melting-point materials such as plastics, aluminum, certain glasses and some electronic materials such as GaAs as a deposition substrate, when hydrogenic diamond CVD techniques are employed in Step 3 (Section 2.3).

Second, for growing diamond films the substrate material should have a thermal expansion coefficient comparable with that of diamond, since at the high growth temperatures currently used, a substrate will tend to expand, and thus the diamond coating will be grown upon and bonded directly to an expanded substrate. Upon cooling, the substrate will contract back to its room temperature size, whereas the diamond coating, with its very small expansion coefficient, will be relatively unaffected by the temperature change, causing the diamond film to experience significant compressive stresses from the shrinking substrate, leading to bowing of the sample, and/or cracking, flaking or even delamination of the entire film [85]. However, a nondiamond deposition surface for growing diamond tool handle molecules, starting from surface-bound tooltip molecule nuclei, should incorporate the maximum possible thermal expansion mismatch between the substrate and diamond, producing thermal stresses upon cooling that can facilitate tool separation from the nondiamond deposition surface in Step 4 (Section 2.4).

Third, a mismatch in the crystal lattice constant [86, 87] between the diamond comprising the tool handle molecule and the nondiamond substrate greatly reduces the bonding opportunities between handle molecule and substrate, during handle molecule growth (Section 2.3). An extensive interfacial misfit also facilitates tool separation from the nondiamond deposition surface in Step 4 (Section 2.4).

Fourth, in order to form adherent diamond films it is a customary requirement that the substrate material should be capable of forming a carbide layer to a certain extent, since diamond CVD on nondiamond substrates usually involves the formation of a thin carbide interfacial layer upon which the diamond then grows. The carbide layer is viewed as a “glue” which promotes diamond growth and aids its adhesion by (partial) relief of interfacial stresses caused by lattice mismatch and substrate contraction [85]. However, the ideal nondiamond deposition surface for growing diamond tool handle molecules, starting from surface-bound tooltip molecule nuclei, is a substrate that resists or prohibits carbide formation. The absence of carbide on the nondiamond deposition surface (a) discourages downgrowth of the tool handle molecule into the substrate, (b) helps maintain the isolation of the finished tooltip apex, and (c) facilitates tool separation from the nondiamond deposition surface in Step 4 (Section 2.4). On the basis of carbide exclusion, potential substrate materials including metals, alloys and pure elements can be subdivided into three broad classes [85, 88], in descending order of preference for the present invention:

     (1) Carbide Exclusion. Metals such as Cu, Sn, Pb, Ag and Au, as well as non-metals such as Ge and sapphire/alumina (Al2O3), have little or no solubility or reaction with C. These materials do not form a carbide layer, and so any diamond layer that might try to form will not adhere well to the surface (which is known as a way to make free-standing diamond films, as the films will often readily delaminate after deposition). These are the best materials for a deposition surface upon which to grow detachable diamond tool handle molecules nucleated by surface-bound tooltip molecules. Unwanted natural nucleation centers are unlikely to arise on polished non-pretreated surfaces and downgrowth from the tooltip molecule seed or the growing tool handle structure, towards the substrate, will be resisted by these surfaces.

     (2) Carbon Solvation. Metals such as Pt, Pd, Rh, Ni, Ti and Fe exhibit substantial mutual solubility or reaction with C (all industrially important ferrous materials such as iron and stainless steel cannot be diamond coated using simple CVD methods) [85]. During CVD, a substrate composed of these metals acts as a carbon sink whereupon deposited carbon dissolves into the surface, forming a solid solution. This dissolution transports large quantities of C into the bulk, rather than remaining at the surface where it can promote diamond nucleation [85]. Often diamond growth on the surface only begins after the substrate is completely saturated with carbon, with carbide finally appearing on the surface, by which time the tool handle molecule may already have grown sufficiently large as a single diamond crystal atop a surface-bound tooltip molecule.

     (3) Carbide Formation. Metals such as Ti, Zr, Hf, V, Nb, Ta, Cr, Mo, W, Co, Ni, Fe, Y, Al, and certain other rare-earth metals can form carbide during CVD. In some metals, such as Ti, the interfacial carbide layer continues growing during diamond deposition and can become hundreds of microns thick. Non-metals such as B and Si, and Si-containing compounds such as SiO2, quartz and Si3N4, also form carbide layers, and substrates composed of carbides themselves, such as SiC, WC and TiC, are particularly amenable to diamond deposition [85]. Surface nucleation rates (cm-2 hr-1) on stable carbide-forming substrates (Si, Mo, W) are 10-100 times higher than on carbide-resistant substrates [89], and surface nucleation density (cm-2) on Mo is about 10 times higher than on other carbide-forming substrates (Si, Ni, Ti, Al) under similar deposition conditions [90]. If used as polished non-pretreated deposition surfaces for diamond tool handle growth, these materials should only sparsely produce competing diamond crystal nucleation centers during hydrogenic CVD processes. (Diamond cannot be epitaxially grown directly on silicon or GaAs substrates [91].) However, carbon dimers that are present in the feedstock gases during low-temperature nonhydrogenic CVD can insert into Si and SiO2 surfaces, readily producing silicon carbide [18]. Additionally, as the CVD process continues, carbide-forming materials may permit some unwanted downgrowth from the surface-bound tooltip molecule or growing tool handle structure, towards the substrate. Note that bombardment of surfaces, particularly refractory metal surfaces such as tungsten, with fullerene ions having energies from about 0.0025-250 MeV results in implantation of carbon and the formation of surface or subsurface carbides [11].

Dimer Release Criterion. In addition to these four basic factors, a fifth criterion in the choice of deposition substrate material is that the tooltip molecule should bind the C2 dimer more strongly than the deposition surface, so that when the finished tool is pulled away from the deposition surface in Step 4 (Section 2.4), the dimer will stay attached to the tool and not remain on the deposition surface. If the dimer stays with the tool, then the result is a tool with an active tip ready to perform diamond mechanosynthesis. If the dimer remains on the deposition surface, the result is a dimerless “discharged” tool which must be recharged with C2 dimer by some additional process [38] before the tool can be used for diamond mechanosynthesis.

A full computational simulation of the interaction between complete modeled deposition surfaces and the DCB6-Ge tooltip has not yet been done. However, a preliminary evaluation has examined the energy minima of a tooltip that is first joined to a deposition surface through the dimer (EJ) and is then pulled away from the deposition surface, for Dimer-on-Tooltip (EDoT) and Dimer-on-Surface (EDoS) configurations, where the “surfaces” are crudely modeled as follows: C (diamond), Si, Ge, Sn, and Pb surface as a single nonterminated 10-atom adamantane-like cage, with the tooltip dimer bonded to 2 adjacent cage atoms; Cu surface as 4 Cu atoms arranged in a square, with the tooltip dimer bonded to 2 adjacent Cu atoms; Al2O3 as a single 5-atom chain of alternating Al and O atoms, with the tooltip dimer bonded to the two Al atoms; and C (graphite) as a 3×3 (unit cells) flat single-plane sheet with all perimeter C atoms immobilized. The quantity (EDoS – EDoT), tabulated in the rightmost column of Table 8 for each surface, is negative if the dimer prefers to stick to the surface after the tooltip has been pulled away from the surface, and is positive if the dimer prefers to stick to the tooltip after the tooltip has been pulled away from the surface, the desired result. (This is only a crude analysis because the quantity (EDoS – EDoT) really informs us only as to whether the total process of charged tooltip deposition plus discharged tooltip retraction is endo- or exothermic, not the reaction direction or preference.) Since surfaces composed of the larger-radius Ag and Au atoms should bind the dimer less strongly than Cu, it appears that all “carbide exclusion” deposition surface materials listed in Table 7 (with the possible exception of Cu, whose (EDoS – EDoT) is slightly negative; Table 8), and graphite, at least tentatively satisfy this additional dimer-release criterion. Note that a release energy (EJ – EDoT) < 0 for all deposition surface in Table 8 suggests a thermodynamic preference for a decapped tooltip molecule to bind to the deposition surface.

Taking all five factors into account (Tables 7 and 8), “carbide exclusion” materials are the optimal substrate for diamond handle molecule crystal growth, and thus constitute the preferred embodiment of this invention. Graphene sheets (e.g., graphite, carbon nanotubes) may also be used with nonhydrogenic CVD processes, since atomic hydrogen etches graphene, although there exists a preferential epitaxial lattice registry relationship between the diamond C(111) and graphite (0001) surfaces, and similarly between the diamond C(110) and graphite (1120) surfaces [84], which might encourage non-tooltip-molecule nucleation. Furthermore, any conventional substrate material suitable for the deposition of CVD diamond thereon may be employed as the substrate utilized in the present invention, though perhaps with decreased efficiency or convenience. Thus the substrate material could be a metal, a metal carbide, a metal nitride, or a ceramic – e.g., silicon carbide, tungsten carbide, molybdenum, boron, boron nitride, niobium, graphite, copper, aluminum nitride, silver, iron, steel, nickel, silicon, alumina, or silica [5], or combinations thereof including cermets such as Al2O3-Fe, TiC-Ni, TiC-Co, TiC-TiN, or B4C-Fe systems [110]. Finally, specialized surface treatments may be applied to the deposition surface in order to suppress natural nucleation – for example, ion implantation of Ar+ ions (3 x 1015 ions/cm2 at 100 KeV) on silicon substrate is known to decrease nucleation density [111].

2.2.2 Tooltip Attachment Method A: Ion Bombardment in Vacuo

Tooltip molecules may be bonded to the deposition surface in the desired orientation via low-energy ion bombardment of the deposition surface in vacuo, creating a low density of preferred diamond nucleation sites. This is similar to the recognized pretreatment method of (for example) As+ ion implantation (1014 ions/cm2 at 100 KeV) on silicon substrate [112, 113] which yields a typical nucleation density of 105-106 nuclei/cm2, up from 104 in the absence of such ion implantation treatment [84]. Ion-beam implantation of C+ ions to form diamond-like carbon (DLC) films on various atomically clean substrates in chambers maintained at <10-9 torr are well-known [114-118, 137], including gold [118] and copper [119] surfaces, and halogen atoms have been partially substituted for hydrogen in DLC deposited on metal substrate in photosensor applications [120].

The specifics of Attachment Method A in the present invention are as follows. First, capped tooltip molecules (Section 2.1) are supplied to an ionization source. A vapor of capped tooltip molecules is created by heating in a vacuum chamber (e.g., C60 has a vapor pressure of 0.001 torr at 500 oC [17]). The vaporized capped tooltip molecules are next ionized by at least one of the procedures of laser ablation, electron bombardment, electron attachment, or photoionization. The capped tooltip molecule ions are then electrostatically accelerated to form a low-energy, highly dilute tooltip molecule ion beam, a well-known technology [121]. The ion beam is then directed in a scanning pattern across the deposition surface in vacuo. Upon striking the surface, the tooltip molecule ions (Figure 7A) may partially fragment with the release of the capping group, producing dangling bonds at the C2 dimer which can then insert into the substrate surface (Figure 7B). This beam energy transferred to the tooltip molecule upon impact should not significantly exceed 7.802 eV, the minimum energy required to entirely remove the C2 dimer from an uncapped DCB6-Ge tooltip molecule [36]. (This is considerably lower than the 10-80 eV ions studied by Sinnott et al [145] to functionalize carbon nanotubes (CNTs) by similar means, the 10-300 eV C+ ion beams used to grow diamond-like carbon films on various substrates [118], and the >250 eV needed to fragment fullerene ions into free C2 dimer radicals [11].) Another outcome is that only one capping group is released, bonding the tooltip molecule to the surface with only one bond through the C2 dimer (Figure 7C). Table 9 shows that this 1-bond outcome is energetically comparable to the 2-bond outcome, in the case an iodine cap and a graphite surface. Yet another possible outcome is that the tooltip molecule bonds to the surface at its base through either one (Figure 7D) or two (Figure 7E) bonds, releasing an H or H2, respectively, though neither base-bonding outcome is energetically preferred compared to the desired dimer-bonding outcomes.

Figure 7. Schematic of iodine-capped DCB6-Ge tooltip molecule (A) impacting 3×3 unit-cell graphite surface in desired orientation, (B) bonding to surface and releasing capping group as an I2 molecule, or alternatively, (C) bonding to surface with only one bond through the C2 dimer with release of one I atom, (D) one bond to surface through tooltip molecule base with release of one H atom, or (E) two bonds to surface through tooltip molecule base with release of one H2 molecule

Capping group removal energies from an isolated DCB6-Ge tooltip molecule for a variety of capping groups are estimated computationally (using semi-empirical AM1) as ranging from 1.9-7.4 eV (Table 10), as, for example, 3.554 eV for two iodine capping atoms, 4.728 eV for two amine capping groups, or 7.453 eV for two hydroxyl capping groups. These required energies would be halved when only one capping group is removed during tooltip molecule ion impact with the surface.

However, the removal energy for a single passivating hydrogen atom in the base of the tooltip molecule is 3.519 eV for an H atom removed from the bottom of the tooltip molecule base, comparable to many of the capping group removal energies listed in Table 10. Given the random orientation of tooltip molecules upon their arrival at (and impact with) the deposition surface, the sweep of a dilute beam of tooltip molecule ions across the surface will result in a thin scattering of tooltip molecules attached to the surface in a variety of orientations–some bound by two bonds to the uncapped dimer (as desired), others bound by only one bond to a partially uncapped dimer, and others bound directly to the tooltip molecule base in various orientations. Simple inspection of potential impact geometries suggests that energy transfer primarily into the dimer capping group upon impact is most probable if the tooltip molecule arrives at the deposition surface within (conservatively) ±20_o of vertical, in tip-down orientation. Therefore the probability of such arrival (assuming a random distribution of tooltip molecule ion orientations in the beam) and hence the probability of a dimer-bonded tooltip molecule (having either 1 or 2 bonds to the surface through the C₂ dimer) is roughly (40^o/360^o)2 ~ 1%, among all tooltip molecules that become bonded to the deposition surface.

Given a ~1% success rate, after the bombardment process and prior to the commencement of Step 3 the surface should be scanned by SPM to find and record the positions of those few tooltip molecules that are bound to the surface in the desired orientation. Depending upon the number density achieved, undesired tooltip molecule nucleation sites might simply be avoided during tool detachment in Step 4. Surface editing is another approach. Due to the low surface nucleation density (Section 2.2.1), after the aforementioned mapping procedure it may be possible to selectively detach and remove from the surface all attached misoriented tooltip molecules that are detected, e.g., using focused ion beam, electron beam, or NSOM photoionization, subtractively editing the deposition surface prior to commencing CVD in Step 3. A second alternative to subtractive editing is additive editing, wherein FIB deposition of new substrate atoms on and around the misoriented tooltip molecule can effectively bury it under a smooth mound of fresh substrate, again preventing nucleation of diamond at that site during Step 3. A third corrective procedure is reparative editing, wherein the methods described in Attachment Method B (Section 2.2.3) are employed to fully uncap the only partially uncapped tooltip molecule which has become bonded to the deposition surface (through only one carbon atom of the C₂ dimer) during the ion bombardment process of Attachment Method A. The result of this editing is that in Step 3, diamond handle structures will grow only on properly-oriented surface-bound tooltip molecules.

The ability of a chemisorbed (covalently bonded) tooltip molecule to migrate across a deposition surface in vacuo depends strongly upon the chemical structure of both tooltip molecule and the deposition surface material, and temperature. For example, spontaneous surface migration of gold atoms on gold surfaces is well known, though this mobility is greatly reduced at low temperatures and possibly also by alloying with silver or in combinations with other carbide resistant substrate materials. On the other hand, Larsson [122] estimates that during conventional diamond CVD on diamond substrate the acetylide radical (C2_H) has an energy barrier to migration of 3.6 eV across a clean diamond C(111) surface and the methyl radical (CH₃) has an even higher energy barrier to migration of 3.7 eV; on C(100), estimates for migration barriers range from 1.3-1.9 eV for methylene (CH₂) radicals [123, 124], 1.1-2.7 eV for methyl radicals [123, 125], and 1.7 eV for ethylene (C=CH₂) radicals [124]. Taking migration time from the Arrhenius equation as t_migrate^-1 ~ (k_BT/h) exp(-E_mig/k_BT), where h = 6.63 x 10^-34 J-sec (Planck’s constant) and k_B = 1.381 x 10^-23 J/K (Boltzmann’s constant), then at T = 300 K and E_mig = 1.1-2.7 eV, t_migrate ~ 5 x 10⁵ sec – 3 x 10³² sec on diamond substrate, which is very slow. Tooltip molecules have ten times as many atoms per molecule as the aforementioned radicals, hence should exhibit much slower surface migrations at any given temperature.

2.2.3 Tooltip Attachment Method B: Surface Decapping in Vacuo

Tooltip molecules may be bonded to the deposition surface in the desired orientation by non-impact dispersal and weak physisorption on the deposition surface, followed by tooltip molecule decapping via targeted energy input producing dangling bonds at the C₂ dimer which can then bond into the deposition surface in vacuo, again creating a low density of preferred diamond nucleation sites (Figure 8).

Figure 8. Schematic of iodine-capped DCB6-Ge tooltip molecule (A) dispersed on 3×3 unit-cell graphite surface in desired orientation, (B) absorbing targeted energy sufficient to decap the tooltip molecule in vacuo, releasing the capping group as two iodine ions or as an I₂ molecule, and (C) bonding to the deposition surface

(A)	(B)	(C)

The specifics of Attachment Method B in the present invention are as follows.

First, capped tooltip molecules are dispersed and physisorbed onto the deposition surface by any of several methods. These methods may include (but is not limited to): (1) spin coating, in which a suspension of capped tooltip molecules is applied to the center of a spinning wafer of smooth deposition surface material, and subsequently dispersed across the wafer surface; (2) dip coating, in which a wafer of smooth deposition surface material is dipped into a suspension of capped tooltip molecules and slowly withdrawn; or (3) spray coating, in which a suspension of capped tooltip molecules is applied to the wafer of smooth deposition surface material as a fine spray. All three methods have been successfully employed commercially to apply onto a smooth silicon wafer a dilute coating of 100-200 nm diamond particles to a number density of ~1 µm^-2 (~10⁸ cm^-2), starting with a suspension of 1 gm diamond particles in 1 liter of isopropanol [126-128], ethanol [82], or methanol [129]. In another analogous application [130], a layer of hydrocarbon molecules is applied to a substrate by the Langmuir-Blodgett technique, whereupon the surface is irradiated with a laser to decompose the layer of molecules at the surface without influencing the substrate; after decomposition the carbon atoms rearrange on the substrate surface to form a DLC film.

It is well-known that simple adamantane (C₁₀H₁₆), though having one of the highest melting points (542 K) of any hydrocarbon, "sublimes readily at atmospheric pressure and room temperature." [60] The enthalpy of sublimation for adamantane is DH_subl = 58,810 J/mole (~0.61 eV/molecule) [131] and the triple point for adamantane is T_triple = 733 K at P_triple = 2.7 GPa [132, 133], hence from the Clausius-Clapeyron equation the partial pressure of solid adamantane (P_adam) may be estimated as: ln(P_adam) = ln(P_triple) + (DH_subl / R) (T_triple^-1 – T_adam^-1) = 31.37 – (7077 T_adam^-1), where R = 8.31 J/mole-K (universal gas constant). At T_adam = 77 K (LN₂ temperature), the partial pressure of adamantane is only 5 x 10^-32 atm, or ~1 sublimed adamantane molecule per 200,000 m³ of volume at equilibrium, entirely negligible. However, at 300 K, P_adam = 0.024 atm, or ~1 sublimed adamantane molecule per 1700 nm³ of volume at equilibrium, a substantial sublimation rate.

The capped triadamantane tooltip molecule, being a larger molecule and containing two or more heavy atoms, should be less easily sublimed under ambient conditions. However, these molecules have not yet been synthesized nor are their precise thermodynamic properties known. Taking adamantane as the worst-case scenario, the surface dispersal conditions most certain to work consist of a suspension of capped tooltip molecules in a liquid nitrogen (LN₂) carrier fluid, dispersed onto a smooth deposition surface which is maintained at or slightly below 77 K, the boiling point of LN₂. After applying the suspension to the deposition surface, the surface temperature may be temporarily elevated to slightly above 77 K to drive off the chemically inert LN₂ carrier fluid, leaving only capped tooltip molecules dispersed in vacuo on the cold deposition substrate surface in the energetically preferred equilibrium position shown in Figure 8A. If the selected capped tooltip molecules have a low or negligible sublimation rate at room temperature, then other higher-temperature suspension fluids may be used which are easily evaporatable and compatible with the underlying substrate, i.e., chemically nonreactive with the underlying substrate material(s). For example, fullerenes including C₆₀ and C₇₀ have been dispersed onto silicon, silica, and copper surfaces at room temperature using an evaporatable carrier fluid (e.g., toluene), then employed as growth nuclei for microwave plasma diamond film CVD [82].

Second, the capping group must be induced to debond from the C₂ dimer in the tooltip molecule via excitation of the =C-cap bond. Some crude methods will not work. For example, if the capping atom is iodine, this atom has a large mass and hence a low frequency of vibration in a C-I bond (e.g., ~5.0 x 10¹² Hz at 350 K), so the absorption of a single IR photon of this frequency would add only ~0.02 eV to the bond, which is insufficient to break it. From Table 10, ~1.777 eV is required to break each of the two C-I bonds constituting the capping group of a DCB6-Ge tooltip molecule. This energy corresponds to the absorption of a single 430 THz (~7000 Å) visible red photon. Laser photoexcitation, photodissociation or photofragmentation [11] is commonly used in atom-selective bond breaking to selectively control a chemical reaction, e.g., the photodissociation of iodine atoms from iodopropane ions [134]. The requisite bond-breaking energy can be provided by a beam of electrons, noble element ions, or other energetic neutrals [135-137] directed towards the cooled deposition surface where the capped tooltip molecules reside. Viewed from above in its preferred orientation relative to the deposition surface, the iodine capped tooltip molecule has a cross-sectional area of ~44.42 Ų of which ~5.05 Ų represents the cross-sectional area of the iodine capping group, hence the beam of photons or ions carrying the debonding energy will strike the capping group, on average, ~10% of the time that they strike a tooltip molecule at all. Much more selectively, an STM tip can be scanned over the cold deposition surface specifically to break the C-I bond via ~1.5 eV single tunneling electrons [138-140]. For instance, the STM-mediated positionally-controlled single-molecule dissociation of an iodine atom from individual molecules of copper surface-physisorbed iodobenzene (C₆H₅I) and diiodobenzene (C₆H₄I₂) has been demonstrated experimentally by Hla et al [140]; in the inelastic tunneling regime, lower-energy electrons can also be injected via a resonance state between tip/substrate and the target molecule, breaking the weak C-I bond in iodobenzene without breaking the stronger C-C or C-H bonds [140].

Third, once the capping group has been removed and the dangling bonds have been exposed from the C₂ dimer, these bonds can form strong attachments with the deposition substrate surface, thus affixing the tooltip molecule to the deposition surface in the desired tip-down orientation. The energetics of the bond-by-bond decapping procedure for an iodine-capped DCB6-Ge tooltip molecule on a 3×3 unit-cell graphite surface is estimated in Figure 9 using semi-empirical AM1 simulations which included four unattached atoms (2H, 2I) to permit total atom count to remain constant throughout all substitutions. After each iodine capping atom is removed, the conversion of the dangling C₂ dimer bond to a new covalent bond between dimer and deposition surface appears to be energetically favored by 1.574 eV for the first bond and by 1.284 eV for the second bond. However, the presence of stray H or I ions can poison this reaction. For example, the dangling dimer bonds will bond to any H ions that are present, in preference to bonding with the deposition surface, so hydrogen must be excluded from the vicinity of the tooltip molecules during this stage of the process. It would be helpful to include a hydrogen getter in the vacuum chamber to absorb any hydrogens that become separated from the tooltip base. Stray iodine ions have a similar effect so it is helpful to include an intermittent positive-voltage getter plate inside the chamber to periodically attract and collect negative iodine ions as they are released from the tooltip caps. However, if the number of purposely decapped iodine atoms or accidentally debonded hydrogen atoms is on the order of ~10⁵ cm^-2 (Section 2.2.1 and Table 6) in a relatively large vacuum chamber, then an encounter between such stray atoms and a surface-bound tooltip molecule, even in the absence of any countermeasures, should be an exceedingly rare event.

Figure 9. Estimated energetics of the iodine-capped DCB6-Ge tooltip molecule decapping process on 3×3 unit-cell graphite surface, using semi-empirical AM1

The process of energy transfer to the tooltip molecule for the purpose of releasing the capping iodine atoms might also accidentally debond a hydrogen atom from the adamantane base of the tooltip molecule. The energetics of this dehydrogenation during various phases of the bond-by-bond decapping procedure for an iodine-capped DCB6-Ge tooltip molecule on a 3×3 unit-cell graphite surface is estimated in Figures 10, 11, and 12 using semi-empirical AM1 and including four unattached atoms (2H, 2I) to permit atom count to remain constant during all substitutions.

In the case of a tooltip molecule having no bonds to the surface through the C₂ dimer (Figure 10), that loses one hydrogen atom in the side position of the base, the tooltip molecule has a large energy barrier of 1.319 eV against bonding to the deposition surface through the dangling bond. Unless a stray H or I atom impinges at high velocity and recombines, the dehydrogenated tooltip molecule will remain on the deposition surface in the unreacted state and can later be sublimated off the deposition surface by gentle heating.

Figure 10. Estimated energetics of a dehydrogenation of the base of the iodine-capped DCB6-Ge tooltip molecule during the decapping process on 3×3 unit-cell graphite surface, using semi-empirical AM1 (0 eV = lowest-energy configuration), for a tooltip molecule having no bonds to the surface (at bottom left)

In the case of a tooltip molecule having one bond to the surface through the C₂ dimer (Figure 11), that loses one hydrogen atom in the side position of the base, the tooltip molecule has only a small energy barrier (0.063 eV) against bonding to the deposition surface through the dangling bond, so this unwanted double bonding is likely to occur even at LN₂ temperatures and cannot later be reversed via gentle heating. Since the barrier is of order ~k_BT, the configuration change will occur about equally in both directions, producing approximately equal populations of 1-bonded and 2-bonded configurations of tooltip molecules that have lost a single H atom in the side position of the base. These unwanted configurations can be observed by SPM and edited out as previously described. In the unlikely event that a stray H atom impinges and recombines, before the new bond to the deposition surface is established, the original hydrogenated tooltip molecule will be restored.

Figure 11. Estimated energetics of a dehydrogenation of the base of the iodine-capped DCB6-Ge tooltip molecule during the decapping process on 3×3 unit-cell graphite surface, using semi-empirical AM1, for a tooltip molecule with one bond to the surface (at bottom left)

In the case of a tooltip molecule having two bonds to the surface through the C₂ dimer (Figure 12), that loses one hydrogen atom in the side position of the base, the tooltip molecule has a strong energy preference (2.277 eV) to bond again to the deposition surface through the dangling bond, making a total of 3 bonds to the surface, a configuration that must be removed by post-process editing, or mapped and avoided. As before, the unlikely prior recombination of a stray H atom restores the original hydrogenated tooltip molecule, but impingement of a stray H or I atom before dehydrogenating the base can partially debond the properly 2-bonded tooltip molecule from the deposition surface. While the activation energy barrier to this reaction may be large, even preventative, the existence of such pathways emphasizes the need to minimize the number of stray H and I atoms that are present in the vacuum chamber during the tooltip molecule attachment process.

Figure 12. Estimated energetics of a dehydrogenation of the base of the iodine-capped DCB6-Ge tooltip molecule during the decapping process on 3×3 unit-cell graphite surface, using semi-empirical AM1, for a tooltip molecule with two bonds to the surface (at left)

Once a tooltip molecule has established at least one strong bond to the deposition surface, its surface mobility should be extremely low (Section 2.2.2). However, prior to such bonding these molecules are only physisorbed to the surface. Isolated pairs of iodine-capped DCB6-Ge tooltip molecules placed in tip-to-tip, tip-to-base, tip-to-side, and base-to-base orientations show weak energy barriers (calculated using semi-empirical AM1) between these configurations of only 0.05-0.09 eV (vs. 0.04 eV for (300 K) room temperature, 0.007 eV for (77 K) LN2 temperature), with just a slight preference for the base-to-base orientation. Tooltip molecules placed near each other and tooltip molecules placed several molecule widths apart in the same orientation show almost no energetic preference with separation distance, so tooltip molecules should be distributed randomly across the cold deposition surface. By varying the choices of tooltip molecule, capping group, deposition surface materials, and deposition surface temperature, the speed of tooltip molecule migration across the deposition surface can be made almost arbitrarily slow.

The enthalpy of sublimation for molecular iodine (I2) is DHsubl = 60,800 J/mole (~0.63 eV/molecule) and the vapor pressure over the solid is 6060 Pa at 100 oC [141], hence from the Clausius-Clapeyron equation the partial pressure of solid iodine (Piodine) may be estimated as: ln(Piodine) ~ 28.32 – (7316 Tiodine-1). At Tiodine = 77 K (LN2 temperature), the partial pressure of iodine is only 1 x 10-34 atm but at room temperature (Tiodine = 300 K) the partial pressure Piodine = 0.0005 atm, hence any stray iodine that remains physisorbed to the deposition surface after the completion of the decapping procedure may be driven off by gentle heating and sublimation.

2.2.4 Tooltip Attachment Method C: Solution Chemistry

Tooltip molecules may be bonded to the deposition surface in the preferred orientation using the techniques of conventional solution-phase chemical synthesis, creating a low density of preferred diamond nucleation sites (Figure 13).

The specifics of Attachment Method C in the present invention are as follows.

First, the deposition surface is functionalized with an appropriate functionalization group. For illustrative purposes, Figure 13A shows a section of (10,0) single-walled carbon nanotube (CNT) with a functional group “X” attached at the para- isomer positions (1 and 4) in one of the 6-carbon rings in the graphene surface. A capped tooltip is shown above this surface. For this invention, the functionalized deposition surface could be a flat graphene surface (i.e., graphite), or could be a functionalized non-graphene surface such as silicon, germanium, gold, and so forth (see Table 7). Graphite is attacked by strong oxidizing agents (such as sulfuric + nitric acid, or by chromic acid) [142], allowing the random surface functionalization of graphene; also, the chemical functionalization of fullerenes is well-studied [143-148]. Since site-specific functionalization may not be not strictly required in all cases, e-beam irradiation of dilutely surface-dispersed moieties, ion-beam implantation of functional-group ions, electrochemical functionalization [149, 150], or other related techniques could be employed in some cases to attach functional groups on the deposition surface at very high dilution, e.g., at 1 micron separations. However, direct chemical modification of surfaces via SPM tip [39, 140] enables the functionalization of the deposition surface at specific atomic sites, in cases where this is necessary.

Figure 13. Attachment of tooltip molecule to graphene deposition surface via solution phase combination of capping group and surface functionalization group

(A)	(B)	(C)
(D)	(E)	(F)

Second, conventional techniques of chemical synthesis are employed to establish conditions in solution phase whereby the tooltip molecule capping group, illustrated in Figure 13A by iodine, combines with the deposition surface functionalization group, here illustrated as "X", resulting in the removal of both I and X, leaving the tooltip molecule chemically bound to the deposition surface across two bonds at the carbon C₂ dimer as shown in Figure 13B – much like the standard esterification reaction wherein an alcohol molecule having a terminal –OH group combines with a second organic acid molecule having a terminal –H group, creating a C-C covalent bond between the two molecules (an ester) with the release of an H₂O in the process. It is possible that a specific convenient alkenation reaction can be found in the standard chemical synthesis literature, perhaps as an analog to the synthesis pathways for bicyclooctene (Figure 13C) or more directly as an analog to methods that may already be known for the alkenation (ethenation) of graphite, CNTs, or other deposition surfaces such as Si, Ge, or Au. The attachment reaction could be enhanced in the case of a nanotube deposition surface by using a kinked CNT, then anticipating the tooltip to preferentially attach at the kink site where CNTs are most reactive [151].

Density functional theory (DFT) analysis [152] has considered cycloadditions of dipolar molecules to the C(100)-(2×1) diamond surface. Experiments [153] have demonstrated the [2+4] cycloaddition of benzyne (C6H4) to polycyclic aromatics such as anthracene, forming triptycene (Figure 13D). DFT studies [154, 155] of the possible cycloaddition reaction of ortho-benzyne molecules to the graphene walls of carbon nanotubes have been done (Figures 13E and 13F). There have also been experimental investigations of solution-phase cycloaddition of organic molecules to semiconductor surfaces [156] and studies of diamondlike carbon films grown in organic solution [157] or grown via the electrolysis of acetates in solution phase [158]. Hoke et al [159] and others [160] have examined the reaction path for ortho-benzyne with C₆₀ and C₇₀ that leads to the [2+2] cycloaddition product in which benzyne adds across one of the interpentagonal bonds, forming a cyclobutene ring.

Most directly on point as prior art, Giraud et al [161-163] have synthesized 2,2-divinyladamantane (DVA), a single-cage adamantane molecule with two vinyl (-CH=CH₂) groups bonded to the same carbon atom in the cage, then dispersed this molecule onto a polished hydrogen-terminated Si(111) surface. Upon exposure to UV irradiation, photochemical double hydrosilylation occurs, fixing the adamantane molecule through two -C-C- tethers to two adjacent silicon atoms on the Si(111) surface with minimal steric strain. A rinse with ethanol, deionized water, and a 10 minute sonication with dichloromethane removed all ungrafted or physisorbed DVA. All adamantane molecules that become tethered to the surface via two bonds adopt the identical geometric orientation relative to the surface. Giraud et al [162] note that formation of the C-Si bond between the adamantane molecule and the silicon surface can be achieved by adapting any one of several commonly known techniques, including radical mediated hydrosilylation of olefins with molecular silanes [165-167], photochemical hydrosilylation of olefins with trichlorosilane [168], or hydrosilylation of olefins catalyzed by transition metal complexes [169-173].

2.3 STEP 3: Attach Handle Structure to Tooltip Molecule

STEP 3. Attach a large handle molecule or other handle structure to the deposition surface-bound tooltip molecule created in Step 2. There are two general methods that may be used to accomplish this: nanocrystal growth (Section 2.3.1) and direct handle bonding (Section 2.3.2).

2.3.1 Handle Attachment Method A: Nanocrystal Growth

In Method A, a bulk diamond deposition process (see below) is applied simultaneously to the entire tooltip-containing deposition surface (e.g., ~1 cm²) created in Step 2. The adamantane (diamond nanocrystal) base of each bound tooltip molecule serves as a nucleation seed from which a large diamond crystal will grow outward, in preference to growth on areas of the deposition surface where tooltip nucleation seed molecules are absent (Figure 14). Deposition should proceed until a sufficient quantity of bulk diamond crystal has grown outward and around the tooltip seed molecule such that the tooltip and its newly grown handle can be securely grasped by a MEMS-scale manipulator mechanism. The deposition process should be halted before adjacent growing crystals merge into a single film. As noted in Section 2.2, the number density of tools on the surface is controlled by limiting the number density of tooltip seed molecules attached to the deposition surface during Step 2. As distinguished from the more complex ex post strategy of chemically attaching a capped tooltip molecule to a larger prefabricated handle molecule, in the process described here the handle is grown directly onto the surface-bound tooltip, creating an optimally rigid and durable unitary mechanosynthetic tool structure. Alternatively and less preferred, the growing diamond crystal handle structure can be covalently bonded to some other appropriate large rigid structure such as a CNT, tungsten, or diamond-shard AFM tip, or an EBID/FIB-deposited metal or carbon column, e.g., by growing a vertical column of DLC atop the properly oriented tooltip molecule using a focused beam of hydrocarbon or C⁺ ions [114-118].

Figure 14. Multiply twinned diamond crystal growth during hot-filament assisted CVD. Photos courtesy of John C. Angus, Case Western Reserve University [174]

The most useful bulk deposition process is conventional diamond CVD, wherein micron/hour or faster deposition rates are typically demonstrated experimentally. The initial deposition rate onto the starting seed may be slow, but this rate should rapidly increase as more of the diamond handle structure is laid down during the deposition process which will require times on the order of hours. Traditional high-temperature CVD uses a large excess of atomic hydrogen which will etch a graphite or graphene surface, but CVD diamond can be deposited slowly at temperatures as low as 280-350 ^oC if necessary using the nonhydrogenic Argonne Lab C₆₀/C₂-dimer approach [175, 176] (Section 1.1(C)) which uses very little atomic H, in which case graphene etching would no longer be a serious problem. (Thermal suppression of nucleation at 1000 ^oC has been discussed by McCune [3].)

Does the CVD process deposit sp³-bonded diamond, not sp²-bonded graphite, onto such a tiny nucleation seed as the triadamantane base structure of the tooltip molecule? Conditions in vapor deposition of thin films require a critical nucleus size only on the order of a few atoms [177]. Under these conditions the free energy of formation of a critical nucleus may be negative [177] and the surface energy contribution may cause a reverse effect on the graphite-diamond phase stability [178, 179], a situation called nonclassical nucleation process [177]. Simple thermodynamic calculations by Badziag et al [180] and others [178, 179] have confirmed that hydrogen-terminated diamond nuclei <3 nm in diameter should have a lower energy than hydrogen-terminated graphite nuclei with the same number of carbon atoms, and that for surface bonds terminated with H atoms, diamonds smaller than ~3 nm are energetically favored over polycyclic aromatics (the precursors to graphite).

In 1983, Matsumato and Matsui [19], and later in 1990, Sato [20] and Olah [21], suggested that hydrocarbon cage molecules such as adamantane, bicyclooctane, tetracyclododecane, hexacyclopentadecane, and dodecahedrane could possibly serve as embryos for the homogeneous nucleation of diamond in gas phase. The adamantane molecule (C₁₀M/sub>H₁₆) is the smallest combination of carbon atoms possessing the diamond unit cage structure, i.e., three six-member rings in a chair conformation. The tetracyclododecane and hexacyclopentadecane molecules represent twinned diamond embryos that were proposed as precursors to the fivefold twinned diamond microcrystals prevalent in CVD diamond films – from simple atomic structure comparisons, the diamond lattice is easily generated from these cage compounds by simple hydrogen abstraction followed by carbon addition [7]. However, in one experiment adamantane placed on a molybdenum deposition surface during acetylene-oxygen combustion CVD failed to nucleate diamond growth [181], possibly due to “a fast transformation of adamantane on molybdenum to molybdenum carbide under diamond growth conditions.”

The first successful demonstration of the ability of surface-bound single-cage adamantane molecules to serve as nucleation seeds for diamond CVD was achieved experimentally by the Giraud group [161-164] during 1998-2001. In this process, a special seed molecule – 2,2-divinyladamantane (DVA), a single-cage adamantane with two vinyl (-CH=CH₂) groups bonded to the same carbon atom in the cage–is synthesized using conventional solution phase techniques [161], then dispersed onto a polished hydrogen-terminated Si(111) surface. When a surface prepared in this way is subjected to microwave plasma CVD using an H₂-rich 1% CH₄ feedstock gas at 40 mbar and 850 ^oC for 2 hours, only a few diamond grains are observed during subsequent SEM inspection, with a nucleation density below ~10⁴ cm-² [163]. However, when the surface is additionally exposed to UV irradiation from a xenon arc lamp for 24 hours prior to CVD, photochemical double hydrosilylation occurs, fixing the seed molecule via two -C-C- tethers to two adjacent silicon atoms on the Si(111) surface with minimal steric strain. With the seed molecule thus tethered to the silicon surface, the CVD process is then run again as previously described, this time resulting in a diamond nucleation density that rises to ~10⁹ cm^-2 and producing a very homogeneous diamond size of ~2 microns [163] (indicating essentially all adamantane-based nucleations), as shown in Figure 15.

Figure 15. SEM photograph of uniform 2-micron diamond crystals grown by MPCVD using surface-tethered single-cage adamantane molecules as nucleation seeds on a Si(111) surface; image courtesy of Luc Giraud [163]

Giraud et al [163] notes that although the treatment should densely cover the surface with covalently bound adamantane seed molecules, “the subsequent CVD plasma conditions will remove all the singly and presumably a few doubly attached molecules. The fact that nucleated diamonds were effectively obtained…shows the stability of grafted DVA in the nucleation conditions. All the samples treated without…UV…suffered no nucleation. This nucleation method therefore offers, on top of the advantage of flexibility and mildness, the possibility of photolithographic nucleation: diamonds adopt a homogeneous spatial repartition in the center of the irradiated region, with a well-faceted shape due to their cubic structure, while nucleation density sharply decreases to ~5 x 10⁶ cm^-2 on the brink of the irradiated region without even using a light mask.” In sum, doubly bonded adamantane seed molecules nucleate the growth diamond “handle” crystals, whereas singly bonded or unbonded seed molecules are removed by the hot CVD process and thus produce no crystal growth.

Even though the core of the tooltip molecule is iceane (the unit cell of hexagonal diamond or lonsdaleite) and not pure adamantane as in conventional cubic diamond crystal, lonsdaleite can also be grown experimentally [73-76]. The Raman spectra of lonsdaleite has been reported [182] and detected in localized stacking defect domains in textured CVD films [183]. Crystals of hexagonal diamond have been prepared in both static and shock high-pressure laboratory experiments [184, 185], and directly from cubic diamond [186]. Lonsdaleite can also be reliably synthesized [187] using rf-assisted plasma CVD and pure acetylene gas as the carbon source with no hydrogen – Roul et al [188] reports that crystallites grown on Si(100) substrates consisted mainly of polytypes of hexagonal diamond with a little cubic diamond and a few higher-order hydrocarbon phases, and others have found diamond polytypes in CVD diamond films [189]. Both cis and trans boat-boat bicyclodecane and related multiply-twinned compounds have been suggested as possible lonsdaleite nucleators based on the presence of both boat and chair hexagonal carbon rings [190, 191]. Twinning – the stacking of alternating (as in lonsdaleite) or arbitrarily-ordered re-entrant and intersecting chair and boat planes – is commonly seen in CVD diamond [191-195]. A semi-empirical theoretical analysis of the lonsdaleite structure by Burgos et al [196] gives results in reasonable accord with the limited experimental data. L.V. Zhigilei et al [197] note that intermediate states during the reconstruction of the C(111) surface of cubic diamond can lead to growth processes which result in the formation of a stacking fault, or twin plane [198-200], which could in turn produce lonsdaleite [201], and other transition mechanisms have been proposed [202].

As noted by Battaile et al [203], experimentally grown CVD diamond crystallites can exhibit C(100) and C(111) facets [204-206]. The C(110) surfaces are not usually observed (except in (110)-oriented homoepitaxy [207, 208]) because they grow much faster than the C(111) and C(100) faces [204, 210], hence are normally terminated by (100) and (111) facets. Diamond deposition rates in a hot-filament CVD reactor at 1200 K from methyl radical are typically 1.3-2.0 µm/hr for C(110) [209, 210] but only 0.5 µm/hr for C(111) and just 0.4-0.5 µm/hr for C(100) [209-212]. With the tooltip molecule bound to the deposition surface in the preferred orientation (i.e., inverted), the C(110) plane is angled at 45^o from vertical, leaning away from the vertical centerline; the C(100) plane is also angled at 45^o from vertical, but leans toward the vertical centerline; the C(111) plane goes straight up along the centerline. So under CVD deposition, the tool handle structure will grow fastest outward at 45^o. The C(100) plane will be buried inside the tool, and the tool handle crystal will exhibit C(110) facets on the sides and a C(111) facet on the top. (Plasma CVD diamond crystallites grown on Si(100) wafers also display a combination of C(111) and C(110) facets [6].) Note that while lonsdaleite has a repeating structure, here we should expect only a single twinning fault at the centerplane, not a series of repeating twinnings. However, geometry dictates that the detached tool cannot be concave on its active face, and would at worst be flat, hence even at minimum can serve as a primitive tool to experimentally demonstrate positionally controlled diamond mechanosynthesis.

Diamond films have been formed by immersing a substrate in a fluid medium comprising a carbon-containing precursor and irradiating the substrate with a laser to pyrolyze the precursor, a technique that could also be adapted to grow diamond handle structures onto isolated surface-bound tooltip molecules. For example, Hacker et al [213] describe a process in which gas containing an aliphatic acid or an aromatic carboxylic anhydride that vaporizes without decomposition is passed over a substrate and irradiated with a focused high-powered pulsed laser, depositing a diamond film. In the process disclosed by Neifeld [214], the substrate is immersed in a liquid containing carbon and hydrogen, e.g. methanol, and a laser pulse is then directed through the liquid coating to heat the substrate. The liquid is pyrolyzed and carbon material from the pyrolyzed liquid grows on the substrate to form a diamond coating on the substrate. Yu [130] applies a hydrocarbon layer to a substrate by the Langmuir-Blodgett technique, then irradiates the surface with a laser (or e-beam, x-rays, etc.) to decompose the layer of molecules at the surface without influencing the substrate; after decomposition, the carbon atoms rearrange on the surface of the substrate to form a DLC film. Bovenkerk et al [4] proposes using an unusual dual gas approach to CVD in which, for example, a hydrogen (H₂) or methane (CH₄) feedstock gas is alternated with a carbon tetraiodide (CI₄) feedstock gas, with each exposure resulting in the deposition of a new diamond monolayer on an existing diamond substrate, and alternative lower-temperature CVD gas chemistries are being investigated such as use of CO₂-based [215] or halogen-containing [216] gas mixtures. Finally, laser heating of solid CO₂ at 30-80 GPa pressure causes the molecule to decompose into oxygen and diamond, revealing a new region of the CO₂ phase diagram with a boundary having a negative P-T slope [217].

There are several other lesser-known alternatives to CVD, ion beam deposition, and laser pyrolysis which might also be adapted for growing the handle structure onto the surface-bound tooltip molecule. Diamond film prepared by physical vapor deposition has been described by Namba et al [218]. Liquid-phase diamond synthesis in boiling benzene or in molten lead was reported as early as 1905 [219], and more recently, a 2% yield of diamond from carbon tetrachloride in liquid sodium at 700^oC [220] and the electrochemical growth of diamond films below 50^oC in liquid ethanol [157] and in solutions of ammonium acetate in liquid acetic acid [158], and also the hydrothermal synthesis of diamond [221].

A final consideration is the overall temperature stability of the bound tooltip molecule under the conditions of CVD growth and related processes. One concern is that the tooltip molecule might destabilize if heated to CVD temperatures. Pure adamantane graphitizes at >480^oC [60], and early thermodynamic equilibrium calculations [222, 223] showed that these and similar low molecular weight hydrocarbons are not stable at high temperatures (>600^oC) in the harsh CVD environment. Another concern is that at elevated temperatures, the tooltip molecule might debond from the deposition surface. However, the work of the Giraud group [161-164] with the 2,2-divinyladamantane nucleation molecule for diamond CVD confirms experimentally that adamantane molecules having two tethers to a silicon deposition surface can survive at least 2 hours of CVD conditions at 850^oC without destabilizing or detaching from the surface, although adamantanes with only one or no bonds to the surface evidently may be detached or destroyed at these temperatures. Table 8 gives the release energy (E^J – E^DoT) for a decapped tooltip molecule bound to a Ge deposition surface as ~4.7 eV. If the activation energy (reaction barrier) is of similar magnitude, then from the Arrhenius equation (Section 2.2.2) the mean detachment time for a decapped tooltip molecule bound to a Ge deposition surface at 850^oC is ~5 x 10⁷ sec (>1 year). For some deposition surface materials the tooltip release energy (and reaction barrier) can be considerably lower, so it may be necessary to employ a lower-temperature CVD process to obtain an acceptably long thermal detachment time for some substrates. Successful low-temperature CVD of diamond crystallites or DLC films have been reported at temperatures as low as 250-750 ^oC [224], 280-350 ^oC [175, 176], 300-500 ^oC [116], 350-600 ^oC [128], >400 ^oC [110], and <500 ^oC [10].

2.3.2 Handle Attachment Method B: Direct Handle Bonding

In Method B, an SPM-manipulated dehydrogenated diamond shard having a flat or convex tip is brought down vertically onto a surface upon which tooltip molecules are attached. Retraction of the tip pulls the tooltip molecule off the surface, yielding a finished tool for diamond mechanosynthesis consisting of a tooltip molecule mounted on the diamond shard with an active C₂ dimer exposed at the tip, as illustrated in Figure 16.

Figure 16. Extraction of surface-bound tooltip molecule via bonding to vertically inserted and retracted dehydrogenated diamond C(110) probe manipulated via SPM

The specific sequence of events is as follows:

(1) Prepare tooltip molecules. Bond tooltip molecules to the deposition surface in the preferred orientation, as described in Step 2 (Section 2.2).

(2) Mount diamond AFM tip. Mount a diamond shard as the working tip of an AFM. The apex of the shard should be flat or convex in cross-section, and the apical tip surface of the shard should expose the diamond C(110) crystal face.

(3) Depassivate AFM tip. The AFM tip is baked in vacuo at >1300 K to completely dehydrogenate the entire diamond shard, including most importantly its C(110) apical tip surface. The C(110) surface does not reconstruct during thermal depassivation [225].

(4) Lower tip onto surface. The depassivated diamond shard tip is positioned perpendicular to the deposition surface upon which the tooltip molecules are affixed in the preferred orientation. The shard tip is then lowered toward the deposition surface (Figure 16A), in vacuo at room temperature.

(5) Bind shard to tooltip molecule. As the apical tip surface of the diamond shard reaches and contacts the deposition surface, the many dangling bonds at the C(110) crystal face of the apical tip surface bond with several carbon atoms in the base of a tooltip molecule, displacing several passivating hydrogen atoms which migrate to nearby dangling bonds on the diamond shard apical tip surface (Figure 16B).

(6) Retract tip from surface. The diamond shard is retracted from the deposition surface in the vertical direction. The tooltip molecule is more strongly bonded to the shard, so the vertical retraction of the shard causes the two bonds to the deposition surface through the C₂ dimer to break (Figure 16C), creating an active C2 dimer radical exposed at the apical tip surface of the shard. The diamond shard is now an active tool that can be employed in diamond mechanosynthesis.

The process for manufacturing a mechanosynthetic tool via Method B is much inferior to the Method A process for a number of reasons. First, in Method B, after contacting the surface it will be uncertain how many, if any, tooltip molecules have become bonded to the apical tip surface of the diamond shard probe. Second, after the bonds to the deposition surface through the C₂ dimer have been broken, the tooltip molecule is free to rotate and may form additional bonds between the tooltip molecule base and the depassivated apical tip surface, most likely carrying the tooltip molecule out of its vertical orientation and placing it in some unknown, possibly useless, orientation. Third, if the tooltip molecule is bonded to the diamond shard probe through only a minimal number of bonds then the tool may be far less rigid than the solid crystalline tool created by Method A, and thus may be incapable of transmitting the full range of magnitudes and directions of forces that may be required in mechanosynthetic operations. Finally, if the tooltip molecule is bonded to the diamond shard probe through bonds in various numbers and different crystallographic positions, then the position, vibrations, and other important characteristics of the tool will be far less predictable than the tool created by Method A, and the positional uncertainty of dimer placement may be much greater, possibly unacceptably high for many applications, even if the tool is operated at LN₂ or lower temperatures. Nevertheless, Method B is a considerably easier process from an experimental standpoint and so it may be possible to manufacture early, though less capable, mechanosynthetic tools in this manner.

2.4 STEP 4: Separate Finished Tool from Deposition Surface

STEP 4. Mechanically grasp and break away the diamond crystal-handled tool from the deposition surface, in vacuo. The covalent bond between the tooltip (through the C₂ dimer) and the surface will mechanically break (Table 8), yielding either a tool with a naked carbon dimer attached (i.e., a charged, active mechanosynthetic tool; Figure 17A) or a tool with no dimer attached (i.e., a “discharged” tool needing recharge, e.g., with acetylene; Figure 17B). Ideally, handle diamond near the tooltip forms only weak van der Waals bonds to the deposition surface, so tool breakaway produces few or no unwanted dangling bonds near the active tip. If deemed necessary, each tool can be further machined or shaped via laser-, e-beam-, or ion-beam-ablation to provide any desired aspect ratio for the finished tool, or to provide any necessary larger-scale features on the handle surface such as slots, grooves, or ridges, prior to separation of the tool from the surface. This toolbuilding process should work for any carbon dimer deposition tooltip of similar type, as long as the capping group and the deposition surface are judiciously chosen for each case. Note also that the discharged dimer deposition tool can often be employed as a dimer removal tool [38], at least in the case of isolated dimers on a mechanosynthetic workpiece, permitting limited rework capability during subsequent mechanosynthetic operations using the tools produced by the present invention.

Figure 17. Idealized mechanosynthetic tool handle structure (passivating hydrogen atoms not shown)

Following the completion of Step 3 but prior to the commencement of Step 4, the mechanosynthetic tools grown on the deposition surface in Step 3 may be stably stored indefinitely at room temperature under an inert atmosphere. Prior to the commencement of Step 4, the deposition surface containing the bound tools should be baked in vacuo at a temperature high enough to drive off any physisorbed impurities that may have accumulated on the surface or handle structure during storage, but at a temperature low enough to avoid significant dehydrogenation of the diamond handle crystal. Hydrogen desorption becomes measurable at 800-1100 K for the C(111) diamond surface [226], 1400 K for the C(110) surface [227], and possibly as low as 623 K for the C(100) surface [228]. Taking T_bake = 600 K and the dimer-to-surface C-C bond energy E_bond = 556 zJ [32], then the minimum thermal detachment time is given by the Arrhenius equation as t_detach ~ [(kBT_bake/h) exp(-E_bond/kBT_bake)]-1 = 1.1 x 10¹⁶ sec, where h = 6.63 x 10^-34 J-sec (Planck’s constant) and k_B, = 1.381 x 10^-23 J/K (Boltzmann’s constant).

The minimum force required to break a C-C bond in a characteristic bond cleavage time of ~0.1 ns at 300 K is estimated as ~4.4 nN and ~4.0 nN for a C-Si bond, and the threshold stress for breaking two C-C bonds “mechanically constrained to cleave in a concerted process” is ~6 nN per bond [32]. Hence the force required to simultaneously break both of the bonds between the two tooltip dimer carbon atoms and the two deposition surface atoms to which they are attached, during tool separation in Step 4, is likely on the order of 8-12 nN. However, a much larger van der Waals attraction may exist between the diamond tool handle crystal and the deposition surface. For example, two opposed hydrogenated diamond C(111) surfaces equilibrate at ~2.3 Å separation, according to a simple molecular mechanics (MM+) simulation. Assuming no additional covalent bonds have formed between tool and deposition surface except through the C₂ dimer at the tooltip, two planar surfaces of area A ~ 1 µm² with Hamaker constant H ~ 300 zJ (i.e., diamond, Si, Ge, graphite, metal surfaces) separated by a distance s ~ 2.3 Å experience an attractive force [32, 93] of F ~ HA/12ps³ ~ 650,000 nN. Even if the contact interface is only 100 nm² the attractive force is still F ~ 65 nN, an order of magnitude larger than the force required to break each of the two covalent bonds between deposition surface and C₂ dimer. The separation force required to snap the finished tool free from the deposition surface, assuming no rogue covalent bonds, is therefore on the order of 10²-10⁶ nN. For comparison, the force of gravity on a 1 µm³ diamond crystal is ~0.00003 nN and the force from a 10,000-g shock impact acceleration (e.g., dropping object on concrete floor) produces a lateral accelerative force of only 0.3 nN.

Additionally, the flexural strength of diamond is 23 times greater than that of silicon, permitting much greater forces to be applied to the tool handle element without breakage; if the diamond handle crystal should contact the substrate which it overhangs, its low coefficient of static friction ensures that the diamond crystal will not adhere to the substrate [18]. Note that in one combustion CVD experiment with adamantane-seeded diamond growth on Mo (a carbide-forming surface; Table 7) [181], it was observed that “the diamond crystals show a low adhesion on the molybdenum substrate.” Differential thermal expansion during post-CVD cooling causes the built tool and the deposition surface to shrink differently, creating stresses and possibly prematurely breaking off the tool; a similar technique allows a grown diamond film to separate as an integral diamond sheet on cooling.

The need to securely grip and apply forces against mechanical resistance during the tool separation process, while retaining precise positional knowledge in all coordinate and rotational axes, imposes specific operational requirements for the gripper and manipulator system. Since the bondlength between C₂ dimer and deposition surface is ~1.5 Å, and since these bonds cannot tolerate excessive stretching before breaking, the manipulator system should have a repeatable positioning resolution of at least DR_min ~ 2 Å. Subsequent mechanosynthetic operations on diamond surfaces will likely require repeatable positional accuracies of at least 0.5 Å, and in some cases as little as 0.2 Å [38, 235], or about tenfold better than for mere tool separation alone. Since handle crystals are of slightly different size, shape, and orientation, it is also important to avoid excessively rotating the handle as it is being grasped in preparation for tool separation from the deposition surface. A handle crystal of radius R_handle = 1 mm and a minimum allowable displacement of DR_min = 2 Å implies a minimum allowable rotation of Dq_min = sin-1(DR_min/R_handle) ~ 200 µrad, or 20 µrad for mechanosynthesis operations where DR = 0.2 Å. A further requirement is the ability of the manipulator to apply incremental forces along various translational or rotational vectors of DF_min = 10²-10⁶ nN.

The Zyvex S100 Nanomanipulator [229] achieves a rotational accuracy of Dq_S100 = 2 µrad << Dq_min = 20-200 µrad, as required. The S100 grippers provide a maximum gripping force of 550,000 nN ~ DF_min, which should be adequate in most cases. However, the repeatable positional accuracy of the S100 is only 50 Å, or 25 times coarser than the ~2 Å required for controlled tool separation and ~100-250 times coarser than the 0.2-0.5 Å required for accurate mechanosynthesis [235]. The Klocke Nanotechnik Nanomanipulator claims 20 Å step sizes and 10 Å positional accuracy without backlash [230], still not quite good enough. Nevertheless, in a somewhat different context scanning with AFM tips may be undertaken with the ~0.1 Å accuracy that would be required during room temperature mechanosynthesis operations. By premeasuring the exact positions of all viable tooltip molecules attached to the deposition surface, and then carefully tracking all positional and rotational motions that are subsequently applied to the tool, the exact 3D spatial position of the active tool dimer may be continuously estimated with sufficient accuracy.

Once the completed mechanosynthetic tool has been detached from the deposition surface, the exposed C₂ dimer radical is extremely chemically active. According to an AM1 simulation, an activated DCB6-Ge tooltip is energetically preferred to combine with incident O₂ molecules by 6.7 eV and with incident N₂ molecules by 2.8 eV, the principal constituents of air, the most likely environmental contaminant. Since any laboratory vacuum is imperfect, stray atoms, ions, and molecules will populate the vacuum chamber at some low concentration and will eventually impinge upon an unused active tooltip, reacting with it and rendering it useless for further mechanosynthetic work.

Using the standard formula for molecular incident rate [231], the mean lifetime ttool of an active DCB6-Ge tooltip exposed to vacuum with a partial pressure Patm of contaminant molecules having molar mass _molar (kg/mole) at temperature T, is given by: t_tool = (N_hits V_molar / A_target P_atm N_A) (p M_molar / 2 k_BT N_A)^1/2 (seconds), where the number of encounters between an active tooltip and a contaminant molecule that are required to deactivate the tooltip is taken as N_hits = 1, the molar gas volume V_molar = 22.4141 x 10^-3 m³-atm/mole, Atarget ~ 2 Ų is the cross-sectional area of the exposed C₂ dimer impact target (analogous to the room temperature dimer atom positional uncertainty footprint described in [38]), T = 77 K (LN₂ temperatures), NA = 6.023 x 10²³ molecules/mole (Avogadro’s number), and kB = 1.381 x 10^-23 J/K (Boltzmann’s constant). Expressing pressure as P_torr = 760 P_atm in torr and rearranging terms, then P_torr = (2.2 x 10^-6) / t_tool (torr) for hydrogen atoms (H) having molar mass M_molar = 1 x 10^-3 kg/mole; P_torr = (1.2 x 10^-5) / t_tool (torr) for nitrogen molecules (N₂) having molar mass M_molar = 28 x 10^-3 kg/mole and P_torr = (1.3 x 10^-5) / t_tool (torr) for oxygen molecules (O₂) having molar mass M_molar = 32 x 10^-3 kg/mole, the two most likely contaminant molecules from the ambient environment. To ensure a mean tooltip lifetime of t_tool = 1000 sec requires maintaining a partial pressure Ptorr = 2.2 x 10^-9 torr for H atoms, P_torr = 1.2 x 10^-8 torr for N₂ and P_torr = 1.3 x 10^-8 torr for O₂. Ultrahigh vacuums (UHV) of 10^-7-10^-10 torr have been commonly accessible experimentally for many decades [232], and vacuums as high as 10^-15 torr have been created in the laboratory [233]. Note that a vacuum of 10^-9 torr inside an enclosed 10,000 cubic micron box contains, on average, far less than one contaminant molecule – usually making, in effect, a perfect vacuum and allowing, in principle, an unrestricted tooltip lifetime.

Links to Additional Papers by Robert Freitas:

Kinematic Self-Replicating Machines

Theoretical Analysis of a Carbon-Carbon Dimer Placement Tool for Diamond Mechanosynthesis Theoretical Analysis of Diamond

Mechanosynthesis. Part I. Stability of Mediated Growth of Nanocryalline Diamond C(110) Surface

References

1. Tohru Inoue, Masaya Kadono, Akiharu Miyanaga, “Method for forming diamond and apparatus for forming the same,” U.S. Patent 5,360,477, 1 November 1994.

2. Nobuo Setaka, "Process for producing diamond powder by shock compression," U.S. Patent 4,377,565, 22 March 1983.

3. Robert C. McCune, Ronald J. Baird, "Making diamond composite coated cutting tools," U.S. Patent 4,919,974, 24 April 1990.

4. Harold P. Bovenkerk, Thomas R. Anthony, James F. Fleischer, William F. Banholzer, "CVD diamond by alternating chemical reactions," U.S. Patent 5,302,231, 12 April 1994.

5. Thomas R. Anthony, James F. Fleischer, "Smooth surface CVD diamond films and method for producing same," U.S. Patent 5,523,121, 4 June 1996.

6. Dieter M. Gruen, Thomas G. McCauley, Dan Zhou, Alan R. Krauss, "Tailoring nanocrystalline diamond film properties," U.S. Patent 6,592,839, 15 July 2003.

7. Y. Matsui, A. Yuuki, M. Sahara, Y. Hirose, "Flame structure and diamond growth mechanism of acetylene torch," Jpn. J. Appl. Phys. Part 1 28(1989):1718-1724.

8. D. M. Gruen, S. Liu, A. R. Krauss, J. Luo, X. Pan, "Fullerenes as Precursors for Diamond Film Growth Without Hydrogen or Oxygen Additions," Appl. Phys. Lett. 64(1994):1502-1504.

9. Dieter M. Gruen, Shengzhong Liu, Alan R. Krauss, Xianzheng Pan, "Diamond film growth from fullerene precursors," U.S. Patent 5,620,512, 15 April 1997.

10. Dieter M. Gruen, Alan R. Krauss, "Method for the preparation of nanocrystalline diamond thin films," U.S. Patent 5,772,760, 30 June 1998.

11. Dieter M. Gruen, "Conversion of fullerenes to diamond," U.S. Patent 5,209,916, 11 May 1993.

12. Dieter M. Gruen, "Conversion of fullerenes to diamond," U.S. Patent 5,328,676, 12 July 1994.

13. Dieter M. Gruen, "Conversion of fullerenes to diamond," U.S. Patent 5,370,855, 6 December 1994.

14. Dieter M. Gruen, "Conversion of fullerenes to diamonds," U.S. Patent 5,462,776, 31 October 1995.

15. Dieter M. Gruen, Alan R. Krauss, Shengzhong Liu, Xianzheng Pan, Christopher D. Zuiker, "Diamond film growth argon-carbon plasmas," U.S. Patent 5,849,079, 15 December 1998.

16. Dieter M. Gruen, Alan R. Krauss, Ali Erdemir, Cuma Bindal, Christopher D. Zuiker, "Smooth diamond films as low friction, long wear surfaces," U.S. Patent 5,989,511, 23 November 1999.

17. "Directed Energy Interactions with Surfaces: Fullerenes As Precursors for Diamond Film Growth," Chemistry Division, Argonne National Laboratory, accessed 29 December 2003; http://chemistry.anl.gov/surfaces/fullerenes.html

18. Alan R. Krauss, Dieter M. Gruen, Michael J. Pellin, Orlando Auciello, "Ultrananocrystalline diamond cantilever wide dynamic range acceleration/vibration/pressure sensor," U.S. Patent 6,422,077, 23 July 2002.

19. R.J.H. Klein-Douwel, J.J. ter Meulen, "Spatial distributions of atomic hydrogen and C2 in an oxyacetylene flame in relation to diamond growth," J. Appl. Phys. 83(1 May 1998):4734-4745; http://www.mlf.sci.kun.nl/publ/1998/H_C2.pdf

20. S. Matsumoto, Y. Matsui, "Electron microscopic observation of diamond particles grown from the vapour phase," J. Mater. Sci. 18(1983):1785-1793.

21. Y. Sato, Japan Review in New Diamond (English version), Japan New Diamond Forum, 1990, p. 5.

22. Jeremy E. Dahl, Robert M. Carlson, Shenggao Liu, "Diamondoid-containing materials in microelectronics," U.S. Patent Application 20020130407, 19 September 2002.

23. Thomas A. Plaisted, Susan B. Sinnott "Hydrocarbon thin films produced from adamantane–diamond surface deposition: Molecular dynamics simulations," J. Vac. Sci. Technol. A 19(January/February 2001):262-266; http://dx.doi.org/10.1116/1.1335683

24. M. Matsuura, K. Murakami, Y. Inaki, T. Yamamoto, "Diamond-like-carbon thin films deposited from adamantane," Bull. Res. Inst. Electronics Shizuoka Univ. 23(1988):47-56. In Japanese.

25. Charles B. Musgrave, Jason K. Perry, Ralph C. Merkle, William A. Goddard III, "Theoretical studies of a hydrogen abstraction tool for nanotechnology," Nanotechnology 2(1991):187-195; http://www.zyvex.com/nanotech/Habs/Habs.html

26. Michael Page, Donald W. Brenner, "Hydrogen abstraction from a diamond surface: Ab initio quantum chemical study using constrained isobutane as a model," J. Am. Chem. Soc. 113(1991):3270-3274.

27. Michael Page, Donald W. Brenner, "Ab initio quantum chemical study of hydrogen abstraction from isobutane constrained to model a diamond surface," in Russell Messier, Jeffrey T. Glass, James E. Butler, Rustum Roy, eds., Proceedings of the Second International Conference, New Diamond Science and Technology, Materials Research Society, Pittsburgh, PA, 1991, pp. 45-50.

28. Xiao Yan Chang, Martin Perry, James Peploski, Donald L. Thompson, Lionel M. Raff, "Theoretical studies of hydrogen-abstraction reactions from diamond and diamond-like surfaces," J. Chem. Phys. 99(15 September 1993):4748-4758.

29. Susan B. Sinnott, Richard J. Colton, Carter T. White, Donald W. Brenner, "Surface patterning by atomically-controlled chemical forces: molecular dynamics simulations," Surf. Sci. 316(1994):L1055-L1060.

30. D.W. Brenner, S.B. Sinnott, J.A. Harrison, O.A. Shenderova, "Simulated engineering of nanostructures," Nanotechnology 7(1996):161-167; http://www.zyvex.com/nanotech/nano4/brennerAbstract.html and http://www.zyvex.com/nanotech/nano4/brennerPaper.pdf

31. A. Ricca, C.W. Bauschlicher Jr., J.K. Kang, C.B. Musgrave, "Hydrogen abstraction from a diamond (111) surface in a uniform electric field," Surf. Sci. 429(1999):199-205.

32. K. Eric Drexler, Nanosystems: Molecular Machinery, Manufacturing, and Computation, John Wiley & Sons, New York, 1992; http://www.zyvex.com/nanotech/nanosystems.html

33. Ralph C. Merkle, “A proposed ‘metabolism’ for a hydrocarbon assembler,” Nanotechnology 8(1997):149-162; http://www.zyvex.com/nanotech/hydroCarbonMetabolism.html

34. Stephen P. Walch, Ralph C. Merkle, "Theoretical studies of diamond mechanosynthesis reactions," Nanotechnology 9(September 1998):285-296;

35. Fedor N. Dzegilenko, Deepak Srivastava, Subhash Saini, "Simulations of carbon nanotube tip assisted mechano-chemical reactions on a diamond surface," Nanotechnology 9(December 1998):325-330.

36. Ralph C. Merkle, Robert A. Freitas Jr., "Theoretical analysis of a carbon-carbon dimer placement tool for diamond mechanosynthesis," J. Nanosci. Nanotechnol. 3(August 2003):319-324; http://www.rfreitas.com/Nano/DimerTool.htm and http://www.rfreitas.com/Nano/JNNDimerTool.pdf

37. Jingping Peng, Robert A. Freitas Jr., Ralph C. Merkle, "Theoretical Analysis of Diamond Mechanosynthesis. Part I. Stability of C2 Mediated Growth of Nanocrystalline Diamond C(110) Surface," J. Comput. Theor. Nanosci. 1(March 2004):62-70. http://www.MolecularAssembler.com/JCTNPengMar04.pdf

38. David J. Mann, Jingping Peng, Robert A. Freitas Jr., Ralph C. Merkle, "Theoretical Analysis of Diamond Mechanosynthesis. Part II. C2 Mediated Growth of Diamond C(110) Surface via Si/Ge-Triadamantane Dimer Placement Tools," J. Comput. Theor. Nanosci. 1(March 2004):71-80. http://www.MolecularAssembler.com/JCTNMannMar04.pdf

39. Wilson Ho, Hyojune Lee, "Single bond formation and characterization with a scanning tunneling microscope," Science 286(26 November 1999):1719-1722; http://www.physics.uci.edu/~wilsonho/stm-iets.html

40. A. Herman, "Towards mechanosynthesis of diamondoid structures: I. Quantum-chemical molecular dynamics simulations of sila-adamantane synthesis on hydrogenated Si(111) surface with the STM," Nanotechnology 8(September 1997):132-144.

41. A. Herman, "Towards mechanosynthesis of diamondoid structures. II. Quantum-chemical molecular dynamics simulations of mechanosynthesis on an hydrogenated Si(111) surface with STM," Modelling Simul. Mater. Sci. Eng. 7(January 1999):43-58; A. Herman, “Computational nanotechnology of silicon structures: a challenge far beyond 2000,” TASK Quarterly 1(July 1997):9-20.

42. Noriaki Oyabu, Oscar Custance, Insook Yi, Yasuhiro Sugawara, Seizo Morita, "Mechanical vertical manipulation of selected single atoms by soft nanoindentation using near contact Atomic Force Microscopy," Phys. Rev. Lett. 90(2 May 2003):176102; http://link.aps.org/abstract/PRL/v90/e176102

43. O. Marti, B. Drake, and P. K. Hansma, "Atomic force microscopy of liquid-covered surfaces: Atomic resolution images," Appl. Phys. Lett. 51(17 August 1987):484-486; http://content.aip.org/APPLAB/v51/i7/484_1.html

44. G. Tanasa, O. Kurnosikov, C.F.J. Flipse, J.G. Buijnsters, W.J.P. van Enckevort, "Diamond deposition on modified silicon substrates: Making diamond atomic force microscopy tips for nanofriction experiments," J. Appl. Phys. 94(1 August 2003):1699-1704; http://content.aip.org/JAPIAU/v94/i3/1699_1.html

45. Sacharia Albin, Jianli Zheng, John B. Cooper, Weihai Fu, Arnel C. Lavarias, "Microwave plasma chemical vapor deposited diamond tips for scanning tunneling microscopy," Appl. Phys. Lett. 71(10 November 1997):2848-2850.

46. E. Oesterschulze, W. Scholz, Ch. Mihalcea, D. Albert, B. Sobisch, W. Kulisch, "Fabrication of small diamond tips for scanning probe microscopy application," Appl. Phys. Lett. 70(27 January 1997):435-437; http://content.aip.org/APPLAB/v70/i4/435_1.html

47. G.J. Germann, G.M. McClelland, Y. Mitsuda, M. Buck, H. Seki, "Diamond force microscope tips fabricated by chemical vapor deposition," Rev. Sci. Instrum. 63(1 September 1992):4053-4055; http://content.aip.org/RSINAK/v63/i9/4053_1.html

48. Geoffrey J. Germann, Sidney R. Cohen, Gabi Neubauer, Gary M. McClelland, Hajime Seki, D. Coulman, "Atomic scale friction of a diamond tip on diamond (100) and (111) surfaces," J. Appl. Phys. 73(1 January 1993):163-167; http://content.aip.org/JAPIAU/v73/i1/163_1.html

49. Eric P. Visser, Jan W. Gerritsen, Willem J. P. van Enckevort, Herman van Kempen, "Tip for scanning tunneling microscopy made of monocrystalline, semiconducting, chemical vapor deposited diamond," Appl. Phys. Lett. 60(29 June 1992):3232-3234; http://content.aip.org/APPLAB/v60/i26/3232_1.html

50. J.H. Hafner, C.L. Chueng, C.M. Lieber, "Growth of nanotubes for probe microscopy tips," Nature 398(1999):761-762.

51. C.L. Cheung, J.H. Hafner, C.M. Lieber, "Carbon nanotube atomic force microscopy tips: Direct growth by chemical vapor deposition and application to high-resolution imaging," Proc. Natl. Acad. Sci. (USA) 97(2000):3809-3813.

52. Natalya Shcherbinina, "Carbon Nanotube Tips for Atomic Force Microscopy," 28 June 2001; http://cmliris.harvard.edu/html_natalya/research/probes/tip.html

53. E.W. Krahe, R. Mattes, K.-F. Tebbe, H.G. v. Schnering, G. Fritz, "Formation of organosilicons compounds. 47. The crystal and molecular structure of 1,3,5,7-tetramethyl-tetrasila-adamantane," Z. Anorg. Allg. Chem. 393(1972):74-80. In German.

54. R. Mattes, "Structure of organosilicon compounds. III. Vibrational spectra of 1,3,5,7-tetraethyl-silaadamantane," J. Molec. Structure 16(April 1973):53-58. In German.

55. G. Fritz, G. Marquardt, "Formation of organosilicons compounds. LIII. Novel carbosilanes by pyrolysis of Si(CH3)4 and their isolation," Z. Anorg. Allg. Chem. 404(March 1974):1-37. In German.

56. Stephan Pawlenko, Organosilicon Chemistry, Walter de Gruyter, New York, 1986.

57. Cecil L. Frye, Jerome M. Kosowski, Donald R. Weyenberg, "1,3,5,7-tetrasilaadamantanes. A facile synthesis via catalyzed ligand redistribution," J. Am. Chem. Soc. 92(21 October 1970):6379-6380.

58. M. Anthony McKervey, John J. Rooney, "Catalytic routes to adamantane and its homologues," in George A. Olah, ed., Cage Hydrocarbons, John Wiley & Sons, New York, 1990, pp. 39-64.

59. R.C. Bingham, P.v.R. Schleyer, Chemistry of Adamantanes: Recent Developments in the Chemistry of Adamantane and Related Polycyclic Hydrocarbons, Springer-Verlag, New York, 1971.

60. Raymond C. Fort, Jr., Adamantane: The Chemistry of Diamond Molecules, Marcel Dekker, New York, 1976.

61. Evgenii Ignatevich Bagrii, Adamantany: Poluchenie, Svoistva, Primenenie (Adamantanes: Preparation, Properties, and Application), Nauka Press, Moscow, 1989.

62. George A. Olah, ed., Cage Hydrocarbons, John Wiley & Sons, New York, 1990.

63. Paul von Rague Schleyer, "My thirty years in hydrocarbon cages: From adamantane to dodecahedrane," in George A. Olah, ed., Cage Hydrocarbons, John Wiley & Sons, New York, 1990, pp. 1-38.

64. W.D. Graham, P. von R. Schleyer, "Diamond lattice hydrocarbons: spiro[adamantane-2,2’-adamantane]," Tetrahedron Lett. 12(1972):1179-1180.

65. E. Boelema, J. Strating, Hans Wynberg, "Spiro[adamantane-2,2’-adamantane]," Tetrahedron Lett. (1972):1175-1177.

66. W. David Graham, Paul von R. Schleyer, Edward W. Hagaman, Ernest Wenkert, "[2]Diadamantane, the first member of a new class of diamondoid hydrocarbons," J. Am. Chem. Soc. 95(22 August 1973):5785-5786.

67. Chris A. Cupas, Paul von R. Schleyer, David J. Trecker, "Congressane," J. Am. Chem. Soc. 87(20 February 1965):917-918.

68. Van Zandt Williams, Jr., Paul von R. Schleyer, Gerald Jay Gleicher, Lynn B. Rodewald, "Triamantane," J. Am. Chem. Soc. 88(20 August 1966):3862-3863.

69. O. Vogl, B.C. Anderson, D.M. Simons, "Synthesis of hexaoxadiamantanes," Tetrahedron Lett. (1966):415-418.

70. William Burns, Thomas R.B. Mitchell, M. Anthony McKervey, John J. Rooney, George Ferguson, Paul Roberts, "Gas-phase reactions on platinum. Synthesis and crystal structure of anti-tetramantane, a large diamondoid fragment," J. Chem. Soc. Chem. Commun. (1976):893-895.

71. William Burns, M. Anthony McKervey, Thomas R.B. Mitchell, John J. Rooney, "A new approach to the synthesis of diamondoid hydrocarbons: synthesis of anti-tetramantane," J. Am. Chem. Soc. 100(1978):906-911.

72. Mingzuo Shen, Henry F. Schaefer III, Congxing Liang, Jenn-Huei Lii, Norman L. Allinger, Paul von Rague Schleyer, "Finite Td symmetry models for diamond: From adamantane to superadamantane (C35H36)," J. Am. Chem. Soc. 114(1992):497-505.

73. Chris A. Cupas, Leonard Hodakowski, "Iceane," J. Am. Chem. Soc. 96(10 July 1974):4668-4669.

74. David P.G. Hamon, Garry F. Taylor, "A Synthesis of Tetracyclo[5,3,1,1^2.6,0^4.9]dodecane (Iceane)," Tetrahedron Lett. (1975):155-158; David P.G. Hamon, Garry F. Taylor, "A Synthesis of Tetracyclo[5,3,1,1^2.6,0^4.9]dodecane (Iceane)," Aust. J. Chem. 29(1976):1721-1734.

75. David P.G. Hamon, Colin L. Raston, Garry F. Taylor, Jose N. Varghese, Allan H. White, "Crystal Structure of Tetracyclo[5,3,1,1^2.6,0^4.9]dodecane (Iceane)," Aust. J. Chem. 30(1977):1837-1840.

76. P.D. Ownby, "First commercial source of hexagonal diamond powder," submitted, 2004; http://www.umr.edu/~ownby/publications.html

77. S. Fahy, S.G. Louie, "High-pressure structural and electronic properties of carbon," Phys. Rev. B 36(1987):3373-3385.

78. G. Laqua, H. Musso, W. Boland, R. Ahlrichs, "Force Field Calculations (MM2) of Carbon Lattices," J. Am. Chem. Soc. 112(1990):7391-7392.

79. P.D. Ownby, X. Yang, J. Liu, "Calculated x-ray diffraction data for diamond polytypes," J. Amer. Ceramic Soc. 75(1992):1876-1883.

80. Pierluigi Mercandelli, Massimo Moret, Angelo Sironi, "Molecular mechanics in crystalline media," Inorg. Chem. 37(1998):2563-2569.

81. M. Nomura, P. von R. Schleyer, A.A. Arz, "Alkyladamantanes by rearrangement from diverse starting materials," J. Am. Chem. Soc. 89(5 July 1967):3657-3659.

82. Zhu Feng, Marilee Brewer, Ian Brown, Kyriakos Komvopoulos, "Pretreatment process for forming a smooth surface diamond film on a carbon-coated substrate," U.S. Patent 5,308,661, 3 May 1994.

83. A.A. Morrish, Pehr E. Pehrsson, "Effects of surface pretreatments on nucleation and growth of diamond films on a variety of substrates," Appl. Phys. Lett. 59(22 July 1991):417-419; http://content.aip.org/APPLAB/v59/i4/417_1.html

84. Huimin Liu, David S. Dandy, "Studies on nucleation process in diamond CVD: An overview of recent developments," Diam. Relat. Mater. 4(1995):1173-1188; http://navier.engr.colostate.edu/pubs/DRM1-Full.pdf. See also: Huimin Liu, David S. Dandy, Diamond Chemical Vapor Deposition: Nucleation and Early Growth Stages, Noyes Publications, Park Ridge, New Jersey, 1995; http://navier.engr.colostate.edu/pubs/BookSummary.pdf

85. Paul W. May, "Diamond thin films: a 21st-century material," Phil. Trans. R. Soc. Lond. A 358(2000):473-495; http://www.chm.bris.ac.uk/pt/diamond/pdf/rscreview.pdf

86. Wei Zhu, Peichun Yang, Jeffrey T. Glass, "Method of fabricating oriented diamond films on nondiamond substrates and related structures," U.S. Patent 5,449,531, 12 September 1995.

87. Hiromu Shiomi, Naoji Fujimori, "Method for producing single crystal diamond film," U.S. Patent 5,387,310, 7 February 1995.

88. B. Lux, R. Haubner, "Nucleation and growth of low-pressure diamond," in R.E. Clausing, L.L. Horton, J.C. Angus, P. Koidl, eds., Diamond and Diamond-like Films and Coatings, Plenum Press, New York, 1991, p. 579-609.

89. B.V. Spitzyn, L.L. Bouilov, B.V. Derjaguin, J. Crystal Growth 52(1981):219.

90. J.W. Kim, Y.J. Baik, K.Y. Eun, in Y. Tzeng, M. Yoshikawa, M. Murakawa, A. Feldman, eds., Applications of Diamond Films and Related Materials, Elsevier Sci. Publ., New York 1991, p. 399.

91. Takahiro Imai, Naoji Fujimori, "Thin film single crystal diamond substrate," U.S. Patent 4,863,529, 5 September 1989.

92. R. Hultgren, P.D. Desai, D.T. Hawkins, M. Gleiser, K.K. Kelley, Selected Values of the Thermodynamic Properties of Binary Alloys, American Society for Metals, Metals Park, OH, 1973.

93. Robert A. Freitas Jr., Nanomedicine, Volume I: Basic Capabilities, Landes Bioscience, Georgetown, TX, 1999, Appendix A. http://www.nanomedicine.com/NMI/AppendixA.htm

94. H.O. Pierson, Handbook of Carbon, Graphite, Diamond and Fullerenes, Noyes Publications, Park Ridge, New Jersey, 1993; J.E. Field, The Properties of Diamond, Academic Press, London, 1979.

95. Michael J. Mehl, "Tight binding parameters for the elements," U.S. Naval Research Laboratory (NRL), 25 July 2002; http://cst-www.nrl.navy.mil/bind/index.html

96. Robert C. Weast, Handbook of Chemistry and Physics, 49th Edition, CRC Press, Cleveland OH, 1968.

97. E.A. Brandes, Smithells Metals Reference Book, 6th edition, Butterworth & Co, London, 1983.

98. "WebElements" website, 2003; http://www.webelements.com/webelements/elements/text/<Symbol>/heat.html

99. "Sapphire Properties Table," MarkeTech International Inc., Port Townsend WA, 21 February 2002; http://www.mkt-intl.com/sapphires/sapphphotos.htm

100. S.M. Sze, Physics of Semiconductor Devices, Wiley Interscience Publications, New York, 1981, pp. 848-849; http://www.veeco.com/learning/learning_lattice.asp

101. http://www.impex-hightech.de/Quartz.html; http://www.argusinternational.com/quartz.html

102. J. Shackelford, W. Alexander, The CRC Materials Science and Engineering Handbook, CRC Press, Boca Raton FL, 1992.

103. http://www.accuratus.com/silinit.html

104. http://www.lucasmilhaupt.com/htmdocs/brazing_support/everything_about_brazing/materials_comp_chart.html

105. M.E. Bachlechner, A. Omeltchenko, A. Nakano, R.K. Kalia, P. Vashishta, A. Madhukar, P. Messina, "Multimillion-atom molecular dynamics simulation of atomic level stresses in Si(111)/Si3N4(0001) nanopixels," Appl. Phys. Lett. 72(20 April 1998):1969-1971.

106. D.A. Papaconstantopoulos, M.J. Mehl, "First-principles study of superconductivity in high-pressure boron," 3 July 2003; http://arxiv.org/pdf/cond-mat/0111385

107. G.V. Samsonov, The Oxide Handbook, IFI/Plenum Data Corporation, New York, 1973.

108. http://www.umsl.edu/~fraundor/rworld/msa99.pdf

109. A.V. Postnikov, P. Entel, "Ab initio molecular dynamics and elastic properties of TiC and TiN nanoparticles," University of Duisburg, 2003; http://www.thp.uni-duisburg.de/Paper/Postnik/tin_final.pdf

110. Toshimichi Ito, "Method for synthesis of diamond and apparatus therefor," U.S. Patent 4,869,924, 26 September 1989.

111. K. Kobayashi, M. Kumagai, S. Karasawa, T. Watanabe, F. Togashi, J. Cryst. Growth 128(1993):408.

112. S.J. Lin, S.L. Lee, J. Hwang, C.S. Chang, H.Y. Wen, "Effects of local facet and lattice damage on nucleation of diamond grown by microwave plasma chemical vapor deposition," Appl. Phys. Lett. 60(30 March 1992):1559-1561; http://content.aip.org/APPLAB/v60/i13/1559_1.html

113. K. Hirabayashi, Y. Taniguchi, O. Takamatsu, T. Ikeda, K. Ikoma, N. Iwasaki-Kurihara, "Selective deposition of diamond crystals by chemical vapor deposition using a tungsten-filament method," Appl. Phys. Lett. 53(7 November 1988):1815-1817; http://content.aip.org/APPLAB/v53/i19/1815_1.html

114. S. Aisenberg, R. Chabot, "Ion-beam deposition of thin films of diamondlike carbon," J. Appl. Phys. 42(1971):2953.

115. V.E. Strelnitskii, I.I. Aksenov, S.I. Vakula, V.G. Padakula, Sov. Phys. Tech. Phys. 23(1978):222.

116. Mutsukazu Kamo, Seiichiro Matsumoto, Yoichiro Sato, Nobuo Setaka, "Method for synthesizing diamond," U.S. Patent 4,434,188, 28 February 1984.

117. Seiichiro Matsumoto, Mototsugu Hino, Yusuke Moriyoshi, Takashi Nagashima, Masayuki Tsutsumi, "Method for synthesizing diamond by using plasma," U.S. Patent 4,767,608, 30 August 1988.

118. John W. Rabalais, Srinandan R.Kasi, "Chemically bonded diamond films and method for producing same," U.S. Patent 4,822,466, 18 April 1989.

119. T.P. Ong, Fulin Xiong, R.P.H. Chang, C.W. White, "Nucleation and growth of diamond on carbon-implanted single crystal copper surfaces," J. Mater. Res. 7(September 1992):2429-2439; http://www.mrs.org/publications/jmr/jmra/1992/sep/P02429.PDF

120. Shuji Iino, Hideo Hotomi, Izumi Osawa, Mitsutoshi Nakamura, "Photosensitive member with hydrogen-containing carbon layer," U.S. Patent 4,743,522, 10 May 1988.

121. Kiyoshi Morimoto, Toshinori Takagi, "Ion beam deposition apparatus," U.S. Patent 4,559,901, 24 December 1985.

122. Karin Larsson, "Migration of species on a diamond (111) surface," in J.L. Davidson, W.D. Brown, A. Gicquel, B.V. Spitsyn, J.C. Angus, eds., Proc. Fifth Intl. Symp. on Diamond Materials, The Electrochemical Society, Pennington, NJ, 1998, pp. 247-253.

123. S.P. Mehandru, Alfred B. Anderson, "Adsorption of H, CH3, CH2 and C2H2 on 2 x 1 restructured diamond (100)," Surf. Sci. 248(June 1991):369-381.

124. Michael Frenklach, Sergei Skokov, "Surface migration in diamond growth," J. Phys. Chem. 101(1997):3025-3036.

125. E.J. Dawnkaski, D. Srivastava, B.J. Garrison, "Time dependent Monte Carlo simulations of H reactions on the diamond (001)(2×1) surface under chemical vapor deposition conditions," J. Chem. Phys. 102(15 June 1995):9401-9411; http://galilei.chem.psu.edu/pdf/147bjg.pdf

126. Frank Jansen, Mary A. Machonkin, "Processes for the preparation of polycrystalline diamond films," U.S. Patent 4,925,701, 15 May 1990.

127. Allen R. Kirkpatrick, "Diamond films and method of growing diamond films on nondiamond substrates," U.S. Patent 5,082,359, 21 January 1992.

128. Michael J. Ulczynski, Donnie K. Reinhard, Jes Asmussen, "Process for depositing adherent diamond thin films," U.S. Patent 5,897,924, 27 April 1999.

129. Japanese Patent Application, Abstract No. 2138395, June 1987; cited in Frank Jansen, Mary A. Machonkin, "Processes for the preparation of polycrystalline diamond films," U.S. Patent 4,925,701, 15 May 1990.

130. Bing-Kun Yu, "Preparation of diamond and diamond-like thin films," U.S. Patent 5,273,788, 28 December 1993.

131. James Chickos, Donald Hesse, Sarah Hosseini, Gary Nichols, Paul Webb, "Sublimation enthalpies at 298.15 K using correlation gas chromatography and differential scanning calorimetry measurements," Thermochimica Acta 313(1998):101-110; http://www.umsl.edu/~jscumsl/JSCPUBS/seale.pdf

132. C.W.F.T. Pistorius, H.C. Snyman, Z. Physik Chem. 43(1964):278; C.W.F.T. Pistorius, H.A. Resing, Mol. Cryst. Liq. Cryst. 5(1969):353.

133. Ilham Mokbel, Kvetoslav Ruzicka, Vladimir Majer, Vlastimil Ruzicka, Madeleine Ribeiro, Jacques Jose, Milan Zabransky, "Phase equilibria (v-s, v-l) for three compounds of petroleum interest: 1-phenyldodecane, (5a)-cholestane, adamantane," Fluid Phase Equilibria 169(2000):191-207.

134. Sang Tae Park, Sang Kyu Kim, Myung Soo Kim, "Observation of conformation-specific pathways in the photodissociation of 1-iodopropane ions," Nature 415(17 January 2002):306-308.

135. G. Lucovsky, P. D. Richards, R. J. Markura, paper presented at Workshop of Dielectric Systems for III-V Semiconductors, San Diego CA, 26-27 May 1984.

136. P.D. Richard, R.J. Markunas, G. Lucovsky, G.G. Fountain, A.N. Mansour, D.V. Tsu, "Remote plasma enhanced CVD deposition of silicon nitride and oxide for gate insulators in (In, Ga)AS FET devices," J. Vac. Sci. Technology A 3(May-June 1985):867-872.

137. Chandra V. Desphandey, Rointan F. Bunshah, Hans J. Doerr, "Process for making diamond, doped diamond, diamond-cubic boron nitride composite films," U.S. Patent 4,816,291, 28 March 1989.

138. S.W. Hla, L. Bartels, G. Meyer, K.-H. Rieder, "Inducing all steps of a chemical reaction with the scanning tunneling microscop tip: Towards single molecule engineering," Phys. Rev. Lett. 85(2000):2777-2780; http://www.phy.ohiou.edu/~hla/7.pdf

139. Saw-Wai Hla, Gerhard Meyer, Karl-Heinz Rieder, "Inducing single-molecule chemical reactions with a UHV-STM: A new dimension for nano-science and technology," Chem. Phys. Chem. 2(2001):361-366; http://plato.phy.ohiou.edu/~hla/HLA2001-1.pdf

140. Saw-Wai Hla, Karl-Heinz Rieder, "STM control of chemical reactions: single-molecule synthesis," Annu. Rev. Phys. Chem. 54(June 2003):307-330; http://www.phy.ohiou.edu/~hla/HLA-annualreview.pdf

141. William L. Masterton, Emil J. Slowinski, Chemical Principles, Second Edition, W.B. Saunders Co., Philadelphia, PA, 1969, pp. 215, 299.

142. Robert A. Freitas Jr., Nanomedicine, Volume IIA: Biocompatibility, Landes Bioscience, Georgetown, TX, 2003, p. 64. http://www.nanomedicine.com/NMIIA/15.3.3.3.htm#p6

143. R. Taylor, D.R.M. Walton, "The chemistry of fullerenes," Nature 363(1993):685.

144. F. Diederich, C. Thilgen, "Covalent fullerene chemistry," Science 271(19 January 1996):317-323.

145. B. Ni, S.B. Sinnott, "Chemical functionalization of carbon nanotubes through energetic radical collisions," Phys. Rev. B 61(2000):R16343-R16346.

146. Sarbajit Banerjee, Stanislaus S. Wong, "Functionalization of Carbon Nanotubes with a Metal-Containing Molecular Complex," Nano Letters 2(2002):49-53.

147. Keun Soo Kim, Kyung Ah Park, Hyun Jin Kim, Dong Jae Bae, Seong Chu Lim, Young Hee Lee, Jae Ryong Kim, Ju-Jin Kim, Won Bong Choi, "Band Gap Modulation of a Carbon Nanotube by Hydrogen Functionalization," J. Korean Phys. Soc. 42(February 2003):S137-S142; http://nanotube.skku.ac.kr/data/paper/KSKim_JKPS.pdf

148. Sarbajit Banerjee, Michael G.C. Kahn, Stanislaus S. Wong, "Rational Chemical Strategies for Carbon Nanotube Functionalization," Chem. Eur. J. 9(2003):1898-1908.

149. Frank J. Owens, Zafar Iqbal, "Electrochemical Functionalization Of Carbon Nanotubes With Hydrogen," 23rd Army Science Conference, Session L Poster Summaries: Nanotechnology, LP-11, 2002; http://www.asc2002.com/summaries/l/LP-11.pdf

150. J.L. Bahr, J.P. Yang, D.V. Kosynkin, M.J. Bronikowski, R.E. Smalley, J.M. Tour, "Functionalization of carbon nanotubes by electrochemical reduction of aryl diazonium salts: A bucky paper electrode," J. Am. Chem. Soc. 123(2001):6536-6542; http://smalley.rice.edu/rick’s publications/JACS123-6536.pdf

151. Kevin D. Ausman, Henry W. Rohrs, MinFeng Yu, Rodney S. Ruoff, "Nanostressing and mechanochemistry," Nanotechnology 10(September 1999):258-262.

152. X. Lu, X. Xu, N. Wang, Q. Zhang, "A DFT study of the 1,3-dipolar cycloadditions on the C(100)-2 x 1 surface," J. Org. Chem. 67(25 January 2002):515-520.

153. R. W. Hoffmann, Dehydrobenzene and Cycloalkynes, Verlag Chemie-Academic Press, New York, 1967.

154. Richard Jaffe, Jie Han, Al Globus, "Formation of Carbon Nanotube Based Gears: Quantum Chemistry and Molecular Mechanics Study of the Electrophilic Addition of o-Benzyne to Fullerenes, Graphene, and Nanotubes," First Electronic Molecular Modeling & Graphics Society Conference, 1996; http://www.nas.nasa.gov/Groups/Nanotechnology/publications/MGMS_EC1/quantum/paper.html

155. Al Globus, Richard Jaffe, "NanoDesign: Concepts and Software for a Nanotechnology Based on Functionalized Fullerenes," First Electronic Molecular Modeling & Graphics Society Conference, 1996; http://www.nas.nasa.gov/Groups/Nanotechnology/publications/MGMS_EC1/NanoDesign/paper.html

156. R.J. Hamers, S.K. Coulter, M.D. Ellison, J.S. Hovis, D.F. Padowitz, M.P. Schwartz, C.M. Greenlief, J.N. Russell Jr., "Cycloaddition chemistry of organic molecules with semiconductor surfaces," Acc. Chem. Res. 33(September 2000):617-624.

157. Yoshikatsu Namba, "Attempt to grow diamond phase carbon films from an organic solution," J. Vac. Sci. Technol. A 10(September/October 1992):3368-3370.

158. P. Aublanc, V.P. Novikov, L.V. Kuznetsova, M. Mermoux, "Diamond synthesis by electrolysis of acetates," Diam. Rel. Mater. 10(March-July 2001):942-946.

159. Steven H. Hoke, Jay Molstad, Dominique Dilettato, Mary Jennifer Jay, Dean Carlson, Bart Kahr, R. Graham Cooks, "Reaction of Fullerenes and Benzyne," J. Org. Chem. 57(11 September 1992):5069-5071.

160. M.S. Meier, G.W. Wang, R.C. Haddon, C.P. Brock, M.A. Lloyd, J.P. Selegue, "Benzyne adds across a closed 5-6 ring fusion in C70: Evidence for bond delocalization in fullerenes," J. Am. Chem. Soc. 120(1998):2337-2342

161. L. Giraud, V. Huber, T. Jenny, "2,2-Divinyladamantane: a new substrate for the modification of silicon surfaces," Tetrahedron 54(1998):11899-11906.

162. E. Leroy, O. M. Kuttel, L. Schlapbach, L. Giraud, T. Jenny, "Chemical vapor deposition of diamond growth using chemical precursor," Appl. Phys. Lett. 73(24 August 1998):1050-1052.

163. Anne Giraud, Titus Jenny, Eric Leroy, Olivier M. Kuttel, Louis Schlapbach, Patrice Vanelle, Luc Giraud, "Chemical nucleation for CVD diamond growth," J. Am. Chem. Soc. 123(2001):2271-2274.

164. Liliana Dumitrescu Buforn, Eberhard Blank, "Diamond nucleation on chemically modified silicon using HFCVD," Paper G/PII.35, Session X: Multilayer Coatings, Symposium G, E-MRS Spring Meeting 2003, 10-13 June 2003; http://www-emrs.c-strasbourg.fr/2003SPRING/2003ABSTRACTS/2003_G_ABS.PDF

165. L.H. Sommer, E.W. Pietrusza, F.C. Whitmore, J. Am. Chem. Soc. 69(1947):188.

166. C. Chatgilialoglu, "Organosilanes as radical-based reducing agents in synthesis," Acc. Chem. Res. 25(1992):188-194.

167. B. Kopping, C. Chatgilialoglu, M. Zehnder, B. Giese, J. Org. Chem. 57(1992):3994-4000.

168. W.I. Bevan, R.N. Haszeldine, J. Middleton, A.E. Tipping, J. Chem Soc., Perkin Trans. 1(1974):2305-2309.

169. K. Yamamoto, T. Hayashi, M. Kumada, J. Organomet. Chem. 28(1971):C37-C38.

170. J.L. Speier, "Homogeneous Catalysis Hydrosilation by Transition Metals," in F.G.A. Stones, R. West, eds., Advances in Organometallic Chemistry, Vol. 17, Academic Press, New York, 1979, pp. 407-447.

171. K. Tamao, T. Nakajima, R. Sumiya, H. Arai, N. Higuchi, Y. Ito, J. Am. Chem. Soc. 108(1986):6090-6093.

172. M. Tanaka, Y. Uchimari, H.J. Lautenschlager, Organometallics 10(1991):16-18.

173. L.N. Lewis, J. Stein, R.E. Colborn, Y. Gao, J. Dong, "The chemistry of fumarate and maleate inhibitors with platinum hydrosilylation catalysts," J. Organomet. Chem. 521(1996):221-227.

174. John C. Angus, Phillip W. Morrison, "Diamond Lab," Case Western Reserve University, 1998; http://web.archive.org/web/19981206235656/http://k2.scl.cwru.edu/cse/eche/faculty/angus/diamond.htm

175. K. Bando, K. Kamo, T. Ando, Y. Sato, "Deposition of diamond crystal at substrate temperature lower than 500 oC," in Russell Messier, Jeffrey T. Glass, James E. Butler, Rustum Roy, eds., Proceedings of the Second International Conference, New Diamond Science and Technology, Materials Research Society, Pittsburgh, PA, 1991, pp. 467-472.

176. Argonne National Laboratory, "Diamond Films for Microelectromechanical Systems (MEMS)"; http://www.techtransfer.anl.gov/techtour/diamondmems.html

177. M. Tomellini, "Evidence for nonclassical nucleation at solid surfaces in diamond deposition from the gas phase," J. Mater. Res. 8(July 1993):1596-1604.

178. A.R. Badzian, R.C. DeVries, "Crystallization of diamond from the gas phase: Part I," Mater. Res. Soc. Bull. 23(1988):385-400.

179. A.R. Badzian, T. Badzian, R. Roy, R. Messier, K.E. Spear, "Crystallization of diamond crystals and films by microwave assisted CVD (Part II)," Mater. Res. Soc. Bull. 23(April 1988):531-548.

180. P. Badziag, W.S. Verwoerd, W.P. Ellis, N.R. Greiner, "Nanometre-sized diamonds are more stable than graphite," Nature 343(1990):244-245.

181. Burak Atakan, Karsten Lummer, Katharina Kohse-Hoinghaus, "Diamond deposition in acetylene-oxygen: nucleation and early growth on molybdenum substrates for different pretreatment procedures," Phys. Chem. Chem. Phys. 1(1999):3151-3156; http://www.rsc.org/ej/CP/1999/F9901945.PDF

182. D.S. Knight, W.B. White, "Characterization of diamond films by Raman spectroscopy," J. Mater. Res. 4(March-April 1989):385-393.

183. L. Fayette, M. Mermoux, B. Marcus, F. Brunet, P. Germi, M. Pernet, L. Abello, G. Lucazeau, J. Garden, "Analysis of the fine structure of the Raman line and x-ray reflection profiles for textured CVD diamond films," Diam. Rel. Mater. 4(1995):1243-1250.

184. F.P. Bundy, J.S. Kasper, "Hexagonal diamond – A new form of carbon," J. Chem. Phys. 46(1 May 1967):3437-3446.

185. R.E. Hanneman, H.M. Strong, F.P. Bundy, "Hexagonal diamonds in meteorites: implications," Science (24 February 1967):995-997.

186. Hongliang He, T. Sekine, T. Kobayashi, "Direct transformation of cubic diamond to hexagonal diamond," Appl. Phys. Lett. 81(22 July 2002):610-612; http://content.aip.org/APPLAB/v81/i4/610_1.html

187. D.V. Fedoseev, V.L. Bukhovets, I.G. Varshavskaya, A.V. Lavrentev, B.V. Derjaguin, "Transition of graphite into diamond in a solid phase under the atmospheric pressure," Carbon 21(1983):237-241.

188. B.K. Roul, B.B. Nayak, P.K. Mishra, B.C. Mohanty, "Diamond and diamond-like-carbon growth on Si(100) by hot filament-assisted RF plasma CVD," J. Mater. Synth. Proc. 7(1999):281-288.

189. S. Bhargava, H.D. Bist, S. Sahli, M. Aslam, H.B. Tripathi, "Diamond polytypes in the chemical vapor deposited diamond films," Appl. Phys. Lett. 67(December 1995):1706-1708.

190. J.C. Angus, F.A. Buck, M. Sunkara, T.F. Groth, C.C. Hayman, R. Gat, "Diamond growth at low pressures," MRS Bull. 1989(October 1989):38-47.

191. John C. Angus, Cliff C. Hayman, “Low-pressure, metastable growth of diamond and "diamondlike’ phases," Science 241(19 August 1988):913-921.

192. R.E. Clausing, L. Heatherly, K.L. More, G.M. Begun, "Electron microscopy of the growth features and crystal structures of filament-assisted CVD diamond films," Surf. Coatings Technol. 39/40(1989):199-210.

193. B.E. Williams, J.T. Glass, R.F. Davis, K. Kobashi, "The analysis of defect structures and substrate/film interfaces of diamond thin films," J. Cryst. Growth 99(1990):1168-1176.

194. Keiji Hirabayashi, Noriko Iwasaki Kurihara, Naoto Ohtake, Masanori Yoshikawa, "Size dependence of morphology of diamond surfaces prepared by DC arc plasma jet chemical vapor deposition," Jpn. J. Appl. Phys. 31(February 1992):355-360.

195. R.C. DeVries, "Synthesis of diamond under metastable conditions," Annu. Rev. Mater. Sci. 17(1987):161-176.

196. E. Burgos, E. Halac, H. Bonadeo, "A semi-empirical potential for the statics and dynamics of covalent carbon systems," Chem. Phys. Lett. 298(18 December 1998):273-278.

197. L.V.Zhigilei, D.Srivastava, and B.J.Garrison, "Intermediate metastable structure of the C{111}/(1×1)H-C{111}/(2×1) surface phase transition," Phys. Rev. B 55(1997):1838-1843; http://galilei.chem.psu.edu/pdf/155bjg.pdf

198. B.E. Williams, J.T. Glass, R.F. Davis, K. Kobashi, K.L. More, in J.P. Dismukes, ed., Proc. First Intl. Symp. On Diamond and Diamond-like Films, Electrochemical Society, New York, 1989, p. 202.

199. John C. Angus, Alberto Argoitia, Roy Gat, Zhidan Li, Mahendra Sunkara, Long Wang, Yaxin Wang, "Chemical vapour deposition of diamond," in A. Lettington, J.W. Steeds, eds., Thin Film Diamond, Chapman and Hall, London, 1994, pp. 1-14; see also in: Phil. Trans. R. Soc. London A 342(1993):195-208.

200. C. Wild, R. Kohl, N. Herres, W. Muller-Sebert, P. Koidl, "Oriented CVD diamond films: twin formation, structure and morphology," Diam. Rel. Mater. 3(April 1994):373-381.

201. K.E. Spear, M. Frenklach, in K.E. Spear, J.P. Dismukes, eds., Synthetic Diamond: Emerging CVD Science and Technology, John Wiley & Sons, New York, 1993, pp. 243-304.

202. H. Sowa, E. Koch, "A proposal for a transition mechanism from the diamond to the lonsdaleite type," Acta Crystallogr. A 57(July 2001):406-413.

203. C.C. Battaile, D.J. Srolovitz, I.I. Oleinik, D.G. Pettifor, A.P. Sutton, S.J. Harris, J.E. Butler, "Etching effects during the chemical vapor deposition of (100) diamond," J. Chem. Phys. 111(1 September 1999):4291-4299; http://www.princeton.edu/~pmi/srolgroup/publications/JCP04291.pdf

204. K.E. Spear, "Diamond, ceramic coating of the future," J. Am. Ceram. Soc. 72(1989):171-191.

205. Andrzej Badzian, Teresa Badzian, "Diamond homoepitaxy by chemical vapor deposition," Diam. Rel. Mater. 2(31 March 1993):147-157.

206. C. Wild, P. Koidl, W. Muller-Sebert, H. Walcher, R. Kohl, N. Herres, R. Locher, R. Samlenski, R. Brenn, "Chemical vapour deposition and characterization of smooth {100}-faceted diamond films," Diam. Rel. Mater. 2(1993):158-168.

207. G. Janssen, J.J. Schermer, W.J.P. van Enckevort, L.J. Giling, "On the occurrence of (113) facets on CVD-grown diamond," J. Cryst. Growth 125(November 1992):42-50.

208. K.A. Snail, Z.P. Lu, R. Weimer, J. Heberlein, E. Pfender, L.M. Hanssen, "Confirmation of (113) facets on diamond grown by chemical vapor deposition," J. Cryst. Growth 137(April 1994):676-679.

209. C.J. Chu, M.P. D’Evelyn, R.H. Hauge, J.L. Margrave, "Mechanism of diamond growth by chemical vapor deposition on diamond (100), (111), and (110) surfaces: Carbon-13 studies," J. Appl. Phys. 70(1 August 1991):1695-1705.

210. C.J. Chu, R.H. Hauge, J.L. Margrave, M.P. D’Evelyn, "Growth kinetics of (100), (110), and (111) homoepitaxial diamond films," Appl. Phys. Lett. 61(21 September 1992):1393-1395.

211. S.S. Lee, D.W. Minsek, D.J. Vestyck, P. Chen, "Growth of diamond from atomic hydrogen and a supersonic free jet of methyl radicals," Science 263(18 March 1994):1596-1598.

212. R.E. Rawles, W.G. Morris, M.P. D’Evelyn, in D.L. Dreifus, A. Collins, T. Humphreys, K. Das, P.E. Pehrsson, eds., Diamond for Electronic Applications, Symp. Proc. 416, Materials Research Society, Pittsburgh, PA, 1996, pp. 13-18.

213. Nigel P. Hacker, George W. Tyndall, III, "Deposition of diamond films," U.S. Patent 4,948,629, 14 August 1990.

214. Richard A. Neifeld, "Method of preparing a thin diamond film," U.S. Patent 4,954,365, 4 September 1990.

215. T.P. Mollart, K.L. Lewis, "Optical-quality diamond growth from CO2-containing gas chemistries," Diam. Relat. Mater. 8(March 1999):236-241.

216. Marcus Asmann, Joachim Heberlein, Emil Pfender, "A review of diamond CVD utilizing halogenated precursors," Diam. Relat. Mater. 8(1 January 1999):1-16.

217. O. Tschauner, H.K. Mao, R.J. Hemley, "New transformations of CO(2) at high pressures and temperatures," Phys. Rev. Lett. 87(13 August 2001):075701.

218. Y. Namba, Jin Wie, T. Mohri, E.A. Heidarpour, "Large grain size thin films of carbon with diamond structure," J. Vac. Sci. Technol. A 7(January-February 1989):36-39.

219. C.V. Burton, "Artificial diamonds," Nature 72(24 August 1905):397.

220. Yadong Li, Yitai Qian, Hongwei Liao, Yi Ding, Li Yang, Cunyi Xu, Fangqing Li, Guien Zhou, "A reduction-pyrolysis-catalysis synthesis of diamond," Science 281(10 July 1998):246-247.

221. S. Feng, R. Xu, "New materials in hydrothermal synthesis," Acc. Chem. Res. 34(March 2001):239-247.

222. Stephen A. Godleski, Paul von Rague Schleyer, Eiji Osawa, Todd Wipke, "The systematic prediction of the most stable neutral hydrocarbon isomer," Prog. Phys. Org. Chem. 13(1981):63-117.

223. Stephen E. Stein, "Diamond and graphite precursors," Nature 346(9 August 1990):517.

224. Donald E. Patterson, Robert H. Hauge, C. Judith Chu, John L. Margrave, "Halogen-assisted chemical vapor deposition of diamond," U.S. Patent 5,071,677, 10 December 1991.

225. P.G. Lurie, J.M. Wilson, "The diamond surface. I. The structure of the clean surface and the interaction with gases and metals," Surf. Sci. 65(1977):453-475.

226. K. Bobrov, B. Fisgeer, H. Shechter, M. Folman, A. Hoffman, "Thermally-programmed desorption (TPD) of deuterium from Di(111) surface: presence of two adsorption states," Diam. Rel. Mater. 6(April 1997):736-742.

227. B.B. Pate, "The diamond surface: Atomic and electronic structure," Surf. Sci. 165(1986):83-142.

228. G.R. Brandes, A.P. Mills, Jr., "Work function and affinity changes associated with the structure of hydrogen-terminated diamond (100) surfaces," Phys. Rev. B 58(15 August 1998):4952-4962.

229. "Zyvex’s S100 Nanomanipulator System," http://www.zyvex.com/Products/S100_Faq.html; "Zyvex Microgrippers," http://www.zyvex.com/Products/Grippers.html

230. Klocke Nanotechnik, “Manipulators: Univeral Tools with 1 Nanometer Resolution,” http://www.nanomotor.de/p_nanomanipulator.htm; “SEM-Manipulators,” http://www.nanomotor.de/pdf/Compare_e_lo.PDF; “Processing Material in Electron Microscopes: Nanomanipulation With Several D.O.F.” http://www.nanomotor.de/aa_processing.htm

231. W.C. Gardiner Jr., Rates and Mechanisms of Chemical Reactions, Benjamin, New York, 1969.

232. W.F. Brunner Jr., T.H. Batzer, Practical Vacuum Techniques, Reinhold Publishing, New York, 1965, p. 124.

233. "Vacuum Pumps," McGraw Hill Encyclopedia of Science and Technology, Vol. 19, 1992, p. 128.

234. Robert A. Freitas Jr., Ralph C. Merkle, Diamond Surfaces and Diamond Mechanosynthesis, Landes Bioscience, Georgetown, TX, 2006. In preparation. http://www.MolecularAssembler.com/DSDM.htm. See also: Robert A. Freitas Jr., Ralph C. Merkle, "A Minimal Toolset for Positional Diamond Mechanosynthesis,"J. Comput. Theor. Nanosci. (2005). Submitted.

235. Jingping Peng, Robert A. Freitas Jr., Ralph C. Merkle, John N. Randall, George D. Skidmore, "Theoretical Analysis of Diamond Mechanosynthesis. Part III. Positional C2 Deposition on Diamond C(110) Surface using Si/Ge/Sn-based Dimer Placement Tools," J. Comput. Theor. Nanosci. (2005). Submitted.

© 2003-2004 Robert Freitas. Reprinted with permission.

Karl Sarnow says:

Welcome to the Xplora scheduled chat with Ray Kurzweil about robots and artificial intelligence.

My name is Karl Sarnow and I am member of the Xplora team, trying to enable teachers of mathematics and science to give fascinating science lessons.

It is a great pleasure and honour to have Ray Kurzweil here in the chat. The topic itself is really fascinating and inspiring. But having one of the early birds of AI applications personally here is nothing more but wonderful.

Cordially welcome Ray.

Ray Kurzweil says:

Hi, glad to be here.

Karl Sarnow says:

We now start with the procedure as discussed in the mail in advance. Ray will start with a short introduction and then the others follow with a short sentence.

Ray Kurzweil says:

Yes, here is something I wrote recently that is a brief introduction to my latest book, The Singularity is Near, When Humans Transcend Biology:

Ray Kurzweil says:

So what is the Singularity?

Within a quarter century, nonbiological intelligence will match the range and subtlety of human intelligence. It will then soar past it because of the continuing acceleration of information-based technologies, as well as the ability of machines to instantly share their knowledge. Intelligent nanorobots will be deeply integrated in our bodies, our brains, and our environment, overcoming pollution and poverty, providing vastly extended longevity, full-immersion virtual reality incorporating all of the senses (like “The Matrix”), "experience beaming” (like “Being John Malkovich”), and vastly enhanced human intelligence. The result will be an intimate merger between the technology-creating species and the technological evolutionary process it spawned.

And that’s the Singularity?

No, that’s just the precursor. Nonbiological intelligence will have access to its own design and will be able to improve itself in an increasingly rapid redesign cycle. We’ll get to a point where technical progress will be so fast that unenhanced human intelligence will be unable to follow it. That will mark the Singularity.

When will that occur?

I set the date for the Singularity—representing a profound and disruptive transformation in human capability—as 2045. The nonbiological intelligence created in that year will be one billion times more powerful than all human intelligence today.

Why is this called the Singularity?

The term “Singularity” in my book is comparable to the use of this term by the physics community. Just as we find it hard to see beyond the event horizon of a black hole, we also find it difficult to see beyond the event horizon of the historical Singularity. How can we, with our limited biological brains, imagine what our future civilization, with its intelligence multiplied trillions-fold, be capable of thinking and doing? Nevertheless, just as we can draw conclusions about the nature of black holes through our conceptual thinking, despite never having actually been inside one, our thinking today is powerful enough to have meaningful insights into the implications of the Singularity. That’s what I’ve tried to do in this book.

Okay, let’s break this down. It seems a key part of your thesis is that we will be able to capture the intelligence of our brains in a machine.

Indeed.

So how are we going to achieve that?

We can break this down further into hardware and software requirements. In the book, I show how we need about 10 quadrillion (1016) calculations per second (cps) to provide a functional equivalent to all the regions of the brain. Some estimates are lower than this by a factor of 100. Supercomputers are already at 100 trillion (1014) cps, and will hit 1016 cps around the end of this decade. Several supercomputers with 1 quadrillion cps are already on the drawing board, with two Japanese efforts targeting already on the drawing board, with two Japanese efforts targeting 10 quadrillion cps around the end of the decade. By 2020, 10 quadrillion cps will be available for around $1,000. Achieving the hardware requirement was controversial when my last book on this topic, The Age of Spiritual Machines, came out in 1999, but is now pretty much of a mainstream view among informed observers. Now the controversy is focused on the algorithms.

And how will we recreate the algorithms of human intelligence?

To understand the principles of human intelligence we need to reverse-engineer the human brain. Here, progress is far greater than most people realize. The spatial and temporal (time) resolution of brain scanning is also progressing at an exponential rate, roughly doubling each year, like most everything else having to do with information. Just recently, scanning tools can see individual interneuronal connections, and watch them fire in real time. Already, we have mathematical models and simulations of a couple dozen regions of the brain, including the cerebellum, which comprises more than half the neurons in the brain. IBM is now creating a simulation of about 10,000 cortical neurons, including tens of millions of connections. The first version will simulate the electrical activity, and a future version will also simulate the relevant chemical activity. By the mid 2020s, it’s conservative to conclude that we will have effective models for all of the brain.

So at that point we’ll just copy a human brain into a supercomputer?

I would rather put it this way: At that point, we’ll have a full understanding of the methods of the human brain. One benefit will be a deep understanding of ourselves, but the key implication is that it will expand the toolkit of techniques we can apply to create artificial intelligence. We will then be able to create nonbiological systems that match human intelligence in the ways that humans are now superior, for example, our pattern recognition abilities. These superintelligent computers will be able to do things we are not able to do, such as share knowledge and skills at electronic speeds.

By 2030, a thousand dollars of computation will be about a thousand times more powerful than a human brain. Keep in mind also that computers will not be organized as discrete objects as they are today. There will be a web of computing deeply integrated into the environment, our bodies and brains.

Okay, that should do it. Sorry it had to be cut into a number of pieces.

Karl Sarnow says:

Okay, lets start with an introduction. I propose in the order people appear on the right. So Alexa will be the first one.

Alexa Joyce says:

I’m Alexa Joyce, I’m project manager of Xplora. I’m a biologist originally but now working in technology.

Donelle Batty says:

Hello I’m Tom Steele and I’m a student from Riverside High School from Tasmania. I’m be representing the Pegasus project during this chat

Donelle Batty says:

(sorry I may be abit tired it is 1AM over here)

Karl Sarnow says to Donelle Batty:

Youre welcome Tom. A real student is fine. James now?

Ray Kurzweil says:

Where are you?

James Whipple says:

Hi I’m James, I’m a design student who has been interested in the singularity for several years.

Damon Zucconi says:

Sorry I was reading… I’m Damon Zucconi, a student at Maryland Institute College of Art in Interactive Media.

Karl Sarnow says:

I am a physicist with a PhD in biophysics, teaching mathematics, physics and computer science at a German Gymnasium for about 30 years. Now I am seconded to the European Schoolnet to help setup Xplora.

Matt Neil says:

I’m Matt I am in Australia 2, I work in technology as a IT Architect designing systems for large corporates, the organisation I work for is busy building a grid mesh for global computing. In my last role I was working for a health organising changing the face of radiology from traditional photograpghy to digital and I have a passion for what we are talking about

Ray Kurzweil says:

I’m Ray Kurzweil, an inventor, author, and futurist. Delighted to be with all of you, including all the students "listening" in.

Sally H says:

I am Sally. I am in Loughborough, UK. I work in IT (eLearning) and Home Educate my 15 year old son. He is very interested in robots, the future of computing, as well as computer gaming!

Karl Sarnow says:

Ok, now we know who you are (almost).

Karl Sarnow says:

Now for the questions, only one per person please.

Karl Sarnow says:

(for this round).

Donelle Batty says:

Maybe I should start with a simple one

Karl Sarnow says:

Just go ahead

James Whipple says:

i will ask after Donelle’s question

Donelle Batty says:

Some of the students wish to know what inspired you to create all of the fantastic things you’ve created and they also wish to know what troubles you encountered as they wish to be inventors themselves.

Ray Kurzweil says:

I’ve had the idea of being an inventor since I was 5. What is exciting and inspiring about being an inventor is the link between dry formulas on a black board (the invention) and transformations in people’s lives. As for challenges, the biggest issue is timing. Most inventors get their inventions to "work" but most of the time the timing is wrong, so it fails in the market place. That’s why I got into tracking technology trends over 30 years ago.

Ray Kurzweil says:

One more comment on Donnelle’s question: the most gratifying project I’ve been involved in has been reading machines for the blind. I introduced the first one 30 years ago, and have been involved ever since. We just introduced a print-to-speech reading machine that fits in your pocket—it’s 10,000 times smaller and lighter than the first one.

Donelle Batty says:

Thanks.

Matt Neil says:

Ray my question is around how we are going to transfer/copy/replicate those already higher congnitive functions of the mind, such as ESP that may actually lay dormant in the mind, that we do not understand and from observation may look like garbage as they are not normally activated. Are these functions inherent? And if we copy the code will these extras services come along with it?

Ray Kurzweil says:

WRG Matt’s question, we’re now embarked on a grand project to "reverse-engineer" the human brain. We’re in the early stages but the progress will exponential not linear. So we will ultimately understand in precise terms how the brain performs its functions, including ESP to the extent that it can actually do that. We will routinely have brain-to-brain communication when we have nanobots in our brains that are on the Internet. We already have simulations of 20 regions of the brain that perform well on tests compared to human function, for example, the cerebellum which comprises more than half the neurons in the brain, and 15 regions of the auditory cortext. There are several hundred more regions to go. (end of response to Matt).

Matt Neil says to Ray Kurzweil:

Cheers

James Whipple says:

Will there be room for the human ego in a post-singularity society, or will we be led to less and less individuation as our interconnections grow? How much of the brain’s baggage will we want to take with us as we integrate with machines?

Ray Kurzweil says:

WRG James’ question, I suppose we’ll have to reconsider what ego means. It’s not always a bad thing if people are driven to perform creative work. Machines can have it both ways—they can be individuals and they can also merge to form one larger intelligence. Humans "merge" also in societies but not with the same ease. We’ll still have ego and conflicting agendas but we’ll have more capability.

James Whipple says:

Interesting, reminds me of Howard Bloom’s "Global Brain" book. Thanks, Ray.

Ray Kurzweil says:

As for the brain’s baggage—we will ultimately have the "source code" for our intelligence—we’ll have a precise description of its algorithms and be able to modify them. We’ll have to proceed with caution, of course, but there will be obvious dysfunctions we’d like to fix. (end of response to James).

Sally H says:

With the known vulnerabilities in current O/S software and computer security still being a nightmare—do you think that this vision of the future could be scarily nightmarish, and can you see a way that this would be countered?

Ray Kurzweil says:

WRG Sally’s question, there are definitively downsides to all 3 overlapping revolutions—G (genetics or biotech), N (nanotech) and R (robotics which really refers to strong AI, AI at the human level). We have the downside of "G" now—the potential for bioengineered biological viruses. I wrote (coauthored with Bill Joy) an op-ed piece in the New York Times recently criticizing the US Goverment for publishing the genome of the 1918 flu virus, for example, as it could aid bioterrorists. WRG software, I actually think we’re doing reasonably well. We have "mission critical" software running intensive care units in hospitals, flying and landing airplanes, running factories, etc., and this software almost nevere fails. We do know how to create reliable software. And with regard to software pathogens (software viruses, etc.), we’re also reasonably keeping pace. Our technological "immune system" responds generally within hours of a new type of attack. (end of response to Sally).

Karl Sarnow says:

You mention “We’ve already created simulations of ~ 20 regions (out of several hundred) of the brain”. Do you mean computer programs, that behave like the brain regions? How can you test that? Is there any interface to living beings?

Ray Kurzweil says:

WRG Karl’s question, yes these are computer programs—computer simulations and yes, these are tested—for example by applying psychoacoustic tests to the simulation and applying the same tests to human auditory perception. It does not prove that the simulations are perfect, and undoubtedly they are not, but it shows we are moving in the right direction.

Ray Kurzweil says:

WRG interfacing to living beings, we have of course a variety of neural implants now—cochlear implants, and implants for Parkinson’s patients which replaced diseased brain tissue. The latest generation of the Parkinson’s implant allows you to download new software to your neural implant from outside the patient. There are stroke patients who have an implant that can now communicate to their computer and by extension to the rest of the world and to control their environment. (end)

Alexa Joyce says:

yes exactly I’m rather worried to read about the re-creation of the live virus by bio-engineering too

Karl Sarnow says:

Did everybody have his question?

Matt Neil says to Ray Kurzweil:

Not a question but an interuption—are you using speech to text for todays session?

Jan Kapoun says:

Hi, Ray. I was interested in Martin Rees´ book "Our Final Hour", especially the part about dangers of new technology. Have you read the book? What do you think about it?

Ray Kurzweil says:

WRG Kapound’s question:

Ray Kurzweil says:

I alluded earlier to the downsides—the perils—of GNR. I’m familiar with Rees’ book. He talks about these perils as well as natural ones, like an asteroid hitting Earth. On that last one, this happens infrequently (at least a big one) and I’m confident we’ll have the technology to blast it out of the sky before that happens. The more daunting challenges are the downsides of the self-replicating technology we’re creating. I mentioned bioterrorists creating a modified biological virus. When we have full nanotechnology manufacturing, there will ultimately be the potential for self-replicating nanotechnology. There are strategies for dealing with these issues. The issue for society is one of priorities. We need to put a higher priority on the defenses—I gave testimony to the U.S. Congress on a proposal for a $100 billion program to develop new technologies (like RNA interference) to combat new biological viruses. President Bush recently proposed a $7 billion program for this—it’s a start but not enough. We also need to be smart about not disseminating overtly dangerous information. No one proposes putting the design of an atom bomb on the web, so why put the design of a killer virus there? (end)

Jan Kapoun says:

Thank you. But is it certain that we will avoid a situation, such as the example in Michael Crichton´s Prey?

Ray Kurzweil says:

Now with regard Crichton’s prey.

Ray Kurzweil says:

He was writing about self-replicating nanotech, although he mixed it up with biological viruses and there are some scientifically unrealistic aspects to his novel, but the basic danger of self-replicating nanotech is—or I should say will be—real. The existential danger we face now is with biological viruses, we don’t yet have full molecular nanotechnology assembly. When we do the basic message for society is the same—which is to put more stones on the defensive side of the scale by developing explicitly defensive technologies. I describe some stratgies in Singularity is Near. There are also ethical guidelines that need to be followed by responsible practitioners. The Foresight Institute, founded by nanotechnology pioneer Eric Drexler has articulated a set of these. (end)

Damon Zucconi says:

With technologies such as the Fritz chip coming into play do you think that the fragmentation of THE computer (the Internet) is something that is possible and a legitimate concern for the iminent singularity?

Ray Kurzweil says:

WRG Damon’s question—by fragmentation I assume you mean decentralization. This is a very good trend. Decentralized technologies are more stable. We are moving to a "world wide mesh" in which computing and communications will be distributed among billions of devices in a flexible and self-organizing manner. (end)

Damon Zucconi says:

I think I meant fragmentation more like cut of government controlled networks.

Damon Zucconi says:

cut-off rather

Ray Kurzweil says:

I’m not using speech to text—I type faster.

Ray Kurzweil says:

WRG Damon’s question—do you mean attempts by the Chinese government to control the web? I think these will fail—they may nominally "control" overt expression of a few sensitive political issues, but people will work around them. There is already an enormous explosion of expression on Chinese web sites including a healthy and exploding blogger community. This is a democratizing force. I wrote in the 1980s that the decentralized communication technologies that would emerge would ultimately destroy the Soviet Union, which it did. That 1991 coup against Gorbachev failed not because of Yeltsin standing on a tank but because of the clandestine network of fax machines and early email using teletype machines. I mentioned this to Gorbachev recently at a lunch I had with him and he heartedly agreed. Of course, anything to put Yeltsin down.

Alexa Joyce says:

Do you think then, by extension of what you say about bio-viruses, in the next few years we’ll see teenage bio-hackers the same way we have young hackers on the Internet now?

Ray Kurzweil says:

WRG Alex’s concern:

Yes, it will ultimately get easier and easier to do this kind of work, so the answer is we need to develop a (very) rapid response system that can combat ANY new biological virus whether natural (like bird flu) or bioengineered. The good news is that the tools to accomplish this are coming into place. RNA interference can turn any gene off—we send pieces of RNA in as a medication, it latches on to the messenger RNA expressing a gene and destroys it.

This has been effective for stopping biological viruses. I described a plan in which we would have a rapid response system that could quickly sequence a virus, create an RNAi medication and gear up production, all in a manner of days. We have the tools to do this but we need to put it in place. There are other protective ideas as well.

Alexa Joyce says:

You mean antisense RNA here to block the genes?

Alexa Joyce says:

This concept of an open source biology community is very intriguing.

Ray Kurzweil says:

Wel first to clarify in response to Joyce—there are two competing technologies that can block the messenger RNA expressing a gene – antisense technology and RNA interference (RNAi). RNAi works very well, antisense technology has been disappointing.

Ray Kurzweil says:

WRG Joyces comment—yes there will be open source biology. Everything of importance will ultimately be information. Even manufacturing products. In the 2020s we’ll be able to manufacture almost anything we need/want with our own table top manufacturing devices—and there will be open source versions of designs—sneakers, meals, etc. (end)

Karl Sarnow says:

I have read your PP with great intrest. In slide 56 there are a lot of abbreviations. Is there some information available what these abbreviations mean? Could you give us a pointer?

Ray Kurzweil says:

Which slide was that?

Karl Sarnow says:

It is about reverse engineering the brain and shows a diagramm with many 3-letters abbreviations.

Damon Zucconi says:

Reverse engineering the human brain.

Ray Kurzweil says:

See endnote 96 for chapter 5 of Singularity is Near.

Karl Sarnow says to Ray Kurzweil:

Ok.

Ray Kurzweil says:

Page 546. The main text discussion is on pages 183-185. Chapter 4 is about reverse engineering the brain.

Ray Kurzweil says:

Did I miss a question?

Karl Sarnow says:

I don’t thinks so, what say the others?

Jan Kapoun says:

Your great book, "The Age of Intelligent Machines," celebrated 15 years this year. Would you change something in it now?

Ray Kurzweil says:

I think it actually does track quite well. Obviously some things are off by a few years—I would have said 1997 for a computer taking the world chess championship not 1998. This brings up the issue as to whether or not we can predict the future. The common wisdom is that we cannot. But there are certain measures of information technology—price-performance, capacity, bandwidth, etc.—that are very predictable. And its not just computer devices, but information technology is deeply influencing everything of value. So we can anticipate many scenarios quite accurately. We might wonder how can this be? Specific projects are indeed not predictable. yet the overall impact is predictable. We see a similar phenomenon in thermodynamics—the path of each particle in a gas is unpredictable, yet the overall properties of the gas—made of a vast number of chaotic unpredictable particles—is very predictable according to the laws of thermodynamics, to a high degree of precision. So it is with information technology, also a complex and chaotic system. (end)

Matt Neil says to Ray Kurzweil:

Which companies do you think will get the jump those coming from the bio health side or those coming from the tech molecular computing side.

Ray Kurzweil says:

There’s clearly a role for both. If you look at how biology is done now it is becoming an information technology. It used to be hit and miss—we would just find something that happened to work with no theory of operation, using "drug discovery." Now we’re actually learning the precise information processes underlying biological processes like atherosclerosis (the cause of most heart disease) and also gaining the means to reprogram these processes away from disease. Biochemical simulators are playing a big role. Drug development is already quite specialized with smaller companies doing the "risk removal" of specific treatments, then doing deals with the larger pharma companies. (end)

Karl Sarnow says:

I am not sure about this chat room, but I assume it is kicking us out at exactly 15h00 Brussels time. But I would not be happy to finish the session, before saying a big, big, thank you to Ray. It was very inspiring to read your answers and the questions from all of you. You will be able to read an edited version of the chat on Xplora. Thanks a lot again and I hope to see/read you again at Xplora somehow.

Good bye and thanks.

Karl

James Whipple says:

Thanks!

Alexa Joyce says:

Actually I think we can carry on if Ray is still happy to take a couple more questions…

Ray Kurzweil says:

I’ve got time for one or two more.

Karl Sarnow says:

I just hear that the tool will probably not kick us out, but nobody will be able to get in. So feel free to continue. I will record until the end of the debate.

Jan Kapoun says:

Thanks, Ray! It was a nice experience to chat with you.

James Whipple says:

I’m sure you know about people such as Hugo de Garis’ pessimistic visions of society’s reaction to a technological singularity. How will society react to change it can barely keep track of? How can the transition to be smoothed out?

Ray Kurzweil says:

WRG de Garis, as I said before I am concerned with the downsides. Bill Joy’s pessimistic piece in WIRED stemmed from a conversation we had in 1998 and his reading Age of Spiritual Machines. I do think that de Garis particular scenario does not make sense. He envisions a war between the "cosmists" (those who have enhanced themselves by merging with nonbiological intelligence) and the "terrists" (those who have not). It’s kind of absurd – like a war between those who use cell phones and those who don’t, or between the Amish and the armed forces. There are certainly concerns about "strong AI" run amok, but a war between those eschewing technology and those embracing it does not make sense. Such a "war" would be a non starter. (end)

Donelle Batty says:

I was just wondering about the use of nanobots to acheive immortality becasue I. Wouldn’t this create many more problems such as over-populization.

James Whipple says:

I agree!

Matt Neil says to Ray Kurzweil:

Yes and on that how many tablets are you taking a day for longevity—some say its in the hundres!

Ray Kurzweil says:

WRG over population:

Ray Kurzweil says:

That would be a problem if we had radical life extension and NO other changes. But nanotechnology will also enable us to create any physical product we will need from inexpensive raw materials being reorganized by massively parallel computerized processes using table top nanotech fabricators (2020s scenario). We’ll be able to meet the needs of any conceivable size biological population. I describe in Singularity is Near a scenario for energy—by capturing just 3% of 1% of the sunlight that falls on the Earth we can meet the projected energy needs of 2030. We’ll be able to do with nanoengineered solar panels and store the energy in highly decentralized nanoengineered fuel cells. (end)

Ray Kurzweil says:

Finally, last question, wrg supplements:

Ray Kurzweil says:

I do take a lot of supplements (about 250 pills a day) to "reprogram" my biochemistry. I take a lot of tests (50 or 60 blood levels) every few months to see how I’m doing. And I’m doing fine. I had type II diabetes 22 years ago but for 20 years have had no indication of this. My cholesterol many years ago was 280, but its been 130 for a long time. And all my other levels are relatively ideal. And according to biological aging tests, I was biologically 38 when I was chronolically 40. Now that I’m chronologically 57, I come out about 40 biologically. So there may be controversy about the validity of these biological aging tests, but I do a lot of other testing and feel I’m doing well. People may think this is a lot of trouble to go to, but actually I think it’s a lot more trouble to get sick. For young people in their 20s and 30s, they only need to stay reasonably healthy and perhaps take a good multivitamin. But for my contemporaries, people in their 50s and 60s, if they really want to be in good shape when we have these dramatic new technologies from biotech and nanotech, then they need to be aggressive to reprogram their biochemistry now.

Ray Kurzweil says:

Thanks again for chatting—enjoyed it a great deal!

© 2005 xplora. Reprinted with permission.

Click here for audio recording.

President Berkey, trustees, esteemed faculty, honored graduates, proud parents and guests, it’s a pleasure to be here. It’s a great honor to receive this distinction. Congratulations to all of you. I’ve long been an admirer of WPI and this is a terrific way to start your career. Actually judging by the practical experience you’ve had and the entrepreneurship which is blossoming on this campus you’ve already started your career.

A commencement is a good time to reflect on the future, on your future, and I’ve actually spent a few decades thinking about the future, trying to model technology trends. I suppose that’s one reason you asked me to share my ideas with you on what the future will hold, which will be rather different and empowering in terms of our ability to create knowledge, more so than many people realize.

I started thinking about the future and trying to anticipate it because of my interest in being an inventor myself. I realized that my inventions had to make sense when I finished a project, which would be three or four years later, and the world would be a different place. Everything would be different—the channels of distribution, the development tools. Most inventions, most technology projects fail not because the R&D department can’t get it to work—if you read business plans, 90 percent of those groups will do exactly what they say if they’re given the opportunity yet 90 percent of those projects will still fail because the timing is wrong. Not all the enabling factors will be in place when they’re needed. So realizing that, I began to try to model technology trends attempting to anticipate where technology will be. This has taken on a life of its own. I have a team of 10 people that gathers data in many different fields and we try to build mathematical models of what the future will look like.

Now, people say you can’t predict the future. And for some things that turns out to be true. If you ask me, “Will the stock price of Google be higher or lower three years from now?” that’s hard to predict. What will the next wireless common standard be? WiMAX, G-3, CDMA? That’s hard to predict. But if you ask me, “What will the cost of a MIPS of computing be in 2010?” or, “How much will it cost to sequence a base pair of DNA in 2012?” or, “What will the special and temporal resolution of non-invasive brain scanning be in 2014?,” I can give you a figure and it’s likely to be accurate because we’ve been making these predictions for several decades based on these models. There’s smooth, exponential growth in the power of these information technologies and computation that goes back a century—very smooth, exponential growth, basically doubling the power of electronics and communication every year. That’s a 50 percent deflation rate.

The same thing is true in biology. It took us 15 years to sequence HIV. We sequenced SARS in 31 days. We’ll soon be able to sequence a virus in just a few days’ time. We’re basically doubling the power of these technologies every year.

And that’s going to lead to three great revolutions that sometimes go by the letters GNR: genetics, nanotechnology and robotics. Let me describe these briefly and talk about the implications for our lives ahead.

G, genetics, which is really a term for biotechnology, means that we are gaining the tools to actually understand biology as information processes and reprogram them. Now, 99 percent of the drugs that are on the market today were not done that way. They were done through drug discovery, basically finding something. “Oh, here’s something that lowers blood pressure.” We have no idea why it works or how it works and invariably it has lots of side effects, similar to primitive man and woman when they discovered their first tools. “Oh, here’s a rock, this will make a good hammer.” But we didn’t have the means of shaping the tools to actually do a job. We’re now understanding the information processes underlying disease and aging and getting the tools to reprogram them.

We have little software programs inside us called genes, about 23 thousand of them. They were designed or evolved tens of thousands of years ago when conditions were quite different. I’ll give you just one example. The fat insulin receptor gene says, “Hold on to every calorie because the next hunting season may not work out so well.” And that’s a gene we’d like to reprogram. It made sense 20 thousand years ago when calories were few and far between. What would happen if we blocked that? We have a new technology that can turn genes off called RNA interference. So when that gene was turned off in mice, these mice ate ravenously and yet they remained slim. They got the health benefits of being slim. They didn’t get diabetes, didn’t get heart disease or cancer. They lived 20 to 25 percent longer while eating ravenously. There are several pharmaceutical companies who have noticed that might be a good human drug.

There’s many other genes we’d like to turn off. There are genes that are necessary for atherosclerosis, the cause of heart disease, to progress. There are genes that cancer relies on to progress. If we can turn these genes off, we could turn these diseases off. Turning genes off is just one of the methodologies. There are new forms of gene therapy that actually add genes so we’ll not just have designer babies but designer baby boomers. And you probably read this Korean announcement a couple of days ago of a new form of cell therapy where we can actually create new cells with your DNA so if you need a new heart or new heart cells you will be able to grow them with your own DNA, have them DNA-corrected, and thereby rejuvenate all your cells and tissues.

Ten or 15 years from now, which is not that far away, we’ll have the maturing of these biotechnology techniques and we’ll dramatically overcome the major diseases that we’ve struggled with for eons and also allow us to slow down, stop and even reverse aging processes.

The next revolution is nanotechnology, where we’re applying information technology to matter and energy. We’ll be able to overcome major problems that human civilization has struggled with. For example, energy. We have a little bit of sunlight here today. If we captured .03 percent, that’s three ten-thousandths of the sunlight that falls on the Earth, we could meet all of our energy needs. We can’t do that today because solar panels are very heavy, expensive and inefficient. New nano-engineered designs, designing them at the molecular level will enable us to create very inexpensive, very efficient, light-weight solar panels, store the energy in nano-engineered fuel cells, which are highly decentralized, and meet all of our energy needs.

The killer app of nanotechnology is something called nanobots, basically little robots the size of blood cells. If that sound very futuristic, there are four major conferences on that already and they’re already performing therapeutic functions in animals. One scientist cured Type-1 diabetes with these blood cell-sized nano-engineered capsules.

In regard to the 2020s, these devices will be able to go inside the human body and keep us healthy by destroying pathogens, correcting DNA errors, killing cancer cells and so on and even go into the brain, and interact with our biological neurons. If that sounds futuristic, there are already neural implants that are FDA-approved so there are people walking around who have computers in their brains and the biological neurons in their vicinity are perfectly happy to interact with these computerized devices. And the latest generation of the neural implant for Parkinson’s disease allows the patients to download new software to their neural implant from outside the patient. By the 2020s, we’ll be able to greatly enhance human intelligence, provide full immersion virtual reality, for example, from within the nervous system using these types of technologies.

And finally R, which stands for robotics, which is really artificial intelligence at the human level, we’ll see that in the late 2020s. By that time this exponential growth of computation will provide computer systems that are more powerful than the human brain. We’ll have completed the reverse engineering of the human brain to get the software algorithms, the secrets, the principles of operation of how human intelligence works. A side benefit of that is we’ll have greater insight into ourselves, how human intelligence works, how our emotional intelligence works, what human dysfunction is all about. We’ll be able to correct, for example, neurological diseases and also expand human intelligence. And this is not going to be an alien invasion of intelligent machines. We already routinely do things in our civilization that would be impossible without our computer intelligence. If all the AI programs, narrow AI, that’s embedded in our economic infrastructure were to stop today, our human civilization would grind to a halt. So we’re already very integrated with our technology. Computer technology used to be very remote. Now we carry it in our pockets. It’ll soon be in our clothing. It’s already begun migrating into our bodies and brains. We will become increasingly intimate with our technology.

The implications of all this is we will extend human longevity. We’ve already done that. A thousand years ago, human life expectancy was about 23. So most of you would be senior citizens if this were taking place a thousand years ago. In 1800, 200 years ago, human life expectancy was 37. So most of the parents here, including myself, wouldn’t be here. It was 50 years in 1900. It’s now pushing 80. Every time there’s been some advance in technology we’ve pushed it forward.: sanitation, antibiotics. This biotechnology revolution will expand it again. Nanotechnology will solve problems that we don’t get around to with biotechnology. We’ll have dramatic expansion of human longevity.

But actually life would get boring if we were sitting around for a few hundred years—we would be doing the same things over and over again—unless we had radical life expansion. And this technology will also expand our opportunities, expand our ability to create and appreciate knowledge. And creating knowledge is what the human species is all about. We’re the only species that has knowledge that we pass down from generation to generation. That’s what you’ve been doing for the last four years. That’s what you will continue doing indefinitely. We are expanding exponentially human knowledge and that is really what is exciting about the future.

I was told that commencement addresses should have a vision, which I’ve tried to share with you, and some practical advice. And my practical advice is that creating knowledge is what will be most exciting in life. And in order to create knowledge you have to have passion. So find a challenge that you can be passionate about, and there many of them that are worthwhile. And if you’re passionate about a worthwhile challenge, you can find the ideas to overcome that challenge. Those ideas exist and you can find them. And persistence usually pays off. You’ve all had timed tests where you had two or three hours to complete a test. But the tests in life are not timed. If you need an extra hour you can take it. Or an extra day, an extra week, an extra year, an extra decade. You’re the only one that will determine your own success or failure. Thomas Edison tried thousands of filaments to get his light bulb to work and none of them worked. And he easily could have said, “I guess all those skeptics who said that a practical light bulb was impossible were right.” Obviously he didn’t do that. You know the rest of the story.

If you have a challenge that you feel passionately about that’s really worthwhile, then you should never give in. To quote Winston Churchill, “Never give in. Never give in. Never, never, never, never, in nothing great or small, large or petty, never give in.”

Congratulations once again. This is a great achievement. I wish all of you long lives—very long lives—of success, creativity, health and happiness. And may the Force be with you.

© 2005 KurzweilAI.net

“But with Two Stars for Peace, the young people of Israel and Palestine can look forward to a future when they can travel freely throughout the United States, get their education in any part of the United States, or they can travel back and forth between Israel and Palestine. They can look forward to a future of instead of warring armies, everybody is part of a single United States army. The young people have no vested interest in the past of bickering and hostility. It’s depressing. But Two Stars for Peace gives them a way to have a good life.”

Click here to hear Martine Rothblatt’s appearance on Sirius satellite radio. (2.8 MB MP3 file, 3 minutes)

The history of technology is one of disruption and exponential growth, epitomized in Moore’s law, and generalized to many basic technological capabilities that are compounding independently from the economy. More than a niche subject of interest only to chip designers, the continued march of Moore’s Law will affect all of the sciences, just as nanotech will affect all industries. Thinking about Moore’s Law in the abstract provides a framework for predicting the future of computation and the transition to a new substrate: molecular electronics. An analysis of progress in molecular electronics provides a detailed example of the commercialization challenges and opportunities common to many nanotechnologies.

Introduction to Technology Exponentials:

Despite a natural human tendency to presume linearity, accelerating change from positive feedback is a common pattern in technology and evolution. We are now crossing a threshold where the pace of disruptive shifts is no longer inter-generational and begins to have a meaningful impact over the span of careers and eventually product cycles.

As early stage VCs, we look for disruptive businesses run by entrepreneurs who want to change the world. To be successful, we have to identify technology waves early and act upon those beliefs. At DFJ, we believe that nanotech is the next great technology wave, the nexus of scientific innovation that revolutionizes most industries and indirectly affects the fabric of society. Historians will look back on the upcoming epoch with no less portent than the Industrial Revolution.

The aforementioned are some long-term trends. Today, from a seed-stage venture capitalist perspective (with a broad sampling of the entrepreneurial pool), we are seeing more innovation than ever before. And we are investing in more new companies than ever before.

In the medium term, disruptive technological progress is relatively decoupled from economic cycles. For example, for the past 40 years in the semiconductor industry, Moore’s Law has not wavered in the face of dramatic economic cycles. Ray Kurzweil’s abstraction of Moore’s Law (from transistor-centricity to computational capability and storage capacity) shows an uninterrupted exponential curve for over 100 years, again without perturbation during the Great Depression or the World Wars. Similar exponentials can be seen in Internet connectivity, medical imaging resolution, genes mapped and solved 3D protein structures. In each case, the level of analysis is not products or companies, but basic technological capabilities.

In his forthcoming book, Kurzweil summarizes the exponentiation of our technological capabilities, and our evolution, with the near-term shorthand: the next 20 years of technological progress will be equivalent to the entire 20^th century. For most of us, who do not recall what life was like one hundred years ago, the metaphor is a bit abstract. In 1900, in the U.S., there were only 144 miles of paved road, and most Americans (94%+) were born at home, without a telephone, and never graduated high school. Most (86%+) did not have a bathtub at home or reliable access to electricity. Consider how much technology-driven change has compounded over the past century, and consider that an equivalent amount of progress will occur in one human generation, by 2020. It boggles the mind, until one dwells on genetics, nanotechnology, and their intersection. Exponential progress perpetually pierces the linear presumptions of our intuition. “Future Shock” is no longer on an inter-generational time-scale.

The history of humanity is that we use our tools and our knowledge to build better tools and expand the bounds of our learning. We are entering an era of exponential growth in our capabilities in biotech, molecular engineering and computing. The cross-fertilization of these formerly discrete domains compounds our rate of learning and our engineering capabilities across the spectrum. With the digitization of biology and matter, technologists from myriad backgrounds can decode and engage the information systems of biology as never before. And this inspires new approaches to bottom-up manufacturing, self-assembly, and layered complex systems development.

Moore’s Law:

Moore’s Law is commonly reported as a doubling of transistor density every 18 months. But this is not something the co-founder of Intel, Gordon Moore, has ever said. It is a nice blending of his two predictions; in 1965, he predicted an annual doubling of transistor counts in the most cost effective chip and revised it in 1975 to every 24 months. With a little hand waving, most reports attribute 18 months to Moore’s Law, but there is quite a bit of variability. The popular perception of Moore’s Law is that computer chips are compounding in their complexity at near constant per unit cost. This is one of the many abstractions of Moore’s Law, and it relates to the compounding of transistor density in two dimensions. Others relate to speed (the signals have less distance to travel) and computational power (speed x density).

So as to not miss the long-term trend while sorting out the details, we will focus on the 100-year abstraction of Moore’s Law below. But we should digress for a moment to underscore the importance of continued progress in Moore’s law to a broad set of industries.

Importance of Moore’s Law:

Moore’s Law drives chips, communications and computers and has become the primary driver in drug discovery and bioinformatics, medical imaging and diagnostics. Over time, the lab sciences become information sciences, modeled on a computer rather than trial and error experimentation.

NASA Ames shut down their wind tunnels this year. As Moore’s Law provided enough computational power to model turbulence and airflow, there was no longer a need to test iterative physical design variations of aircraft in the wind tunnels, and the pace of innovative design exploration dramatically accelerated.

Eli Lilly processed 100x fewer molecules this year than they did 15 years ago. But their annual productivity in drug discovery did not drop proportionately; it went up over the same period. “Fewer atoms and more bits” is their coda.

Accurate simulation demands computational power, and once a sufficient threshold has been crossed, simulation acts as an innovation accelerant over physical experimentation. Many more questions can be answered per day.

Recent accuracy thresholds have been crossed in diverse areas, such as modeling the weather (predicting a thunderstorm six hours in advance) and automobile collisions (a relief for the crash test dummies), and the thresholds have yet to be crossed for many areas, such as protein folding dynamics.

Long Term Abstraction of Moore’s Law:

Unless you work for a chip company and focus on fab-yield optimization, you do not care about transistor counts. Integrated circuit customers do not buy transistors. Consumers of technology purchase computational speed and data storage density. When recast in these terms, Moore’s Law is no longer a transistor-centric metric, and this abstraction allows for longer-term analysis.

The exponential curve of Moore’s Law extends smoothly back in time for over 100 years, long before the invention of the semiconductor. Through five paradigm shifts—such as electro-mechanical calculators and vacuum tube computers—the computational power that $1000 buys has doubled every two years. For the past 30 years, it has been doubling every year.

Each horizontal line on this logarithmic graph represents a 100x improvement. A straight diagonal line would be an exponential, or geometrically compounding, curve of progress. Kurzweil plots a slightly upward curving line—a double exponential.

Each dot represents a human drama. They did not realize that they were on a predictive curve. Each dot represents an attempt to build the best computer with the tools of the day. Of course, we use these computers to make better design software and manufacturing control algorithms. And so the progress continues.

One machine was used in the 1890 Census; one cracked the Nazi Enigma cipher in World War II; one predicted Eisenhower’s win in the Presidential election. And there is the Apple ][, and the Cray 1, and just to make sure the curve had not petered out recently, I looked up the cheapest PC available for sale on Wal*Mart.com, and that is the green dot that I have added to the upper right corner of the graph.

And notice the relative immunity to economic cycles. The Great Depression and the World Wars and various recessions do not introduce a meaningful delay in the progress of Moore’s Law. Certainly, the adoption rates, revenue, profits and inventory levels of the computer companies behind the various dots on the graph may go though wild oscillations, but the long-term trend emerges nevertheless.

Any one technology, such as the CMOS transistor, follows an elongated S-shaped curve of slow progress during initial development, upward progress during a rapid adoption phase, and then slower growth from market saturation over time. But a more generalized capability, such as computation, storage, or bandwidth, tends to follow a pure exponential—bridging across a variety of technologies and their cascade of S-curves.

If history is any guide, Moore’s Law will continue on and will jump to a different substrate than CMOS silicon. It has done so five times in the past, and will need to again in the future.

Problems With the Current Paradigm:

Intel co-founder Gordon Moore has chuckled at those who have predicted the imminent demise of Moore’s Law in decades past. But the traditional semiconductor chip is finally approaching some fundamental physical limits. Moore recently admitted that Moore’s Law, in its current form, with CMOS silicon, will run out of gas in 2017.

One of the problems is that the chips are getting very hot. The following graph of power density is also a logarithmic scale:

This provides the impetus for chip cooling companies, like Nanocoolers, to provide a breakthrough solution for removing 100 Watts per square centimeter. In the long term, the paradigm has to change.

Another physical limit is the atomic limit—the indivisibility of atoms. Intel’s current gate oxide is 1.2nm thick. Intel’s 45nm process is expected to have a gate oxide that is only 3 atoms thick. It is hard to imagine many more doublings from there, even with further innovation in insulating materials. Intel has recently announced a breakthrough in a nano-structured gate oxide (high k dielectric) and metal contact materials that should enable the 45nm node to come on line in 2007. None of the industry participants has a CMOS roadmap for the next 50 years.

A major issue with thin gate oxides, and one that will also come to the fore with high-k dielectrics, is quantum mechanical tunneling. As the oxide becomes thinner, the gate current can approach and even exceed the channel current so that the transistor cannot be controlled by the gate.

Another problem is the escalating cost of a semiconductor fab plant, which is doubling every three years, a phenomenon dubbed Moore’s Second Law. Human ingenuity keeps shrinking the CMOS transistor, but with increasingly expensive manufacturing facilities—currently $3 billion per fab.

A large component of fab cost is the lithography equipment that patterns the wafers with successive sub-micron layers. Nanoimprint lithography from companies like Molecular Imprints can dramatically lower cost and leave room for further improvement from the field of molecular electronics.

We have been investing in a variety of companies, such as Coatue, D-Wave, FlexICs, Nantero, and ZettaCore that are working on the next paradigm shift to extend Moore’s Law beyond 2017. One near term extension to Moore’s Law focuses on the cost side of the equation. Imagine rolls of wallpaper embedded with inexpensive transistors. FlexICs deposits traditional transistors at room temperature on plastic, a much cheaper bulk process than growing and cutting crystalline silicon ingots.

Molecular Electronics:

The primary contender for the post-silicon computation paradigm is molecular electronics, a nano-scale alternative to the CMOS transistor. Eventually, molecular switches will revolutionize computation by scaling into the third dimension—overcoming the planar deposition limitations of CMOS. Initially, they will substitute for the transistor bottleneck on an otherwise standard silicon process with standard external I/O interfaces.

For example, Nantero employs carbon nanotubes suspended above metal electrodes on silicon to create high-density nonvolatile memory chips (the weak Van der Waals bond can hold a deflected tube in place indefinitely with no power drain). Carbon nanotubes are small (~10 atoms wide), 30x stronger than steel at 1/6 the weight, and perform the functions of wires, capacitors and transistors with better speed, power, density and cost. Cheap nonvolatile memory enables important advances, such as “instant-on” PCs.

Other companies, such as Hewlett Packard and ZettaCore, are combining organic chemistry with a silicon substrate to create memory elements that self-assemble using chemical bonds that form along pre-patterned regions of exposed silicon.

There are several reasons why molecular electronics is the next paradigm for Moore’s Law:

• Size: Molecular electronics has the potential to dramatically extend the miniaturization that has driven the density and speed advantages of the integrated circuit (IC) phase of Moore’s Law. In 2002, using a STM to manipulate individual carbon monoxide molecules, IBM built a 3-input sorter by arranging those molecules precisely on a copper surface. It is 260,000x smaller than the equivalent circuit built in the most modern chip plant.

For a memorable sense of the difference in scale, consider a single drop of water. There are more molecules in a single drop of water than all transistors ever built. Think of the transistors in every memory chip and every processor ever built—there are about 100x more molecules in a drop of water. Sure, water molecules are small, but an important part of the comparison depends on the 3D volume of a drop. Every IC, in contrast, is a thin veneer of computation on a thick and inert substrate.

• Power: One of the reasons that transistors are not stacked into 3D volumes today is that the silicon would melt. The inefficiency of the modern transistor is staggering. It is much less efficient at its task than the internal combustion engine. The brain provides an existence proof of what is possible; it is 100 million times more efficient in power/calculation than our best processors. Sure it is slow (under a kHz) but it is massively interconnected (with 100 trillion synapses between 60 billion neurons), and it is folded into a 3D volume. Power per calculation will dominate clock speed as the metric of merit for the future of computation.

• Manufacturing Cost: Many of the molecular electronics designs use simple spin coating or molecular self-assembly of organic compounds. The process complexity is embodied in the synthesized molecular structures, and so they can literally be splashed on to a prepared silicon wafer. The complexity is not in the deposition or the manufacturing process or the systems engineering. Much of the conceptual difference of nanotech products derives from a biological metaphor: complexity builds from the bottom up and pivots about conformational changes, weak bonds, and surfaces. It is not engineered from the top with precise manipulation and static placement.

• Low Temperature Manufacturing: Biology does not tend to assemble complexity at 1000 degrees in a high vacuum. It tends to be room temperature or body temperature. In a manufacturing domain, this opens the possibility of cheap plastic substrates instead of expensive silicon ingots.

• Elegance: In addition to these advantages, some of the molecular electronics approaches offer elegant solutions to non-volatile and inherently digital storage. We go through unnatural acts with CMOS silicon to get an inherently analog and leaky medium to approximate a digital and non-volatile abstraction that we depend on for our design methodology. Many of the molecular electronic approaches are inherently digital, and some are inherently non-volatile.

Other research projects, from quantum computing to using DNA as a structural material for directed assembly of carbon nanotubes, have one thing in common: they are all nanotechnology.

Why the term “Nanotechnology”?

Nanotech is often defined as the manipulation and control of matter at the nanometer scale (critical dimensions of 1-100nm). It is a bit unusual to describe a technology by a length scale. We certainly didn’t get very excited by “inch-o-technology.” As venture capitalists, we start to get interested when there are unique properties of matter that emerge at the nanoscale, and that are not exploitable at the macroscale world of today’s engineered products. We like to ask the startups that we are investing in: “Why now? Why couldn’t you have started this business ten years ago?” Our portfolio of nanotech startups have a common thread in their response to this question—recent developments in the capacity to understand and engineer nanoscale materials have enabled new products that could not have been developed at larger scale.

There are various unique properties of matter that are expressed at the nanoscale and are quite foreign to our “bulk statistical” senses (we do not see single photons or quanta of electric charge; we feel bulk phenomena, like friction, at the statistical or emergent macroscale). At the nanoscale, the bulk approximations of Newtonian physics are revealed for their inaccuracy, and give way to quantum physics. Nanotechnology is more than a linear improvement with scale; everything changes. Quantum entanglement, tunneling, ballistic transport, frictionless rotation of superfluids, and several other phenomena have been regarded as “spooky” by many of the smartest scientists, even Einstein, upon first exposure.

For a simple example of nanotech’s discontinuous divergence from the “bulk” sciences, consider the simple aluminum Coke can. If you take the inert aluminum metal in that can and grind it down into a powder of 20-30nm particles, it will spontaneously explode in air. It becomes a rocket fuel catalyst. The energetic properties of matter change at that scale. The surface area to volume ratios become relevant, and even the inter-atomic distances in a metal lattice change from surface effects.

Innovation from the Edge:

Disruptive innovation, the driver of growth and renewal, occurs at the edge. In startups, innovation occurs out of the mainstream, away from the warmth of the herd. In biological evolution, innovative mutations take hold at the physical edge of the population, at the edge of survival. In complexity theory, structure and complexity emerge at the edge of chaos—the dividing line between predictable regularity and chaotic indeterminacy. And in science, meaningful disruptive innovation occurs at the inter-disciplinary interstices between formal academic disciplines.

Herein lies much of the excitement about nanotechnology: in the richness of human communication about science. Nanotech exposes the core areas of overlap in the fundamental sciences, the place where quantum physics and quantum chemistry can cross-pollinate with ideas from the life sciences.

Over time, each of the academic disciplines develops its own proprietary systems vernacular that isolates it from neighboring disciplines. Nanoscale science requires scientists to cut across the scientific languages to unite the isolated islands of innovation.

Nanotech is the nexus of the sciences.

In academic centers and government labs, nanotech is fostering new conversations. At Stanford, Duke and many other schools, the new nanotech buildings are physically located at the symbolic hub of the schools of engineering, computer science and medicine.

Nanotech is the nexus of the sciences, but outside of the science and research itself, the nanotech umbrella conveys no business synergy whatsoever. The marketing, distribution and sales of a nanotech solar cell, memory chip or drug delivery capsule will be completely different from each other, and will present few opportunities for common learning or synergy.

Market Timing:

As an umbrella term for a myriad of technologies spanning multiple industries, nanotech will eventually disrupt these industries over different time frames—but most are long-term opportunities. Electronics, energy, drug delivery and materials are areas of active nanotech research today. Medicine and bulk manufacturing are future opportunities. The NSF predicts that nanotech will have a trillion dollar impact on various industries inside of 15 years.

Of course, if one thinks far enough in the future, every industry will be eventually revolutionized by a fundamental capability for molecular manufacturing—from the inorganic structures to the organic and even the biological. Analog manufacturing becomes digital, engendering a profound restructuring of the substrate of the physical world.

The science futurism and predictions of potential nanotech products has a near term benefit. It helps attract some of the best and brightest scientists to work on hard problems that are stepping-stones to the future vision. Scientists relish in exploring the frontier of the unknown, and nanotech embodies the inner frontier.

Given that much of the abstract potential of nanotech is a question of “when” not “if”, the challenge for the venture capitalist is one of market timing. When should we be investing, and in which sub-sectors? It is as if we need to pull the sea of possibilities through an intellectual chromatograph to tease apart the various segments into a timeline of probable progression. That is an ongoing process of data collection (e.g., the growing pool of business plan submissions), business and technology analysis, and intuition.

Two touchstone events for the scientific enthusiasm for the timing of nanotech were the decoding of the human genome and the dazzling visual images from the Scanning Tunneling Microscope (e.g., the arrangement of individual Xenon atoms into the IBM logo). They represent the digitization of biology and matter, symbolic milestones for accelerated learning and simulation-driven innovation.

And more recently, nanotech publication has proliferated, much like the early days of the Internet. Beside the popular press, the number of scientific publications on nanotech has grown 10x in the past ten years. According to the U.S. Patent Office, the number of nanotech patents granted each year has skyrocketed 3x in the past seven years. Ripe with symbolism, IBM has more lawyers than engineers working on nanotech.

With the recent codification of the National Nanotech Initiative into law, federal funding will continue to fill the pipeline of nanotech research. With $847 million earmarked for 2004, nanotech was a rarity in the tight budget process; it received more funding than was requested. And now nanotech is second only to the space race for federal funding of science. And the U.S. is not alone in funding nanotechnology. Unlike many previous technological areas, we aren’t even in the lead. Japan outspends the U.S. each year on nanotech research. In 2003, the U.S. government spending was one fourth of the world total.

Federal funding is the seed corn for nanotech entrepreneurship. All of our nanotech portfolio companies are spin-offs (with negotiated IP transfers) from universities or government labs, and all got their start with federal funding. Often these companies need specialized equipment and expensive laboratories to do the early tinkering that will germinate a new breakthrough. These are typically lacking in the proverbial garage of the entrepreneur at home.

And corporate investors have discovered a keen interest in nanotechnology, with internal R&D, external investments in startups, and acquisitions of promising companies, such as AMD’s recent acquisition of the molecular electronics company Coatue.

Despite all of this excitement, there are a fair number of investment dead-ends, and so we continue to refine the filters we use in selecting companies to back. Every entrepreneur wants to present their business as fitting an appropriate timeline to commercialization. How can we guide our intuition on which of these entrepreneurs are right?

The Vertical Integration Question:

Nanotech involves the reengineering of the lowest level physical layer of a system, and so a natural business question arises: How far forward do you need to vertically integrate before you can sell a product on the open market? For example, in molecular electronics, if you can ship a DRAM-compatible chip, you have found a horizontal layer of standardization, and further vertical integration is not necessary. If you have an incompatible 3D memory block, you may have to vertically integrate to the storage subsystem level, or further, to bring product to market. That may require industry partnerships, and will, in general, take more time and money as change is introduced farther up the product stack. 3D logic with massive interconnectivity may require a new computer design and a new form of software; this would take the longest to commercialize. And most startups on this end of the spectrum would seek partnerships to bring their vision to market. The success and timeliness of that endeavor will depend on many factors, including IP protection, the magnitude of improvement, the vertical tier at which that value is recognized, the number of potential partners, and the degree of tooling and other industry accommodations.

Product development timelines are impacted by the cycle time of the R&D feedback loop. For example, outdoor lifetime testing for organic LEDs will take longer than in silico simulation spins of digital products. If the product requires partners in the R&D loop or multiple nested tiers of testing, it will take longer to commercialize.

The “Interface Problem”:

As we think about the startup opportunities in nanotechnology, an uncertain financial environment underscores the importance of market timing and revenue opportunities over the next five years. Of the various paths to nanotech, which are 20-year quests in search of a government grant, and which are market-driven businesses that will attract venture capital? Are there co-factors of production that require a whole industry to be in place before a company ships product?

As a thought experiment, imagine that I could hand you today any nanotech marvel of your design—a molecular machine as advanced as you would like. What would it be? A supercomputer? A bloodstream submarine? A matter compiler capable of producing diamond rods or arbitrary physical objects? Pick something.

Now, imagine some of the complexities: Did it blow off my hand as I offer it to you? Can it autonomously move to its intended destination? What is its energy source? How do you communicate with it?

These questions draw the “interface problem” into sharp focus: Does your design require an entire nanotech industry to support, power, and “interface” to your molecular machine? As an analogy, imagine that you have one of the latest Pentium processors out of Intel’s wafer fab. How would you make use of the Pentium chip? You then need to wire-bond the chip to a larger lead frame in a package that connects to a larger printed circuit board, fed by a bulky power supply that connects to the electrical power grid. Each of these successive layers relies on the larger-scale precursors from above (which were developed in reverse chronological order), and the entire hierarchy is needed to access the potential of the microchip.

For molecular nanotech, where is the scaling hierarchy?

Today’s business-driven paths to nanotech diverge into two strategies to cross the “interface” chasm—the biologically inspired bottom-up path, and the top-down approach of the semiconductor industry. The non-biological MEMS developers are addressing current markets in the micro-world while pursuing an ever-shrinking spiral of miniaturization that builds the relevant infrastructure tiers along the way. Not surprisingly, this is very similar to the path that has been followed in the semiconductor industry, and many of its adherents see nanotech as inevitable, but in the distant future.

On the other hand, biological manipulation presents myriad opportunities to effect great change in the near-term. Drug development, tissue engineering, and genetic engineering are all powerfully impacted by the molecular manipulation capabilities available to us today. And genetically modified microbes, whether by artificial evolution or directed gene splicing, give researchers the ability to build structures from the bottom up.

The Top Down “Chip Path”:

This path is consonant with the original vision of physicist Richard Feynman (in his 1959 lecture at Caltech) of the iterative miniaturization of our tools down to the nano scale. Some companies, like Zyvex, are pursuing the gradual shrinking of semiconductor manufacturing technology from the micro-electro-mechanical systems (MEMS) of today into the nanometer domain of NEMS. SiWave engineers and manufactures MEMS structures with applications in the consumer electronics, biomedical and communications markets. These precision mechanical devices are built utilizing a customized semiconductor fab.

MEMS technologies have already revolutionized the automotive industry with airbag sensors and the printing sector with ink jet nozzles, and are on track to do the same in medical devices, photonic switches for communications and mobile phones. In-Stat/MDR forecasts that the $4.7 billion of MEMS revenue in 2003 will grow to $8.3 billion by 2007. But progress is constrained by the pace (and cost) of the semiconductor equipment industry, and by the long turnaround time for fab runs. Microfabrica in Torrance, CA, is seeking to overcome these limitations to expand the market for MEMS to 3D structures in more materials than just silicon and with rapid turnaround times.

Many of the nanotech advances in storage, semiconductors and molecular electronics can be improved, or in some cases enabled, by tools that allow for the manipulation of matter at the nanoscale. Here are three examples:

• Nanolithography

Molecular Imprints is commercializing a unique imprint lithographic technology developed at the University of Texas at Austin. The technology uses photo-curable liquids and etched quartz plates to dramatically reduce the cost of nanoscale lithography. This lithography approach, recently added to the ITRS Roadmap, has special advantages for applications in the areas of nano-devices, MEMS, microfluidics, optical components and devices, as well as molecular electronics.

• Optical Traps

Arryx has developed a breakthrough in nano-material manipulation. They generate hundreds of independently controllable laser tweezers that can manipulate molecular objects in 3D (move, rotate, cut, place), all from one laser source passing through an adaptive hologram. The applications span from cell sorting, to carbon nanotube placement, to continuous material handling. They can even manipulate the organelles inside an unruptured living cell (and weigh the DNA in the nucleus).

• Metrology

Imago’s LEAP atom probe microscope is being used by the chip and disk drive industries to produce 3D pictures that depict both chemistry and structure of items on an atom-by-atom basis.  Unlike traditional microscopes, which zoom in to see an item on a microscopic level, Imago’s nanoscope analyzes structures, one atom at a time, and "zooms out" as it digitally reconstructs the item of interest at a rate of millions of atoms per minute.  This creates an unprecedented level of visibility and information at the atomic level.

Advances in nanoscale tools help us control and analyze matter more precisely, which in turn, allows us to produce better tools.

To summarize, the top-down path is designed and engineered with:

• Semiconductor industry adjacencies (with the benefits of market extensions and revenue along the way and the limitation of planar manufacturing techniques)

• Interfaces of scale inherited from the top

The Biological Bottom Up Path:

In contrast to the top-down path, the biological bottom up archetype is:

• Grown via replication, evolution, and self assembly in a 3D, fluid medium

• Constrained at interfaces to the inorganic world

• Limited by learning and theory gaps (in systems biology, complexity theory and the pruning rules of emergence)

• Bootstrapped by a powerful pre-existing hierarchy of interpreters of digital molecular code.

To elaborate on this last point, the ribosome takes digital instructions in the form of mRNA and manufactures almost everything we care about in our bodies from a sequential concatenation of amino acids into proteins. The ribosome is a wonderful existence proof of the power and robustness of a molecular machine. It is roughly 20nm on a side and consists of only 99 thousand atoms. Biological systems are replicating machines that parse molecular code (DNA) and a variety of feedback to grow macro-scale beings. These highly evolved systems can be hijacked and reprogrammed to great effect.

So how does this help with the development of molecular electronics or nanotech manufacturing? The biological bootstrap provides a more immediate path to nanotech futures. Biology provides us with a library of pre-built components and subsystems that can be repurposed and reused, and scientists in various labs are well underway in re-engineering the information systems of biology.

For example, researchers at NASA Ames are taking self-assembling heat shock proteins from thermophiles and genetically modifying them so that they will deposit a regular array of electrodes with a 17nm spacing. This could be useful for patterned magnetic media in the disk drive industry or electrodes in a polymer solar cell.

At MIT, researchers are using accelerated artificial evolution to rapidly breed M13 bacteriophage to infect bacteria in such a way that they bind and organize semiconducting materials with molecular precision.

At IBEA, Craig Venter and Hamilton Smith are leading the Minimal Genome Project. They take the Mycoplasma genitalium from the human urogenital tract, and strip out 200 unnecessary genes, thereby creating the simplest organism that can self-replicate. Then they plan to layer new functionality on to this artificial genome, such as the ability to generate hydrogen from water using the sun’s energy for photonic hydrolysis.

The limiting factor is our understanding of these complex systems, but our pace of learning has been compounding exponentially. We will learn more about genetics and the origins of disease in the next 10 years than we have in all of human history. And for the minimal genome microbes, the possibility of understanding the entire proteome and metabolic pathways seems tantalizingly close to achievable. These simpler organisms have a simple “one gene: one protein” mapping, and lack the nested loops of feedback that make the human genetic code so rich.

Hybrid Molecular Electronics Example:

In the near term, there are myriad companies who are leveraging the power of organic self-assembly (bottom up) and the market interface advantages of top down design. The top down substrate constrains the domain of self-assembly.

Based in Denver, ZettaCore builds molecular memories from energetically elegant molecules that are similar to chlorophyll. ZettaCore’s synthetic organic porphyrin molecule self-assembles on exposed silicon. These molecules, called multiporphyrin nanostructures, can be oxidized and reduced (electrons removed or replaced) in a way that is stable, reproducible, and reversible. In this way, the molecules can be used as a reliable storage medium for electronic devices. Furthermore, the molecules can be engineered to store multiple bits of information and to maintain that information for relatively long periods of time before needing to be refreshed.

Recall the water drop to transistor count comparison, and realize that these multiporphyrins have already demonstrated up to eight stable digital states per molecule.

The technology has future potential to scale to 3D circuits with minimal power dissipation, but initially it will enhance the weakest element of an otherwise standard 2D memory chip. The ZettaCore memory chip looks like a standard memory chip to the end customer; nobody needs to know that it has “nano inside.” The I/O pads, sense amps, row decoders and wiring interconnect are produced with a standard semiconductor process. As a final manufacturing step, the molecules are splashed on the wafer where they self-assemble in the pre-defined regions of exposed metal.

From a business perspective, the hybrid product design allows an immediate market entry because the memory chip defines a standard product feature set, and the molecular electronics manufacturing process need not change any of the prior manufacturing steps. The inter-dependencies with the standard silicon manufacturing steps are also avoided given this late coupling; the fab can process wafers as they do now before spin coating the molecules. In contrast, new materials for gate oxides or metal interconnects can have a number of effects on other processing steps that need to be tested, which introduces delay (as was seen with copper interconnect).

For these reasons, ZettaCore is currently in the lead in the commercialization of molecular electronics, with a working megabit chip, technology tested to a trillion read/write cycles, and manufacturing partners. In a symbolic nod to the future, Intel co-founder Les Vadasz (badge #3), has just joined the Board of Directors of ZettaCore. He was formerly the design manager for the world’s first DRAM, EPROM and microprocessor.

Generalizing from the ZettaCore experience, the early revenue in molecular electronics will likely come from simple 1D structures such as chemical sensors and self-assembled 2D arrays on standard substrates, such as memory chips, sensor arrays, displays, CCDs for cameras and solar cells.

IP and business model:

Beyond product development timelines, the path to commercialization is dramatically impacted by the cost and scale of the manufacturing ramp. Partnerships with industry incumbents can be the accelerant or albatross for market entry.

The strength of the IP protection for nanotech relates to the business models that can be safely pursued. For example, if the composition of matter patents afford the nanotech startup the same degree of protection as a biotech startup, then a “biotech licensing model” may be possible in nanotech. For example, a molecular electronics company could partner with a large semiconductor company for manufacturing, sales and marketing, just as a biotech company partners with a big pharma partner for clinical trials, marketing, sales and distribution. In both cases, the cost to the big partner is on the order of $100 million, and the startup earns a royalty on future product sales.

Notice how the transaction costs and viability of this business model option pivots around the strength of IP protection. A software business, on the other end of the IP spectrum, would be very cautious about sharing their source code with Microsoft in the hopes of forming a partnership based on royalties.

Manufacturing partnerships are common in the semiconductor industry, with the “fabless” business model. This layering of the value chain separates the formerly integrated functions of product conceptualization, design, manufacturing, testing, and packaging. This has happened in the semiconductor industry because the capital cost of manufacturing is so large. The fabless model is a useful way for a small company with a good idea to bring its own product to market, but the company then has to face the issue of gaining access to its market and funding the development of marketing, distribution, and sales.

Having looked at the molecular electronics example in some depth, we can now move up the abstraction ladder to aggregates, complex systems, and the potential to advance the capabilities of Moore’s Law in software.

Systems, Software, and other Abstractions:

Unlike memory chips, which have a regular array of elements, processors and logic chips are limited by the rats’ nest of wires that span the chip on multiple layers. The bottleneck in logic chip design is not raw numbers of transistors, but a design approach that can utilize all of that capability in a timely fashion. For a solution, several next generation processor companies have redesigned “systems on silicon” with a distributed computing bent; wiring bottlenecks are localized, and chip designers can be more productive by using a high-level programming language, instead of wiring diagrams and logic gates. Chip design benefits from the abstraction hierarchy of computer science.

Compared to the relentless march of Moore’s Law, the cognitive capability of humans is relatively fixed. We have relied on the compounding power of our tools to achieve exponential progress. To take advantage of accelerating hardware power, we must further develop layers of abstraction in software to manage the underlying complexity. For the next 1000-fold improvement in computing, the imperative will shift to the growth of distributed complex systems. Our inspiration will likely come from biology.

As we race to interpret the now complete map of the human genome, and embark upon deciphering the proteome, the accelerating pace of learning is not only opening doors to the better diagnosis and treatment of disease, it is also a source of inspiration for much more powerful models for computer programming and complex systems development.

Biological Muse:

Many of the interesting software challenges relate to growing complex systems or have other biological metaphors as inspiration. Some of the interesting areas include: Biomimetics, Artificial Evolution, Genetic Algorithms, A-life, Emergence, IBM’s Autonomic Computing initiative, Viral Marketing, Mesh, Hives, Neural Networks and the Subsumption architecture in robotics. The Santa Fe Institute just launched a BioComp research initiative.

In short, biology inspires IT and IT drives biology.

But how inspirational are the information systems of biology? If we took your entire genetic code–the entire biological program that resulted in your cells, organs, body and mind–and burned it into a CD, it would be smaller than Microsoft Office. Just as images and text can be stored digitally, two digital bits can encode for the four DNA bases (A,T,C and G) resulting in a 750MB file that can be compressed for the preponderance of structural filler in the DNA chain.

If, as many scientists believe, most of the human genome consists of vestigial evolutionary remnants that serve no useful purpose, then we could compress it to 60MB of concentrated information. Having recently reinstalled Office, I am humbled by the comparison between its relatively simple capabilities and the wonder of human life. Much of the power in bio-processing comes from the use of non-linear fuzzy logic and feedback in the electrical, physical and chemical domains.

For example, in a fetus, the initial inter-neuronal connections, or "wiring" of the brain, follow chemical gradients. The massive number of inter-neuron connections in an adult brain could not be simply encoded in our DNA, even if the entire DNA sequence was dedicated to this one task. There are on the order of 100 trillion synaptic connections between 60 billion neurons in your brain.

This incredibly complex system is not ‘installed’ like Microsoft Office from your DNA. It is grown, first through widespread connectivity sprouting from ‘static storms’ of positive electro-chemical feedback, and then through the pruning of many underused connections through continuous usage-based feedback. In fact, at the age of 2 to 3 years old, humans hit their peak with a quadrillion synaptic connections, and twice the energy burn of an adult brain.

The brain has already served as an inspirational model for artificial intelligence (AI) programmers. The neural network approach to AI involves the fully interconnected wiring of nodes, and then the iterative adjustment of the strength of these connections through numerous training exercises and the back-propagation of feedback through the system.

Moving beyond rules-based AI systems, these artificial neural networks are capable of many human-like tasks, such as speech and visual pattern recognition with a tolerance for noise and other errors. These systems shine precisely in the areas where traditional programming approaches fail.

The coding efficiency of our DNA extends beyond the leverage of numerous feedback loops to the complex interactions between genes. The regulatory genes produce proteins that respond to external or internal signals to regulate the activity of previously produced proteins or other genes. The result is a complex mesh of direct and indirect controls.

This nested complexity implies that genetic re-engineering can be a very tricky endeavor if we have partial system-wide knowledge about the side effects of tweaking any one gene. For example, recent experiments show that genetically enhanced memory comes at the expense of enhanced sensitivity to pain.

By analogy, our genetic code is a dense network of nested hyperlinks, much like the evolving Web. Computer programmers already tap into the power and efficiency of indirect pointers and recursive loops. More recently, biological systems have inspired research in evolutionary programming, where computer programs are competitively grown in a simulated environment of natural selection and mutation. These efforts could transcend the local optimization inherent to natural evolution.

But therein lies great complexity. We have little experience with the long-term effects of the artificial evolution of complex systems. Early subsystem work can be deterministic of emergent and higher-level capabilities, as with the neuron (witness the Cambrian explosion of structural complexity and intelligence in biological systems once the neuron enabled something other than nearest-neighbor inter-cellular communication. Prior to the neuron, most multi-cellular organisms were small blobs).

Recent breakthroughs in robotics were inspired by the "subsumption architecture" of biological evolution—using a layered approach to assembling reactive rules into complete control systems from the bottom up. The low-level reflexes are developed early on, and remain unchanged as complexity builds. Early subsystem work in any subsumptive system can have profound effects on its higher order constructs. We may not have a predictive model of these downstream effects as we are developing the architectural equivalent of the neuron.

The Web is the first distributed experiment in biological growth in technological systems. Peer-to-peer software development and the rise of low-cost Web-connected embedded systems give the possibility that complex artificial systems will arise on the Internet, rather than on one programmer’s desktop. We already use biological metaphors, such as viral marketing to describe the network economy.

Nanotech Accelerants: quantum simulation and high-throughput experimentation:

We have already discussed the migration of the lab sciences to the innovation cycles of the information sciences and Moore’s Law. Advances in multi-scale molecular modeling are helping some companies design complex molecular systems in silico. But the quantum effects that underlie the unique properties of nano-scale systems are a double-edged sword. Although scientists have known for nearly 100 years how to write down the equations that an engineer needs to solve in order to understand any quantum system, no computer has ever been built that is powerful enough to solve them. Even today’s most powerful supercomputers choke on systems bigger than a single water molecule.

This means that the behavior of nano-scale systems can only be reliably studied by empirical methods—building something in a lab, and poking and prodding it to see what happens.

This observation is distressing on several counts. We would like to design and visualize nano-scale products in the tradition of mechanical engineering, using CAD-like programs. Unfortunately this future can never be accurately realized using traditional computer architectures. The structures of interest to nano-scale scientists present intractable computational challenges to traditional computers.

The shortfall in our ability to use computers to shorten and cheapen the design cycles of nano-scale products has serious business ramifications. If the development of all nano-scale products fundamentally requires long R&D cycles and significant investment, the nascent nanotechnology industry will face many of the difficulties that the biotechnology industry faces, without having a parallel to the pharmaceutical industry to shepherd products to markets.

In a wonderful turn of poetic elegance, quantum mechanics itself turns out to be the solution to this quandary. Machines known as quantum computers, built to harness some simple properties of quantum systems, can perform accurate simulations of any nano-scale system of comparable complexity. The type of simulation that a quantum computer does results in an exact prediction of how a system will behave in nature—something that is literally impossible for any traditional computer, no matter how powerful.

Once quantum computers become available, engineers working at the nano-scale will be able to use them to model and design nano-scale systems just like today’s aerospace engineers model and design airplanes—completely virtually—with no wind tunnels (or their chemical analogues).

This may seem strange, but really it’s not. Think of it like this: conventional computers are really good at modeling conventional (that is, non-quantum) stuff—like automobiles and airplanes. Quantum computers are really good at modeling quantum stuff. Each type of computer speaks a different language.

Based in Vancouver, Canada, D-Wave is building a quantum computer using aluminum-based circuits. The company projects that by 2008 it will be building thumbnail-sized chips with more computing power than the aggregate total of all computers on the planet today and ever built in history, when applied to simulating the behavior and predicting the properties of nano-scale systems—highlighting the vast difference in capabilities of quantum and conventional computers. This would be of great value to the development of the nanotechnology industry. And it’s a jaw-dropping claim. Professor David Deutsch of Oxford summarized: “Quantum computers have the potential to solve problems that would take a classical computer longer than the age of the universe.”

While any physical experiment can be regarded as a complex computation, we will need quantum computers to transcend Moore’s law into the quantum domain to make this equivalence realizable. In the meantime, scientists will perform experiments. Until recently, the methods used for the discovery of new functional materials differed little from those used by scientists and engineers a hundred years ago. It was very much a manual, skilled labor-intensive process. One sample was prepared from millions of possibilities, then it was tested, the results recorded and the process repeated. Discoveries routinely took years.

Companies like Affymetrix, Intematix and Symyx have made major improvements in a new methodology: high throughput experimentation. For example, Intematix performs high throughput synthesis and screening of materials to produce and characterize these materials for a wide range of technology applications. This technology platform enables them to discover compound materials solutions more than one hundred times faster than conventional methods. Initial materials developed have application in wireless communications, fuel cells, batteries, x-ray imaging, semiconductors, LEDs, and phosphors.

Combinatorial materials discovery replaces the old traditional method by generating a multitude of combinations—possibly all feasible combinations—of a set of raw materials simultaneously. This "Materials Library" contains all combinations of a set of materials, and they can be quickly tested in parallel by automated methods similar to those used in the combinatorial chemistry and the pharmaceutical industry. What used to take years to develop now only takes months.

Timeline:

Given our discussion of the various factors affecting the commercialization of nanotech-nologies, how do we see them sequencing?

• Early Revenue

- Tools and bulk materials (powders, composites). Several revenue stage and public companies already exist in this category.

- 1D chemical and biological sensors. Out of body medical sensors and diagnostics

- Larger MEMS-scale devices

• Medium Term

- 2D Nanoelectronics: memory, displays, solar cells

- Hierarchically-structured nanomaterials

- Hybrid Bio-nano, efficient energy storage and conversion

- Passive drug delivery & diagnostics, improved implantable medical devices

• Long Term

- 3D Nanoelectronics

- Nanomedicine, therapeutics, and artificial chromosomes

- Quantum computers used in small molecule design

- Machine-phase manufacturing

- The safest long-term prediction is that the most important nanotech developments will be the unforeseen opportunities, something that we could not predict today.

In the long term, nanotechnology research could ultimately enable miniaturization to a magnitude never before previously seen, and could restructure and digitize the basis of manufacturing—such that matter becomes code. Like the digitization of music, the importance is not just in the fidelity of reproduction, but in the decoupling of content from distribution. New opportunities arise once a product is digitized, such as online music swapping—transforming an industry.

With replicating molecular machines, physical production itself migrates to the rapid innovation cycle of information technology. With physical goods, the basis of manufacturing governs inventory planning and logistics, and the optimal distribution and retail supply chain has undergone little radical change for many decades. Flexible, low-cost manufacturing near the point of consumption could transform the physical goods economy, and even change our notion of ownership—especially for infrequently used objects.

These are some profound changes to the manufacturing of everything, which ripples through the fabric of society. The science futurists have pondered the implications of being able to manufacture anything for $1 per pound. And as some of these technologies couple tightly to our biology, it will draw into question the nature and extensibility of our humanity.

Genes, Memes and Digital Expression:

These changes may not be welcomed smoothly, especially with regard to reengineering the human germ line. At the societal level, we will likely try to curtail “genetic free speech” and the evolution of evolvability. Larry Lessig predicts that we will recapitulate the 200-year debate about the First Amendment to the Constitution. Pressures to curtail free genetic expression will focus on the dangers of “bad speech”, and others will argue that good genetic expression will crowd out the bad, as it did with mimetic evolution (in the scientific method and the free exchange of ideas). Artificial chromosomes with adult trigger events can decouple the agency debate about parental control. And, with a touch of irony, China may lead the charge.

We subconsciously cling to the selfish notion that humanity is the endpoint of evolution. In the debates about machine intelligence and genetic enhancements, there is a common and deeply rooted fear about being surpassed—in our lifetime. When framed as a question of parenthood (would you want your great grandchild to be smarter and healthier than you?), the emotion often shifts from a selfish sense of supremacy to a universal human search for symbolic immortality.

Summary:

While the future is becoming more difficult to predict with each passing year, we should expect an accelerating pace of technological change. We conclude that nanotechnology is the next great technology wave and the next phase of Moore’s Law. Nanotech innovations enable myriad disruptive businesses that were not possible before, driven by entrepreneurship.

Much of our future context will be defined by the accelerating proliferation of information technology—as it innervates society and begins to subsume matter into code. It is a period of exponential growth in the impact of the learning-doing cycle where the power of biology, IT and nanotech compounds the advances in each formerly discrete domain.

So, at DFJ, we conclude that it is a great time to invest in startups. As in evolution and the Cambrian explosion, many will become extinct. But some will change the world. So we pursue the strategy of a diversified portfolio, or in other words, we try to make a broad bet on mammals.

© 2003 Steve T. Jurvetson. Reprinted with permission.

THE ADVENT OF THE home molecular biology laboratory is not far off. While there is no Star Trek “Tricorder” in sight, the physical infrastructure of molecular biology is becoming more sophisticated and less expensive every day. Automated commercial instrumentation handles an increasing fraction of laboratory tasks that were once the sole province of doctoral level researchers, reducing labor costs and increasing productivity. This technology is gradually moving into the broader marketplace as laboratories upgrade to new equipment. Older, still very powerful instruments are finding their way into wide distribution, as any cursory tour of eBay will reveal.¹ These factors are contributing to a proliferation that will soon put highly capable tools in the hands of both professionals and amateurs worldwide. There are obvious short term risks from increased access to DNA synthesis and sequencing technologies, and the general improvement of technologies used in measuring and manipulating molecules will soon enable a broad and distributed enhancement in the ability to alter biological systems. The resulting potential for mischief or mistake causes understandable concern—there are already public calls by scientists and politicians alike to restrict access to certain technologies, to regulate the direction of biological research, and to censor publication of some new techniques and data. It is questionable, however, whether such efforts will increase security or benefit the public good. Proscription of information and artifacts generally leads directly to a black market that is difficult to monitor and therefore difficult to police. A superior alternative is the deliberate creation of an open and expansive research community, which may be better able to respond to crises and better able to keep track of research whether in the university or in the garage.

FACTORS DRIVING THE BIOTECH REVOLUTION

The development of powerful laboratory tools is enabling ever more sophisticated measurement of biology at the molecular level. Beyond its own experimental utility, every new measurement technique creates a new mode of interaction with biological systems. Moreover, new measurement techniques can swiftly become means to manipulate biological systems. Estimating the pace of improvement of representative technologies is one way to illustrate the rate at which our ability to interact with and manipulate biological systems is changing.

For example, chemically synthesized DNA fragments, or oligonucleotides, can be used in DNA computation, in the fabrication of gene expression arrays (“gene chips”), and to make larger constructs for genetic manipulation. Mail-order oligonucleotides were with much fanfare recently used to build a functional poliovirus genome from constituent molecules for the first time.² The rate at which DNA synthesis capacity is changing is thus a measure of the improvement in our ability to manipulate biological systems and biological information. Similarly, improvements in DNA sequencing capabilities are a measure of our ability to read biological information; in particular the ability to proofread the results of DNA synthesis. Here I refer to such technology, whether instrument or molecule, as “biological technology.”

THE PACE OF TECHNOLOGICAL CHANGE THROUGH THE PRISM OF MOORE’S LAW

Figure 1 contains estimates of potential daily productivity of DNA synthesis and sequencing based on commercially available instruments, including the time necessary to prepare samples. There have been only a few generations of instruments—there is thus a limited amount of data for examination. These estimates are not intended to absolutely quantify a rate of change, but rather to capture the essence of the trends. Several technologies used in protein structure determination show similar trends (Figure 2), suggesting a general rapid improvement of biological technologies. As a reference, Moore’s Law, which describes the doubling time of the number of transistors on microchips, is also shown in Figure 1.

FIG. 1. On this semi-log plot, DNA synthesis and sequencing productivity are both increasing at least as fast as Moore?s Law (upwards triangles). Each of the remaining points is the amount of DNA that can be processed by one person running multiple machines for one eight hour day, defined by the time required for preprocessing and sample handling on each instrument. Not included in these estimates is the time required for sequence analysis. For comparison, the approximate rate at which a single molecule of E. coli DNA Polymerase III replicates DNA is shown (dashed horizontal line), referenced to an eight-hour day.

Sample processing time and cycle time per run for instruments in production are based on the experience of the scientific staff of the Molecular Sciences Institute and on estimates provided by manufacturers. ABI synthesis and sequencing data and Intel transistor data courtesy of those corporations. Pyrosequencing data courtesy of Mostafa Ronaghi at the Stanford Genome Technology Center. GeneWriter data courtesy of Glen Evans, Egea Biosciences. Projections are based on instruments under development.

Comparing anything to Moore’s Law is already a cliché, but doing so remains a useful device to gauge our expectations of how other technologies will affect socioeconomic change. This comparison starts with the observation that chip doubling times are a consequence of the planning intrinsic to the semiconductor and computer industry.³ Moore’s Law is primarily a function of the capital cost and resource allocation necessary to build chip fabrication plants. In addition, for much of the last thirty years there was feedback between the ability to design new chips and the computational power of the chips used in the design process.

We can now see the beginnings of a similar effect in the development of biological technologies. For example, enzymes optimized for laboratory conditions are used in the preparation of DNA for sequencing, where earlier sequencing technologies were part of characterizing and modifying those enzymes. Recombinant proteins are used every day to elucidate interactions between proteins within organisms, and that information is already being used to design and build new protein networks. Enzymes are directly used in a process known as Pyrosequencing,⁴ and its performance (Figure 1) is an indication of what will happen when we begin to manipulate biology, using biology, on a large scale at many levels of complexity.

FIG. 2. The dramatic improvement in the time required to determine protein structures is evidence of a general trend towards increased productivity in biological technologies. Many of the technologies used in finding protein structures are used widely in biology for other purposes. Raw estimates of time to collect and crystallize recombinant proteins, to take x-ray data, and to build structural models were compiled by Richard Yu (The Molecular Sciences Institute, Berkeley, CA) based on his experience and a survey of five additional crystallographers. From these estimates, the shortest time and mean time to find protein structures were computed. The time required for each step can vary significantly depending upon the protein. For example, successful crystallization may take anywhere between hours and months of effort. The difference between the estimates of the average time to structure and the shortest time to structure illustrates the difficulty in absolutely quantifying productivity.

Other observers have compared increases in the total number of sequenced genes to Moore’s law. But this mixes proverbial apples and oranges because total sequencing productivity is a measure of total industrial capacity (the number of sequencing instruments produced and in operation) whereas the number of transistors is ostensibly a measure of the potential productivity improvements enabled by each individual computer. The total number of sequenced genes is more analogous to the total number of computer chips in existence, or possibly the total number of computational operations enabled by those chips. Comparing Moore’s Law to estimates of the daily productivity of one person at a biology laboratory bench is appropriate because that productivity determines how much benefit, or havoc, one person can generate.

An alternative statement of Moore’s Law is “Computational resources for a fixed price double every 18 months.” Assuming for a moment that the cost of appropriately skilled labor has remained constant, the units of the vertical axis in Figure 1, “bases synthesizedâ…Ðf=GET http://www.neurosurgery-online.com/b day,” match the metric of resource cost, which is explicitly labor in this case. Note that this assumption is too conservative. Labor costs associated with sequencing have actually fallen as bench top laboratory techniques that once required a doctorate’s worth of experience have been replaced by automated processes that can be monitored by a technician with only limited training (see below). The capability of individuals has improved dramatically over the last 15 years.

The rapid increase in sequencing productivity is the primary reason that the private effort by Celera was able to sequence the human genome so quickly. Money was always available to buy many slow machines—this was, after all, the original plan for the publicly funded genome project—but coordinating the effort and paying for the labor to run those machines was prohibitively expensive for a private project. The advent of new technology provided an opportunity for a new approach, which Celera seized. Only when sequencing instruments became sufficiently automated that labor was reduced to loading samples, whereupon one person could shepherd several machines and the total task could be completed in an interesting time interval, was a commercial effort possible. This required highly centralized sequence production facilities in order to minimize the number of instruments, given their individual high cost. This infrastructure is similar to that of microchip fabrication plants, otherwise known as “chip fabs.”

However, because sequencing instruments are much closer to commodities than are the plasma etchers and vapor deposition systems used in microchip production, it is not at all obvious that the current centralized model will be relevant to the future of biological technology. On the contrary, because there has been to date only limited feedback between biological discovery and the technology it enables, it seems more likely that low cost, highly capable instrumentation will be broadly distributed. Sequencing machines are already widespread in laboratories and there is clear demand for faster, cheaper instruments.

More significantly, the long term distribution and development of biological technology is likely to be largely unconstrained by economic considerations. While Moore’s Law is a forecast based on understandable large capital costs and projected improvements in existing technologies, which to a great extent determined its remarkably constant behavior, current progress in biology is exemplified by successive shifts to new technologies. These technologies share the common scientific inheritance of molecular biology, but in general their implementations as tools emerge independently and have independent scientific and economic impacts. For example, the advent of gene expression chips spawned a new industrial segment with significant market value. Recombinant DNA, gel and capillary sequencing, and monoclonal antibodies have produced similar results. And while the cost of chip fabs has reached upwards of one billion dollars per facility and is expected to increase, there is good reason to expect that the cost of biological manufacturing and sequencing will only decrease. Indeed, the continuing costs of sequencing (expendables such as reagents) have fallen exponentially over the time period covered by Figure 1.⁵ Lander et al., state in Nature that by 2000 the total costs of sequencing had fallen by a factor of 100 in ten years, with costs falling by a factor of 2 approximately every eighteen months.⁶ With the caveat that there are only limited data to date, it does appear that the total cost of sequencing and synthesis are falling exponentially (Figure 3).

These trends—successive shifts to new technologies and increased capability at decreased cost—are likely to continue. In the fifteen years that commercial sequencers have been available, the technology has progressed, using the simple metric in Figure 1, from labor intensive gel slab based instruments, through highly automated capillary electrophoresis based machines, to the partially enzymatic Pyrosequencing process. These techniques are based on chemical analysis of many copies of a given sequence. New technologies under development are aimed at directly reading one copy at a time by directly measuring physical properties of molecules, with a goal of rapidly reading genomes of individual cells. These include efforts to measure differences in ion currents due to size variations between bases as DNA is electrophoresed through a small pore,⁷ and measuring differences in force between complimentary bases as double stranded DNA is unzipped by pulling it apart with an atomic force microscope.⁸ While physically-based sequencing techniques have historically faced technical difficulties inherent in working with individual molecules, an expanding variety of measurement techniques applied to biological systems will likely yield methods capable of rapid direct sequencing.

FIG. 3. Rough estimates of the cost of synthesis and raw sequencing per base. Only very limited data are available. Estimates of synthesis costs are from John Mulligan, Blue Heron Biotechnology. Historical costs of sequencing are generally not available in the literature, have not been publicized by federally funded Genome Centers, and are, in general, surprisingly hard to come by:44(1) from Lander et al.;6 (2) from Dan Rokhsar, UC Berkeley; (3) approximate current commercial rate.

ANTICIPATING THE FUTURE OF SYNTHESIS AND SEQUENCING

A rough extrapolation of the curves (as opposed to their tangents) in Figure 1 suggests that by 2010 a single person will be able to sequence or synthesize 10¹⁰ bases a day. These potential productivity numbers should be compared to the three billion bases (3 X 10⁹) in the human genome. Note that while automation may make this productivity level technologically feasible, costs may prohibit reaching it (see Figure 3). Even if actual technological developments do not sustain current trends, the drive towards automation and integration will certainly continue, enhancing distribution. This is the explicit goal of numerous commercial endeavors, particularly those intent on producing the tantalizing “lab on a chip.” Tools of this kind will be particularly powerful in the context of the current labor-consuming processes involved in preparing samples for sequence or expression analysis or in purifying them after synthesis. The “microdiagnostics” company Cepheid, for example, will soon begin selling its GeneXpert Platform, which includes technology that spans sample preparation, purification, and detection of pathogen DNA, potentially reducing sample analysis from days to minutes.^9,10 Scientists, clinicians, first responders, epidemiologists, biological weapons inspectors, and biological weapons producers will appreciate these capabilities equally.

While it is still early in the development of such platforms, they promise to be another important shift in technology, perhaps helping realize the trends in Figure 1. If those trends are born out, within a decade a single person at the lab bench could sequence or synthesize all the DNA describing all the people on the planet many times over in an eight-hour day, even given profligate human reproduction. Alternatively, one person could sequence his or her own DNA within seconds.

Despite the fantastic nature of these numbers, there is no physical reason why sequencing an individual human genome should take longer than a few minutes. Sequencing a billion bases in a thousand seconds would require querying each base for only a microsecond, which is well within the measurement capability of many physical systems. Inexpensive disk drives, for example, already read the state of magnetic domains at upwards of a billion times a second. Although storage media is an example of a mature technology, it is also an indication of the sort of interaction that will be possible with biological systems. Indeed, it seems unwise to assume limits on potential applications of our newly developing ability to manipulate and probe matter at the scale of individual molecules. Every week there are exciting new examples of imaging and manipulation of molecules, or small objects such as carbon nanotubes, each pushing back previously imagined limits. Figure 1 illustrates how fast an individual enzyme can copy DNA, and hybrid techniques utilizing physical measurement of the activity of individual enzymes may provide extremely rapid sequencing.¹¹ Yet at some point, despite ever increasing speed, sequencing capabilities will likely reach an asymptote in utility— how fast is fast enough?

This raises the question of how much longer effort put into developing rapid sequencing technology will be a wise investment. The greater challenge is sensitivity—biology comes in units of single cells, which is the level we must work at to reprogram biological systems and deal with many diseases. Cancer is one such disease. Generally it is not a whole organ or tissue that becomes cancerous, but rather one cell that breaks loose from its developmental pathway—due to chance mutations, changes in the environment, or infection—and runs amok. Similarly, many infections essentially begin with attacks by individual pathogens on individual cells, even if many simultaneous such events are necessary to produce full-blown symptoms. This is of obvious concern for scientists and clinicians interested in novel pathogens, both natural and artificial. Yet no technology currently emerging from the bench can sequence the genome of a single cell without amplification steps that introduce significant errors (though many academic labs and companies promise this ability soon). Most current technology, particularly that applied to determining interactions between proteins, requires a large number of cells and thus produces data that is an average over the states of those cells. Investigating the metabolic or proteomic state of cells (without the use of genetic modification) is similarly often predominantly limited to large samples.

Regardless of the direction of technological development, the synthesis and sequencing capabilities available to an individual in the next decade will be impressive, greatly facilitating the task of manipulating biological systems. The cost of each instrument should generally decrease, following the trend of similar commodities, suggesting that the infrastructure of biological technology will be highly distributed. One indication of this trend is that the parts for a DNA synthesizer—mostly plumbing and off-the-shelf electronics—can now be purchased for approximately $10,000. The assembly effort and monetary sum are similar to that expended by many car and computer hobbyists, and both the parts list and design information sufficient to assemble the synthesizer are available online.¹²

Despite existing infrastructure that provides for downloading sequences directly into a synthesizer, possession of a DNA synthesizer does not a new organism make. Current chemical synthesis produces only short runs of DNA. Although ingenuity and care are required to assemble full-length genes, the techniques are already described in the scientific literature. Moreover, there is significant economic motivation to make such assembly routine, and multiple companies have been founded to sort out the relevant manufacturing details and to take advantage of the growing demand for long synthetic DNA sequences. Many of these companies provide synthetic DNA via mail based on sequences submitted over the Web, and not all such companies screen ordered sequences against sequences of known pathogens and biological toxins.¹³ Even if great care is taken to limit the commercial synthesis of DNA from pathogens or toxins, it is unlikely the chemical tricks and instrumentation that companies develop in the course of building their businesses will remain confined within their walls. Eventually, efficient synthesis will be possible using instruments assembled at home. The diffusion of synthesis capability into the garage will no doubt be slowed by the fact that some the reagents used in chemical DNA synthesis are controlled substances. However, history demonstrates that regulating the synthesis of even complex compounds does not greatly inhibit illicit production (see below). The requisite techniques are in fact already highly distributed.

THE PROLIFERATION OF SKILLS AND MATERIALS IS INEVITABLE

Beyond information about writing DNA from scratch, extensive instructions on standard chemistry and molecular biology techniques are available on the Web, notably detailed descriptions of PCR (polymerase chain reaction) and other important DNA manipulation procedures. While some skills are still highly specialized, basic know-how is permeating the educational process.¹⁴ For several years community colleges have offered courses of study aimed at providing the biotech industry with skilled technicians. A case in point: when it was founded in 1990, the sequencing facility at the Whitehead Institute Center for Genome Research employed primarily scientists with doctorates. Over the years these PhD’s were gradually replaced by masters degrees, then bachelors and associates degrees. Now many of the staff have completed only a six month qualification course at local community college or are recent Tibetan immigrants who received training in basic skills at the Institute.¹⁵ These technicians are educated in all the steps necessary to shepherd DNA from incoming sample to outgoing sequence information, including generating bacteria containing DNA from other organisms. This point bears repeating: Creating genetically modified organisms is now the province of immigrants with little formal education. More sophisticated practical knowledge is available to many AP Biology students in high school. Pointing the way into the future, several universities now teach a Molecular Biology for Engineers class. Exploring the limits of this trend is a class taught at MIT wherein students ranging from undergraduates to post-docs design and test new genetic circuits.¹⁶ Successful designs will be included in a databook of biological parts.¹⁷

Where design expertise exceeds practical experience, commercially available kits include recipes that allow moving genes between organisms by following simple recipes. The process might be slightly more complicated than baking cookies, but it is for the most part less complicated than making wine or beer. This broad distribution of biological technology naturally leads to questions of how it will be applied. Our society is just beginning discussions about the role of genetic modifications and the applications of cellular cloning.

More important, perhaps, is the debate over regulation of research and who will be permitted access to which biological technologies. But it is unlikely that regulation of materiel or skills will produce an increase in public safety. The industrial demand alone for skilled biotechnology workers has increased 14–17% per year for the last decade, and many of these workers come from overseas.¹⁸ Not all these workers will remain in this country, and it is safe to say many of those who leave will make use of their skills elsewhere. If we decide to try to limit the practice of certain methods, it will be unrealistic to try to centrally monitor every skilled person in this or any other country. We certainly cannot simply “unteach” the relevant skills to prevent unauthorized use, and any action to limit the proliferation of skills would cripple that portion of the U.S. economy reliant upon biological technologies.

Perhaps more problematic than distributed skills will be ubiquitous materials. The widespread distillation of alcohol during the Prohibition period in the U.S. and the proliferation of modern illegal drug synthesis labs both illustrate the principle that outlawing chemical products merely leads to black markets more difficult to observe and regulate than open markets.

Effective regulation relies on effective enforcement, which in turn requires effective detection. The extent of illegal drug production in the United States and previous failures to detect illicit biological weapons production gives some indication of the relevant challenges of detection and enforcement. Approximately 8,000 clandestine drug laboratories were seized in the U.S. in 2001, with the vast majority of those being “Mom and Pop” operations producing less than five kilograms per day.¹⁹ Yet despite the large number of seizures (which has on average remained constant for the last decade) illegal drug use is apparently still rising.²⁰ This failure of enforcement, and the detection failure demonstrated when Western intelligence services failed to uncover the existence of extensive bioweapons programs in the former Soviet Union and Iraq,²¹ provide explicit challenges to the notion that the risks posed by mistakes or mischief resulting from biological technologies can be mitigated through regulation.

Given the potential power of biological technologies, it is worth considering whether open markets are more, or less, desirable than the inevitable black markets that would emerge with regulation. Those black markets would be, by definition, beyond regulation. More importantly, in this case, they would be opaque.

The real threat from distributed biological technologies lies neither in their development nor use, per se, but rather that biological systems may be the subject of accidental or intentional modification without the knowledge of those who might be harmed. Because this may include significant human, animal, or plant populations, it behooves us to maximize our knowledge about what sort of experimentation is taking place around the world. Unfortunately (though understandably), the first response to incidents such as the anthrax attacks in the fall of 2001 is to attempt to improve public safety through means that paradoxically often limit our capabilities to gather such information.²²

THE FALSE PROMISE OF REGULATION

Some view as an immediate threat the proliferation of technologies useful in manipulating biological systems: Passionate arguments are being made that research should be slowed and that some research should be avoided altogether. “Letting the genie out of the bottle” is a ubiquitous concern, one that has been loudly voiced in other fields over the years and is meant to set off alarm bells about biological research.

A favorite rhetorical device in this discussion is the comparison of nuclear technologies with biological technologies. Success in limiting the development and spread of nuclear technologies is taken to mean similar feats are possible with biological technologies. But this sort of argument fails to consider the logistical, let alone ethical, differences between embargoing raw fissionable materials used in nuclear or radiological weapons and embargoing biological technology or even biology itself.

Regulation of the development of nuclear weapons has been successful only because access to raw fissionable materials has, fortunately, been relatively easy to restrict. However, both the knowledge and tools necessary to construct rudimentary weapons have for decades been highly distributed. It is arguable that, with some effort, construction of a rudimentary nuclear device is within the capabilities of most physics and engineering college graduates who have access to a basic machine shop. Building nuclear devices is thus theoretically quite feasible but physically difficult, even for the knowledgeable, because the raw materials are simply not available. Yet the raw stuff of biology has always been readily at hand, and our schools and industries are now equipping students with the skills to manipulate biological systems through powerful and distributed technology. Because skills are already widespread and will only become more so, altering and reverse engineering biological systems will become both easier and more common. Regulation can do little to alter this trend.

If strict regulation held promise of real protection, it would be well worth considering. But regulation is inherently leaky, and it is more often a form of management than blanket prohibition. Certainly no category of crime has ever been eliminated through legal prohibition. In this light, we must ask how many infringements of potential regulation of biological technologies we are willing to risk. Further, will the threat of sanctions such as imprisonment ever be enough to dissuade infringement? Given the potential damage wrought by misuses of the technology, we may never be satisfied that such sanctions would constitute a repayment of debt to society, the fundamental tenet of our criminal justice system. The damages may always exceed any punishment meted out to those deemed criminal. These considerations come down to how we choose to balance the risks and consequences of infringement against whatever safety may be found in regulation and attempts at enforcement. More important than this tenuous safety, however, is the potential danger of enforced ignorance. In the end, we must decide not whether we are willing to risk damages caused by biological technology, but whether limiting the general direction of biological research in the coming years will enable us to deal with the outcome of mischief or mistake. We must decide if we are willing to take the risk of being unprepared.

There are currently calls to limit research in the United States on the basic biology of many pathogens to preempt their use as bioweapons,²² and the possession and transport of many pathogens was legislated into criminality by the Patriot Act.²³ The main difficulty with this approach is not that it assumes the basic biology of pathogens is static—which because of either natural variation or human intervention it is not—but rather that it assumes we have already catalogued all possible natural pathogens, that we already know how to detect and defeat known and unknown pathogens, and that rogue elements will not be able to learn how to manipulate pathogens and toxins on their own. These assumptions are demonstrably false. Pathogens ranging from HIV to M. tuberculosis to P. falciparum (which causes malaria) have successfully evolved to escape formerly effective treatments. New human pathogens are constantly emerging, which as in the case of SARS might be identified quickly but require much longer to develop treatments against. In the last century governments and independent organizations alike have developed and used biological weapons. Restricting our own research will merely leave us less prepared for the inevitable emergence of new natural and artificial biological threats. Moreover, it is naive to think we can successfully limit access to existing pertinent information within our current economic and political framework.

As is clear from recent efforts to limit peer-to-peer file sharing on the Internet, in today’s environment strict prohibition of information flow can only be achieved by quarantine—unplugging wires and blocking wireless transmission. Thwarted by the difficulty of such endeavors, music conglomerates have resorted to flooding file servers with corrupted files (camouflage),²⁴ and requesting the legal authority to engage in preemptive cracking of file trader’s computers (sabotage).²⁵

Neither strategy is likely to be a long term solution of controlling information for the music industry, and similar efforts to regulate biological technologies are bound to be more difficult still. Attempting to maintain control of information and instrumentation will be a futile task in light of the increasingly sophisticated biological technologies blossoming around the world.

While the most advanced research and instrumentation developments may occur first in fully industrialized countries such as the US, where export might be controlled, other countries are developing a skill base that will enable broad domestic utilization of biological technologies. China has an aggressive program in plant biotechnology, and as of 2002 plans to increase funding 400% by 2005.²⁶ This energetic investment also exists in the Chinese private sector, and the national scientific establishment is attempting to lure foreign trained scientists to return with lucrative financial packages.²⁷ India is in the process of tripling funding to its national biotech center,²⁸ and is promoting the development and use of genetically modified crops throughout Asia.²⁹ Singapore has for many years made a practice of recruiting foreign scientists.³⁰ Taiwan is investing large amounts in biotechnology³¹ and is seeking citizens to return home to build up biotechnology in academia and industry.³² A Brazilian coalition recently demonstrated sophisticated domestic use of biological technologies by successfully sequencing the plant pathogen X. fastidiosa in 2000.³³

Given these developments in the context of the increase in individual capabilities and the independent reduction in cost, it is unrealistic to think biological technologies can be isolated within the borders of officially sanctioned countries. Even if such a regime were implemented, it would merely include those countries that already have a particular technology. We can do little to take technology away from those in whose hands it was developed and resides. The best strategy going forward is in fact to encourage such efforts at all levels in an open environment.

WHAT SHOULD (AND SHOULD NOT) BE DONE

If regulation is not merely an ineffective option but will actually be an impediment to security, how can we attempt to mitigate coming risks? The goal is clearly to counter both mistakes in the laboratory and weapons created from biological components and, ideally, to make such threats irrelevant before they become a problem.

It may be many decades before our understanding of biology provides for the requisite rapid detection, analysis, and response. Fortunately, it is also probably true that we have some time to prepare before both technology and skills become truly pervasive. In the meantime, we can lay the groundwork for an increase in security with dramatically improved communication and focused technology development.

We should focus on three challenges:

1) We should resist the impulse to restrict research and the flow of information. Ignorance will help no one in the event of an emergent threat and, given the pace and proliferation of biological technologies, the likelihood of threats will increase in coming years. Among the greatest threats we face is that potentially detrimental work will proceed while we sit on our hands. If we are not ourselves pushing the boundaries of what is known about how pathogens work or ways to manipulate them, we are by definition at a disadvantage. Put simply, it will be much easier to keep track of what is in the wind if we don’t have our heads in the sand.

2) The best way to keep apprised of the activities of both amateurs and professionals is to establish open networks of researchers, perhaps modeled on the Open Source Software (OSS) movement, and potentially sponsored by the government during their embryonic phases. The Open Source development community thrives on constant communication and plentiful free advice. This behavior is common practice for professional biology hackers, and it is already evident on the Web amongst amateur biology hackers.¹⁴ This represents an opportunity to keep apprised of current research in a distributed fashion. Anyone trying something new will require advice from peers and may advertise at least some portion of the results of their work. As is evident from the ready criticism leveled at miscreants in online forums frequented by software developers (Slashdot, Kuro5hin, etc.), people are not afraid to speak out when they feel the work of a particular person or group is substandard or threatens the public good. Thus our best potential defense against biological threats is to create and maintain open networks of researchers at every level, thereby magnifying the number of eyes and ears keeping track of what is going on in the world.

3) Because human intelligence gathering is, alas, demonstrably inadequate for the task at hand, we should develop technology that enables pervasive environmental monitoring. The best way to detect biological threats is using biology itself, in the form of genetically modified organisms. Unlike the production and deployment of chemical weapons or fissile materials, which can often be monitored with remote sensing technologies such as aerial and satellite reconnaissance, the initial indication of biological threats may be only a few cells or molecules. This small quantity may already be a lethal dose and can be very hard to detect using physical means. Alternatively, “surveillance bugs” distributed in the environment could transduce small amounts of cells or molecules into signals measurable by remote sensing. The organisms might be modified to reproduce in the presence of certain signals, to change their schooling or flocking behavior, or to alter their physical appearance. Candidate “detector platforms” span the range of bacteria, insects, plants, and animals. Transgenic zebrafish³⁴ and nematodes³⁵ have already been produced for this purpose, and there is some progress in producing a generalized system for detecting arbitrary molecules using signal transduction pathways in bacteria.³⁶

None of these recommended goals will be trivial to accomplish. Considerable sums have already been spent over the last five decades to understand biological systems at the molecular level, much of this in the name of defeating infectious disease. While this effort has produced considerable advances in diagnosing and treating disease, we should now redouble our efforts. We have entered an era when the ability to modify biological systems is becoming widespread in the absence of an attendant ability to remediate potential mistakes or mischief. Maintaining safety and security in this context will require concerted effort, and an immediate, focused governmental R&D investment would be a good start. Although “bug to drug in twenty four hours” sounds much flashier than “bug to drug in six to eight weeks,” the latter is the more realistic timeline to shoot for—even if it is a decade or more away—and this goal may serve as an organizational focus for an endeavor organized and sponsored by the government.

Previous governmental efforts to rapidly develop technology, such as the Manhattan and Apollo Projects, were predominantly closed, arguably with good reason at the time. But we live in a different era and should consider an open effort that takes advantage of preexisting research and development networks. This strategy may result in more robust, sustainable, distributed security and economic benefits.^14,37 Note also that though both were closed and centrally coordinated, the Manhattan and Apollo Projects were very different in structure. The Apollo Project took place in the public eye, with failures plainly writ in smoke and debris in the sky. The Manhattan Project, on the other hand, took place behind barbed wire and was so secret that very few people within the US government and military knew of its existence. This is not the ideal model for research that is explicitly aimed at understanding how to modify biological systems. Above all else, let us insist that this work happens in the light, subject to the scrutiny of all who choose to examine it.

The only way we will be able to keep track of the fruits of biological technologies, regardless of merit, is a combination of ubiquitous measurement and networks of people. For several decades, the Soviet Union employed tens of thousands of people in research, testing, and production of biological weapons.³⁸ During that time, the USSR was the primary focus of intelligence agencies in the West, and, despite the size of the Soviet bioweapons project, none of those agencies was able to provide conclusive evidence of the project’s existence. The extent of biological weapons development and deployment in Iraq during the early 1990′s was also an unpleasant surprise.³⁹ A more integrated worldwide community of professionals and amateurs might provide earlier and more accurate warning of such developments.

Beyond their innate intelligence gathering capability, open and distributed networks of researchers would provide a flexible and robust workforce for developing technology. This resource could be employed in rapid reaction to emerging threats and in the development of a response that might include assembling novel compounds or organisms. The rudiments of such a system were demonstrated during the recent SARS outbreak, but much more is required.⁴⁰ One lesson from the OSS community is that even distributed technology development that starts at the grass roots level eventually requires some centralized leadership and coordination.⁴¹ This is often provided by a strong-willed individual, though increasingly independent foundations are formed to coordinate work, gather and distribute funds, and disseminate results.⁴²

Some may consider several decades of experience with open source software insufficient as an organizational model to serve as a basis for a response to biological threats. The best model may in fact be found in the history of biology itself. In order to bring the focus of an Apollo Project to the task at hand, the traditions of open discourse amongst academics and the sharing of reagents and biological stocks might be strengthened and adapted. Hoarding of results or materials should be strongly discouraged, and in fact sharing information and stocks should be required. It may be prudent to write down these guidelines in documents with legal standing, if only to give added weight to peer pressure. To be sure, this might be viewed as a form of self-regulation, but it would be in the context of open markets rather than black markets emerging under regulation from above. These agreements might be structured so that voluntary participation would provide ready access to information or reagents otherwise difficult to procure, thereby encouraging participation but not outlawing the activities of those who choose to remain independent. New or existing foundations might take these agreements in hand to provide coordination analogous to that cropping up in the OSS community. There is already some structure of this sort extant in the biological community, with organizations such as the American Cancer Society, the Wellcome Trust, the Bill and Melinda Gates Foundation, amongst many others, providing funding for meetings, journals, physical infrastructure, and particular directions of research.

Finally, the best argument for encouraging the adoption of Open Source organizational principles in amateur, academic, and industrial contexts is that the resulting technology may be considerably more robust and bug free.⁴³ This goal is nowhere more important than in the burgeoning enterprise of manipulating life at the genetic level. Creating international networks that coordinate an Open Source Biology may be the most important step we can take to improve our security in the coming decades.

CONCLUSION

Our ability to manipulate biological systems is rapidly improving and this naturally raises concerns both about how relevant technology will be applied and about potential consequent dangers. The straightforward answer is that those dangers are real and considerable. We may view this as a threat or an opportunity. The common response to a perceived threat is to reduce the likelihood of it coming to fruition, an effort that often takes the form of regulation. However, the argument for strict regulation of biological technologies is misleading and therefore dangerous. Fear of potential hazards should be met with increased research and education rather than closing the door on the profound positive impacts of biological technology.

We could err disastrously in the short term by restricting the development of science and technology, thereby stunting our ability to respond to natural or artificial threats. Restriction of research could leave us woefully unprepared to deal with mistakes or mischief. I am not suggesting that all regulation is without merit, but rather that rules and restrictions will not eliminate problems; they never have. Given the power of biological technologies, the only way to ensure safety in the long run is to push research and development as fast as possible.

We should maintain an open environment as possible and make sure that we move rapidly beyond the point where we can alter systems without the ability to understand them or learn to fix them. Improving such capabilities will also aid in diagnosing and treating rapidly emerging natural pathogens. The existing technology lag between our ability to manipulate and our ability to detect, understand, and remediate must be eliminated with all haste. Regulation or proscription of either science or technology is unlikely to ease the way forward. In the dark we cannot see the road ahead, navigate, or avoid collisions with either natural or artificial hazards.

Regardless of the outcome of the debate explored above, the stage is set for remarkable change. We have clearly entered a period in which our understanding of biological systems is itself producing new biologically based technologies. These in turn lead to new insight and new technologies, further enhancing our ability to understand and manipulate biological systems. The demand for more capable technology is both broad and deep suggesting that, as the trend to increasingly sophisticated yet less expensive instrumentation continues, biological technology will become ever more commoditized. The resulting wide distribution will further accelerate discovery and invention.

ACKNOWLEDGMENTS

The author wishes to thank Drew Endy, Roger Brent, Stewart Brand, Freeman Dyson, Sydney Brenner, Rik Wehbring, Brad Smith, and Sarah Keller for thoughtful conversations, and Richard Yu, Robert Waterston, Dan Rokhsar, Mostafa Ronaghi, Glen Evans, and John Mulligan for estimates of cost and/or productivity.

__________________________________________________________

¹See http://listings.ebay.com/pool1/listings/list/all/category 11811/index.html.

²Cello J, Paul AV, Wimmer E. Chemical Synthesis of Po­liovirus cDNA: Generation of Infectious Virus in the Absence of Natural Template. Science 2002. 297(5583): p. 1016–1018.

³Moore, G. Cramming more components onto integrated circuits. Electronics 1965. 38(8).

⁴Ronaghi M. Pyrosequencing Sheds Light on DNA Sequencing. Genome Research 2001. 11(1): p. 3–11.

⁵Editorial. Genome Technology, 2001.

⁶Lander ES, et al., Initial sequencing and analysis of the human genome. Nature 2001. 409(6822): p. 860–921.

⁷Meller A, et al. Rapid nanopore discrimination between single polynucleotide molecules. PNAS, 2000. 97(3): p. 1079–1084.

⁸Bockelmann U, Thomen P, Heslot F. Unzipping DNA with High Sequence Resolution. European Biophysics Journal 2000. 29(4–5): p. 249.

⁹Jones M, et al. Rapid and Sensitive Detection of Mycobacterium DNA using Cepheid SmartCycler and Tube Lysis System. Clinical Chemistry, 2001. 47(10): p. 1917–1918.

¹⁰MT Taylor PB, Joshi R, Kintz GA, Northrup MA. Fully Automated Sample Preparation for Pathogen Detection Performed in a Microfluidic Cassette. Micro Total Analysis Sys­tems 2001: p. 670–672.

¹¹Braslavsky I, et al. Sequence information can be obtained from single DNA molecules. Proc Natl Acad Sci U S A 2003. 100(7): p. 3960–4.

¹²http://innovation.swmed.edu/Instrumentation/mermade_ oligonucleotide_synthesi.htm.

¹³John Mulligan, Personal Communication.

¹⁴Carlson R. Open-Source Biology And Its Impact on Industry. IEEE Spectrum 2001.

¹⁵Carrie Sougnez, Personal Communication.

¹⁶See http://student.mit.edu/searchiap/iap-4968.html and http:// web.mit.edu/synbio/www/iap/.

¹⁷See http://biobricks.ai.mit.edu/.

¹⁸Sevier ED, Dahms AS. The role of foreign worker scientists in the US biotechnology industry. Nat Biotechnol 2002. 20(9): p. 955–6.

¹⁹http://www.dea.gov/concern/drug_trafficking.html.

²⁰http://www.dea.gov/statistics.html.

²¹Pearson GS. How to make microbes safer. Nature 1998. 394(6690): p. 217–8.

²²Knight J. Biodefence boost leaves experts worried over laboratory safety. Nature 2002. 415(6873): p. 719–20.

²³Check E. Law sends laboratories into pathogen panic. Nature 2003. 421(6918): p. 4.

²⁴Chmielewski DC. Music industry swamps swap networks with phony files, in Mercury News 2002: San Jose.

²⁵Bridis T. Senator favors really punishing music thieves, in Tribune 2003: Chicago.

²⁶Huang J, et al. Plant Biotechnology in China. Science, 2002. 295: p. 674–678.

²⁷Breithaupt H. China’s leap forward in biotechnology. EMBO Rep, 2003. 4(2): p. 111–3.

²⁸Taylhardat AR, Falaschi A. Funding assured for India’s international biotechnology centre. Nature 2001. 409(6818): p. 281.

²⁹Jayaraman KS. India promotes GMOs in Asia. NatBiotechnol 2002. 20(7): p. 641–2.

³⁰Singapore attracts foreign talent. Nature 1998. 394: p. 604.

³¹Swinbanks D, Cyranoski D. Taiwan backs experience in quest for biotech success. Nature, 2000. 407(6802): p. 417–26.

³²Cyranoski D. Taiwan: Biotech vision. Nature 2003. 421: p. 672–673.

³³Simpson AJ, et al. The genome sequence of the plant pathogen Xylella fastidiosa. The Xylella fastidiosa Consortium of the Organization for Nucleotide Sequencing and Analysis. Nature 2000. 406(6792): p. 151–7.

³⁴Amanuma K, et al. Transgenic zebrafish for detecting mutations caused by compounds in aquatic environments. Nat Biotechnol 2000. 18(1): p. 62–5.

³⁵David HE, et al. Construction and evaluation of a transgenic hsp16-GFP-lacZ Caenorhabditis elegans strain for environmental monitoring. Environ Toxicol Chem 2003. 22(1): p. 111–8.

³⁶Looger LL, et al. Computational design of receptor and sensor proteins with novel functions. Nature 2003. 423(6936): p. 185–90.

³⁷R. Carlson and R. Brent, Letter to DARPA on Open Source Biology, October 2000, http://www.molsci.org/~rcarlson/DARPA_OSB_Letter.html.

³⁸Alibek K, Handelman S. Biohazard: the chilling true story of the largest covert biological weapons program in the world, told from the inside by the man who ran it. 1st ed. 1999, New York: Random House. xi, 319 p., [8] p. of plates.

³⁹Seelos C. Lessons from Iraq on bioweapons. Nature 1999. 398(6724): p. 187–8.

⁴⁰Pearson H, et al. SARS: What have we learned? Nature 2003. 424(6945): p. 121–126.

⁴¹For a very readable introduction to the structure of the Open Source community see http://www.theinquirer.net/?article10114 and http://www.theinquirer.net/?article10222.

⁴²For example, http://www.mozillafoundation.org/.

⁴³Ball P. Openness makes software better sooner. Nature Science Update June 25, 2003, http://www.nature.com/nsu/030623/030623-6.html.

⁴⁴Robert Waterston, Personal Communication.

__________________________________________________________

Address reprint requests to:

Robert Carlson Department of Electrical Engineering University of Washington Seattle, WA 98195

E-mail: rcarlson@molsci.org

© 2003 Mary Ann Liebert, Inc. Reprinted with permission.

This material is posted here with permission of the IEEE. Such permission of the IEEE does not in any way imply IEEE endorsement of any of KurzweilAI.net’s products or services. Internal or personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution must be obtained from the IEEE by writing to pubs-permissions@ieee.org. By choosing to view this document, you agree to all provisions of the copyright laws protecting it. Thank you.

In 50 years, you may be reading IEEE Spectrum on a leaf. The page will not actually look like a leaf, but it will be grown like a leaf. It will be designed for its function, and it will be alive. The leaf will be the product of intentional biological design and manufacturing.

Rather than being constantly green, the cells on its surface will contain pigments controlled by the action of something akin to a nervous system. Like the skin of a cuttlefish, the cells will turn color to form words and images as directed by a connection to the Internet of the day. Given the speed with which the cuttlefish changes its pigment, these pages may not change fast enough to display moving images, but they will be fine for the written word. Each page will be slightly thicker than the paper Spectrum is now printed on, making room for control elements (the nervous system) and circulation of nutrients. When a page ages, or is damaged, it will be easily recycled. It will be fueled by sugar and light.

Many of the artifacts produced in 50 years and used in daily living will have a similar appearance and a similar origin. The consequences of mature biological design and manufacturing will be widespread, and will affect all aspects of the economy, including energy and resource usage, transportation, and labor. Today, electronic paper and similar display technologies are just around the corner, but in the long run they will not be able to compete with the products of inexpensive, distributed biological manufacturing.

Growing engineered leaves for display devices may seem a complex biological engineering feat, but foundations for the technology are already being laid. Structurally simple replacement human tissues are currently being grown in the laboratory on frameworks of suture material. Projects to grow functional human heart tissue, and eventually a whole heart, are under way, with a timeline for completion of 10 years.

Genomic parts list

Within those 10 years, the genomes of many organisms will be sequenced, providing a parts list for the proteins forming the structural and control elements in those organisms. Biologists, engineers, and physicists are already collaborating on models that will help us understand how those parts work and fit together. The goal for these models is quantitative prediction of the behavior of biological systems, which will have profound implications for the understanding of basic biology and for improving human health.

Beyond initial biomedical consequences, models that can be used to predict the effects of perturbations to existing biological systems will become de facto design tools, providing an infrastructure for creating new technologies based on biology. When we can successfully predict the behavior of designed biological systems, then an intentional biology will exist. With an explicit engineering component, intentional biology is the opposite of the current, very nearly random applications of biology as technology.

For instance, the present debate over genetically modified foods is more indicative of the poorly planned use of an immature technology than a failure of the technology itself. At present we simply can’t predict the effects of tinkering with a system as complex as crops and their pests. But as with the progression of every other human technology, from fire, to bridges, to computers, biological engineering will improve with time. Quantitative models for simple systems like viral infections of bacteria and yeast signal transduction pathways are already being tested. Computational methods developed in those efforts will soon be applied to higher plants and animals. It is a short step from successful prediction to design and the beginning of industrial applications.

Yet even before the advent of true biological design, more general lessons from biology are already transforming our economy. The potential impact on industrial practices of learning from biology is enormous and is explored in the book Natural Capitalism, by Paul Hawken and Amory and L. Hunter Lovins (Little, Brown, London, 1999).

The authors point out that structuring business practices along biological lines can significantly improve the bottom line. The human circulatory system, for instance, is optimized to minimize the work required to pump blood throughout the body. The majority of industrial pumping systems, however, are optimized to minimize the cost of the pipes during construction. This means smaller pipes are used, requiring large pumps that use vastly more energy than necessary.

Similarly, in the human pumping system, the heart has to work too hard when arteriosclerosis reduces the diameter of blood vessels. These vessels then require maintenance in the form of an angioplasty. Industrial pumping systems are designed with built-in arteriolosclerosis, and fixing them requires rebuilding from the ground up. Paying careful attention to several hundred million years of nature’s trial-and-error design experience will save human industry considerable energy and resources.

A living industrial infrastructure

Borrowing a design aesthetic for industrial function from nature is just the beginning. The living world will also become part of our industrial infrastructure. Nature has already discovered how to fabricate materials and to finesse chemistry in ways that are the envy of human engineers and chemists. Many companies, both established and startup, are now focusing on harvesting enzymes from organisms in the environment for use in industrial processes.

Popular examples of high-strength materials fabricated by biology at low temperature, pressure, and energy cost are spider silk and abalone shell. Yet increased resource efficiency and biomaterials are only the first steps in a revolution in manufacturing. Beyond using biology as a model for the structure and function of industrial production, the year 2050 will see humans using biology as the means of production itself.

Whereas most manufacturing today is highly centralized and materials are transported long distances throughout the assembly process, in the year 2050 human industry will use distributed and renewable manufacturing based upon biology. Renewable manufacturing means that biology will be used to produce many of the physical things we use every day.

In early implementation, the organism of choice is likely to be yeast or a bacterium. The physical infrastructure for this type of manufacturing is inherently flexible: it is essentially the vats, pumps, and fluid-handling capacity found in any brewery. Production runs for different products would involve seeding a vat with a yeast strain containing the appropriate genetic instructions and then providing raw materials.

To be sure, there will always be applications and environments in which biological fabrication is not the best option, and it is not clear how complex the fabrication task can be, but biology is capable of fabrication feats impossible for any current or envisioned human technology to emulate. In some ways, this scheme sounds a bit like Eric Drexler’s nanotechnological assemblers, except that we already have functional nanotechnology—it’s called biology.

The transformation to an economy based on biological manufacturing will occur as technical manipulations become easier with practice and through a proliferation of workers with the appropriate skills. Biological engineering will proceed from profession, to vocation, to avocation, because the availability of inexpensive, quality DNA sequencing and synthesis equipment will allow participation by anyone who wants to learn the details. In 2050, following the fine tradition of hacking automobiles and computers, garage biology hacking will be well under way.

Considerable information is already available on how to manipulate and analyze DNA in the kitchen. A recent Scientific American Amateur Scientist column provided instructions for amplifying DNA through the polymerase chain reaction (PCR), and a previous column dealt with analyzing DNA samples on homemade electrophoresis equipment. The discussion was immediately picked up in a slashdot.org thread where participants provided tips for improving the yield of the PCR process.

More detailed, technical information can be found in any university biology library in Current Protocols in Molecular Biology, which contains instructions on how to perform virtually every task needed in modern molecular biology. This printed compendium has recently joined the myriad resources maintained on-line by universities and government agencies, thereby becoming all the more accessible. Open-source biology is already becoming a reality.

As the "coding" infrastructure for understanding, troubleshooting, and, ultimately, designing biology develops, DNA sequencers and synthesizers will become less expensive, faster, and ever simpler to use. These critical technologies will first move from academic labs and large biotechnology companies to small businesses, and eventually to the home garage and kitchen. Many standard laboratory techniques that once required a doctorate’s worth of knowledge and experience to execute correctly are now used by undergraduates in a research setting with kits containing color-coded bottles of reagents. The recipes are easy to follow.

This change in technology represents a democratization of sorts, and it illustrates the likely changes in labor structure that will accompany the blossoming of biological technology.

Distributed biological manufacturing

The course of labor in biological technology can be charted by looking at the experience of the computer and Internet industries. Many startup companies in Silicon Valley have become contract-engineering efforts, funded by venture capital, where workers sign on expecting the company will be sold within a few years, whereupon they will find a new assignment. The leading edge of the biological technology revolution could soon look the same. However, unlike today’s integrated circuits, where manufacturing infrastructure costs have now reached upward of US $1 billion per facility, the infrastructure costs for renewable biological manufacturing will continue to decline. Life, and all the evolutionarily developed technology it utilizes, operates at essentially room temperature, fueled by sugars. Renewable, biological manufacturing will take place anywhere someone wants to set up a vat or plant a seed.

Distributed biological manufacturing will be all the more flexible because the commodity in biotechnology is today becoming information, rather than things. While it is still often necessary to exchange samples through the mail, the genomics industry has already begun to derive income from solely selling information about gene expression, or which genes are turned on in a particular population of cells.

In a few decades it will be the genomic sequence that is sent between labs, there to be re-synthesized and expressed as needed. It is already possible to synthesize sufficient DNA to build a bacterial genome from scratch in a few weeks using chemical means. Over the coming decades, that time will be reduced to days, and then to hours, eventually via the development of directed, template-free, enzymatic synthesis—a DNA "synthase."

It is possible that the evolution of open-source biology will be delayed by retrenchment on the part of corporations trying to protect intellectual property. However, the future model of biology as a technological instrument of any corporation can be found by simply looking at the way life currently makes use of biological technology. Only very rarely is it the case that advantage is conferred on an organism via a biochemically unique enzyme or pathway.

The toolbox of biochemistry, the parts list—"the kernel," to stretch the software analogy—is shared by all organisms on the planet. In general, organisms differ from one another because of their order of gene expression or because of relatively subtle perturbations to protein structures common to all forms of terrestrial life. That is, innovation in the natural world in some sense has always followed the idea of a service and flow economy. If the environment is static, only when an organism figures out how to use the old toolbox to provide itself, or another organism, with a new service is advantage conferred.

The analogy to future industrial applications of biology is clear: When molecular biologists figure out the kernel of biology, innovation by humans will consist of tweaking the parts to provide new services. Because of the sheer amount of information, it is unlikely that a single corporate entity could maintain a monopoly on the kernel. Eventually, as design tasks increase in number and sophistication, corporations will have to share techniques and this information will inevitably spread widely, reaching all levels of technical ability—the currency of the day will be innovation and design. As with every other technology developed by humans, biological technology will be broadly disseminated.

Bypassing conventional infrastructure

As open-source biological manufacturing spreads, it will be adopted quickly in less developed economies to bypass the first world’s investment in industrial infrastructure. Given the stressed state of natural resources throughout much of the developing world, it will not be possible for many of those countries to attain first-world standards of living with industrial infrastructure as wasteful as that of the United States. The developing world simply cannot afford industrial and energy inefficiency.

A short cut is to follow the example of the growing wireless-only communications infrastructure in Africa and to skip building systems to transport power and goods. It is already clear that distributed power generation will soon become more efficient than are centralized systems. Distributed manufacturing based upon local resources will save transportation costs, simplify customization, require less infrastructure investment, and, as a result, will likely cost less than centralized manufacturing.

Distributed biological manufacturing is the future of the global economy. With design and fabrication power spread throughout the world to the extent suggested here, it is necessary to consider possible dangers. The simple answer is that those dangers are real and considerable.

This technology enables the creation of new organisms potentially pathogenic to humans, or to animals and plants upon which we rely. It is already clear that the social and biological consequences of extending human life span and human germline engineering will consume considerable public debate time over the next few decades. Moreover, the underlying infrastructure and methods are already so widespread that no one country will be able to manipulate the development of biological technology by controlling the research within its borders.

But fear of potential hazards should be met with increased research and education, rather than closing the door on the profound positive impacts that distributed biological technology will have on human health, human impacts on the environment, and increasing standards of living around the world.

Technology based on intentional, open-source biology is on its way, whether we like it or not, and the opportunity it represents will just begin to emerge in the next 50 years.

This essay won a Silver Award in The Economist/Shell World in 2050 essay competition held last year.

Technology has always been a double-edged sword, bringing us longer and healthier life spans, freedom from physical and mental drudgery, and many new creative possibilities on the one hand, while introducing new and salient dangers on the other. Technology empowers both our creative and destructive natures. Genetic engineering is in the early stages of enormous strides in reversing disease and aging processes.

Ubiquitous nanotechnology, now about two decades away, will continue an exponential expansion of these benefits. These technologies will create extraordinary wealth, thereby overcoming poverty, and enabling us to provide for all of our material needs by transforming inexpensive raw materials and information into virtually any type of product. Lingering problems from our waning industrial age will be overcome. We will be able to reverse remaining environmental destruction.

Nanoengineered fuel cells and solar cells will provide clean energy. Nanobots in our physical bodies will destroy pathogens, remove debris such as misformed proteins and protofibrils, repair DNA, and reverse aging. We will be able to redesign all of the systems in our bodies and brains to be far more capable and durable. And that’s only the beginning.

There are also salient dangers. The means and knowledge exists in a routine college bioengineering lab to create unfriendly pathogens more dangerous than nuclear weapons. Unrestrained nanobot replication ("unrestrained" being the operative word here) would endanger all physical entities, biological or otherwise. As for "unfriendly" AI, that’s the most daunting challenge of all because intelligence is inherently the most powerful force in the Universe.

Awareness of these dangers has resulted in calls for broad relinquishment. Bill McKibben, the environmentalist who was one of the first to warn against global warming, takes the position that we have sufficient technology and that further progress should end. In his latest book titled "Enough: Staying Human in an Engineered Age," he metaphorically compares technology to beer and writes that "one beer is good, two beers may be better; eight beers, you’re almost certainly going to regret." McKibben’s metaphor comparing continued engineering to gluttony misses the point, and ignores the extensive suffering that remains in the human world, which we will be in a position to alleviate through sustained technological progress.

Another level of relinquishment, one recommended in Bill Joy’s Wired magazine cover story, would be to forego certain fields–nanotechnology, for example–that might be regarded as too dangerous. But such sweeping strokes of relinquishment are equally untenable. Nanotechnology is simply the inevitable end result of the persistent trend towards miniaturization that pervades all of technology. It is far from a single centralized effort, but is being pursued by a myriad of projects with many diverse goals.

Abandonment of broad areas of technology will only push them underground, where development would continue unimpeded by ethics and regulation. In such a situation, it would be the less-stable, less-responsible practitioners (e.g., terrorists) who would have all the expertise.

The siren calls for broad relinquishment are effective because they paint a picture of future dangers as if they were released on today’s unprepared world. The reality is that the sophistication and power of our defensive technologies and knowledge will grow along with the dangers. When we have "gray goo" (unrestrained nanobot replication), we will also have "blue goo" ("police" nanobots that combat the "bad" nanobots). The story of the 21st century has not yet been written, so we cannot say with assurance that we will successfully avoid all misuse. But the surest way to prevent the development of the defensive technologies would be to relinquish the pursuit of knowledge in broad areas. This was the primary moral of the novel Brave New World.

Consider software viruses. We have been able to largely control harmful software virus replication because the requisite knowledge is widely available to responsible practitioners. Attempts to restrict this knowledge would have created a far less stable situation. Responses to new challenges would have been far slower, and it is likely that the balance would have shifted towards the more destructive applications (that is, the software pathogens). Stopping the "GNR" technologies is not feasible, at least not without adopting a totalitarian system, and pursuit of such broad forms of relinquishment will only distract us from the vital task in front of us. In terms of public policy, the task at hand is to rapidly develop the defensive steps needed, which include ethical standards, legal standards, and defensive technologies. It is quite clearly a race. There is simply no alternative. We cannot relinquish our way out of this challenge.

There have been useful proposals for protective strategies, such as Ralph Merkle’s "broadcast" architecture, in which replicating entities need to obtain replication codes from a secure server. We need to realize, of course, that each level of protection will only work to a certain level of sophistication.

The "meta" lesson here is that we will need to place society’s highest priority during the 21st century on continuing to advance the defensive technologies and to keep them one or more steps ahead of destructive misuse. In this way, we can realize the profound promise of these accelerating technologies, while managing the peril.

The revolutionary promise of molecular nanotechnology (MNT) has become a part of society’s expectations for the future. This technology will provide nanomedicine breakthroughs that could cure cancer and extend life spans, bring abundance without environmental harm, and provide clean sources of energy. These ideas are part of the vision that launched the field of nanotechnology.

Although the U.S. National Nanotechnology Initiative (NNI) currently supports a host of valuable projects, it is excluding work explicitly directed toward MNT. In an effort to distance the field from fears that might threaten funding, the leading NNI spokesman, Dr. Richard Smalley, has declared that molecular assemblers are impossible. This stance has opened a vast gap, creating a world in which students interested in pursuing MNT research lack sponsorship, while lab groups and start-up companies working toward MNT goals must hide their intentions. By declaring molecular assembly technology to be impossible, detractors have tried to relegate it to fringe status.

Fortunately, this erroneous situation is beginning to change, in part because the extended Foresight community refuses to let this important issue be dismissed. We now have a unique opportunity to seize the momentum. Richard Smalley has responded to my challenge, and the ensuing exchange—the Dec. 1 cover story of the American Chemical Society’s magazine, Chemical & Engineering News—may mark a tipping point, but only if it is seen—and properly understood—by a wider audience, and if it is properly translated into action.

WHAT YOU CAN DO!

I urge you to read the Foresight press release. Read the full exchange and then consider what part you can play in adding to the momentum. The detractors of MNT have shown the power of disinformation; it’s time they saw what well-informed people can do.

Some suggestions:

1.     SPEAK UP: make others aware of what’s going on. Forward the press release and the exchange. Write a letter to the editor of your favorite publication, attaching these materials and requesting coverage of this important issue. Raise issues and answer naysayers though message boards and blogs. Show the opposition our numbers and knowledge.

2.     ELEVATE THE DEBATE: shift the discussion on molecular assemblers and molecular manufacturing from rhetoric and metaphors to science and research. Demand proof from those dismissing the accomplishments to date. Give someone influential a copy of Nanosystems (chapters 1 and 2 are on the web). Refer them to the work of Ralph Merkle, Robert Freitas and others.

3.     GET MORE ACTIVE: request seminars and classes on related topics. Transform your next social event or book group to focus on these issues. Become more engaged with Foresight :

• Help match the challenge grant
• Tell us about yourself and your skills
• Consider how you can help with plans for Foresight’s next phase     

4. ABOVE ALL, TAKE ACTION

Regardless of what avenues you may choose, make your voice and intentions heard. Our future is counting on you.

Christine Peterson,
President
Foresight Institute

Foresight Institute, PO Box 61058, Palo Alto, CA 94306 USA
tel +1 650 917 1122 fax +1 650 917 1123

Foresight is the leading public interest organization in nanotechnology. A non-profit educational think tank, Foresight Institute was founded in 1986.

1. Introduction

The utility of a new technology depends on many factors, including the difficulty of development and the ease and cost of use.  Most technologies require significant additional work to form useful products.  Previous theoretical work in molecular nanotechnology has provided only incomplete and fragmentary answers to the question of how molecular nanotechnologic devices can be used in practice.  Although it appears that fabrication systems can be built on a nanometer scale (Drexler, 1992), small devices will be difficult to use directly in many applications.  Several designs have been proposed in more or less detail (Drexler, 1986, 1992; Bishop, 1996; Merkle, 1997a; Hall, 1999; Freitas and Merkle, in press) for parallel control of many small fabricators to make a large product.  Other proposals (Hall, 1993) combine many small products to create a large product.  However, each of these proposals has provided insufficient detail to allow estimation of their practical difficulty and utility.

This paper builds on previous proposals to describe an architecture for combining large numbers of programmable mechanochemical fabricators into a manufacturing system, or nanofactory, capable of producing a wide range of human-scale products.  The proposed system is described in sufficient detail to allow estimation of nanofactory mass, volume, power requirements, reliability, fabrication time, and product capability and cost, as simple functions of the properties of the mechanochemical fabricator component.  Bootstrapping a human-scale system from a sub-micron system is also discussed.  Discussion of product design issues and nanofactory manufacturing capability demonstrates that the nanofactory should be able to efficiently fabricate duplicates of itself as well as larger versions.  This proposal differs from previous proposals in that, with the exception of mechanochemical component fabrication, design of the nanofactory should be within the reach of present-day engineering; physical structures and functional requirements are described in sufficient detail that remaining problems should be within the capability of current engineering practice to solve.  In particular, the design considers all transport and manipulation requirements for raw materials and product components, as well as control, power, and cooling issues.

This exploration can provide a basis for estimating the practical value and difficulty of developing a nanofactory.  As noted in (Merkle, 1999), even a primitive sub-micron mechanochemical fabricator may produce valuable products.  The question at hand is whether, once such a device is developed, it is feasible and worthwhile to adapt such devices into a nanofactory.  Since no complete designs, or even complete parameter sets, exist for a mechanochemical fabricator, this question cannot be answered fully at this time.  However, the results of the present paper can be applied to a wide range of hypothetical fabricator parameters.  As fabricator designs are proposed in increasing detail, these results will become increasingly useful in predicting the capabilities of a nanofactory based on such designs.  Issues of product design and manufacture are examined in order to establish that the nanofactory is capable of fabricating duplicates and larger versions of itself.  The time required to bootstrap a human-scale factory from a nano-scale fabricator cannot be estimated with any certainty, since bootstrapping will require time for debugging and redesign as well as for fabrication of larger versions.  However, the minimum time required for fabrication can be estimated, and the design developed here is simple enough that debugging and redesign may be fairly simple and rapid.

The paper is arranged in several sections.  Section 2 surveys previous work toward manufacturing systems relying on mechanochemistry and producing human-scale products.  Section 3 describes two innovations required for efficient operation of the nanofactory architecture.  Section 4 describes the nanofactory architecture, including a highly reliable production module incorporating several thousand mechanochemical fabricators and a scalable convergent assembly and transport architecture for integrating large numbers of production modules.  Section 5 covers issues of product design to establish that the nanofactory is designable and buildable by itself.  Section 6 discusses computer control of the nanofactory.  Section 7 covers product performance.  Section 8 provides calculations relating nanofactory performance and characteristics to fabricator performance and characteristics.  Section 9 summarizes the paper.  Appendix A is a computer program that implements repetitive calculations for probability, size of components, and pressure in cooling channels.  Appendix B is a brief discussion of the suitability of a proposed fabricator design (Merkle, 1999) for the nanofactory architecture.

2. Background

In order to create useful products with molecular manufacturing, several steps are required.  Large products cannot be built by a single small fabricator.  Even at a million atoms per second, building a gram of product would take more than a billion years.  Building a large product requires a system implementing several steps.  First, molecules must be reacted under positional control by fabricators to form parts.  Second, the parts must be combined into nanosystems.  Third, the nanosystems must be combined into products, either by physical attachment or by distributed control.  Many authors have considered one or more of these steps, but none has described a complete factory system.  

2.1. Mechanochemistry

As used in this paper, mechanochemistry refers to the process of inducing covalent bond formation or breaking under controlled conditions by mechanical motion.  As discussed in (Drexler, 1992, chap. 8 & 9), mechanochemistry performed in a well-controlled environment appears sufficient to fabricate small devices from covalently bonded carbon (diamondoid).  Merkle (1997d, 1998) describes additional reactions that could be used to build diamondoid products—a complete hydrocarbon "metabolism" capable of refreshing the molecular deposition tools, and Merkle and Freitas (2003) have analyzed a specific diamond mechanosynthesis tool in detail.  The present design assumes that some such chemistry is possible in practice, and will have been characterized to some extent in the process of building a working fabricator.  Diamondoid fabrication chemistry need not be completely understood—a basic set of a few reliable deposition reactions, with motions parameterized to account for edges and other discontinuities, should be sufficient to build bulk diamond.

In order to focus on nanofactory architecture, the present work does not consider mechanochemical operations in detail.  Instead, the design assumes the existence of a small programmable mechanochemical fabricator.  To simplify architectural considerations, the fabricator is assumed to be self-contained: it must be capable within a small volume of performing all mechanical motions necessary to fabricate parts from feedstock and assemble them into small devices of complexity comparable to itself.  

2.2. Mechanochemical fabricator designs

Several proposed devices appear to be capable of performing reliable mechanochemical operations with sufficient flexibility for self-duplication.  These include the robot arm described by Drexler (1992, sec. 13.4), the double tripod described by Merkle (1997c), the molecular mill described by Drexler (1992, sec. 13.3), and the "parts synthesizer" described by Hall (1999).  Additionally, biological or hybrid systems have been proposed (Bradbury, 2003) in which organic synthesis is used to build relatively large chemical components.  Each of these systems is attractive for various reasons.

The robot arm requires several different mechanical components, including small gears, triply-threaded toroidal worm drives, and several types of cylindrical sliding interfaces.  Each of these components may require significant atom-level design.  In addition, the robot arm requires a control system involving rotational motion on several drive rods.  Nanometer-scale clutches have not been designed in detail.  In order to provide results relevant to early fabricator and nanofactory development, this paper does not assume that devices of such complexity can be built.   Hall’s parts synthesizer requires separate assembly robots to deliver chemicals, and the power/control mechanism is not specified.

Systems relying on many biologically-based feedstock molecules require separate synthesis and assembly areas, which may have quite different environmental requirements.  In addition, they may require nontrivial transport mechanisms to prevent premature reaction of the feedstock molecules.  Finally, such systems do not appear to permit the fabrication of diamondoid structures.

The molecular mill is an attractive concept for several reasons.  It does not require explicit control of each mechanochemical operation, thus greatly increasing efficiency over the other three proposals.  Operations can also be quite fast, since merely moving a belt a short distance is sufficient to accomplish a mechanochemical operation.  However, each reactive encounter mechanism, or station, in a molecular mill performs only one mechanochemical operation.  Although each station is efficient, in the sense of processing a mass equal to its own in a short time and with little energy wasted, a large number of stations would be required to fabricate all the parts needed to build a nanofactory, let alone the desired range of products.  This number has not been quantified.  Additional design would be required to explain how a number of stations performing different mechanochemical operations and using different parts can produce all the required parts for self-replication given that only one chemical operation is performed at each station.  Although this does not contradict the possibility of a self-replicating set of mills, it indicates that the set may be large and difficult to design.  In addition, the set may grow unpredictably when required to produce additional parts for the non-fabricator portions of the nanofactory, and may need significant modification if the design of a part must be changed.  Accordingly, a mill solution is not used in this paper.  However, it should be noted that a combination of a mill making small blocks and a double tripod capable of both joining blocks seamlessly (Drexler, 1992, sec. 9.7.3) and performing detailed mechanochemistry may be fast, flexible, and relatively easy to design, and would be preferable to the simple robotic manipulator implicitly assumed for this baseline design.

The current design effort is based loosely on a double-tripod "assembler" discussed by Merkle (1999).  Merkle’s assembler is self-contained, simple to control, and approximately the right size for a basic factory or product building block.  Every effort has been made to avoid depending on any specific property of Merkle’s design.  However, the claimed feasibility of this design serves as inspiration for the present effort to integrate designs with comparable functionality into a monolithic nanofactory.  

2.3. Parts assembly, scaling, and product integration

Several nanotech manufacturing designs have been proposed that could be used to build large products.  For example, Bishop (1996) describes an "Overtool" composed of multiple "active cells" and "gantry cells" which can both do mechanochemistry and encompass and manipulate a large product.  This design is incomplete, lacking description of control algorithms and internal communications.  Hall (1999) describes a system of robots and framework components that can in theory scale to large size and then make large products.  However, feedstock delivery and system control are not specified, and products it can fabricate are not described.  Drexler (1992, chap. 14) describes a system in a fair amount of detail, including estimates of volume, mass, and replication time.  However, this description does not include the assembly operations used, robotics required, or control of those robotics.  Additionally, the system uses molecular mills, which have not been studied in detail, and the physical layout of the system is specified only in general terms.  This work, though seminal and inspiring, does not permit detailed estimation of the technological sophistication required to design such a system.  Merkle (1997a) described a variant of that system, including fabrication time and the suggestion of assembling products from large sub-blocks.  However, he did not calculate the power requirements, describe the internal control mechanisms, or discuss product design issues or the feasibility of self-replication in any detail.

Large-scale cooperative designs are not well understood today, and directing them may be expected to be difficult.  Hall (1993) describes a "utility fog" composed of many small identical robots.  Such a system would be relatively simple to manufacture, requiring no large-scale assembly.  Hall suggests that the fog could use any of several fairly simple algorithms to simulate solid objects.  However, he also does not consider how to power the product/object, and the control algorithms are not worked out in detail.  Also, his fog is quite weak for its mass, at least compared to a more strongly fastened diamondoid product.

The purpose of the nanofactory is to build strong, functionally rich, monolithic, human-scale products that are easy to design and use.  Several innovations described in this paper allow a nanofactory design to be presented in detail for the first time.  The nanofactory itself is intended to be in the set of possible products.  The paper focuses on early development and on demonstrably feasible designs, so does not include some obvious but currently speculative techniques for improving performance.  The design is deliberately simple, especially in minimizing the amount of mechanochemical design needed in addition to the preexisting fabricator.  The major design effort focuses on mechanical and digital design, physical layout, and fault tolerance.

Mechanical design methodology has achieved great competence in the transformation of mechanical motion and force.  Many devices have been developed to accomplish this, such as cams and followers, rack and pinion drives, planetary and differential gears, and pantographs.  Drexler (1992, chap. 10) demonstrated that many of these devices can be translated directly to nanometer scale.  However, many of Drexler’s designs use a specialized arrangement of surface atoms, and sometimes of internal atoms.  Such devices would require individual chemical design.  To avoid the unknown but potentially large effort involved in developing new chemical synthesis for new mechanical structures, the present design does not generally assume the use of mechanical features smaller than ~1 nm.  Such a design is referred to as "bulk diamond", meaning that simply specifying a suitable volumetric design to be filled with diamond lattice is sufficient to specify a part with the required mechanical function.

Although a wide range of sensing and feedback technologies have been developed at the macro scale, some of them do not work at the nanometer scale (e.g. optics and electromagnets; see Drexler, 1992, sec. 2.4), and others may require excessive volume or complexity.  In general, this design effort avoids sensing in favor of predictability and reliability.  Digital logic is useful for performing repetitive functions, doing precise calculations, and selecting among alternatives using well-specified criteria.  Software systems that interact with mechanical systems may be hampered by sensor data that are not well specified.  Accordingly, this design does not make much use of feedback, or require software to deal with "fuzzy" situations.  Almost all aspects of nanofactory operation are deterministic; this mirrors the (theoretically) deterministic nature of the mechanochemical technique (Drexler, 1992, sec. 6.3).  Because the factory layout is extremely repetitive and strictly hierarchical, issues in controlling a large number of fabricators and robots can be reduced to controlling a single fabricator or robot, plus simple iteration.

As previously noted, this paper builds on work which demonstrates that a nanofactory is conceptually feasible.  The design presented here is sufficiently detailed that the feasibility of each part of it can be assessed.  However, it is sufficiently general that it can accommodate a variety of mechanochemical systems.  Where design principles are well understood, details are not supplied.  For example, the use of a gantry crane is specified in order to demonstrate the existence of robotics capable of doing the job required, and to allow approximate calculation of the mass of those components.  The drive mechanism of the gantry crane is not specified; however, given a design tolerance of 1 nm and the presumed feasibility of motors  as small as 50 nm in diameter (see Section 8.2), it is clear that such a mechanism can be designed in a wide range of sizes; at the present level of design, it is not necessary to examine work on industrial robotics.

The extreme conservatism that is appropriate for a feasibility demonstration is less appropriate for a preliminary engineering study.  For example, it is conservative to assume that each individual part will require individual design at the atomic scale.  However, it is reasonable to assume that in the case of a rod with bumps regularly spaced on it, the rod can be extended and bumps can be added or removed without requiring detailed redesign.  Thus the use of mechanical digital logic designs (Drexler, 1992, chap. 12) is assumed to require a mechanochemical design effort for only a fixed and relatively small number of parts.  Likewise, the quantitative sections of this paper choose typical or reasonable values instead of extreme or pessimistic values.  

2.4. Nanofactory overview

The nanofactory system described here incorporates a large number of fabricators under computer control.  In a single product cycle, each fabricator produces one nanoblock, approximately the same size as the fabricator.  The blocks are then joined together, eight sub-blocks making one block twice as big.  This process is repeated until eight large blocks are produced, and finally joined in an arrangement that is not necessarily cubical.  The output of multiple product cycles may be combined to produce large products.  The production system is arranged in a three-dimensional hierarchical branching structure (see Section 4.3) which allows the sub-block assembly to be done by machinery of appropriate size.  Eight factories of a given size can be combined to form one larger factory; the 64 blocks produced are joined into eight blocks twice as big.  The design is easily scalable to tabletop size, with a ~1 meter factory producing eight ~5 cm blocks per product cycle.  As discussed in Section 8.4, depending on the capabilities of the mechanochemical fabricator, the time required for a product cycle will be conveniently measured in hours.  The blocks need not be solid cubes, and their interior may be quite complex.  As discussed in Section 5.1.5, products can be unfolded after manufacture, greatly increasing the range of possible product structures and allowing products to be much larger than the nanofactory that produced them.

The exact size of the nanoblock is unimportant.  For this design, a 200-nm cube is convenient: it is large enough to contain a simple 8086-equivalent CPU, a microwatt worth of electrostatic motors/generators, a shaft carrying 0.4 watts (Freitas, 1999, sec. 6.4.3.4), or the Merkle assembler (1999), but small enough to be fabricated quickly and to survive background radiation for a useful period of time.  As discussed in Section 6, the partitioning of the product into nanoblocks, and the use of relatively large sub-blocks at each step, allows the use of relatively simple robotics and control algorithms in the nanofactory.  As discussed in Section 5, such division also simplifies product design without imposing many practical limits on product complexity.  (W. Ware points out that a combination of tetrahedra and truncated tetrahedra is also space-filling, and that this may be more compatible with the tetrahedral diamond matrix.)

At the smallest scale, the organization of the factory changes to allow simpler distribution of feedstock, cooling, power, and control, and simpler error handling.  A production module consists of one computer and a few thousand fabricators.  It produces a few blocks, a few microns in size, by combining a few thousand nanoblocks.  These rectilinear production modules incorporate a few block assembly stages.  They are combined into the smallest factories, which are also rectilinear—and so on to any size desired.  At each stage, product blocks are delivered through the center of the smallest face, allowing compact stacking of multiple modules or stages.  The stages are stacked on either side of a gathering/assembly tube which contains simple robotics to join the incoming product blocks into larger blocks and deliver them out the end of the tube.  Two stacks of stages, plus the tube in between, constitute the next higher level stage.

The nanofactory design is highly repetitive: each input (sub-factory, or substage) to a stage is identical.  Thus only one design is required for each level, regardless of the number of substages at that level.  Since each stage joins eight blocks to form one block with twice the linear dimension, 19 sizes of stage (4 internal to the production module) are required to progress from a 200-nm nanoblock to a 10.5-cm product.  (One additional stage, a simplified gathering stage, is used to transition from production modules to gathering/assembly stages.)  Most of these stages perform identical block-joining operations.  The design of one stage may be used with minor modification for several similar stage sizes.

A nanofactory built with primitive fabricators and control systems may use a lot of power.  It will be cooled by a fluid with suspended encapsulated ice particles (Drexler, 1992, sec. 11.5).  Thus the temperature of the nanofactory will be a uniform 0 C (273 K).  This is significant for the energy used by digital logic (Section 8.2) and for aligning and joining large blocks (Section 3.2.1).

The control architecture of the nanofactory, like the physical arrangement, is strictly hierarchical.  Instructions can be distributed from central computers directly to the computers that directly control the fabricators.  All error detection and correction takes place either within a single nanocomputer or within a production module controlled by a single nanocomputer, and error reporting and compensation are not required beyond the production module.  There is no need for communication between any two computers at the same level; a simple tree architecture can be used to send all required data (Section 8.1).  Exotic or complex control algorithms, networking architectures, and operating systems are not required.

The size, mass, energy requirement, and duplication time of this nanofactory design depend heavily on the properties of the fabricator.  Sections 8.2, 8.3, and 8.4 quantify these relations.  With the assumptions made in those sections, a tabletop nanofactory (1x1x1/2 meters) might weigh 10 kg or less, produce 4 kg of diamondoid (~10.5 cm cube) in 3 hours, and require as little as fifteen hours to produce a duplicate nanofactory.  

3. Components and Innovations

To provide a workable design for a simple first-generation nanofactory made from primitive fabricators, several innovations are described.  The linear ratchet drive proposed by Drexler (1992, sec. 16.3.2) is extremely inefficient.  Section 3.1 describes a thermodynamically efficient stepping drive that is applicable to all stepping actuators.  The problem of how to join small components into a large product has been greatly simplified by designing a mechanical fastening system, described in Section 3.2.1.  It has only two moving parts, requires no insertion force or actuation, preserves much of the strength of the unbroken material, is easy to grip and handle, and is tolerant of alignment errors.  With the addition of one actuator, the joint can be made reversible to aid in product unfolding (Section 5.1.5).  Section 3.2.2 describes a variety of press-fit connections for conveying power, signal, and fluid between nanoblocks.

The fabricator used in the nanofactory is unspecified; the nanofactory design is sufficiently general that a wide variety of possible fabricator designs can be incorporated.  The reader may find it helpful to study Merkle’s "assembler" (1999) (Appendix B) as a prototype.  The only requirements are that the fabricator must be capable of producing a variety of products of size and complexity equal to itself from soluble feedstock molecules, and that it use digital deterministic control, which implies that the mechanochemical processes must be highly reliable.  Since only a few reactions will be sufficient to make a wide variety of molecular shapes, and since error detection and correction will be difficult if not impossible in early broadcast-architecture assemblers, these requirements do not greatly reduce the generality of the present design.  (Unreliable operations can be retried multiple times even in a deterministic system; see for example Drexler, 1992, sec. 13.3.1c).  

3.1. A Thermodynamically Efficient Stepping Drive

Figure 1: Pin Drive

A mechanical device driven by a sequence of simple digital commands will have an internal state that changes with each command, and must be maintained without error or slippage.  Thermal noise injects constant vibration into the system, requiring strong latching mechanisms.  A simple latching mechanism is a ratchet with a strong spring, as proposed by Drexler and used by Merkle.  A stepping drive can be built from two ratchets, and early assembler designs that are controlled by simple external signals may make extensive use of such drives.  Such a mechanism is extremely inefficient, since the energy used to compress the spring (at least 100 kT [Boltzmann's constant times ambient temperature] to overpower room-temperature thermal noise) is lost each time the ratchet moves to the next tooth.  However, when a fabricator is connected to a digital logic system, the fabricator no longer needs internal state-maintenance mechanisms, and the device can be made far more efficient. (Digital logic, including gates and registers, can be made thermodynamically efficient.)

An efficient drive with functional characteristics similar to the ratchet drive is the pin drive.  (See Figure 1.)  In this design, pins are inserted into equally spaced holes (or notches) in the moving bar to assure its position at all times.  The latching pin moves in and out but not sideways, and is used to hold the bar still.  The driving pin moves in and out, and also is moved sideways by a bar actuator over a distance equal to the spacing of the holes.  The pins are similar in structure and function to the rods in Drexler’s rod logic design.  To move the bar one step, the bar actuator is moved to one end of its range, pulling the driving pin into alignment with a hole.  The driving pin is inserted.  The latching pin is withdrawn.  Then the bar actuator is moved to the other end of its range, bringing the next hole into alignment with the latching pin, which is then inserted.  Finally, the driving pin is withdrawn and then moved back to its original position.  Smaller step sizes may be obtained by additional offset pins, or by a second vernier drive, similar to the vernier ratchet drive described by Drexler (1992, sec. 16.3.2), with slightly different hole spacing and bar actuator range of motion from the main drive.  Larger step sizes may be obtained by moving the bar actuator a larger distance in each cycle.

Although the pin drive requires one more actuator than the ratchet drive—two pins and a bar actuator, instead of two ratchet pawl pullers—it has the advantage that it can move by measured steps in either direction, whereas the two-ratchet drive can only move stepwise in one direction and must retract in a single motion by lifting both ratchets.  (If both pins are lifted simultaneously, the bar can be moved without restriction by a weak return actuator, allowing the same rapid return motion as the ratchet drive.  Note that without careful design, verifying the complete return of the bar will require energy on the order of 100 kT.)  Similar redesign can be applied to any stepping drive mechanism.  As long as the position of the moving member is initially known, it can be moved stepwise, held stiffly against thermal noise at every point, and locked in place in its new position, all without irreversible state transitions.

While the pins are moving, the bar is stationary and the pins may be moved reversibly.  As the bar actuator is moved, the force encountered may vary.  Nevertheless, stiffly imposed motion will be efficient in most cases.  As long as the force profile of the motion does not vary more rapidly per distance than the stiffness of the drive mechanism, and does not vary substantially between forward and backward motion (e.g. due to an irreversible state transition), it does not matter how much force is required to move the bar at each point because the energy will be recovered when the motion is reversed.  This energy recovery requirement implies that the drive mechanism must be able to recover energy from being driven by the bar; such designs are not difficult, and include Drexler’s electrostatic motor/generator (1992, sec. 11.7).  The same argument applies to other types of actuators driven by other digital control mechanisms: as long as the force profile is reversible and is less steep than the stiffness of the drive mechanism, the energy that is put into the system is recoverable.  Note that rapid motion causes the force profile to deviate from reversibility due to various energy dissipation mechanisms.  (The author thanks Eric Drexler for clarifying discussion of thermodynamic reversibility.)  

3.2. Joining Product Blocks

There are several possible ways to join two mechanochemically fabricated objects.  Van der Waals force is an attractive force that develops between any two nearby objects.  For a few unterminated surfaces, covalent chemical bond formation can in theory be used to make a seamless joint.  A wide variety of mechanical joints can be used.  Section 3.2.1 describes a particularly useful strong mechanical joint, and Section 3.2.2 describes several press-fit joints for power, control, and fluid connections between blocks.

At very small separations, two objects experience an attractive force called van der Waals force: simply bring them close together, and they stick.  For two flat diamond surfaces, the force is approximately 1 nanonewton per square nanometer, or 10,000 atm of pressure (Freitas, 1999, sec. 9.3.2).  This is reasonably high, although it provides only a fraction of the strength and stiffness of chemical bonds.  The van der Waals force is the simplest method of joining, it is reversible, and it should provide sufficient strength to keep even kg-scale products from falling apart under their own weight.  This type of joint is convenient and can be used for weak joining of structures that must later be separated.

A diamond surface that is not passivated with an outer layer of hydrogen will be very reactive.  Unterminated diamondoid surfaces forced together should form covalent bonds.  According to Drexler, two (110) surfaces of tetrahedral diamond or two (100) surfaces of hexagonal diamond should bond to each other on contact, forming a seamless joint (Drexler, 1992, secs. 8.6, 9.7.3, and 14.2.1).  Sinnott et al. (1997) report the results of simulations that show bond formation, though not seamless joining.  For other diamond surfaces, or in the case of too-rapid joining, a somewhat weaker joint may form with a lower bond density.  Also, it is currently unknown how much pressure would be required to initiate the process.  Crushing buckyballs to diamond requires 20 GPa, but Drexler states (personal communication, January 24, 2003) that a covalent joint should "zipper" itself if started at an edge or corner, that neon atoms should be able to escape the closing gap and would not interfere with the joining, and that argon is even better in this regard.  If the joint were comparable in structure to amorphous diamond currently made for MEMS, it would have a tensile strength of only 8 GPa (Sullivan, 2002); this is significantly less than diamond’s tensile strength of 60 GPa (measured) to over 100 GPa (calculated, depending on crystal orientation) (Telling et al., 2000).  An additional problem is that radiation damage or stray molecules may cause local surface reconstruction or contamination that may hold surfaces apart and prevent a joint from forming.  This type of joint is usually not reversible, though in theory a carefully designed edge might allow predictable crack formation.  Since seamless covalent joints have not yet been demonstrated, the present nanofactory design does not use this method.  

3.2.1.The Expanding Ridge Joint  

Figure 2: Expanding Ridge Joint

Each mating block face is covered with small "ridges" that are roughly triangular in cross section.  See Figure 2.  All exposed surfaces are non-reactive (e.g. hydrogen-passivated diamond).  The ridges on each face interlock with the ridges on the opposing face.  As the joint is pressed together, the ridges split and expand sideways.  The exposed surfaces of the ridges are not smooth, but are shaped to grip the opposing ridge, with scallops deep enough to form overhangs when viewed perpendicular to the block face.  A scallop is chosen instead of a sawtooth or ratchet profile in order to avoid crack formation at sharp concave angles.  Scallops also make assembly motions smoother, and allow the un-powered assembly described below.  The expansion of the ridge opens a space in its center, which is then filled by a shim which sits above the almost-closed gap between the two halves of the ridge.  Once the shim is in place, the volume of the joint cannot easily be compressed, and the surfaces of the ridges cannot easily slide past each other; pulling apart the joint would require compressing a solid mass of diamond by several percent or breaking at least half of the ridges simultaneously.  If the ridges all run in the same direction, the joint may be able to slide freely.  Crossed ridges will produce a joint that is quite stiff against shear.

The triangular shape of the ridges has several advantages.  First, the area of the base of the triangles (almost the entire area of the block surface) is structurally solid.  (By contrast, a square ridge would waste at least half of the structural strength of the blocks being joined, because the block area adjacent to the tops of the ridges would not contribute to the joint.)  Second, at small scales, van der Waals forces make handling of components difficult because the components stick to any manipulator.  With triangular ridges and narrow ridge tops, the contact area of the surface is much lower, reducing the van der Waals force.  Third, a manipulator can easily be aligned with the ridges.  Small blocks can be picked up by simple contact with a V-channeled manipulator that presents sufficient surface area to form a van der Waals bond of the desired strength, and the manipulator will automatically be pulled into alignment.  A more complex mating pattern could fasten on several ridges at once.  If the ridges are placed at varying angles or spacings, a well designed manipulator/ridge interface can guarantee that a misaligned manipulator cannot form a firm grip.  Likewise, a well designed ridge/ridge interface can guarantee that misaligned blocks will not join incorrectly.

There are at least three ways of mounting the ridge so that a small attractive force between mating ridges will be sufficient to cause the ridge to spread.  The first possibility is to join the ridge to the nanoblock with dovetail joints, permitting it to slide sideways with very low friction.  A simple dovetail joint costs somewhat more than half of the possible joint strength; a stairstepped dovetail joint (which is similar to a completed ridge joint) would recover much of the strength at the cost of additional volume and complexity.  The second possibility is to use a mounting that is strong in tension but flexible in shear, such as thin columns of diamond or buckytubes.  The third possibility, for use in linear stacks of many nanoblocks, is to build a solid structure extending from the base of the ridge all the way through the block to the ridge on the opposing face.  Both ridges would move in tandem, and be locked in place when the shims were dropped on each side.  This might require a mechanism for retaining all participating shims until all joints are pressed together.

The simplest version of the expanding ridge joint requires no actuation to form the joint other than moving the faces together.  As the faces are brought together, just before the final closure, each row of scallops brushes past the inverse row on the opposing ridge.  As the interlocking ridges from each surface interpenetrate, the bulges of the scallops brush past each other, close enough to be attracted by van der Waals force.  This pulls the halves of the ridge apart.  The attraction between passing scallops when the faces nearly touch must be stronger than the intra-ridge attraction, to ensure ridge spreading during the last phase of joint insertion, as the final rows of scallops pass each other.  This is ensured by the use of small spacers to control the van der Waals force holding the intra-ridge gap closed.  (The spreading will become increasingly favorable as the faces approach, and the operation will happen slowly enough to allow equilibration, so thermal noise will not cause the joint to fail to close.)  However, the intra-ridge attraction (between halves of the same ridge) must be strong enough in the initial position to prevent premature operation due to thermal noise.  The half-ridges must require a certain energy, say 100 kT, to pull them apart far enough for the shim to be inserted; the displacement which absorbs this energy cannot be greater than the depth of the scallop.  Note that the required energy is not dependent on any spatial parameter; it is related only to temperature.  However, the attractive force is approximately proportional to surface area, so this condition can be satisfied by a sufficiently long ridge joint.  In other words, regardless of the actual inter-scallop force, an intra-ridge gap can be chosen that will allow the ridge halves to be separated; and regardless of the gap, a sufficiently long ridge will be resistant to premature separation.

Due to the complicated geometry of the scallops, exact calculation of the attractive force between mating ridge halves is beyond the scope of this paper.  An inaccurate calculation is given to permit crude estimation of minimum ridge length.  The formula for attraction between cylinders (Drexler, 1992, Fig. 3.10f) will be applied, treating each scallop as a 0.5-nm radius cylinder separated by 0.3 nm.  (This is inaccurate because it ignores the attractive contribution from the material behind the cylinders, and because the formula’s derivation assumes that cylinder radius is much greater than cylinder separation.)  The inter-scallop potential energy is calculated as 61 zJ per linear nanometer of scallop contact, which corresponds to 2 nm^2 of surface between the two halves of the ridge.  To reduce the intra-ridge potential energy to 50 zJ per scallop-nm or 25 zJ per nm^2, the spacing must be at least 0.6 nm (ignoring the attractive contribution from the spacer) according to the formula in (Drexler, 1992, Fig. 3.10d) which slightly overestimates the attractive force since the ridge is not infinitely thick.

When the gap between the half-ridges is fully open, the shim (which includes a hollow to accommodate the spacer) is pulled into the gap and held there reliably by van der Waals force.  The shim will insert when the ridges have moved apart by a distance equal to the depth of the scallop undercut, in this example 1 nm.  With a 1-nm deep scallop and a 0.6 nm initial gap (thus a 1.6-nm wide shim), the difference in potential energy between 0.6 nm and 1.6 nm spacing is 21.5 zJ/nm^2.  To prevent premature insertion, the intra-ridge potential energy of attraction must differ by 100 kT (260 zJ at 0 C) between closed and open positions.  This requires 12 nm^2 of intra-ridge gap.  If the ridge is 8 nm high (with 4 scallops), then it need only be 1.5 nm long.

The joint may be stiffened by compressing the joint volume.  In this case, extra force may be used to insert the shim into the gap.  (This also allows the gap to be somewhat narrower, reducing non-structural volume.)  A simple design for an electrostatic actuator adds only one moving part.  The shim is blanketed between insulated capacitor plates, one of which is flexible.  Charging the capacitor makes the plates pull together, expelling the shim like a watermelon seed.  The electricity to power the actuator can be delivered through contact with small embedded conductors at the proper time during the convergent assembly process.  The tip of the shim can be tapered to help spread the ridge halves.  Once the shim is expelled, the capacitor plates will adhere to each other by van der Waals force, forming a reliable barrier to hold the shim in the joint even if the capacitor is discharged.

Tension on the joint will tend to expand the entire joint volume sideways.  This can be constrained by surrounding each joint (not each ridge) with a diamond collar sufficient to resist the sideways force generated by a single ridge.  The ridge joint is somewhat less stiff in tension or compression than solid diamond would be, but should be almost as strong: failure requires either significant compression of a large volume of diamond, or the simultaneous failure of many covalent bonds.  Effectively, the entire joint volume except for the depth of the scallops and the width of the shim contributes to the tensile strength, and the entire joint volume except for the shim contributes to the compressive strength.  Shear strength and stiffness depend on the orientation and attachment of the ridges, but can be made quite high perpendicular to the ridge line.  Torsional and bending strength and stiffness can also be made quite high.

The width of the shim is unrelated to the size of the ridge, being equal to the depth of the scallop’s undercut plus the intra-ridge van der Waals gap.  A reasonable lower bound for component size is a ridge composed of four scallops 1 nm deep and offset by ½ nm horizontally and 2 nm vertically.  The height of the ridge is 8 nm, and the footprint of a half-ridge is 3.5 nm (accommodating 0.5 nm of motion to mate with the opposing half-ridge), of which 2 nm contributes structural strength.  The 1.6-nm wide shim adds an additional 0.8 nm of non-structural overhead to each half-ridge; the total joint tensile strength is approximately 47% of solid diamond.  (Shallower scallops will improve this number up to a point; scallops that are too shallow can fail by slipping past each other.)  For reliable operation the ridge must be at least 1.5 nm long.  The smallest joint consists of one half-ridge on each side, only one of which (and its shim) needs to move; the rest of the joint including the mating half-ridge can be solid diamond.  A single joint can potentially have a footprint smaller than 3×6 nm.  Larger ridges can have more scallops, with the size of each scallop (and thus of the shim) staying constant.  For example, a half-ridge 20 nm high with 10 scallops has a footprint of 6.5 nm (plus 0.8 nm for its share of the shim) of which 5 nm is structural, for 68% of diamond strength.  Covering a 200-nm block with 8-nm-high ridges on each side requires 8% of the block volume (ignoring block edges and corners).  However, in a high-strength application that requires ridge joint coverage of the full surface, the block must be nearly solid diamond anyway.

Because the strength of the joint decreases only slightly with smaller size (the decrease is a function of the minimum shim, scallop, and van der Waals gap size), small ridges are mechanically adequate for joining blocks at any scale.  Minimum ridge size is determined by the mechanochemical fabrication process.  The only limitations on block size are the precision of the block-handling machinery and the possibility of unequal expansion of the faces due to temperature differences.  With 200 nm nanoblocks, ridges built in a single block can be up to 100 nm in height, with tops 50 nm apart.  (Note that the blocks will overlap by the height of the ridge.  The change in effective block width during assembly presents issues for the assembly process that are straightforward but beyond the scope of this paper.)  An assembly tolerance of 0.05 micron is somewhat beyond today’s standards; current state of the art for automated pick and place assembly for optical components appears to be around 0.5 micron (Blaze Network Products, 2003).  However, today’s pick and place systems use hardware made with a manufacturing tolerance comparable to its performance.  In contrast, the dimensional precision of the nanofactory’s hardware will be approximately one atomic diameter or less, regardless of scale.  At large scales, single ridges can be assembled from multiple nanoblocks, allowing ridge spacing of multiple microns; this is sufficient even for today’s robotics.

Differences in fabrication processes, assembly processes, and internal structure may cause different blocks to be at different temperatures.  The resulting thermal expansion can cause a misalignment of the ridges.  The volumetric thermal expansion coefficient of diamond is 3.5×10^-6/K (Freitas, 1999, Appendix A); the linear coefficient is one-third that, or 1.2×10^-6/K.  A temperature difference of 1 K thus causes a 200 nm block to expand by a small fraction of an angstrom, while a 10.5-cm surface will expand by 126 nm.  Because diamond is an excellent conductor of heat, passive equilibration may be sufficient.  As long as the displacement is not greater than the ridge spacing, or the ridge pattern does not permit improper joining, the blocks may be pressed together slowly, allowing the temperature to equalize.  Even a rarefied internal atmosphere will also facilitate temperature equalization between nearby faces, though this process may be slow depending on block mass, and the process will be somewhat slower with argon than with neon.  Note that the nanofactory is cooled by phase transition (see Section 8.2), so the cooling fluid will have the same temperature throughout the factory, minimizing potential product temperature differences.  Active compensation might involve sensing the temperature at various points on the surfaces and applying heat to the cooler surface via embedded resistive heaters; this will only be necessary for the few large-scale joints that take place near the end of the assembly process, and the heating process can be initiated in advance to avoid delay.  Embedded mechanical (bi-material) thermostats can allow each region to reach a preset temperature without individual attention.

Because the joints require no external manipulation or assembly force, they can be used to fasten non-bonded parts that are only loosely connected to the main nanoblock.  For example, a structural beam one micron long and 50 nm wide can be constructed in five sections.  Each section will be terminated in ridge joints, and laid across a nanoblock in a position that will place the section ends next to each other during block assembly; van der Waals force will hold the section in place during block manipulation.  When the nanoblocks are assembled, the ridge joints of the beam will join at the same time as the rest of the joints, with no additional effort.  This allows the inclusion of long, thin components in product designs.  Likewise, single nanoblocks can be made in separate pieces joined by van der Waals force.  This allows a block to be pulled apart during the unfolding process, forming multiple walls with large spaces between them.  This can be useful to save mass where only thin walls are needed.  If a block is split into as many as 10 walls or 100 columns, the 20-nm width is sufficient for multiple full-sized ridge joints on each part.  This capability is assumed for interior nanofactory structure.

Joints can be formed after the product is released from the factory, as long as contaminants have been excluded from the joint space.  The factory can manufacture a larger containing balloon for product unfolding, or the joints can be protected individually by a variety of covering mechanisms.  A product can be created in a very compact form, then unfold like a pop-up book or like flat-packed cardboard boxes.  Components can be built in pieces, with lightweight pantographic trusswork to bring the ends together as the product expands; once the ends touch, the strong joint will form. A component can also be made in a "broken" state, with mating surfaces held together on one edge with a small hinge at any desired angle, and the open end protected by a bellows if necessary.  When the component is straightened, the mating surfaces will form the desired strong connection.  A weak and reversible joint can be formed by preventing the shims from entering the gaps between the ridges.  This allows blocks to be loosely connected, then disconnected, and finally reconnected tightly in the same or different configuration.  This may be useful if the unfolding process requires a structure to be produced in its final conformation, then flexed, and finally fastened rigidly.  

3.2.2.Functional joints

A product may contain embedded wires, pipes, rotating rods, nanocomputer logic rods, and polyyne control cables.  All of these may need to make a connection between adjacent nanoblocks.  These connections are generally simple, and cost less than 50% of the performance that would be possible with a seamless design.

Embedded wires can be run up to a flat face, and electrical contact made by tunneling.  Contact can be maintained in the case of joint strain by the use of springy interfaces.  According to measured values for a sample of HOPG (highly ordered pyrolitic graphite), (GE Advanced Ceramics, 2002) graphite is about 5000 times more conductive in-plane than cross-plane (and the in-plane value is 1/50 as good as a typical metal).  The separation of graphite planes is 0.335 nm, about 1/600 of the 200-nm nanoblock width.  This implies that a graphite-graphite tunneling surface of 8 nm^2 per nm^2 of graphite wire, spaced every 200 nm, would only double the total resistance.  To save nanoblock surface area, the tunneling surface can consist of interlocking corrugations.  Because diamond is an excellent insulator, high voltages may be used to compensate for the resistance of graphite.  Some buckytubes may be better conductors.

Control cables and control rods will be built into each nanoblock when it is manufactured, and extend only to its edges.  Tension and/or compression must be transferred between blocks.  Nanocomputer logic rods have ends ~1 nm^2 which can be butted together.  The nanocomputer design uses tensional force of 2 nN (Drexler, 1992, sec. 12.3.3.b) but this can be traded for displacement or compressive force without sacrificing reliability.  Alternatively, the joint area can be increased by a few nm^2 to allow a few nN of tensile force to be transmitted through van der Waals attraction.  Crossing between blocks may require adding extra logic gates to transform and condition the signal; this logic can all be reversible at some cost of time.  Such interfaces will not add significantly to the power requirements or design complexity of a nanocomputer.

Polyyne (carbon chains with alternating single and triple bonds) control cables can be terminated with a small diamondoid plate flush with the nanoblock surface.  When the blocks are joined, the plates will stick by van der Waals force.  Each two atoms of polyyne spans a length of 0.2569 nm and has a compliance of 0.00185 m/N (Casing an Assembler, "Control cables").  A 1-nm diamond cube contains 176 carbon atoms.  A van der Waals interface has a stiffness of >30 N/m per nm^2 (Drexler, 1992, sec. 9.7.1), or a compliance of <0.0334 m/N.  Two hundred nm of polyyne contains 1557 carbon atoms and has a compliance of 1.44 m/N, while 198 nm of polyyne interrupted by two 1-nm diamond cubes interfaced by van der Waals force contains 1893 carbon atoms and has a compliance of 1.47 m/N; the interface increases cable mass by 22% and compliance by 2% (ignoring the hydrogen termination and internal compliance of the diamond cubes) and may introduce resonances into extremely high-speed operations.  The main drawback of the interface is its strength; the tensile strength of a polyyne rod is >6 nN, but the strength of the interface is ~1 nN.  Increasing the interface area allows a stronger and stiffer joint, and for joint areas above a few square nanometers a ridge joint can be used at some cost of mass.

Power can be transmitted by means of thin rotating rods, embedded in the nanoblocks like the control cables and logic rods.  Mating convolutions on rod ends will allow the transmission of torque between ends that are simply pressed together.  If the rod is driven near maximum torque, the interface may need to be somewhat larger than the cross section of the rod.  The bursting speed of a disc decreases in proportion with its radius, while the area increases as the square of the radius; thus a 2x increase in interface area will cause a 1.4x reduction in speed.  In this simple example, power transmission is derated by 40%; however, other mechanical linkages such as a thin belt connecting offset and overlapping rods may permit full speed while delivering full torque.  (The belt can be placed around one rod during nanoblock manufacture and held open by any of a variety of methods.  The other rod can be tapered to slip inside the belt during block assembly.  Interlocking (gear-toothed) rod surfaces will also work but may require significant overlap for reliable torque transmission.)  Rods and shafts larger than a few nm can be joined by ridge joints.  Ridge joints may also serve as a means of chocking the shafts to ensure proper alignment, and then unlocking them during convergent assembly: the shims can be inserted only when the joint is fully closed, and the motion of their insertion can be used to remove a mechanical chock.  Small rods can be controlled by adjacent ridge joints, and large shafts by facial joints with internal shims.

Bearing surfaces for rotating shafts small enough to be embedded in nanoblocks can be built into each nanoblock during construction.  Variations in rod diameter will prevent the rods slipping out of the block prior to convergent assembly.  Large rods pose a special problem for convergent assembly, since they cannot be strongly and permanently fastened to a support or bearing structure.  However, for products up to 10 cm size, a tight-fitting bearing surface between a rod and a housing can provide the necessary adhesion by van der Waals force alone.  Rotational freedom can be constrained by small retractable chocks.  Graphite pads covering the matching surfaces of the blocks constituting the shaft and the blocks constituting the housing can provide a bearing surface even for slightly rough curved surfaces.  However, the boundaries between the pads will be aligned on the moving and bearing surface, and this can create a significant force.  (Twisting one of the surfaces relative to the other would break the alignment, but this will not be possible for cylindrical bearings.)  Order-of-magnitude calculations can be made by treating the boundary gaps as regions of wider spacing between the surfaces, calculating the difference in van der Waals energy between aligned and unaligned regions, and dividing that by the width of the gap to find a force.  Approximating the boundary as a trench 1 nm wide and 0.1 nm deep and the pad spacing as 0.2 nm, and applying the formula from (Drexler, 1992, Fig. 3.10d), indicates an energy difference of 81 zJ per nm^2 in favor of the aligned state, or an average force of 81 pN per linear nm of trench.  One mm^2 of flat sliding surface will contain 5×10^9 nm of trench crossing the direction of motion, creating a force of ~0.4 N.  However, the stiffness of 1 mm^2 of graphite bearing surface is ~3×10^13 N/m, so for many macroscopic applications, bearings may be made small enough that the "roughness" is not a significant problem.  A cylindrical bearing surface cuts across two nanoblock planes and only a fraction of the area contributes to stiffness; these factors increase the number of trenches (and thus the "roughness" force) for a given bearing stiffness by approximately a factor of 4.

Pipes are simply voids in the diamond nanoblocks that are butted together when the blocks are assembled.  A flat, uncompressed interface between nitrogen-terminated diamond (111) surfaces is adequate to exclude helium (Drexler, 1992, sec. 11.4.2a).  If this type of interface proves inadequate in practice (perhaps due to joint flexure, or unavailability of nitrogen termination chemistry), a conical extension of the pipe wall wrapped in one or more layers of graphite to provide a compressive seal and extending into a conical depression in the other block should suffice.  Pipes too large to be contained inside a nanoblock can be sealed by diamond or graphite curtain walls, placed along each seam, to separate the interior of the pipe from the mechanical joint area.  If the nanofactory is filled with inert gas, pipes will also be filled with the gas when they are manufactured.  If this is a problem, one possible solution is to place a collapsed graphite tube inside the pipe, terminating the tube ends at the nanoblock faces with a diamond mating collar thin enough to be flexible.  When the blocks are assembled, the collars join.  When first used, the graphite tube will expand and conform to the walls of the pipe while displaced gas can be vented through small channels.  

4. Nanofactory Architecture

A nanofactory, as conceived here, is a single device containing many mechanochemical fabricators and larger-scale manipulator systems.  The mechanochemical fabricators produce nanoblocks and the manipulator systems join them into a product.  The mechanochemical working space of a nanofactory must contain no stray reactive molecules.  The factory must contain computers to control the machinery; space and mechanisms for convergent assembly; structures for distributing power, chemicals, and cooling fluid; mechanochemical fabricators with space for them to work; and additional space for joining blocks into larger blocks and transporting them through the factory.

The nanofactory is built hierarchically, using only a few scalable designs.  At the lowest level, a few thousand fabricators are arranged in a planar grid.  Their products are picked up and assembled into increasingly large blocks by a series of increasingly large robotic manipulators.  This plus a control computer constitutes a basic, reliable production module.  The production modules are stacked three-dimensionally into gathering stages, which assemble blocks and pass them to higher-level gathering stages.  Finally, the entire factory is enclosed in a suitable casing, with a mechanism to output product without contaminating the workspace.

In Merkle’s convergent assembly architecture (1999) it is suggested that each convergent assembly stage has four inputs, each supplying two blocks to make one output block.  However, this means that each input to the preceding stage must supply four blocks to make those two, and so on.  This is feasible if blocks can be manufactured extremely quickly, or (as in Merkle’s design) fed through a relatively small number of ports efficiently.  The current design, using large nanoblocks requiring minutes or hours to fabricate, uses only one block from each fabricator per product cycle.  This implies that each stage will receive all its blocks in parallel.  In general, then, each stage must have either eight (non-redundant) or nine or ten (redundant) inputs.  (The first gathering stage has only four inputs, to compensate for the eighteen inputs of the final stage in the production module; see below.)

4.1. Mechanochemical functionality

Figure 3: Workstation Grids

Once a self-contained, digitally controlled mechanochemical fabrication system has been developed, the fabricator design can be copied directly from it.  Early systems will presumably use a simple, stiff robot, such as a double tripod (Merkle, 1997c) or Stewart platform.  As noted in Section 8.2, any inefficient ratchet or other state-keeping systems in the fabricator can be replaced with thermodynamically efficient stepping drives.  Even with this improvement, the primitive method of mechanochemistry will cost some efficiency relative to the "mill" type designs analyzed by Drexler (1992, sec. 13.3) and used in his nanofactory design (1992, sec. 14.4).  Because placing each atom or molecule requires a large and complicated motion of the tripod system, the nanofactory will suffer some penalty in both speed and energy use; these penalties are substantial but not crippling.  Mills are not included in this preliminary design because they may require significant additional mechanical and mechanochemical design.

Fabricators will be fastened together edgewise to form the planar array, which divides the coolant volume from the working volume.  Cooling fluid with dissolved feedstock circulates past one side; the products (nanoblocks) are fabricated and released on the opposite side, which is open to the nanofactory’s clean working volume.  A square of nine fabricators (one redundant) forms a stage.  Product blocks are picked up by a three degree of freedom gantry crane manipulator and assembled into a 0.4-micron block.  Likewise, a square of nine of these stages forms the next stage.  This continues through several levels; in the current design, four levels is chosen for suitable redundancy and convenient control.  

4.2. The reliable basic production module

Figure 4: Production Module

A production module fabricates two 3.2 micron product blocks out of up to 8,192 nanoblocks, using a fabricator to produce each nanoblock.  The module is extremely reliable in the face of radiation damage, and is controlled by an integrated nanocomputer.  The overall shape of the module is a rectangular solid ~16x16x12 microns.  The fabricators are placed on two opposite sides, delivering their product nanoblocks to the interior.  The nanocomputer occupies a third side, surrounding the product exit port.  The remaining three sides may be closed by thin walls, but need not be closed at all where two production modules are placed side by side in the nanofactory.  The interior is sparsely filled with gantry crane manipulators to assemble the nanoblocks into larger blocks.  The gantry crane mechanisms, even at the smallest scale, can be implemented as bulk diamond machines—the smallest blocks are 200 nm on a side, and bulk diamond parts can be designed far smaller than that, so not much material or volume will be wasted due to inefficient design constraints.  With the ridge joints, the blocks can be assembled simply by bringing them into contact (Section 3.2.1).

Dear Prof. Smalley,

I’m glad you found my early work stimulating, and applaud your goal of debunking nonsense in nanotechnology. I hope that our exchange will result in broader discussion within the community, and in better understanding of molecular manufacturing as a strategic objective.

In light of the nature of your questions and of misperceptions frequently articulated in the press, I should first sketch the fundamental concepts of molecular manufacturing. These spring from Richard Feynman’s famous 1959 talk, "There’s Plenty of Room at the Bottom," which envisioned using productive machinery–factories–to build smaller factories, leading ultimately to nanomachines building atomically precise products.

Although inspired by biology (where nanomachines regularly build more nanomachines despite quantum uncertainty and thermal motion), Feynman’s vision of nanotechnology is fundamentally mechanical, not biological. Molecular manufacturing concepts follow this lead.

Hence, to visualize how a nanofactory system works, it helps to consider a conventional factory system. The technical questions you raise reach beyond chemistry to systems engineering. Problems of control, transport, error rates, and component failure have answers involving computers, conveyors, noise margins, and failure-tolerant redundancy. These issues are explored in technical depth in my book "Nanosystems: Molecular Machinery, Manufacturing, and Computation" (Wiley/Interscience, 1992), which describes the physical basis for desktop-scale nanofactories able to build atomically precise macroscopic products, including more nanofactories.

These nanofactories contain no enzymes, no living cells, no swarms of roaming, replicating nanobots. Instead, they use computers for digitally precise control, conveyors for parts transport, and positioning devices of assorted sizes to assemble small parts into larger parts, building macroscopic products. The smallest devices position molecular parts to assemble structures through mechanosynthesis–"machine-phase" chemistry.

Machine- and solution-phase chemistry share fundamental physical principles, yet differ greatly. In machine-phase chemistry, conveyors and positioners (not solvents and thermal motion) bring reactants together. The resulting positional control (not positional differences in reactivity) enables reliable site-specific reactions. Bound groups adjacent to reactive groups can provide tailored environments that reproduce familiar effects of solvation and catalysis. Positional control itself enables a strong catalytic effect: It can align reactants for repeated collisions in optimal geometries at vibrational (greater than terahertz) frequencies.

Further, positional control naturally avoids most side reactions by preventing unwanted encounters between potential reactants. Transition-state theory indicates that, for suitably chosen reactants, positional control will enable synthetic steps at megahertz frequencies with the reliability of digital switching operations in a computer. The supporting analysis for this conclusion appears in "Nanosystems" and has withstood a decade of scientific scrutiny.

It should be clear that chemical reactions (whether machine-phase or conventional) need no impossible fingers to control the motion of individual atoms within reactants. As molecules come together and react, their atoms (being "sticky") stay bonded to neighbors, and thus need no separate fingers to hold them. If particular conditions will yield the wrong product, one must either choose different conditions (different positions, reactants, adjacent groups) or choose another synthetic target. Direct positional control of reactants is both achievable and revolutionary; talk of additional, impossible control has been a distraction.

What can be made using mechanosynthesis? Organic and organometallic reactions in solution-phase and chemical vapor deposition systems can, in the hands of skilled chemists, produce a vast diversity of structures. These include all the products of organic synthesis, as well as metals, semiconductors, diamond, and nanotubes. Augmenting such chemistries with positional control of reactants will enable the fabrication of macroscale products containing chemically diverse structures in complex, precise, functional arrangements. Nanofactories based on mechanosynthesis thus will be powerful enablers for a wide range of other nanotechnologies.

Synthetic reactions and molecular machinery of the sort required for nanofactories have parallels in known systems, and have been explored using computational chemistry by Georgia Institute of Technology professor Ralph Merkle and others. The physical realization of nanofactories, however, will require a multistage systems engineering effort. In 1959, Feynman suggested scaling down macroscopic machines. In 2003, the flourishing of nanotechnologies suggests a bottom-up strategy: using self-assembly (and perhaps scanning probes) to build solution-phase molecular machines, using these to gain limited positional control of synthesis, and then leveraging this ability to build systems enabling greater control. Thus, multiple areas of current research (in computational chemistry, organic synthesis, protein engineering, supramolecular chemistry, and scanning-probe manipulation of atoms and molecules) constitute progress toward molecular manufacturing.

However, because it is a systems engineering goal, molecular manufacturing cannot be achieved by a collection of uncoordinated science projects. Like any major engineering goal, it will require the design and analysis of desired systems, and a coordinated effort to develop parts that work together as an integrated whole.

Why does this goal matter? Elementary physical principles indicate that molecular manufacturing will be enormously productive. Scaling down moving parts by a factor of a million multiplies their frequency of operation–and in a factory, their productivity per unit mass–by the same factor. Building with atomic precision will dramatically extend the range of potential products and decrease environmental impact as well. The resulting abilities will be so powerful that, in a competitive world, failure to develop molecular manufacturing would be equivalent to unilateral disarmament.

U.S. progress in molecular manufacturing has been impeded by the dangerous illusion that it is infeasible. I hope you will agree that the actual physical principles of molecular manufacturing are sound and quite unlike the various notions, many widespread in the press, that you have correctly rejected. I invite you to join me and others in the call to augment today’s nanoscale research with a systems engineering effort aimed at achieving the grand vision articulated by Richard Feynman. In this effort, an independent scientific review of molecular manufacturing concepts will be a necessary and long-overdue first step.

Best wishes,

K. Eric Drexler

Nanotechnology pioneer Eric Drexler and Rice University Professor and Nobelist Richard Smalley have engaged in a crucial debate on the feasibility of molecular assembly, which is the key to the most revolutionary capabilities of nanotechnology. Although Smalley was originally inspired by Drexler’s ground-breaking works and has himself become a champion of contemporary research initiatives in nanotechnology, he has also taken on the role of key critic of Drexler’s primary idea of precisely guided molecular manufacturing. This debate has picked up intensity with  publication of several rounds of this dialogue between these two pioneers. First some background:

Background: The Roots of Nanotechnology

Nanotechnology promises the tools to rebuild the physical world, our bodies and brains included, molecular fragment by molecular fragment, potentially atom by atom. We are shrinking the key feature size of technology, in accordance with what I call the “law of accelerating returns,” at the exponential rate of approximately a factor of 4 per linear dimension per decade. At this rate, the key feature sizes for most electronic and many mechanical technologies will be in the nanotechnology range, generally considered to be under 100 nanometers, by the 2020s (electronics has already dipped below this threshold, albeit not yet in three-dimensional structures and not self-assembling). Meanwhile, there has been rapid progress, particularly in the last several years, in preparing the conceptual framework and design ideas for the coming age of nanotechnology.

Most nanotechnology historians date the conceptual birth of nanotechnology to physicist Richard Feynman’s seminal speech in 1959, “There’s Plenty of Room at the Bottom,” in which he described the profound implications and the inevitability of engineering machines at the level of atoms:

“The principles of physics, as far as I can see, do not speak against the possibility of maneuvering things atom by atom. It would be, in principle, possible. . . .for a physicist to synthesize any chemical substance that the chemist writes down. . .How? Put the atoms down where the chemist says, and so you make the substance. The problems of chemistry and biology can be greatly helped if our ability to see what we are doing, and to do things on an atomic level, is ultimately developed – a development which I think cannot be avoided.”

An even earlier conceptual root for nanotechnology was formulated by the information theorist John Von Neumann in the early 1950s with his model of a self-replicating system based on a universal constructor combined with a universal computer. In this proposal, the computer runs a program that directs the constructor, which in turn constructs a copy of both the computer (including its self-replication program) and the constructor. At this level of description, Von Neumann’s proposal is quite abstract — the computer and constructor could be made in a great variety of ways, as well as from diverse materials, and could even be a theoretical mathematical construction. He took the concept one step further and proposed a “kinematic constructor,” a robot with at least one manipulator (arm) that would build a replica of itself from a “sea of parts” in its midst.

It was left to Eric Drexler to found the modern field of nanotechnology, with a draft of his seminal Ph.D. thesis in the mid 1980s, by essentially combining these two intriguing suggestions. Drexler described a Von Neumann Kinematic Constructor, which for its “sea of parts” used atoms and molecular fragments, as suggested in Feynman’s speech. Drexler’s vision cut across many disciplinary boundaries, and was so far reaching, that no one was daring enough to be his thesis advisor, except for my own mentor, Marvin Minsky. Drexler’s doctoral thesis (premiered in his book, Engines of Creation in 1986 and articulated technically in his 1992 book Nanosystems) laid out the foundation of nanotechnology and provided the road map still being pursued today.

Von Neumann’s Universal Constructor, as applied to atoms and molecular fragments, was now called a “universal assembler.” Drexler’s assembler was universal because it could essentially make almost anything in the world. A caveat is in order here. The products of a universal assembler necessarily have to follow the laws of physics and chemistry, so only atomically stable structures would be viable. Furthermore, any specific assembler would be restricted to building products from its sea of parts, although the feasibility of using individual atoms has been repeatedly demonstrated.

Although Drexler did not provide a detailed design of an assembler, and such a design has still not been fully specified, his thesis did provide extensive existence proofs for each of the principal components of a universal assembler, which include the following subsystems:

• The computer: to provide the intelligence to control the assembly process. As with all of the subsystems, the computer needs to be small and simple. Drexler described an intriguing mechanical computer with molecular “locks” instead of transistor gates. Each lock required only 5 cubic nanometers of space and could switch 20 billion times a second. This proposal remains more competitive than any known electronic technology, although electronic computers built from three-dimensional arrays of carbon nanotubes may be a suitable alternative.
• The instruction architecture: Drexler and his colleague Ralph Merkle have proposed a “SIMD” (Single Instruction Multiple Data”) architecture in which a single data store would record the instructions and transmit them to trillions of molecular-sized assemblers (each with their own simple computer) simultaneously. Thus each assembler would not have to store the entire program for creating the desired product. This “broadcast” architecture also addresses a key safety concern by shutting down the self-replication process if it got out of control by terminating the centralized source of the replication instructions. However, as Drexler points out[1], a nanoscale assembler does not necessarily have to be self-replicating. Given the inherent dangers in self-replication, the ethical standards proposed by the Foresight Institute contain prohibitions against unrestricted self-replication, especially in a natural environment.
• Instruction transmission: transmission of the instructions from the centralized data store to each of the many assemblers would be accomplished electronically if the computer is electronic or through mechanical vibrations if Drexler’s concept of a mechanical computer were used.
• The construction robot: the constructor would be a simple molecular robot with a single arm, similar to Von Neumann’s kinematic constructor, but on a tiny scale. The feasibility of building molecular-based robot arms, gears, rotors, and motors has been demonstrated in the years since Drexler’s thesis, as I discuss below.
• The robot arm tip: Drexler’s follow-up book in 1992, Nanosystems: molecular machinery, manufacturing, and computation, provided a number of feasible chemistries for the tip of the robot arm that would be capable of grasping (using appropriate atomic force fields) a molecular fragment, or even a single atom, and then depositing it in a desired location. We know from the chemical vapor deposition process used to construct artificial diamonds that it is feasible to remove individual carbon atoms, as well as molecular fragments that include carbon, and then place them in another location through precisely controlled chemical reactions at the tip. The process to build artificial diamond is a chaotic process involving trillions of atoms, but the underlying process has been harnessed to design a robot arm tip that can remove hydrogen atoms from a source material and deposit it at desired location in a molecular machine being constructed. In this proposal, the tiny machines are built out of a diamond-like (called “diamondoid”) material. In addition to having great strength, the material can be doped with impurities in a precise fashion to create electronic components such as transistors. Simulations have shown that gears, levers, motors, and other mechanical systems can also be constructed from these carbon arrays. Additional proposals have been made in the years since, including several innovative designs by Ralph Merkle name=”_ednref2″>[2]. In recent years, there has been a great deal of attention on carbon nanotubes, comprised of hexagonal arrays of carbon atoms assembled in three dimensions, which are also capable of providing both mechanical and electronic functions at the molecular level.
• The assembler’s internal environment needs to prevent environmental impurities from interfering with the delicate assembly process. Drexler’s proposal is to maintain a near vacuum and build the assembler walls out of the same diamondoid material that the assembler itself is capable of making.
• The energy required for the assembly process can be provided either through electricity or through chemical energy. Drexler proposed a chemical process with the fuel interlaced with the raw building material. More recent proposals utilize nanoengineered fuel cells incorporating hydrogen and oxygen or glucose and oxygen.

Although many configurations have been proposed, the typical assembler has been described as a tabletop unit that can manufacture any physically possible product for which we have a software description. Products can range from computers, clothes, and works of art to cooked meals. Larger products, such as furniture, cars, or even houses, can be built in a modular fashion, or using larger assemblers. Of particular importance, an assembler can create copies of itself. The incremental cost of creating any physical product, including the assemblers themselves, would be pennies per pound, basically the cost of the raw materials. The real cost, of course, would be the value of the information describing each type of product, that is the software that controls the assembly process. Thus everything of value in the world, including physical objects, would be comprised essentially of information. We are not that far from this situation today, since the “information content” of products is rapidly asymptoting to 100 percent of their value.

In operation, the centralized data store sends out commands simultaneously to all of the assembly robots. There would be trillions of robots in an assembler, each executing the same instruction at the same time. The assembler creates these molecular robots by starting with a small number and then using these robots to create additional ones in an iterative fashion, until the requisite number of robots has been created.

Each local robot has a local data storage that specifies the type of mechanism it is building. This local data storage is used to mask the global instructions being sent from the centralized data store so that certain instructions are blocked and local parameters are filled in. In this way, even though all of the assemblers are receiving the same sequence of instructions, there is a level of customization to the part being built by each molecular robot. Each robot extracts the raw materials it needs, which includes individual carbon atoms and molecular fragments, from the source material. This source material also includes the requisite chemical fuel. All of the requisite design requirements, including routing the instructions and the source material, were described in detail in Drexler’s two classic works.

The Biological Assembler

Nature shows that molecules can serve as machines because living things work by means of such machinery. Enzymes are molecular machines that make, break, and rearrange the bonds holding other molecules together. Muscles are driven by molecular machines that haul fibers past one another. DNA serves as a data-storage system, transmitting digital instructions to molecular machines, the ribosomes, that manufacture protein molecules. And these protein molecules, in turn, make up most of the molecular machinery.

– Eric Drexler

The ultimate existence proof of the feasibility of a molecular assembler is life itself. Indeed, as we deepen out understanding of the information basis of life processes, we are discovering specific ideas to address the design requirements of a generalized molecular assembler. For example, proposals have been made to use a molecular energy source of glucose and ATP similar to that used by biological cells.

Consider how biology solves each of the design challenges of a Drexler assembler. The ribosome represents both the computer and the construction robot. Life does not use centralized data storage, but provides the entire code to every cell. The ability to restrict the local data storage of a nanoengineered robot to only a small part of the assembly code (using the “broadcast” architecture), particularly when doing self-replication, is one critical way nanotechnology can be engineered to be safer than biology.

With the advent of full-scale nanotechnology in the 2020s, we will have the potential to replace biology’s genetic information repository in the cell nucleus with a nanoengineered system that would maintain the genetic code and simulate the actions of RNA, the ribosome, and other elements of the computer in biology’s assembler. There would be significant benefits in doing this. We could eliminate the accumulation of DNA transcription errors, one major source of the aging process. We could introduce DNA changes to essentially reprogram our genes (something we’ll be able to do long before this scenario, using gene-therapy techniques).

With such a nanoengineered system, the recommended broadcast architecture could enable us to turn off unwanted replication, thereby defeating cancer, autoimmune reactions, and other disease processes. Although most of these disease processes will have already been defeated by genetic engineering, reengineering the computer of life using nanotechnology could eliminate any remaining obstacles and create a level of durability and flexibility that goes vastly beyond the inherent capabilities of biology.

Life’s local data storage is, of course, the DNA strands, broken into specific genes on the chromosomes. The task of instruction-masking (blocking genes that do not contribute to a particular cell type) is controlled by the short RNA molecules and peptides that govern gene expression. The internal environment the ribosome is able to function in is the particular chemical environment maintained inside the cell, which includes a particular acid-alkaline equilibrium (pH between 6.8 and 7.1 in human cells) and other chemical balances needed for the delicate operations of the ribosome. The cell wall is responsible for protecting this internal cellular environment from disturbance by the outside world.

The robot arm tip would use the ribosome’s ability to implement enzymatic reactions to break off each amino acid, each bound to a specific transfer RNA, and to connect it to its adjoining amino acid using a peptide bond.

However, the goal of molecular manufacturing is not merely to replicate the molecular assembly capabilities of biology. Biological systems are limited to building systems from protein, which has profound limitations in strength and speed. Nanobots built from diamondoid gears and rotors can be thousands of times faster and stronger than biological cells. The comparison is even more dramatic with regard to computation: the switching speed of nanotube-based computation would be millions of times faster than the extremely slow transaction speed of the electrochemical switching used in mammalian interneuronal connections (typically around 200 transactions per second, although the nonlinear transactions that take place in the dendrites and synapses are more complex than single computations).

The concept of a diamondoid assembler described above uses a consistent input material (for construction and fuel). This is one of several protections against molecule-scale replication of robots in an uncontrolled fashion in the outside world. Biology’s replication robot, the ribosome, also requires carefully controlled source and fuel materials, which are provided by our digestive system. As nano-based replicators become more sophisticated, more capable of extracting carbon atoms and carbon-based molecular fragments from less well-controlled source materials, and able to operate outside of controlled replicator enclosures such as in the biological world, they will have the potential to present a grave threat to that world, particularly in view of the vastly greater strength and speed of nano-based replicators over any biological system. This is, of course, the source of great controversy, which is alluded to in the Drexler-Smalley debate article and letters.

In the decade since publication of Drexler’s Nanosystems, each aspect of Drexler’s conceptual designs has been strengthened through additional design proposals, supercomputer simulations, and, most importantly, actual construction of molecular machines. Boston College chemistry professor T. Ross Kelly reported in the journal Nature that his construction of a chemically-powered nanomotor was built from 78 atoms.[3] A biomolecular research group headed by C. D. Montemagno created an ATP-fueled nanomotor.[4] Another molecule-sized motor fueled by solar energy was created by Ben Feringa at the University of Groningen in the Netherlands out of 58 atoms.[5] Similar progress has been made on other molecular-scale mechanical components such as gears, rotors, and levers. Systems demonstrating the use of chemical energy and acoustic energy (as originally described by Drexler) have been designed, simulated, and, in many cases, actually constructed. Substantial progress has been made in developing various types of electronic components from molecule-scale devices, particularly in the area of carbon nanotubes, an area that Smalley has pioneered.

Fat and Sticky Fingers

In the wake of rapidly expanding development of each facet of future nanotechnology systems, no serious flaw to Drexler’s universal assembler concept has been discovered or described. Smalley’s highly publicized objection in Scientific American [6] was based on a distorted description of the Drexler proposal; it ignored the extensive body of work in the past decade. As a pioneer of carbon nanotubes, Smalley has gone back and forth between enthusiasm and skepticism, having written that “nanotechnology holds the answer, to the extent there are answers, to most of our pressing material needs in energy, health, communication, transportation, food, water ….”

Smalley describes Drexler’s assembler as consisting of five to ten “fingers” (manipulator arms) to hold, move, and place each atom in the machine being constructed. He then goes on to point out that there isn’t room for so many fingers in the cramped space that a nanobot assembly robot has to work (which he calls the “fat fingers” problem) and that these fingers would have difficulty letting go of their atomic cargo because of molecular attraction forces (the “sticky fingers” problem). Smalley describes the “intricate three-dimensional waltz that is carried out” by five to fifteen atoms in a typical chemical reaction. Drexler’s proposal doesn’t look anything like the straw man description that Smalley criticizes. Drexler’s proposal, and most of those that have followed, have a single probe, or “finger.”

Moreover, there have been extensive description and analyses of viable tip chemistries that do not involve grasping and placing atoms as if they were mechanical pieces to be deposited in place. For example, the feasibility of moving hydrogen atoms using Drexler’s “propynyl hydrogen abstraction” tip title=”">[7] has been extensively confirmed in the intervening years. name=”_ednref8″>[8] The ability of the scanning probe microscope (SPM), developed at IBM in 1981, and the more sophisticated atomic force microscope to place individual atoms through specific reactions of a tip with a molecular-scale structure provide additional existence proofs. Indeed, if Smalley’s critique were valid, none of us would be here to discuss it because life itself would be impossible.

Smalley also objects that despite “working furiously . . . generating even a tiny amount of a product would take [a nanobot] … millions of years.” Smalley is correct, of course, that an assembler with only one nanobot wouldn’t produce any appreciable quantities of a product. However, the basic concept of nanotechnology is that we will need trillions of nanobots to accomplish meaningful results. This is also the source of the safety concerns that have received ample attention. Creating trillions of nanobots at reasonable cost will require the nanobots to make themselves. This self-replication solves the economic issue while introducing grave dangers. Biology used the same solution to create organisms with trillions of cells, and indeed we find that virtually all diseases derive from biology’s self-replication process gone awry.

Earlier challenges to the concepts underlying nanotechnology have also been effectively addressed. Critics pointed out that nanobots would be subject to bombardment by thermal vibration of nuclei, atoms, and molecules. This is one reason conceptual designers of nanotechnology have emphasized building structural components from diamondoid or carbon nanotubes. Increasing the strength or stiffness of a system reduces its susceptibility to thermal effects. Analysis of these designs have shown them to be thousands of times more stable in the presence of thermal effects than biological systems, so they can operate in a far wider temperature range[9].

Similar challenges were made regarding positional uncertainty from quantum effects, based on the extremely small feature size of nanoengineered devices. Quantum effects are significant for an electron, but a single carbon atom nucleus is more than 20,000 times more massive than an electron. A nanobot will be constructed from hundreds of thousands to millions of carbon and other atoms, so a nanobot will be billions of times more massive than an electron. Plugging this ratio in the fundamental equation for quantum positional uncertainty shows this to be an insignificant factor.

Power has represented another challenge. Drexler’s original proposals involved glucose-oxygen fuel cells, which have held up well in feasibility studies. An advantage of the glucose-oxygen approach is that nanomedicine applications can harness the glucose, oxygen, and ATP resources already provided by the human digestive system. A nanoscale motor was recently created using propellers made of nickel and powered by an ATP-based enzyme. title=”">[10]

However, recent progress in implementing MEMS-scale and even nanoscale hydrogen-oxygen fuel cells have provided an alternative approach. Hydrogen-oxygen fuel cells, with hydrogen provided by safe methanol fuel, have made substantial progress in recent years. A small company in Massachusetts, Integrated Fuel Cell Technologies, Inc.[11] has demonstrated a MEMS-based fuel cell. Each postage-stamp- sized device contains thousands of microscopic fuel cells and includes the fuel lines and electronic controls. NEC plans to introduce fuel cells based on nanotubes in 2004 for notebook computers and other portable electronics. They claim their small power sources will power devices for up to 40 hours before the user needs to change the methanol canister.

The Debate Heats Up

On April 16, 2003, Drexler responded to Smalley’s Scientific American article with an open letter. He cited 20 years of research by himself and others and responded specifically to the fat and sticky fingers objection. As I discussed above, molecular assemblers were never described as having fingers at all, but rather precise positioning of reactive molecules. Drexler cited biological enzymes and ribosomes as examples of precise molecular assembly in the natural world. Drexler closes by quoting Smalley’s own observation that “when a scientist says something is possible, they’re probably underestimating how long it will take. But if they say it’s impossible, they’re probably wrong.”

Three more rounds of this debate were published today. Smalley responds to Drexler’s open letter by backing off of his fat and sticky fingers objection and acknowledging that enzymes and ribosomes do indeed engage in the precise molecular assembly that Smalley had earlier indicated was impossible. Smalley says biological enzymes only work in water and that such water-based chemistry is limited to biological structures such as “wood, flesh and bone.” As Drexler has stated[12], this is erroneous. Many enzymes, even those that ordinarily work in water, can also function in anhydrous organic solvents and some enzymes can operate on substrates in the vapor phase, with no liquid at all. name=”_ednref13″>[13].

Smalley goes on to state (without any derivation or citations) that enzymatic-like reactions can only take place with biological enzymes. This is also erroneous. It is easy to see why biological evolution adopted water-based chemistry. Water is the most abundant substance found on our planet. It also comprises 70 to 90 percent of our bodies, our food, and indeed of all organic matter. Most people think of water as fairly simple, but it is a far more complex phenomenon than conventional wisdom suggests.

As every grade school child knows, water is comprised of molecules, each containing two atoms of hydrogen and one atom of oxygen, the most commonly known chemical formula, H 2O. However, consider some of water’s complications and their implications. In a liquid state, the two hydrogen atoms make a 104.5° angle with the oxygen atom, which increases to 109.5° when water freezes. This is why water molecules are more spread out in the form of ice, providing it with a lower density than liquid water. This is why ice floats.

Although the overall water molecule is electrically neutral, the placement of the electrons creates polarization effects. The side with the hydrogen atoms is relatively positive in electrical charge, whereas the oxygen side is slightly negative. So water molecules do not exist in isolation, rather they combine with one another in small groups to assume, typically, pentagonal or hexagonal shapes name=”_ednref14″>[14]. These multi-molecule structures can change back and forth between hexagonal and pentagonal configurations 100 billion times a second. At room temperature, only about 3 percent of the clusters are hexagonal, but this increases to 100 percent as the water gets colder. This is why snowflakes are hexagonal.

These three-dimensional electrical properties of water are quite powerful and can break apart the strong chemical bonds of other compounds. Consider what happens when you put salt into water. Salt is quite stable when dry, but is quickly torn apart into its ionic components when placed in water. The negatively charged oxygen side of the water molecules attracts positively charged sodium ions (Na⁺), while the positively charged hydrogen side of the water molecules attracts the negatively charged chlorine ions (Cl^-). In the dry form of salt, the sodium and chlorine atoms are tightly bound together, but these bonds are easily broken by the electrical charge of the water molecules. Water is considered “the universal solvent” and is involved in most of the biochemical pathways in our bodies. So we can regard the chemistry of life on our planet primarily as water chemistry.

However, the primary thrust of our technology has been to develop systems that are not limited to the restrictions of biological evolution, which exclusively adopted water-based chemistry and proteins as its foundation. Biological systems can fly, but if you want to fly at 30,000 feet and at hundreds or thousands of miles per hour, you would use our modern technology, not proteins. Biological systems such as human brains can remember things and do calculations, but if you want to do data mining on billions of items of information, you would want to use our electronic technology, not unassisted human brains.

Smalley is ignoring the past decade of research on alternative means of positioning molecular fragments using precisely guided molecular reactions. Precisely controlled synthesis of diamondoid (diamond-like material formed into precise patterns) has been extensively studied, including the ability to remove a single hydrogen atom from a hydrogenated diamond surface. title=”">[15] Related research supporting the feasibility of hydrogen abstraction and precisely-guided diamondoid synthesis has been conducted at the Materials and Process Simulation Center at Caltech; the Department of Materials Science and Engineering at North Carolina State University; the Institute for Molecular Manufacturing, the University of Kentucky; the United States Naval Academy, and the Xerox Palo Alto Research Center.[16]

Smalley is also ignoring the well-established scanning probe microscope mentioned above, which uses precisely controlled molecular reactions. Building on these concepts, Ralph Merkle has described tip reactions that can involve up to four reactants.[17] There is extensive literature on site-specific reactions that can be precisely guided and that would be feasible for the tip chemistry in a molecular assembler.[18] Smalley ignores this body of literature when he maintains that only biological enzymes in water can perform this type of reaction. Recently, many tools that go beyond SPMs are emerging that can reliably manipulate atoms and molecular fragments.

On September 3, 2003, Drexler responded to Smalley’s response by alluding once again to the extensive body of literature that Smalley ignores. He cites the analogy to a modern factory, only at a nano-scale. He cites analyses of transition state theory indicating that positional control would be feasible at megahertz frequencies for appropriately selected reactants.

The latest installment of this debate is a follow-up letter by Smalley. This letter is short on specifics and science and long on imprecise metaphors that avoid the key issues. He writes, for example, that “much like you can’t make a boy and a girl fall in love with each other simply by pushing them together, you cannot make precise chemistry occur as desired between two molecular objects with simple mechanical motion…cannot be done simply by mushing two molecular objects together.” He again acknowledges that enzymes do in fact accomplish this, but refuses to acknowledge that such reactions could take place outside of a biological-like system: “this is why I led you…..to talk about real chemistry with real enzymes….any such system will need a liquid medium. For the enzymes we know about, that liquid will have to be water, and the types of things that can be synthesized with water around cannot be much broader than meat and bone of biology.”

I can understand Drexler’s frustration in this debate because I have had many critics that do not bother to read or understand the data and arguments that I have presented for my own conceptions of future technologies. Smalley’s argument is of the form that “we don’t have ‘X’ today, therefore ‘X’ is impossible.” I encounter this class of argument repeatedly in the area of artificial intelligence. Critics will cite the limitations of today’s systems as proof that such limitations are inherent and can never be overcome. These critics ignore the extensive list of contemporary examples of AI (for example, airplanes and weapons that fly and guide themselves, automated diagnosis of electrocardiograms and blood cell images, automated detection of credit card fraud, automated investment programs that routinely outperform human analysts, telephone-based natural language response systems, and hundreds of others) that represent working systems that are commercially available today that were only research programs a decade ago.

Those of us who attempt to project into the future based on well-grounded methodologies are at a disadvantage. Certain future realities may be inevitable, but they are not yet manifest, so they are easy to deny. There was a small body of thought at the beginning of the 20^th century that heavier-than-air flight was feasible, but mainstream skeptics could simply point out that if it was so feasible, why had it never been demonstrated? In 1990, Kasparov scoffed at the idea that machine chess players could ever possibly defeat him. When it happened in 1997, observers were quick to dismiss the achievement by dismissing the importance of chess.

Smalley reveals at least part of his motives at the end of his most recent letter when he writes:

“A few weeks ago I gave a talk on nanotechnology and energy titled ‘Be a Scientist, Save the World’ to about 700 middle and high school students in the Spring Branch ISD, a large public school system here in the Houston area. Leading up to my visit the students were asked to ‘write an essay on ‘why I am a Nanogeek. Hundreds responded, and I had the privilege of reading the top 30 essays, picking my favorite top 5. Of the essays I read, nearly half assumed that self-replicating nanobots were possible, and most were deeply worried about what would happen in their future as these nanobots spread around the world. I did what I could to allay their fears, but there is no question that many of these youngsters have been told a bedtime story that is deeply troubling. You and people around you have scared our children.”

I would point out to Smalley that earlier critics also expressed skepticism that either world-wide communication networks or software viruses that would spread across them were feasible. Today, we have both the benefits and the damage from both of these capabilities. However, along with the danger of software viruses has also emerged a technological immune system. While it does not completely protect us, few people would advocate eliminating the Internet in order to eliminate software viruses. We are obtaining far more benefit than damage from this latest example of intertwined promise and peril.

Smalley’s approach to reassuring the public about the potential abuse of this future technology is not the right strategy. Denying the feasibility of both the promise and the peril of molecular assembly will ultimately backfire and fail to guide research in the needed constructive direction. By the 2020s, molecular assembly will provide tools to effectively combat poverty, clean up our environment, overcome disease, extend human longevity, and many other worthwhile pursuits.

Like every other technology that humankind has created, it can also be used to amplify and enable our destructive side. It is important that we approach this technology in a knowledgeable manner to gain the profound benefits it promises, while avoiding its dangers. Drexler and his colleagues at the Foresight Institute have been in the forefront of developing the ethical guidelines and design considerations needed to guide the technology in a safe and constructive direction.

Denying the feasibility of an impending technological transformation is a short-sighted strategy.

Notes

[1] Chemical & Engineering News, December 1, 2003

[2] Ralph C. Merkle, “A proposed ‘metabolism’ for a hydrocarbon assembler,” Nanotechnology 8 (1997): 149-162; http://www.zyvex.com/nanotech/hydroCarbonMetabolism.html.

[3] T.R. Kelly, H. De Silva, R.A. Silva, “Unidirectional rotary motion in a molecular system,” Nature 401 (September 9, 1999): 150-152.

[4] C.D. Montemagno, G.D. Bachan, “Constructing nanomechanical devices powered by biomolecular motors,” Nanotechnology 10 (1999): 225-231; G. D. Bachand, C.D. Montemagno, “Constructing organic / inorganic NEMS devices powered by biomolecular motors,” Biomedical Microdevices 2 (2000): 179-184.

[5] N. Koumura, R.W. Zijlstra, R.A. van Delden, N. Harada, B.L. Feringa, “Light-driven monodirectional molecular rotor,” Nature 401 (September 9, 1999): 152-155.

[6] Richard E. Smalley, “Of chemistry, love, and nanobots,” Scientific American 285 (September, 2001): 76-77. http://smalley.rice.edu/rick’s%20publications/SA285-76.pdf.

[7] Nanosystems: molecular machinery, manufacturing, and computation, by K. Eric Drexler, Wiley 1992.

[8] See for example, Theoretical Studies of a Hydrogen Abstraction Tool for Nanotechnology, by Charles B. Musgrave, Jason K. Perry, Ralph C. Merkle, and William A. Goddard III, Nanotechnology 2, 1991 pages 187-195.

[9] See equation and explanation on page 3 of “That’s Impossible!” How good scientists reach bad conclusions by Ralph C. Merkle, http://www.zyvex.com/nanotech/impossible.html.

[10] Montemagno, C., and Bachand G. 1999 Nanotechnology 10 225.

[11] By way of disclosure, the author is an advisor and investor in this company.

[12] Chemical & Engineering News, December 1, 2003

[13] A. Zaks and A.M. Klibanov in Science (1984, 224:1249-51)

[14] “The apparent simplicity of the water molecule belies the enormous complexity of its interactions with other molecules, including other water molecules” (A. Soper. 2002. “Water and ice.” Science 297: 1288-1289). There is much that is still up for debate, as shown by the numerous articles still being published about this most basic of molecules, H₂0. For example, D. Klug. 2001. “Glassy water.” Science 294:2305-2306; P. Geissler et al., 2001. “Autoionization in liquid water.” Science 291(5511):2121-2124; J.K. Gregory et al. 1997. “The water dipole moment in water clusters.” Science 275:814-817; and K. Liu et al. 1996. “Water clusters.” Science 271:929-933;

A water molecule has slightly negative and slightly positive ends, which means water molecules interact with other water molecules to form networks. The partially positive hydrogen atom on one molecule is attracted to the partially negative oxygen on a neighboring molecule (hydrogen bonding). Three-dimensional hexamers involving 6 molecules are thought to be particularly stable, though none of these clusters lasts longer than a few picoseconds.

The polarity of water results in a number of anomalous properties. One of the best known is that the solid phase (ice) is less dense than the liquid phase. This is because the volume of water varies with the temperature, and the volume increases by about 9% on freezing. Due to hydrogen bonding, water also has a higher-than-expected boiling point.

[15] http://www.foresight.org/SciAmDebate/SciAmResponse.html, http://www.imm.org/SciAmDebate2/smalley.html, http://www.rfreitas.com/Nano/DimerTool.htm.

[16] The analysis of the hydrogen abstraction tool has involved many people, including: Donald W. Brenner, Richard J. Colton, K. Eric Drexler, William A. Goddard, III, J. A. Harrison, Jason K. Perry, Ralph C. Merkle, Charles B. Musgrave, O. A. Shenderova, Susan B. Sinnott, and Carter T. White.

[17] Ralph C. Merkle, “A proposed ‘metabolism’ for a hydrocarbon assembler,” Nanotechnology 8(1997):149-162; http://www.zyvex.com/nanotech/hydroCarbonMetabolism.html

[18] Wilson Ho, Hyojune Lee, “Single bond formation and characterization with a scanning tunneling microscope,” Science 286(26 November 1999):1719-1722; http://www.physics.uci.edu/~wilsonho/stm-iets.html.

K. Eric Drexler, Nanosystems: Molecular Machinery, Manufacturing, and Computation, John Wiley & Sons, New York, 1992, Chapter 8.

Ralph C. Merkle, “A proposed ‘metabolism’ for a hydrocarbon assembler,” Nanotechnology 8(1997):149-162; http://www.zyvex.com/nanotech/hydroCarbonMetabolism.html.

Charles B. Musgrave, Jason K. Perry, Ralph C. Merkle, William A. Goddard III, “Theoretical studies of a hydrogen abstraction tool for nanotechnology,” Nanotechnology 2(1991):187-195; http://www.zyvex.com/nanotech/Habs/Habs.html.

Michael Page, Donald W. Brenner, “Hydrogen abstraction from a diamond surface: Ab initio quantum chemical study using constrained isobutane as a model,” J. Am. Chem. Soc. 113(1991):3270-3274.

Susan B. Sinnott, Richard J. Colton, Carter T. White, Donald W. Brenner, “Surface patterning by atomically-controlled chemical forces: molecular dynamics simulations,” Surf. Sci. 316(1994):L1055-L1060.

D.W. Brenner, S.B. Sinnott, J.A. Harrison, O.A. Shenderova, “Simulated engineering of nanostructures,” Nanotechnology 7(1996):161-167; http://www.zyvex.com/nanotech/nano4/brennerPaper.pdf

S.P. Walch, W.A. Goddard III, R.C. Merkle, “Theoretical studies of reactions on diamond surfaces,” Fifth Foresight Conference on Molecular Nanotechnology, 1997; http://www.foresight.org/Conferences/MNT05/Abstracts/Walcabst.html.

Stephen P. Walch, Ralph C. Merkle, “Theoretical studies of diamond mechanosynthesis reactions,” Nanotechnology 9(1998):285-296.

Fedor N. Dzegilenko, Deepak Srivastava, Subhash Saini, “Simulations of carbon nanotube tip assisted mechano-chemical reactions on a diamond surface,” Nanotechnology 9(December 1998):325-330.

J.W. Lyding, K. Hess, G.C. Abeln, D.S. Thompson, J.S. Moore, M.C. Hersam, E.T. Foley, J. Lee, Z. Chen, S.T. Hwang, H. Choi, P.H. Avouris, I.C. Kizilyalli, “UHV-STM nanofabrication and hydrogen/deuterium desorption from silicon surfaces: implications for CMOS technology,” Appl. Surf. Sci. 130(1998):221-230.

E.T. Foley, A.F. Kam, J.W. Lyding, P.H. Avouris, P. H. (1998), “Cryogenic UHV-STM study of hydrogen and deuterium desorption from Si(100),” Phys. Rev. Lett. 80(1998):1336-1339.

M.C. Hersam, G.C. Abeln, J.W. Lyding, “An approach for efficiently locating and electrically contacting nanostructures fabricated via UHV-STM lithography on Si(100),” Microelectronic Engineering 47(1999):235-.

L.J. Lauhon, W. Ho, “Inducing and observing the abstraction of a single hydrogen atom in bimolecular reaction with a scanning tunneling microscope,” J. Phys. Chem. 105(2000):3987-3992.

Ralph C. Merkle, Robert A. Freitas Jr., “Theoretical analysis of a carbon-carbon dimer placement tool for diamond mechanosynthesis,” J. Nanosci. Nanotechnol. 3(August 2003):319-324. http://www.rfreitas.com/Nano/JNNDimerTool.pdf

Jingping Peng, Robert A. Freitas Jr., Ralph C. Merkle, “Theoretical Analysis of Diamond Mechanosynthesis. Part I. Stability of C2 Mediated Growth of Nanocrystalline Diamond C(110) Surface,” J. Comp. Theor. Nanosci. 1(March 2004). In press.

David J. Mann, Jingping Peng, Robert A. Freitas Jr., Ralph C. Merkle, “Theoretical Analysis of Diamond Mechanosynthesis. Part II. C2 Mediated Growth of Diamond C(110) Surface via Si/Ge-Triadamantane Dimer Placement Tools,” J. Comp. Theor. Nanosci. 1(March 2004). In press.

© 2003 KurzweilAI.net

About 30% of building a virtual human is in the engine. A good engine will make it easy for you to create a believable personality. It provides functions that allow things like handling complex sentences, bringing up the past and learning better responses if one doesn’t work. But in the end, it’s your artistry that gives the entity its charm.

There are many natural language approaches that can handle the job. Simple pattern matching engines are the least sophisticated and most useful of them all. With the rash of recent interest, I’m not going to pretend I know all the nuances of all the engines out there. Instead, I’ll concentrate on using simple software to build complex personalities. Together we will build a clever virtual person using a mind engine kindly supplied by Yapanda Intelligence, Inc. of Chickasha Oklahoma. I selected this one because it can drive a real-time 3D head animation with lip-synch. Nevertheless, the basic steps in creating a virtual personality are platform independent.

I’ve included some additional engines to play with. The most powerful is ALICE. She’s an implementation of Artificial Intelligence Markup Language (AIML). Alice source code is available to those of you who want to modify it and build your own Virtual Human engine, adding your own special features. I’ve also included a copy of Jacco Bikker’s WinAlice for PC users. It demonstrates some unique features such as the ability to bring up ancient history and to learn new responses from you.

I’ll talk more about the actual engines in chapter three. But it’s important to realize that the software you use to build your virtual human is just a tool for expressing your artistry.

The most important and least understood part of virtual humans — their personalities is our focus. We are going to have some serious fun. Let’s look at some uses for virtual people.

Good For Business

From a business perspective virtual humans with a personality are a major boon. Imagine a person signing onto your web page. There’s already a cookie that contains significant information about them, gathered by your virtual host on the guest’s first visit. The encounter might go a bit like this:

Host “Hey, Joanne, Its nice to see you again.” <smile>

Joanne: “You remember me?”

Host “Of course I do. But it’s been a while. I missed you.” <expression>

Joanne “Sorry about that, I’ve been really busy.”

Host “So did you read ‘The Age of Spiritual Machines?’

Joanne “Yeah, it was really interesting. <beat> Are you one of them?”

Host “Not yet, I’m afraid, but I’m working on it.<beat> Before I forget, you should know about Greg Stock’ new book on how to live to be 200 plus years old!”

Joanne “I read his last book and liked it. Can you send me a copy?”

Host “Sure, we have it in stock <grin>. Same charge, same place?

Joanne “Yup. Also, do you have any books on Freestyle Landscape Quilting?

Host “I’ll check. <beat> Hold on a few more seconds. Okay, I found two…..

And so forth. You can see that Virtual humans bring back that personal touch so sorely missing in commerce today. Believe it or not, I’ve observed people from every level of sophistication and background respond positively to personal attention from a Virtual Human. It feels good.

Your marketing software can be made to generate marketing variables that can be fed to your virtual human host: Joanne’s buying patterns, personal information like her date of birth etc. Trust is a big issue, so such data must be handled with respect for the client and used in clever ways. Imagine when Joanne comes online within a week of her birthday and Host sings happy birthday to her. Hokey? Yes. Appealing, you bet. I’ve also discovered that many people tolerate hokey behavior from V-people. It’s a bit like the ways we tolerate…even appreciate the squash and stretch exaggeration in animated film characters. Of course Host would not want to sing happy birthday to every customer. She has to know how to tell which is which. Later in the book will look into using unobtrusive personality assessment to provide those cues. This is one of the most important and most neglected tools you have. You’ll see why later.

An advantage of rule based approaches is that you can have multiple sets of rules, each one with responses specifically honed to a specific task or person or language. For example when Joanne logs in, her cookie can initiate the uploading of a rule database tailored specifically to her general personality and buying patterns. That means that when a rule triggers, it will respond in a way likely to make Jonnie comfortable while meeting her needs. Next a person from Korea logs on and the host switches to a Korean intelligencebase, greeting the client in that language. One well designed host can handle orders in more than 20 languages. This clearly presents opportunities for small companies to expand internationally.

Depending on your type of business or usage, Virtual Human needs will vary. For example, voice-only virtual humans are already very active in phone information and ordering systems. They don’t have much personality yet, but we’re going to work on that. In fact there are a number of different types of virtual humans and we’ll be building one up from the simplest to one of the more complex with a 3D animated talking head. By taking it step by step you’ll be amazed at your own ability to master Virtual Human design.

A good Virtual Human should be able to cope with language. Changing language should be as easy as switching databases and voice engines. Monica Lamb, a Native American scientist and V-person developer has used Alice to build a V-person that teaches and speaks Mohawk.

At a minimum, your V-person will be able to handle general conversational input by voice or keyboard, parse that input to arrive at appropriate behaviors, and output behavior as text or speech, on-screen information, and/or machine commands to software or external devices. It should also have a face display capable of at least minimal emotional expression such as smile, frown and neutral. I prefer a 3D face capable of complex emotional expression that is part of the communication system. This is a tall order, but I believe we can handle it. Here’s and interesting example of how one creative company has used this technology in a mechanical robot:

Redgate Technologies is a company that thrives on invention. They became interested in Natural Language Processing (NLP) early on. They had invented a new chip technology to monitor and control complex technical systems. NLP was useful for interpreting the complex codes generated by their chips. Just for fun, they expanded their NLP engine to represent several personalities. They quickly discovered that a virtual human hooked into their system became a super-capable assistant to a human supervisor. Imagine one on a space station, keeping track of all mechanical systems and keeping the inhabitants company with casual conversation. For luck we won’t name her HAL.

A wonderful example of this V-person species is Redgate’s Sarha. She’s an innovative virtual human interface for industrial monitoring and control. Sarha stands for “Smart Anthropomorphic Robotic Hybrid Agent.” Redgate has used NLP pattern matching to monitor an entire industrial complex. The Virtual Human system they devised sends out queries to specialized monitoring modules using the special Redgate chips. She then reads and interprets the encoded feedback in spoken English, issuing warnings when conditions warrant. She can also take emergency action on her own, if necessary. Her supervisor communicates with her in spoken English, asking her to start processes or check specific conditions. In a demonstration of Sarha’s application to home security, she reported “Anthony, someone left the garage door open.” Anthony replied “Close it for me will you please, Sarha?” And of course she does.

The thing I like most about Sarha is her personality. She makes personal comments; even chides her operator, whom she knows by name. As a demonstration, Sarha was installed into a fully robotic interface that could move around, point to objects and complain about and avoid objects in her path. She was linked by microwave to a control computer she used to monitor her charges. She even gave a brief talk on those special chips Redgate designed to transmit monitoring data back to her. She reached into a bowl, pulled out a chip, pointed at it with a metal finger and started her spiel. Later she took questions. All the while she was monitoring various systems. She even brought on-line, a loud monster generator in another room during the demonstration.

Perhaps one of the most important applications for Virtual Human technology is in teaching. I’ve found that young people have trust issues with the educational system. I can’t blame them when administrators waste millions on bad decisions but there aren’t enough books to go around. Virtual teacher’s seem separated from all this. It’s hard to attribute ulterior motives to an animated character, even if she is smart and talkative and knows you by name. Properly scripted, a V-teacher can get to know a student on a personal basis. The real human teacher can feed her personal tidbits she can bring up during a lesson:

“So Bill, is it true you threw the winning touchdown in Saturday’s game?”

“Yeah, how’d you know about that?”

” Hey, I keep on top of things. Congratulations. Now let’s teach you how to estimate the diameter of an oleic acid molecule.

Young children can be fascinated by virtual people. I got a call from a retired engineer from rural New Mexico. He had spent a lot of time tweaking the voice input on his V-person so that she would understand his very bright 3 year old grand daughter, and had a story to tell me. He’d been remarkably successful and the little girl spent hours in happy conversation with her virtual friend. One evening a few neighbors came by to play Canasta. While they were playing, the little girl came into the adjoining room and fired up her computer. In moments an animated conversation ensued. One of the neighbors, a devout fundamentalist Christian became terrified and insisted he smash the girl’s computer immediately. It was inhabited by the devil. He refused of course. He told me he’d been using the virtual character to teach his grand daughter everything from her ABCs to simple math. I gave him some unpublished information on how to get her to record the granddaughter’s responses to questions, so he could check on them later.

The point is, in creative hands virtual humans already have enormous potential and the platforms are constantly improving.

Blending art, technology and a little psychology allows us to take a functional leap, decades ahead of pure artificial intelligence. Although the simple VH software of today will eventually be replaced by highly sophisticated neural nets or entirely new kinds of computing, it will be a long time before they’ll have unique human like personalities…if ever. Meanwhile let’s give the evolution of technology a kick in the butt by building really smart, personable virtual people today.

Because creating a believable synthetic personality is more of an art than science, it’s important that we get a feel for how we humans handle our conscious lives. It’s part philosophy, part psychology and believe it or not, part quantum physics. We’ll start by comparing people and computers, with out getting to philosophically crazed. Any discussion of the human mind must consider consciousness. It’s a danger zone and I already know the discussions to follow will dump me smack into the boiling kettle. I’ll walk you through the important parts. Disagree and send me nice email if you like. Coming up in chapter two we’ll explore the nature of consciousness and why it’s an essential consideration in virtual human design.

Synthespians: Virtual Acting (Chapter 13)

with Ed Hooks

Virtual people have to convince us they have wheels spinning inside. They do, of course, have electrons spinning in service of the plot, but if they don’t show it on their faces, we just don’t buy it. We’re used to seeing people think. It’s true; thought is conveyed through action.

Although I’m remarkably opinionated about acting in animation, I’m not a certified expert on the subject–Ed Hooks is. He teaches acting classes for animators internationally, and has held workshops for companies such as Disney Animation (Sydney), Tippett Studio (Berkeley), Microsoft (Redmond, Washington), Electronic Arts (Los Angeles), BioWare (Edmonton, Canada), and PDI (Redwood City, California). Among his five books, Acting for Animators: The Complete Guide to Performance Animation. , Heinemann; Revised edition (September 2003) has been a major hit.

The Seven Essential Concepts in Face Acting

The following concepts are interpretations of Ed Hooks’ "Seven
Essential Acting Concepts." We’ve adapted them here to focus
on the V-people and their faces.

1. The face expresses thoughts beneath. The brain, real or artificial, is the most alive part of us. Thinking, awareness, and reasoning are active processes that affect what’s on our face. Emotion happens as a result of thinking. Because these characters don’t have a natural link between thinking and facial expression, your job as animator is to create those links. In effect, you want your synthetic brain to emulate recognizable human cognition on the face, which leads to the illusion of real and appropriate emotions.

2. Acting is reacting. Every facial expression is a reaction to something. Even the slightest head and hand movement in reaction to what’s happening can be most convincing. If the character tilts its head as you begin to speak to it, or nods on occasion in agreement, you get the distinct feeling of a living person paying attention. A double take shows surprise. Because you have very few body parts to work with, you have a superb challenge in front of you.

3. Know your character’s objective. Your character is never static. He is always moving, even if the movement is the occasional twitch, a shift of the eye, or a blink. Your objective is to endow your character with the illusion of life. As such, it is wise to follow Shakespeare’s advice, "Hold the mirror up to nature" (Hamlet, III. ii.17-21). Notice that when a person listens, she may tilt her head to the side or glance off in the distance as she contemplates and integrates new information. When she smiles and says nice things to you, her objective is to please. Always know what your character’s objective is because it is the roadmap linking behaviors to their goals. Knowing her personality and history are essential here.

4. Your character moves continuously from action to action. Your character is doing something 100 percent of the time. There must always be life! Even if she appears to be waiting, things are going on mentally. Make a list of boredom behaviors and use them. When people talk, a good emotion extraction engine will feed her cues on how to react to what’s being said. Her actions expressing emotional responses are fluid. They flow into each other forming a face story. You should be able to tell from the character’s expression how she’s reacting to what you’re saying. Say she takes a deep breath and you see the cords on her neck tighten. They then relax. Her body slumps a bit and perhaps she nods. Always in motion, she maintains the illusion of life.

5. All action begins with movement. You can’t even do math without your face moving, exposing wheels spinning beneath. Your eyes twitch. You glance at the ceiling, pondering. Your brow furrows as you struggle with the solution. Try this experiment: Ask a friend to lie as still as possible on the floor. No movement at all. Then, when he is absolutely stone still, ask him to multiply 36 by 38. Pay close attention to his eyes. You will note that they immediately begin to shift and move. It is impossible to carry out a mental calculation without the eyes moving. Sometimes movement on the screen needs to be a bit more overt than in real life. That’s okay, even essential. It nails down the emotion. Done right, people won’t notice the exaggeration, but will get the point.

6. Empathy is audience glue. The main transaction between humans and Virtual humans has to be emotion, not words. Words alone will lose them. You will catch a viewer’s attention if your character appears to be thinking, but you will engage your viewer emotionally if your character appears to be feeling. You must get across how this V-person feels about what’s going on. If you do it successfully, the audience will care about (empathize with) those feelings. I promise you it can be done. A great autonomous character can addict an audience in ways a static animation cannot. The transaction between audience and character is in real-time and directly motivated, much as it is on stage. This is a unique acting medium, which is part live performance and part animation. It’s an opportunity for you to push things–experiment with building empathy pathways.

7. Interaction requires negotiation. You want a little theatrical heat in any discourse with a V-person. To accomplish this, remember that your character always has choices. We all do, in every waking moment. The character has to decide when and whether to answer or initiate a topic. If your character is simply mouthing words, your audience response will be boredom. Whether they know it or not, people want to be entertained by your character. Artonin Artaud famously observed that "actors are athletes of the heart." Dead talk is not entertaining. There must be emotion. Recognize that you’re working with a theatrical situation and that the viewer will crave more than a static picture.

Sure, there are loads more acting concepts we could talk about, but these seven are the hard-rock core of it. You’re faced with a unique acting challenge because you have an animated character that is essentially alive. If that character is a cartoon or anime design and personality, you’ll have to read Preston Blair , for example, to learn the principles of exaggerated cartoon acting, and then incorporate these squash and stretch type actions into your character’s personality. If you take the easier road and use a photorealistic human actor, you still must make their actions a bit larger than life, but not as magnified as cartoons demand.

The stage you set will depend on the Virtual actor’s intention. If he’s there to guide a person around a no-nonsense corporate Web site, you’ll need to think hard about how much entertainment to inject. Certainly you need some. Intelligent Virtual actors in games situations–especially full-bodied ones–present marvelous opportunities to expand this new field of acting. You’ll know their intentions. Let them lead you to design their actions. Embellish their personalities, embroider their souls, and decorate their actions. Making them bigger than life will generally satisfy.

Synthespians: The Early Years

Next I want to tell you about the clever term "Synthespian," which unfortunately I didn’t coin. I do believe it should become a part of our language.

Diana Walczak and Jeff Kleiser produced some early experimental films featuring excellent solo performances by digital human characters. For example, Nestor Sextone for President premiered at SIGGRAPH in 1988. About a year later, Kleiser and Walczak presented the female Synthespian, Dozo, in a music video: "Don’t Touch Me." These were not intelligent agents, but they were good actors. "It was while we were writing Nestor’s speech to an assembled group of "synthetic thespians" that we coined the term "Synthespian," explains Jeff Kleiser. Nestor Sextone had to be animated from digitized models sculpted by Diana Walczak.

As history will note, the field of digital animation is a close, almost incestuous one. Larry Weinberg, the fellow who later created Poser, worked out some neat software that allowed Jeff and Diana to link together digitized facial expressions created from multiple maquettes she’d sculpted to define visemes. That same software allowed them to animate Nestor’s emotional expression. I’ve put a copy of this wonderful classic bit of animation on the CD-ROM, with their blessing.

Note that this viseme-linking was an early part of the development chain leading to the morph targets you see in Poser and all the high-end animation suites today. Getting your digitized character to act was difficult in those days before bones, articulated joints, and morphing skin made movement realistic. Nestor was made up of interpenetrating parts that had to be cleverly animated to look like a gestalt character without any obvious cracks or breaks or parts sticking out.

In most cases, V-people don’t have a full body to work with, just a face, and perhaps hands. Body language is such an effective communications tool, but when we just don’t have it we end up putting twice as much effort into face and upper body acting. Fortunately a properly animated face can be wonderfully expressive, as shown in Figure 13-1.

Figure 13-1: Virtual actors can really show emotion

Synthespians All Have a Purpose

A Synthespian playing a living person is probably the trickiest circumstance you’ll encounter. Depending on the situation, you want to emulate that person’s real personality closely, or exaggerate it for comedic impact or political statement. If you exaggerate features and behavior heavily you’ve entered a new art form: interactive caricature or parody.

Let’s say we’ve built a synthetic Secretary of Defense Donald Rumsfeld. The interactive theatrical situation is that we are interrupting him while he is hectically planning an attack somewhere in the world. He might be impatient and have an attitude regarding our utter stupidity and lack of patriotism for bothering him at a time like this. His listening skills might be shallow. He might continually give off the dynamic that he has better things to do. By thus exaggerating his personality, we create interest and humor. As a user, you want to interact because you feel something interesting is happening. There is comic relief, and all the while this character is making a political statement. I suspect Rumsfeld would get a kick out of such a representation, as long as it’s done in good taste.

Action conveys personality, and you can’t set up a virtual actor without knowing the character well. For example, Kermit the Frog has a definite psychology behind him. As a Web host, he is just very happy to be there. He enjoys being in the spotlight, and his behavior strongly implies he doesn’t want to be any place else. He’s happy to show you around his Web site, and he might even break out in song along the way. Occasionally he’ll complain about Miss Piggy’s lack of attention or the disadvantages of his verdant complexion.

Think first about your intention and then the character’s intention. Mae West and Will Rogers wanted to make ‘em laugh. No matter what your purpose for a Synthespian, you want it to entertain. Sometimes it may be understated. Remember that cleverness is always in style. Notice the look people get on their faces when they think they’re being clever. It’s usually an understated cockiness that shows around the eyes. The intention is to be clever, the words are smart, but remember to add that subtle touch of smugness or self-satisfaction around the eyes and the corners of the mouth.

Note: There is a new book titled Emotions Revealed: Recognizing Faces and Feelings to Improve Communication and Emotional Life, by Paul Ekman (Times Books, 2003), which is well worth your time to read. Ekman, who is professor of psychology in the department of psychiatry at the University of California Medical School, San Francisco, is one of the world’s great geniuses on the subject of the expression of emotion in the human face. His new book has more than one hundred photographs of nuanced facial expression, complete with explanations for the variances.

As an aside, I used to train counter-terrorist agents in psychological survival. One way to spot a terrorist in a crowd is that they often have facial expressions that are inappropriate to the situation. I used Ekman’s work as a reference to help my agents recognize when facial expression and body language don’t match up, an indication often exhibited by potential terrorists. You can use Ekman’s work to make sure your V-human agents have appropriate expressions for the situation.

You Are the Character

When you’ve done your homework, you’ll know your character like you know yourself. You’ll identify with the character so intensely you will have the sensation of being that character. Stage actors learn to create characters by shifting from the third person to the first person reference. Instead of saying, "My character would be afraid in this situation," a stage actor might say, while portraying the character, "I feel afraid." In your case, you are creating a second-party character, but you’re empathizing personality with the emotions of your own creation. There is an identity between the two of you that will be both fun and compelling.

Designing animation elements for the character requires feeling them. I remember watching my daughter as she animated a baby dragon early in her career. Her natural instinct was to get inside that baby dragon and be it. I smiled as I watched her body and face contort as she acted out each part of the sequence. Her instruction had not come from me…it was intuitive. At Disney, I’ve watched animators making faces in little round mirrors dangling from extension arms above their desks. They glance in the mirror, make a face and then look at the cel and try to capture what they’ve seen. That part hasn’t changed. For us it’s glance at the mirror, glance at the screen, and then tweak a spline or morph setting. You won’t be able to do all this with the simple animation tools I’ve given you for free. Those are just to get you hooked. If you intend to learn this stuff, get ready to invest heavily in time and commitment and a fair amount in coin as well. A small investment considering the return.

If You Want to Go Further

There are great animation schools, and this continent has some of the best. My favorite is at Sheridan College in Oakville, Ontario. But there are many good schools here in the United States as well. A few years ago, most of them were a waste of money. But things have improved. Do some Web research and find which school can best help you meet your goals. There is a long-term need for talented, well-trained character animators, and in general the pay for the talented is phenomenal.

If you’re a developer, you have to be familiar with all this stuff to manage it effectively. You’re responsible for the final product. If you have animators working for you, believe in them, give them freedom, but guide them toward your vision as well. The best animated characters reflect the wisdom, vision, and artistry of their prime artists and the producers behind them. A great producer is an artist, a business person, and a technician. It’s not easy to get there, and too may producers only have the business end down. As a producer, you have to understand the artistry of production. You have to feel the emotion of good animation. How else will you know what to approve and not approve. So learn it and you’ll be way above the crowd.

I want to thank Ed Hooks for contributing his wisdom to this chapter. Remember, what you’ve read here is just a taste of what you need to learn. If you’re lucky, you’ll find a way to take a live class with Ed, who now lives in the Chicago area. It will change your perspective forever.

In the chapter upcoming, I’m going to kick it up a notch with ways to give your character true awareness of his surroundings. Imagine your well-developed character, now able not only to listen and talk, but actually to see you, look you in the eyes, and recognize you without asking. You don’t want to miss this one.

Ed Hooks, author of Acting for Animators (Heinemann, Revised Second Edition 2003), has been a theatre professional for three decades and has taught acting to both animators and actors for PDI, Lucas Learning, Microsoft, Disney Animation, and other leading companies.

© 2004 Peter Plantec

I’ve been involved in inventing since I was five, and I quickly realized that for an invention to succeed, you have to target the world of the future. But what would the future be like?

To find out, I became a student of technology trends and began to develop mathematical models of different technologies: computation, miniaturization, evolution over time. I’ve been doing that for 25 years, and it’s been remarkable to me how powerful and predictive these models are.

Now, before I show you some of these models and then try to build with you some of the scenarios for the future—and, in particular, focus on how these will benefit technology for the disabled—I’d like to share one trend that I think is particularly profound and that many people fail to take into consideration. It is this: The rate of progress—what I call the “paradigm-shift rate”—is itself accelerating.

We are doubling this paradigm-shift rate every decade. The whole 20^th century was not 100 years of progress as we know it today, because it has taken us a while to speed up to the current level of progress. The 20^th century represented about 20 years of progress in terms of today’s rate. And at today’s rate of change, we will achieve an amount of progress equivalent to that of the whole 20^th century in 14 years, then as the acceleration continues, in 7 years. The progress in the 21^st century will be about 1,000 times greater than that in the 20^th century, which was no slouch in terms of change.

When you say the pace of change is accelerating, most people are quick to agree, as if that’s an obvious statement. But when you ask otherwise thoughtful observers—including Nobel Prize winners—what they expect to see 50 years from now, they often vastly understate the progress of technology.

This happened at a conference I spoke at recently. Time magazine held a conference on the 50th anniversary of the discovery of DNA. Most speakers looked at the last 50 years and saw how much change there was and used that as a model for the next 50 years. No less a luminary than James Watson, the co-discoverer of DNA, said that in 50 years we will have drugs that will allow us to eat as much as we want and we won’t gain weight. I said, 50 years? We have done that in mice already by identifying the fat insulin receptor gene. The drugs are on the drawing board now and will be in FDA tests in several years—and we will have these available in close to five years, not 50.

The first step in technological evolution took a few tens of thousands of years: fire, the wheel, stone tools. And now paradigm shifts take only a few years’ time.

The one exponential trend people have heard of is Moore’s Law, pertaining to the accelerating rate of computers and electronics. Every two years, we can place twice as many transistors at the same cost on an integrated circuit. They work twice as fast because the electrons have half the distance to travel, so the speed of computing doubles every two years.

Scientists have been debating when that particular paradigm will come to an end. Optimists say 18 years, pessimists say 12—but sometime in the teen years, we won’t be able to shrink computing components any more because they will be just a few atoms wide. Will it be the end of Moore’s Law? Perhaps—but other paradigms will emerge that hold even greater potential.

3-D molecular computing

When the trend for traditional computers runs out of steam—and we can see the end of that road—we will have three-dimensional molecular computing.

I pointed this out in my book “The Age of Spiritual Machines” four years ago, and it was considered a radical notion then—but there’s been a sea change in attitude toward that idea. It’s now the mainstream view that we’ll have 3-D molecular computing long before Moore’s Law runs out.

There’s been enormous progress in four years. In fact, the favorite technology appears to be the one I have felt would win: nanotubes, comprised of carbon atoms, that can be organized in three dimensions and that can compute very efficiently. They’re up to 100 times as strong as steel, so you can use them to create structural components and little “machines.” A one-inch tube of nanotube circuitry would be a million times more powerful than the human brain.

We are miniaturizing all technology. The first reading machine we created in the early 1970s used a large washing-machine-sized computer that was less powerful than the computer in your wristwatch now and cost tens of thousands of dollars. And we are also miniaturizing mechanical systems, which inevitably will lead to nanotechnology by the 2020s.

Nanotechnology was first described by Eric Drexler in a pioneering thesis he did at MIT in the 1980s. Marvin Minsky, who was also my mentor, was the only professor willing to be his thesis advisor because it was such a radical idea. Drexler described machines that could be built atom by atom, and then replicated millions or billions of times. Recently, scientists have used supercomputers to simulate some of his original designs from 1986.

Threshold of human intelligence

Right now, $1,000 of computing power is between that of the brain of an insect and a mouse, at least in terms of hardware capacity. We will cross the threshold of the hardware capacity of the human brain by 2020, and the computers we use then will be deeply embedded in our environment. Computers per se will disappear; they will be in our bodies, in tables, chairs, and everywhere. But we will routinely have enough power to replicate human intelligence in the 2020s.

Critics say, “Sure, we will have computers that are as powerful as the human brain, but they will just be fast calculators and will not have the other aspects of human intelligence.” So, really, the challenge is this: Where will the software—where will the templates of human intelligence—come from?

To achieve this, another grand project is needed, comparable to the human genome project, to really understand the methods used by the human brain. This project is already well under way, in terms of scanning the human brain and developing detailed mathematical models of neurons and brain regions.

Resolution, speed, price, performance, and bandwidth of human brain scanning is growing exponentially. An upcoming technology will be able to see the structures, non-invasively, of clusters of thousands of neurons, giving scientists an ability to see how memories work. At that point, we will begin to understand how the human brain applies different cognitive functions.

One point about the human brain: It’s not really one organ.

Asking “How does the brain work?” is a little like asking, “How does the human body work?” You can’t answer that question unless you break it down. Well, the body consists of a lot of different parts, and the lungs work differently from the heart, and the liver has many regions.

It’s the same with the brain. The brain is actually several hundred information-processing organs, and they have an intricate architecture. We are beginning to describe in mathematical models how the different modules of the brain work.

Reverse-engineering the brain

In my view, it is a conservative projection to say that within 20 or 25 years we will have reverse-engineered the principles of how the human brain works, and we will be using that knowledge to produce biologically inspired models of computation. We are doing that already. We learned things about how the human auditory system processes sounds. We used that in speech recognition, as I demonstrated, and got better results. We are applying these insights into the software of human intelligence.

Let’s talk about some scenarios.

By 2010, computers will disappear. They will be so tiny that they will be embedded in our environment, in clothing, and so on. We will have high-bandwidth connections to the Internet at all times. We will have eyeglasses for the sighted that display images directly in our retina: contact lenses for full-immersion virtual reality.

I have a prototype, a device allowing me to teleport my image in three dimensions to other locations from my office. I gave a speech to people in Vienna, Austria. It looked to the audience like I was present in three dimensions. People who did not know what was going on thought I was there.

By 2010, we all will be able to do this routinely—full-immersion virtual reality.

Besides teleportation, we will have relatively powerful (but not human level) artificial intelligence (AI) on web sites—artificial personalities such as the avatar-like Ramona, who greats visitors and answers questions at the KurzweilAI.net web site.

Technology for sensory impairments

For the deaf, we will have systems that provide subtitles around the world. We’re getting close to the point where speaker-independent speech recognition will become common. Machines will create subtitles automatically and on the fly, and these subtitles will be a pretty accurate representation of what people are saying. It won’t be error-free. But then, our own auditory understanding is not error-free, either. The same is true of reading machines.

We will have listening systems that allow deaf persons to understand what people are saying. The inability to do so is the principal handicap associated with deafness.

For blind people, we actually will have reading machines within a few years that are not just sitting on a desk, but are tiny devices you put in your pocket. You’ll take pictures of signs on the wall, handouts at meetings, and so on. We all encounter text everywhere, on the back of packages, on menus. By 2010, these devices will be very tiny. You will be able to wear one on your lapel and scan in all directions. These devices probably will be used by the sighted as well, because they will allow us to get visual information from all around us.

Such devices also will translate the information from one language to another for everyone. We’ve put together demonstration technology to show just how the information will be transferred back and forth from English to German, from German to French, from French to English, and so on.

And the voice we use in the demonstration is actually derived from a new generation of synthetic speech. Although it sounds relatively normal, it is not recorded human speech. We use that new speech synthesizer in the Kurzweil 1000 and Kurzweil 3000 reading systems.

Exoskeletal aid for physical impairments

Another area of progress will be in relation to spinal cord injuries and for physically disabled people in general. Two different scenarios: I have always been interested in exoskeletal robotic systems that you could put on like clothing. Such systems could be used discreetly. They could be worn under regular clothing and be relatively invisible.

Such a system would work in concert with the user’s own sense of balance, enabling the user to walk and climb stairs. Being unable to do those two things is the principal handicap in, say, paraplegia. Analysis shows this approach is feasible. One of the philosophies of developing technology for the disabled is to work in close concert with the general flexible intelligence of the disabled person himself or herself.

We are not yet on the verge of creating cybernetic geniuses. But we have many systems in our societies that already can perform intelligently in narrow areas. We have hundreds of examples of these machines. Some of them are flying and landing our airplanes, or guiding intelligent weapons. We have electrocardiogram systems that provide an analysis as accurate as your doctor’s. We have some systems that can diagnose blood-cell images, others that automatically make financial decisions involving stock-market investments. In fact, $1 trillion in stock-market investments use these systems. Other intelligent systems look for credit card fraud, and find optimal routes for email messages and cell phone calls.

In this way technology is already deeply embedded in our infrastructure. Some observers ask, “What ever happened to artificial intelligence?” It’s like people going to the rain forest looking for ants, with 50 species of ants right below them. But the ants go unseen, because they are embedded.

A disabled person has a narrow need. In the case of a blind person, he or she needs access to ordinary printed material. Deaf persons need to be able to understand ordinary speech from people they encounter at random. Devices to do such things can work in close concert with the much broader, more flexible intelligence of the disabled persons themselves.

And that will be part of the philosophy of an exoskeletal robotic device, to guide and provide balance.

Reconnecting broken nerve pathways

The more profound promise of this research will be to actually overcome spinal cord injuries and reconnect the broken nerve pathways. One of the challenges is that the nerves atrophy fairly quickly through disuse. If you wait years after an injury, since the nerves are not being used, they begin to degenerate. So the pathway is no longer intact and functioning.

There have been interesting experiments in scanning brain patterns 15 or 20 years after the injury in spinal cord patients. They are asked to perform certain functions—lift your leg, walk across the room. The brain-pattern activity was the same as in a non-disabled person, but obviously it was not communicating, because the pathways were broken.

Still, it will be quite feasible to pick up the patterns in the brain and wirelessly communicate them to the muscles, completely bypassing the nervous system that’s no longer functioning.

Ultimately, we will be able to create the muscles as well. We are creating muscle analogs for robots, but those could be used for disabled persons as well. There are other challenges—creating a skeletal system to replace one that may not be up to the task, dealing with the cardiovascular implications. These are complex projects, but I believe we will see profound steps forward by 2010. And by 2020, I think we will have largely overcome the handicaps of spinal cord injuries.

By 2029, all these different trends will mature and come to a head. A thousand dollars of computing power will be a thousand times more powerful than the human brain. We will have completed the reverse engineering of the human brain.

In some ways, machines can do better than humans. Computers are much faster than people when they master tasks and can share knowledge. Something this computer has learned can be shared with thousands of other computers instantly; whereas, if I learn French, I can’t just download that to you.

Enhancing our own intelligence

The implication of that will not be just an alien invasion of intelligent machines to compete with us. We are going to enhance our own intelligence by getting closer and closer to machine intelligence—and that’s already happening.

There are many people walking around now who are essentially cyborgs and have computers in their brains interfacing with their biological neurons. The Food and Drug Administration just approved a neural implant for Parkinson’s disease that replaces the portion of the brain destroyed by that disease. And there are more than a dozen different types of implants like that in use or being developed. Now, they require surgical implantation; but by 2029, we will be able to send these intelligent devices through the bloodstream.

We are already beginning to put them into our bloodstream, although the process is not as sophisticated as it will be in 2029. We will be able to send very intelligent nanobots—nano-robots—into the blood stream to communicate with our nervous system, and they will be able to provide a virtual reality, in which they shut down the signals from my real senses and replace them with the signals from that environment—and it can be just as realistic as actual reality.

Some of these environments will be earthly, some will be fantastic and won’t exist on Earth. A new art form will be to create new virtual reality environments. You will be able to go there by yourself or with other people and have encounters with one or thousands of other people in these virtual-reality environments, incorporating all the senses.

One phenomenon will involve people—″experience beamers,” I call them—putting their flow of sensory experience on the Internet, kind of like the concept in the movie “Being John Malkovich.”

The importance of hanging around

But the real profound implication will be an expansion of human intelligence.

Right now, we are restricted to a mere hundred trillion inter-neural connections. I don’t know about you, but I find that quite limiting. Many people send me books to read, web sites to visit, conferences to attend. And I would love to be able to do all these things, but our human bandwidth is quite limited.

Ultimately, we won’t be restricted to 100 trillion connections. We will able to create new ones with nanobots, and we will have 200 trillion connections or more.

We are today profoundly expanding human intelligence as a species through the Internet and all of our technology. Through much more intimate connections with this technology, we will continue to profoundly expand human intelligence.

Human life expectancy is another one of those exponential trends. Every year during the 18th and 19th centuries, we added a few days to the human life expectancy. Now, we are at the intersection of biology and information science.

Today, we are adding about 120 days every year to the human life expectancy. With the full flowering of the biotechnology revolution, within 10 years, we will be adding more than a year to the human life expectancy every year.

So if we can hang in there for another 10 years, we may actually get to experience the full measure of the profound century ahead.

© 2003 KurzweilAI.net

In his detailed analysis “The Lives and Death of Moore’s Law,” Ilkka Tuomi concludes that the “semiconductor industry has not actually followed an exponential growth trend” and “price decreases have not followed an exponential trend.”¹

Tuomi’s conclusions are surprising, to say the least. If correct, I would have to conclude that the one-quarter MIPS computer costing several million dollars that I used at MIT in 1967 and the 1000 MIPS computer that I purchased recently for $2,000 never really existed. Tuomi has an explanation for all this. He writes that “the apparent explosive big bang in semiconductor technology is . . . an illusion.” name="_ednref2">²

If all of this is an illusion, it has been quite an effective one. The reality is that Tuomi’s conclusions defy common sense and clear observation. They are at odds with the historical data on the past, as well as all of the industry road maps for the future.

Tuomi’s approach is to set up a variety of straw men in the form of faulty interpretations of Moore’s Law and then proceed to show how certain data fail to match these incorrect interpretations. When presented with a “forest” of data representing a clear exponential trend, Tuomi often cites a tree with a missing branch (see my discussion below on Tuomi’s critique on my history of the computing trend over the past century). Tuomi supports his contrarian contentions with a list of often conflicting technical terminology and irrelevant historical anecdotes. There are a variety of misconceptions in Tuomi’s two papers, which I detail in this response essay.

Despite a plethora of misconstructions in the data that Tuomi presents, his analysis is nonetheless replete with exponential trends. Tuomi’s remarkable conclusion that the semiconductor industry has not followed an exponential trend is not consistent with his own analysis.

Tuomi repeatedly points out how the advances have not followed the oft-quoted 18-month doubling time of Moore’s Law. Tuomi is correct that the 18-month figure is a mischaracterization. Moore never said it, and it does not necessarily match the data (depending, of course, on what you’re measuring). Tuomi devotes a lot of his paper to showing how the 18-month figure is not correct, at least for certain measurements. He quotes a lot of people, such as R.X. Cringely in 1992, who have gotten it wrong³.

But the fact that many people state Moore’s Law incorrectly does not mean that there is not a correct way to state it, and it certainly does not follow that there has not been exponential growth.

Moore’s Law refers to the continual shrinking of the size of transistors on an integrated circuit, as well as other process and design improvements. This shrinking increases the number of transistors that can be placed on a chip, as well as their speed, resulting in dramatic exponential gains in the price-performance of electronics.

My law of acceleration returns is broader, and refers to ongoing exponential improvements in the price-performance and capacity of information technologies in general, of which Moore’s law is just one example. I provide further explanation and examples below.

Tuomi writes that

“During the four decades of validity often claimed for Moore’s Law the difference between one-year and three-year doubling time means about eight orders of magnitude. In other words, to get the same number of transistors, the slower growth path would require us to buy 100 million chips, instead of one.” href="#_edn4">⁴

I would certainly agree that there is an enormous difference in implications between a one-year and a three-year doubling time. However, there is no such variability in the data, unless one is trying to create confusion. Whether one gets a 36-month doubling time or a 12-month doubling time, or any other doubling time, depends entirely on what is being measured. If measuring something simple like two-dimensional feature size, then the doubling time (toward smaller features) is about 36 months.⁵

This has nothing to do, however, with cost-effectiveness or price-performance, which is what we really care about. The cost per transistor has fallen by half about every 1.6 years. If one takes into consideration all of the levels of improvement including speed, as I describe below, then the doubling time for price-performance is closer to one year.

Regardless of the doubling time, the trends are all exponential, not linear, both in the historical data, and in the roadmaps for the future. Tuomi’s point about the difference between 36 months and 12 months as a doubling time is based entirely on comparing apples to watermelons, or in this case, cell sizes to actual price-performance improvements.

Let’s examine these issues in more detail. I’ll start with my own four-plus decades of experience in this industry. Compare the MIT computer I mentioned above to my current notebook. As a student in 1967, I had access to a multi-million dollar IBM 7094 with 32K (36-bit) words of memory, and a quarter of a MIPS processor speed. I now use a $2,000 personal computer with a quarter billion bytes of RAM and about a thousand MIPS processor speed. The MIT computer was about a thousand times more expensive, so the comparison with regard to the cost per MIPS is a factor of about 4 million to one.

Click for larger version.

This ignores many other advantages of my contemporary computer. Ignoring these other significant factors of improvement, the contemporary computer provides MIPS of processing at a cost that is 2²² lower than the computer I used in 1967. That’s 22 doublings in 36 years, or about 19 months per doubling. If we factor in the increased value of the approximately 2,000 fold greater RAM memory, vast increases in disk storage, the more powerful instruction set of my circa 2003 computer, vast improvements in communication speeds, more powerful software, and other factors, the doubling time comes down even further.

Consider microprocessor history. The Intel 8080 had 5,000 transistors in 1974. The Pentium IV had 42 million transistors in 2000. That’s just about exactly 13 doublings in 26 years, which is a two-year doubling time. Keep in mind that this two-year doubling time takes into consideration only this single factor of the number of transistors. If we also factor in the fact that the smaller Pentium IV transistors operate many times faster and are organized with many layers of circuit innovation, then the overall price-performance improvement is greater than 2¹³. The graph (of number of transistors) of the intervening processors (such as 8086, 286, 386, 486, Pentium, Pentium 2, etc.) shows smooth exponential growth (R² = 0.9873).⁶

Click for larger version

We should also keep in mind that adding transistors to a microprocessor is not the sole or even the primary goal of semiconductor technology. At a certain point in the future, we will have the optimal complexity for a single processor. We will continue to want to improve price-performance, but not necessarily number of transistors in a single microprocessor. Thus the International Technology Roadmap for Semiconductors (ITRS) projects the number of transistors in a single microprocessor to double every 36 months through 2016, but also projects the cost of a single microprocessor to come down such that the cost per transistor in a microprocessor is coming down by half every 24 months. Even this figure ignores the speed improvement factor, which I discuss below.

Data from Dataquest and Intel shows that the average price of a transistor per year went from 1 dollar in 1968 to about 2 x 10^-7 dollars in 2002. That represents an improvement of 5 x 10⁶ (= approximately 2²²) in 34 years. This represents 22 doublings in 34 years, or about 1.6 years per doubling. Again, the trend has been very smooth in intervening years.⁷

Click for larger version

Keep in mind that unlike Gertrude Stein’s roses, it is not the case that a transistor is a transistor. Because transistors have been getting steadily smaller (at an exponential rate), they have been getting faster, so this factor brings down the doubling time for price-performance even further. And there are other levels of innovation that also improve price-performance.

The available data also supports exponential growth in volumes. The number of transistors shipped, according to Instat/MDR, went from 2 x 10⁹ in 1968 to just under 10¹⁸ in 2002, or an increase of about 4 x 10⁸ (= 2²⁹) in 34 years. This represents a doubling time of 1.1 years. name="_ednref8">⁸ Again, this ignores other levels of improvement.

Click for larger version

Now let’s factor in the speed improvement. Interestingly, Tuomi provides us with this exponential trend in his paper⁹:

Tuomi’s Figure 4: Desktop computer processor speed.
Source: Berndt et al., 2000, Table 1.

It is difficult to see the exponential trend in Tuomi’s linear chart, so I provide here our own logarithmic chart of microprocessor clock speed, which shows both the historical data and the ITRS road map name="_ednref10">¹⁰:

 
Click for larger version

With a speed improvement of approximately 10³ in 34 years (1968 to 2002), the cost per transistor cycle decreased by a factor of 5 x 10⁹ (= 2³²), resulting in a doubling time of just over 12 months.¹¹ Even this analysis takes into consideration only semiconductor density and process improvements, and does not take into consideration improvements at higher levels such as processor design (for example, pipelining, parallel instruction execution, and other innovations).

Click for larger version

The theoretical total number of transistor cycles in the world increased by a factor of 4 x 10¹¹ (= 2³⁹) in 34 years, resulting in a doubling time of only 10.4 months.

One could go on for many pages with such analyses, measuring many different dependent measures, and citing data from many sources, all of which show clear exponential growth. Tuomi’s myopic finding that there is no exponential growth in the semiconductor industry is notable, and I admire his tenacity in attempting to prove that the world of information technology is flat (i.e., linear).

One reason technology improves exponentially is that we seek to improve quantifiable measures by multiples rather than by linear increments. About a half century ago, Dr. An Wang and his engineers struggled to add an increment of a thousand bits to his (iron-core-based) RAM storage. Do engineers struggle to add a thousand bits to a memory design today? Or a million bits? Design goals today are more likely to be measured in billions of bits. Goals are always set in a multiplicative relation to the current standard.

Another reason that technology improves exponentially is that we use the (more) powerful tools of one generation of technology to create the next. Early computers were designed pen-on-paper and wired with individual wires and screwdrivers. Today, a chip or computer designer specifies formulae and high-level parameters in a high-level language, and many layers of intermediate design are automatically computed by powerful computer-assisted-design software systems.

For these and related reasons, we see exponential growth not only in memory and computational price-performance, but across the board in information-related technologies. For over two decades, I have been studying key measures of capacity and price-performance in a wide variety of such technologies. The data clearly shows exponential growth that goes far beyond Moore’s Law or computation. We see exponential growth in a broad variety of measures of the capacity and price-performance of information technologies. To provide just a few examples, consider the price-performance of magnetic-disk memory density, which is a phenomenon distinct from semiconductors¹²:

The Internet¹³:

Click for larger version

Brain scanning¹⁴:

Biological technologies such as DNA sequencing¹⁵:

Click for larger version

We also see exponential growth in varied measures of human knowledge, and even in the key feature sizes of technology, both electronic and mechanical.¹⁶ Tuomi himself adds to our extensive list of examples of exponential growth in information when he writes, “According to Price (1986), the number of scientific journals has doubled about every 15 years since 1750, the number of ‘important discoveries’ has doubled every 20 years, and the number of U.S. engineers about every 10 years.”¹⁷

I do agree, however, that for many applications, exponential growth of a capability (such as memory size or processor speed) does not necessarily translate into exponential growth in utility. For many functions, it requires exponential growth in capability to obtain linear gains in functionality. It requires, for example, exponential gains in computing to obtain linear gains in chess ratings. Similarly, we see linear gains in the accuracy of pattern recognition algorithms (for example, speech recognition) with exponential gains in processor speed and memory capacity. However, for inherently exponential problems, linear gains in functionality and performance are very powerful and sufficient to obtain profound benefits.

Tuomi writes that “Exponential growth . . . is very uncommon in real world [sic]. It usually ends when it starts to matter.”¹⁸ It is clear, however, that the information industry (in all of its manifestations) has indeed begun to matter. Moreover, Tuomi provides no basis to conclude that exponential growth in computing has ended or is about to end.

When we are unable to continue to shrink two-dimensional integrated circuits, we will build three-dimensional circuits. Note that this will not be the first paradigm shift in computing because Moore’s Law itself represented not the first but the fifth paradigm to provide exponential growth to computing. Prior to flat integrated circuits, we had electro-mechanical calculators, relay-based computing vacuum tubes, and then discrete transistors. Even though the semiconductor industry road map¹⁹ indicates that we have more than a dozen years left to obtain exponential growth through two-dimensional circuits, there has been enormous progress in recent years in developing early prototypes of three-dimensional circuits.

Based on our current understanding of the physics of computing, the inherent limits to exponential growth of computation and communication are extremely high (that is, the minimum matter and energy required to compute a bit or transmit a bit is extremely low).²⁰ One of the most promising is to create general-purpose electronics using nanotubes, which are hexagonal arrays of carbon atoms. This approach has shown considerable promise in experiments. When fully developed, nanotube-based circuitry has the potential to be many orders of magnitude more powerful than flat integrated circuits. Even nanotubes do not approach the fundamental limits of computing, based on our current understanding of the physics of computation.

In Tuomi’s review of the industry’s history, he notes that on several occasions the industry just happened to be “saved” by special circumstances, for example, the emergence of the calculator and memory chip markets in 1965-68. name="_ednref21">²¹ Tuomi assumes that the industry was just lucky that the invention of these two product categories came at the right time. He writes that in general, “semiconductor technology has evolved during the last four decades under very special economic conditions.”²²

But the introduction of new product categories made feasible by the greater price-performance of each new generation of semiconductor technology is inherently part of the process. More powerful chips, which have been occurring on a very predictable basis, lead to new product categories, which in turn lead to greater volumes.

The pace of this type of innovation has increased in recent years with the rapid introduction of new types of digital products. An inherent aspect of progress in information-based technologies is new paradigms on every level. Often, old problems are not solved directly, but rather are circumvented by the introduction of new paradigms, new applications, and new markets. But to Tuomi, the 50 years of exponential growth attributable to integrated chips (preceded by 50 years of exponential growth from pre-chip technologies) is all a temporary aberration.

It is important to point out that an evolutionary process – whether of technology or biology – is always a matter of special circumstances. However, there are always special circumstances. Whatever it was that hit the Earth that resulted in the demise of the dinosaurs was a special circumstance, one that had profound implications for all species at that time. But the progression of biological evolution was not dependent on that event just happening. In general, evolutionary events happen in “special” ecological niches that are inherently delicate and bounded by distinctive circumstances.

The “special circumstances” that Tuomi refers to in the semiconductor industry have kept Moore’s Law going for half a century, and counting. The acceleration of the price-performance of computation goes back at least a century. Special circumstances are part of the evolutionary process – not a reason to overlook its exponential progression. As Gilda Radner used to say on Saturday Night Live, “it’s always something” – meaning there is always something special about current circumstances.

Tuomi’s analysis is filled with strained analyses that stretch the data to make his contrarian points.

Consider his hand-drawn trend lines on the following graph²³:

Tuomi’s Figure 3: Number of transistors on Intel microprocessors

Tuomi has drawn his chart in a misleading way. For example, it suddenly jumps up at year 21, yet this improvement is not taken into consideration. There are only two outlier points, both around years 20 and 21 (the “x” below the right point of the middle line, and the left most point of the right most line). If one draws a trend line through all of the points, leaving out these two outliers (which, in any event, cancel each other out), one gets a relatively smooth exponential chart. See my previous chart on the number of transistors in Intel microprocessors. Note that both Tuomi’s and my chart leave out the issue of transistor speed and the effect of many design innovations.

Tuomi includes less powerful processors at various points in time that skew the curve. There are always less-powerful versions of processors offered for special markets. The appropriate microprocessor to include at each point in time is the one providing the optimal performance. It is also worth pointing out that the number of transistors in a microprocessor is not the most relevant variable to measure. We are more concerned with the functionality per unit cost.

As noted previously, at some point, there will be an optimal number of transistors to perform the functions of a single processor. At that point, we won’t be interested in increasing the number of transistors in a microprocessor, but we will continue to be interested in improving price-performance. Above, I provided a properly constructed chart on transistors in Intel processors.

If we measure what is really important (overall processor performance), we need to consider speed improvements, among other factors. Tuomi himself provides evidence of the exponential speed improvement in his figure 4 above. Taking speed as well as design innovations into account, we get a doubling time of about 1.8 years for overall processor performance. name="_ednref24">²⁴ This does not include the issue of word length, which has been increasing during this period. Including this factor would bring down the doubling time further.

Click for larger version

 Remarkably, Tuomi prints a similar chart and concludes that “As can be seen from Figure 5, MIPS ratings of Intel processor have not increased exponentially in time.”²⁵

Tuomi’s Figure 5: Processor performance in millions of instructions per second (MIPS) for Intel processors, 1971-1995.

His hand-drawn trend lines (on a logarithmic chart) show exponential growth, with the trend lines jumping up (meaning increased values) at two points. Each straight line on Tuomi’s logarithmic chart is an exponential, but as one goes across the chart (from left to right), each successive straight line (representing exponential growth during that time period) is at an even higher level.

How he can conclude that “this shows no exponential growth” is not explained.

Tuomi cites the following chart²⁶ to make a point about the life cycle of a particular generation of chip:

Tuomi’s Figure 1: Prices and Quantities of 16-kilobit DRAM chips. Source: Grimm, 1998

Tuomi makes the point that one can obtain misleading trends by taking price points at different times in the life cycles of different chips. But this criticism is not valid for any of the charts I have presented, nor those cited from other industry sources. The chart that I provided above for average cost per transistor is exactly that – the average cost for that year. In charts involving different types of chips, prices at the point of production are consistently used. There has been no attempt to compare one point in the life cycle of one chip to a different point in the life cycle of another chip.

However, let’s take a look at what happens if we examine the entire life cycle of multiple generations of semiconductor technology name="_ednref27">²⁷:

Click for larger version

In this logarithmic chart, we can see the life cycle of each generation, and the overall exponential trend in the improvement of price-performance remains clearly evident.

Tuomi spends a lot of time in both papers talking about the “hedonic” model for economic value of various features and “quality” improvements such as increased memory or increased speed. He writes:

 "…if a 100 MHz PC costs today 500 dollars more than a 60 MHz PC, we might assume that if a 100 MHz PC costs today as much as a 60 MHz PC a year ago, technical advance has been worth 500 dollars."²⁸

The hedonic model has little validity. Current software may only make sense for the mainstream specifications, so purchasers would not be willing to pay very much for more memory than they need for the applications they have or intend to use. However, at a later time, when the more sophisticated applications available require more memory, they would be willing to pay for this extra memory, and in fact would not want the computer if it didn’t provide the memory (or other capabilities) necessary to run these applications. Furthermore, certain variations in specifications may appeal only to small niche markets. All of these factors distort this hedonic model methodology.

However, despite these methodological concerns, Tuomi himself cites numerous examples of exponential growth in price-performance based on quality-adjusted prices. He writes:

“The classic study of quality corrected prices in computing is by Chow (1967), who analyzed mainframe rental prices in the 1960s. According to Chow, quality-adjusted prices fell at an average annual growth rate (AAGR) of about -21 percent during the 1960-1965 period. Cole et al. (1986) studied the price declines of different computer components and found that over the 1972-1984 period, the AAGR for computer processors was -19.2 percent using the hedonic prices. . . .Cartwright (1986), in turn, reported an AAGR of -13.8 percent from 1972 to 1984. According to Gordon (1989), quality adjusted mainframe prices fell 22 percent annually from 1951 to 1984.”²⁹

These are all exponential improvements cited by Tuomi. It should also be noted that during this time frame, mainframes started out as the best value, but were no longer remotely close to the best value by the end of the time period. Mainframes maintained artificially high prices to locked-in customers (who eventually escaped the lock-in to minicomputers and then personal computers). Nonetheless, even just considering mainframes, and an insufficient “quality adjustment” methodology, it still shows exponential improvement in price-performance. Tuomi continues:

“Triplett (1989) summarized earlier hedonic studies on mainframe computer prices and reported a ‘best-practice’ quality-adjusted price decline of -27 percent over the 1953-1972 time period. Gordon (1990) then extended his earlier analysis to personal computers and reported 30 percent annual declines from 1981 to 1987. Berndt and Griliches (1990) collected a large sample of data on personal computers and reported 28 percent annual decreases from 1982 to 1988.”

Note that a 30% decline each year = 51% decline (i.e., doubling of price-performance) in 2 years. 28% each year = 48.2% decline in 2 years, all examples of exponential growth.

Tuomi provides even further evidence of exponential improvement:

“Grimm has also calculated price indexes for microprocessors using the same methodology. For microprocessors the decline in price indexes has been considerably faster than for memory chips. During the 1985-1996 period, quality adjusted microprocessor prices dropped at an average annual rate of 35 percent.”

Note that an annual improvement of 35 percent is a doubling time (of price-performance) of less than 24 months. Also, the quality adjustment methodology, as I noted earlier, understates the values for the reasons I cited above. Tuomi’s conclusion is again to state the “the price decreases have not followed an exponential trend.” This conclusion is not consistent with the evidence that Tuomi provides in his own papers.

Tuomi consistently miscalculates these doubling times. For example, he writes:

“on average prices per unit of memory have declined 32 percent per year during the 1978 – 2000 period. This corresponds to a 30 month doubling time.”³⁰

If prices drop 32 percent in one year, a price of $1 would be $0.68 after one year and $0.46 after two years (i.e., falling to less than half in 24 months). A 32 percent decline in price per year corresponds to a doubling time of under 22 months, not 30 months. He makes this mistake repeatedly, so Tuomi’s stated doubling times cannot be relied upon. Regardless of whether the doubling time is calculated correctly or not, this is one of many pieces of evidence cited by Tuomi himself of exponential improvement.

In Tuomi’s paper on my law of accelerating returns, “Kurzweil, Moore, and Accelerating Change³¹,” he describes my thesis as leading “to an apparently infinite speed of change.” I do want to clarify that exponential growth, even double exponential growth, does not lead to infinite rates of change. It nonetheless will lead to greatly transforming rates of change.

Repeatedly, Tuomi cites Moore’s change in description of what has become known as Moore’s Law from his 1965 paper, in which he cited a doubling time of transistors per dollar of one year, to his revised estimate in 1975 of two years. Tuomi describes this as Moore noting “that the speed of technical change was slowing down.” This is a mischaracterization. Moore simply corrected his earlier estimate to a more accurate one (one that has, incidentally, been conservative). He was not saying that it had been one year and was now becoming two years. He was saying that his earlier one-year estimate had been incorrect, and that it had been two years and would remain so.

As I mentioned earlier, when presented with the “forest” of a trend, Tuomi often responds with a comment about a missing branch of a tree. For example, in response to my chart on a century of double exponential growth in computing:

Tuomi responds that “Ceruzzi (1998:71,74) gives $1.6 to about $2 million as the price of a full IBM 7090 installation. Kurzweil uses $3 million. Kurzweil has also moved Babbage’s Analytical Engine about half a century in time, with the explanation that it probably could have been built in 1900. Other authors have argued the machine could have been built using available manufacturing capabilities.”

I would argue with both of Tuomi’s assertions, but even if one removes these two points, it hardly makes a dent in this forest of a trend.

My team of researchers has been adding additional points to this trend that corroborate this double-exponential trend. Note that a straight line on a logarithmic graph represents exponential growth, and that the exponential trend here is itself exponential – it took three years to double the price-performance of computing at the beginning of the twentieth century, two years in the middle, and it is now doubling approximately every one year.

Hans Moravec’s analysis³², which includes additional points from my own chart, also shows the same double-exponential trend:

Click for larger version

Tuomi also writes that “Historical data also reveals that the early computers rarely were working at their theoretical speeds.” This point only strengthens the observation that later computers were more powerful.

Perhaps the most convoluted argument that Tuomi presents is his discussion of the alleged lack of increased resources to the semiconductor industry. He writes “Technological developments in the semiconductor industry are generally viewed as the drivers of progress in computing. According to Kurzweil’s hypothesis, one would expect semiconductor industry to enjoy increasing positive returns that would speed up technical developments in the industry. Indeed, in Kurzweil’s model, the rapid technical developments would be caused by the increase in resources available for developers.”

Remarkably, Tuomi states that there have been no “accelerating increases in its resources.” Yet he cites a report from the World Semiconductor Trade Statistics (WSTS) that “the average year-to-year change in semiconductor shipment value during the 1958 – 2002 period is 18 percent.” This, of course, is exponential growth, so how does Tuomi justify his conclusion that I am incorrect in my assessment that the computer and information industries (which includes the semiconductor industry) have benefited from increased resources?

Tuomi provides this chart name="_ednref33">³³:

Click for larger version

He takes the growth rate of the semiconductor industry and subtracts the growth rate of the U.S. GDP. This is a dubious proposition because the growth rate of the GDP is fueled specifically by technological innovation, particularly in information technologies. So we are subtracting from the growth rate of the semiconductor industry the growth rate of the economy. Yet it is the semiconductor and information technology industries that are primary contributors to the growth of the economy.

The result is nonetheless positive, which Tuomi calculates as averaging 10.8 percent annual growth (over the GDP growth). However, he then takes the derivative (slope) of this curve and notes that it is negative.

From this, he concludes that the semiconductor industry has not enjoyed increased resources. Yet, the industry has grown on average by 18 percent according to Tuomi’s own analysis, and this growth rate exceeds that of the overall economy to which it contributes. The appropriate conclusion of the negative slope in this complex graph is that the growth rate of the economy is catching up to the growth rate of the semiconductor industry. The reason for that is that information technology in general is becoming increasingly pervasive and influential. Information technology itself has gone from 4.2% of the GDP in 1977 to 8.2% in 1998, with the growth rate recently accelerating.³⁴

Click for larger version

Moreover, information technology is increasingly influential on all aspects of the economy. Even those industries that are not explicitly “information technology” are nonetheless deeply influenced by it. We are rapidly moving towards an era in which the dominant portion of the value of most products is represented by their information content. Thus the overall economy is slowly catching up to the rapid growth rates of information-related industries such as the semiconductor industry, specifically because of the effect of these industries.

The bottom line is that resources have increased exponentially in the information technology industries, including the semiconductor industry, and this rate of growth is not slowing down. Moreover, this is the less important part of the story. The more important issue is not merely the increase in dollars, but the very rapid exponential growth of what each dollar buys.

Both hardware and software have increased enormously in power. Today, a semiconductor engineer sits at a powerful computer assisted design station and writes chip specifications in a high-level language. Many layers of intermediate design, up to and including actual chip layouts, are then computed automatically. Compare that to early semiconductor designers who actually blocked out each cell with ink on paper. Or compare that to the early computer designers who wrote out their designs by pen and then built the computers with individual components, wires and screwdrivers.

Tuomi cites the following graph to argue that computers and software investments have not been growing exponentially³⁵:

Click for larger version

First of all, the data on figure 2 above matches a slow exponential more closely than a straight line, which does not match very well. The data starts under the line and ends up over the line.

More importantly, this graph is not plotting the actual computer and software investments, but rather expressing them as a percentage of private fixed investment, which is itself growing exponentially, reflecting the growth of the IT sector as a percentage of the GDP.

Tuomi writes that “. . .an exponential trend defines a technical trajectory that is independent of any external factors. Moore’s Law, in its original form, is basically such a claim. Its exponential form implies that. . .developments in integrated circuits are effectively independent of economic, organizational, social, or any other forces.”³⁶

Tuomi here is misunderstanding the nature of exponential growth in information technology. The development of semiconductors and related computer technologies is not taking place in a vacuum. External factors are very much involved. This is a classical evolutionary process taking place in a competitive environment. If there were no economic value to increased capacities, they would not be developed. Greater capacities and price-performance lead to new capabilities and applications, which in turn result in increased demand. Moreover, the more powerful tools from one generation of technology create the next more powerful generation.

Tuomi’s argument becomes particularly strained when he discusses the purported benefits of analog computing. He writes, “Many mathematical problems that require an infinite number of algorithmic computations can be solved by intelligent humans and by non-algorithmic calculating machines. A classical technical method of doing this has been to use analog computers. Indeed, in many classes of mathematical problems the computational power of an analog computer is infinitely greater than the computation power of conventional digital computers.”

The above statements are illogical. “Humans and non-algorithmic calculating machines” are clearly not able to solve mathematical problems that “require an infinite number of algorithmic computations.” Moreover, the use of an analog computer certainly does not allow one to accomplish this.

It is also completely unjustified to say that an analog computer “is infinitely greater than the computation power of conventional digital computers.” This conclusion derives from the naïve notion that digital computers can only deal with “on” and “off,” and not with shades of gray in between. By using floating point numbers, digital computers can represent fractional values to any desired precision. In fact, digital computers are much more precise than analog computers in doing this.

Prior to World War II, analog computers were popular, and digital computers required the “digital” modifier to distinguish them. But analog computers are so unpopular today that we no longer are required to use the word “digital” before “computer.” Although an analog computer can represent a fractional value, the accuracy of analog components is relatively low and unpredictable. With a digital computer’s floating point numbers, the accuracy is known from the number of bits in the floating point representation.

If desired, one could use thousands of bits in each floating point number (the algorithms for doing this are well understood), which would provide accuracy far exceeding any conceivable analog process. Of course, for most practical applications, 32-bit or 64-bit floating point numbers are quite sufficient, and exceed the accuracy of existing analog computers.

There is an engineering argument that for some applications, such as precisely modeling the nonlinear aspects of human neurons, using transistors in their native analog mode is more efficient. California Institute of Technology Professor Carver Mead has pioneered this approach to doing neuromorphic modeling. There are counter-arguments to this: such analog chips are difficult to design and are not programmable. But regardless of how one settles this particular design issue, Tuomi’s statements about analog computers appear to have no basis.

In conclusion, every time I open the morning paper (which I now usually read online) and look at the specifications and prices in the ads for the latest digital phones, digital cameras, portable electronic games, MP3 players, digital TVs, notebooks, tablets, and pocket computers, among an increasingly diverse set of new product categories, I am reminded of the obvious exponential growth in price-performance that Tuomi insists on denying.

Further Detailed Response

I provide here a detailed response to specific assertions in Tuomi’s thesis. I encourage the reader to read Tuomi’s papers in full to obtain the full context of Tuomi’s statements.

Responses to Ilkka Tuomi’s

“The Lives and Death of Moore’s Law”³⁷

Tuomi: Technical considerations of optimal chip manufacturing costs have been expanded to processor performance, economics of computing, and social development. It is therefore useful to review the various interpretations of Moore’s Law and empirical evidence that could support them.

Such an analysis reveals that semiconductor technology has evolved during the last four decades under very special economic conditions.

Kurzweil: Every paradigm has special conditions. Evolutionary change in either biology or technology always derives from finely tuned conditions and operates at the edge of survival (of a species or a product line).

Tuomi: Several observers have . . . . speculated about the possibility of "the end of Moore’s Law." Often these speculations have concluded by noting that Moore’s Law will probably be valid for at least "a few more generations of technology," or about a decade. An important example is the International Technology Roadmap for Semiconductors (ITRS), which now extends to 2016. This roadmap is generated by a global group of experts and represents their consensus. . . . it notes that within the next 10-15 years "most of the known technological capabilities will approach or have reached their limits."

Kurzweil: These well-publicized limits of Moore’s Law pertain to flat two-dimensional circuits only. Sometime during the second decade of this century, key feature sizes will be a small number of atoms in width, and we won’t be able to shrink them further. At these scales, these circuits are vastly more efficient than the cumbersome electrochemical signaling used in mammalian interneuronal connections, but only in 2-D. We live in a three-dimensional world, and it is clear that we will move into the third dimension. The research accomplishments underlying three-dimensional molecular computing are escalating rapidly, and are ahead of comparable points in history prior to other paradigm shifts. The entire paradigm of Moore’s Law (flat integrated circuits) was not the first but the fifth paradigm to provide exponential growth to computing – each time it became clear that a paradigm would end, research would intensify on the next paradigm.

Tuomi: Speculations on the extended lifetime of Moore’s Law are therefore often centered on quantum computing, bio-computing, DNA computers, and other theoretically possible information processing mechanisms.

Kurzweil: This is the wrong list. Quantum computing, bio-computing and DNA computers, if perfected, would be special-purpose devices. Although prodigious on certain classes of problems, they are not suitable for general-purpose computing. Quantum computing can in theory try every combination of qubit value simultaneously. So for the class of problems in which a solution can be easily tested, such as finding the factors of large numbers to break encryption codes, it is a great technology. But it provides no speed improvement for most conventional computing problems. The primary focus for the sixth paradigm of computing (after electromechanical, relay-based, vacuum-tubes, discrete transistors, and integrated circuits) is three-dimensional molecular computing. I have always favored nanotube-based designs, and these in fact have obtained the most advances in recent research.

Tuomi: The fundamental assumption was that the total manufacturing costs are practically independent of the complexity of the chips. For this to be the case, the engineering and design costs had to be so small that they could be neglected. Indeed, Moore noted that the costs of integrated circuits were totally dominated by packaging costs. In other words, the costs of silicon was fixed and knowledge was free and the only limiting factor in manufacturing costs was the rapidly increasing waste created by deteriorating yields. Moore’s discussion did not explicitly take into account investment costs.

Kurzweil: It is true that engineering of each new generation of chips has become more complex, but there have been countervailing trends that more than offset this. First, CAD sophistication (and the computers to run CAD software on) has substantially increased, allowing increasingly sophisticated chips to be developed in comparable time frames. Also, the number of chips produced in each generation has increased at an exponential rate, allowing development costs to be amortized over an increasingly large volume.

Tuomi: From an economic point of view, Moore’s Law was a rather fascinating law. It implied that the development of integrated circuits was completely determined by manufacturing costs. Moore’s Law, therefore, defines a completely new economy. In this economy, demand is infinite.

Kurzweil: Infinite? This is clearly an oversimplification. Nonetheless, demand has kept pace with continued exponential gains in memory size and computer capabilities, as evidenced by the exponential growth of the semiconductor industry (which Tuomi himself describes as being 18 percent per year) and the overall information technology industry. Below, Tuomi cites some examples of this as accidents that just happened to save the industry at various times. The reality is that the opening of new markets is inherently part of the process. There are many applications today that are waiting for the communication speeds, memory, and computational capacities of future years (such as ubiquitous full-immersion, visual-auditory virtual reality for business and personal encounters, augmented reality and telepresence applications, and many others).

Tuomi: The essence of Moore’s argument had been that it was becoming possible to manufacture increasingly complex integrated circuits and that the price per component was dropping radically. The limiting factor would be efficient amortization of design investments. This could be done in two basic ways: either by making high volumes of single function or by making designs that could be used for many different chips. The first path led to Intel’s focus on memory chips and the latter, a couple of years later, to microprocessors.

Kurzweil: Also by increases in chip volume created through new applications that result from the greater capabilities of each new generation of chip technology.

Tuomi: In his presentation, Moore analyzed the different causes of the exponential development. First, the physical size of the chips had been growing approximately exponentially. In 1975, chip sizes of the most complex chips were about 20 times larger than in 1959. Second, the miniaturization of component dimensions had also evolved at roughly exponential pace. This miniaturization had led to about 32-fold increase in component density in 15 years. The combination of increased chip size and component miniaturization therefore seemed to explain about 640-fold increase in the number of components. According to Moore’s prediction, however, in 1975 chips were supposed to contain more than 640 components. The remaining 100-fold increase Moore associated with "circuit and device cleverness". New technology, such as better isolation of components and innovations in circuit layouts had made it possible to pack more components closer to each other (Moore, 1975).

Kurzweil: Miniaturization of component dimensions results not only in more components per unit size, but also in faster circuits, since the electrons have less distance to travel. In addition, there is innovation on every level in both hardware and software. Beyond just packing more and faster circuitry onto each square millimeter, there have been many innovations in processor design, such as pipelining, register caches, parallel processing, more powerful instruction sets, etc.

Tuomi: Moore revised his original growth rate estimate and proposed that by the end of the decade, the number of components on the most complex chips would double about every two years. Soon after, this prediction became known as "Moore’s Law." According to Moore, the name was coined by Carver Mead (Yang, 2000).

Kurzweil: This was a revision, not an observation of the data changing.

Tuomi: In 1975, Moore implicitly changed the meaning of Moore’s Law. As he had done ten years before, he was still counting the number of components on semiconductor chips. Instead of focusing on optimal cost circuits, however, he now mapped the evolution of maximum complexity of existing chips. Indeed, in an article written a few years later, the famous growth curve is explicitly called "Moore’s Law limit" (Moore, 1979). At that point the growth estimate is presented as the maximum complexity achievable by technology. In Moore’s 1979 paper, which shows a picture with component counts of Intel chips manufactured in 1977 and 1978, most chips fall one, two, or even three orders of magnitude below this limit.

Kurzweil: Although Moore showed a chart like this, plotting the maximum complexity of chip technology at different points in time is not an appropriate way to measure performance. One has to measure performance on the most cost- effective implementation of computing technology at each point in time. So, for example, one would not just measure mainframe performance, because after minicomputers became established, mainframes were not the most cost-effective implementation. The same thing happened to minicomputers when personal computers became established. These older markets only persisted because customers were locked into legacy applications, but these were not the most cost-effective platforms.

Tuomi: . . . .in 1975.. . . Intel introduced the 16-kilobit CCD, the Intel 2416. In the same year Intel also introduced its 2116-chip, a 16-kilobit dynamic random-access memory (DRAM) chip. Such a chip would have contained somewhat over 16,384 transistors, including some control circuitry, and about 16,384 capacitors. Since the mid-1970s, complexity has been counted based on the number of transistors. Moore’s earlier calculations, however, were based on the total number of components.

Kurzweil: This is typical of Tuomi nit-picking. The reality is that regardless of whether one looks at number of components or number of transistors, there has been clear exponential growth.

Tuomi: Moore presented a new exponential growth curve in his 1979 paper. According to it, the man-hours per month required for integrated circuit production was also growing exponentially. Moore went on to note:

"If we assume that the cost in man-hours per month is inflating at 10 per cent per year (a conservative figure considering the need for increased computer support, software, etc.), then the costs double every two years … This cost can be contrasted with manufacturing costs, which are largely independent of device complexity. Whereas manufacturing costs were once dominant and exceeded those of design, the situation is now reversing, with design costs becoming dominant".

Kurzweil: One has to take into consideration the exponential growth in volumes that were also taking place.

Tuomi: Moore also noted that the problems that slowed down the growth of semiconductor complexity in the 1965-1968 period had not been solved. Engineers were still unable to design and define products that would have used silicon efficiently. Instead, the industry was saved by the invention of two product categories where these problems could be avoided. . . .The calculator was an important product because it was a relatively simple system. Merely connecting four integrated circuits (that had about 40 pins) created a calculator. The interconnection problem, therefore, was tractable. As calculators were produced in high volumes, the design costs could be justified. Memory chips, in turn, were easy to design and universal in their functionality, and therefore also high volume products with low design costs.

Kurzweil: Tuomi assumes that the industry just happened to be “saved” by the lucky invention of these two product categories. But the introduction of new product categories made feasible by the greater price-performance of each new generation of semiconductor technology is inherently part of the process. More powerful chips, which have been occurring on a regular basis, lead to new product categories, which in turn lead to greater volumes. The pace of this type of innovation has increased in recent years with the rapid introduction of new types of digital products. In addition, another inherent aspect of progress in information-based technologies is new paradigms on every level. Old problems are often not directly solved – they are circumvented by introducing new paradigms, new applications, and new markets.

Tuomi: Moore himself has noted:

"I never said 18 months. I said one year, and then two years … Moore’s Law has been the name given to everything that changes exponentially. I saw, if Gore invented the Internet, I invented the exponential" (Yang, 2000). The historically inaccurate 18 months doubling time has been extremely widely used. It is possible even to find fictive quotes of Moore’s 1975 presentation saying: "The number of transistors per chip will double every 18 months."

Kurzweil: Tuomi is correct that the 18 month figure is incorrect (for most measures). Moore never said it, and it does not match the data. Tuomi continues to beat this dead horse repeatedly in the rest of this paper.

Tuomi: As noted above, Moore never claimed that the number of components would double every 18 months. The first version, the doubling of components on a chip every year, would mean that the number of components would increase 1024-fold per decade. The second version, doubling every two years, would translate into a much more modest increase of 32 per decade. In fact, the International Technology Roadmap for Semiconductors (ITRS, 2001) uses as the basis of its near-term microprocessor forecasts three-year doubling time. A three-year doubling time means that the number of transistors on a chip increases about nine-fold in a decade.

Kurzweil: The ITRS roadmap shows doubling of the number of bits per DRAM memory chip every two years: 1 Gb in 2003, 2Gb in 2005, and so on up to 64Gb chips in 2015-2016. That represents a 24-month doubling time. If we factor in additional improvements, including faster switching time and anticipated lower chip costs, the doubling time in price-performance will be less than 24 months.

If one looks only at the reduction in feature size in the ITRS roadmap for microprocessors, one gets a doubling of capacity per mm² in 36 months, but this is consistent with the rate of reduction of feature size going back to 1965. Despite this, the doubling time for the number of transistors per microprocessor has been 24 months, as I cited earlier. The cost per transistor has been coming down by half every 19 months. And when we factor in the increases in speed, the cost per transistor cycle has been coming down by half every 13 months.

It is also important to keep in mind that whereas increasing the number of bits in a memory chip increases its utility without limit, there is a limit to the number of transistors that are desirable in a microprocessor. At a certain level of complexity, we would rather concentrate on reducing the cost per microprocessor and using multiple processors than to continue adding complexity to a single processor. The same consideration does not apply to DRAM. When all of these factors are considered, the doubling time for price-performance for microprocessors in the ITRS roadmap is less than 24 months. Intel’s own roadmap is somewhat more aggressive than ITRS.

Tuomi: Over several decades the differences obviously increase dramatically. During the four decades of validity often claimed for Moore’s Law the difference between one-year and three-year doubling time means about eight orders of magnitude. In other words, to get the same number of transistors, the slower growth path would require us to buy 100 million chips, instead of one. So, although a few months more or less in the doubling rate might not appear to be a big deal, actually it is.

Kurzweil: As I pointed at the beginning of my response, there is no confusion between one-year and three-year doubling times. The trends have been very consistent and both the ITRS and Intel road maps project the same rate of exponential growth out through 2016. Whether one gets a 12 month doubling time or a 36 month doubling time depends on what is being measured. If one looks at a single issue such as line width, one gets longer doubling times. If, however, one considers the multiple levels in which innovation takes place, the doubling times are closer to 12 months.

Tuomi: As specific chip types usually have a long lifetime during which the costs and other parameters of the chip change, the ITRS roadmap differentiates four main life-cycle phases. The first is the year of demonstration. This is the year when the feasibility of a specific characteristic, for example the number of transistors on a single chip, is first demonstrated. The second phase, market introduction, usually two or three years later, is defined to occur when the leading manufacturer ships small quantities of engineering samples. The third phase, production, is defined to occur when the leading manufacturer starts shipping the chip in volume quantities and a second manufacturer follows within three months. The lowest cost phase emerges when the production processes have been optimized and competition does its work. For example, the 1-gigabit DRAM was demonstrated in 1997, introduced in 1999, and is expected to be in volume production in 2003. Similarly, the Intel Itanium processor was announced in 1994, was originally planned to be on market in late 1997, but was delayed and became commercially available in 2001. Market researchers currently project that Itanium will garner less than 10 per cent of the market for server computing in 2007 (Markoff and Lohr, 2002).

Kurzweil: Valid methodologies consider the most cost-effective form of memory or processor at each point in time. Obviously one can obtain invalid results by failing to do this. Tuomi makes this mistake later in this essay by concentrating on mainframe trends way past the point that mainframes represented the most cost-effective method of computing. In addition, graphs of processor speeds and density use the date of production so we can compare consistent points in development of each device.

Tuomi: Using this data we can fit an exponential growth curve and see how good the fit is. . . . . According to this simple fit, there were some five million transistors missing from Pentium II in 1997, representing about 70 per cent, and some 7.5 million, or some 18 per cent, too many in the Pentium 4 in 2000. The estimate for Pentium 4 comes relatively close to its actual value as the data, indeed, has been used to fit the exponential, and the final point is heavily weighted. The error would be greater if we were to use previous points to fit the curve and then predict the number of transistors of Pentium 4.

Kurzweil: Tuomi concentrates here on variances of the actual curve (which is clearly exponential and has about a 24 month doubling time) to a “predicted value.” This is irrelevant. It is true that many observers quote an 18 month doubling time for price-performance. 18 months happens to be approximately correct for the price per transistor (we get 19 months), but not accurate for the number of transistors per microprocessor. There seems to be little point in repeatedly making this point. Tuomi is clearly not demonstrating a lack of exponential growth.

Tuomi: One problem . . . . is that the clock speed is not directly related to the amount of information processed. For example, the original Intel 8088 PC microprocessor took on average 12 clock cycles to execute a single instruction. Modern microprocessors can execute three or more instructions per clock cycle.

Kurzweil: The improvements that Tuomi cites only serve to accelerate improvement further. One could also add increases in word and instruction sizes (from 4 bit to 8 to 16 to 32 to 64 to 128) and other improvements.

Tuomi: With the understanding that the clock frequency does not really measure processor performance, one may, however, check how clock frequency has evolved during the lifetime of microprocessors. Information on average processor clock speeds has been collected by Berndt et al. (2000). This is shown in Figure 4 (see above) on processor speed. The data covers distinctly identified personal computers that have been marketed and sold in the U.S. Processor speeds for mobile computers are excluded from Figure 4.

 Figure 4 has several interesting characteristics. First, it should be noted that the data do not directly represent advances in microprocessor technology. It is based on computers that have been marketed to end customers. In that sense it does reflect changes in the actually available processing power. As can be seen from the Figure, until the end of 1980s the increase in the reported processor speed was quite modest. During the first decade, processor speed grew about four-fold and between 1986-1995 somewhat less than 10-fold. In about 1994, the clock speed started to grow much more rapidly.

Kurzweil: Tuomi’s chart 4 (see page 4 above) on processor speed is an exponential (and represents only one factor contributing to processor price-performance).

Tuomi: By 1985, then, demand had started to be less than infinite and the semiconductor industry was not endogenously driven by technology. In 1982, the increase in MIPS ratings stopped for about three and half years instead of following an exponential trend.

Kurzweil: This is not true if one looks at Tuomi’s own chart (see Tuomi’s figure 5 earlier in this essay). Although he does not place any points on the chart during 1982-1985, the points starting in 1985 are at an even higher level than would be expected by extrapolating the 1970-1982 trend line to 1985.

Tuomi: During the last decades, computer clock frequency and the number of instructions per second have become very inaccurate indicators of processor power. Since the Intel 80286 processors shipped in 1982, microprocessors have utilized parallel processing in many alternative forms. By parallelism, more operations can be accomplished within a time unit. For example, the processor can be loaded with instructions that contain several operations, it can have several execution units that process multiple instructions within one cycle, or the processing of operations can be started before the previous operations have finished. All these forms of parallelism have commonly been used since the mid-1980s.

Kurzweil: This is one of many innovations.

Tuomi: Moreover, since the 1990s, processor architectures have increasingly relied on program compilers that detect and optimize parallelism in the source code programs. Indeed, the innovations in compiler technology have been a main driver in processing power improvements.

Kurzweil: This is an unjustified oversimplification. It is only one of many factors.

Tuomi: Computing power is rarely determined by the capabilities of microprocessors. Usually, the microprocessor is connected to external memory and input and output devices with links that are an order of magnitude slower than connections within the chip.

Kurzweil: The capabilities of microprocessors are certainly important as one driving force. Communication busses have always been about an order of magnitude slower than communication within a chip. However, all of the various systems in a computer, such as the hard disk, the communication busses, and other devices, have benefited from exponential improvements in capability.

Tuomi: The attempt to develop measurement systems for computer processing power have made it clear that the definition of computing power depends on the tasks for which the computer is used. Therefore there is no well-defined criterion or data for arguing that computer power would have increased exponentially. On the contrary, it has frequently been argued that most of the increase in computer capabilities has been consumed by software. This is often formulated as Wirths law: "Software gets slower faster than hardware gets faster".

Kurzweil: Try going back and using old software to do common contemporary tasks. We quickly get used to, and rely on, new features and capabilities.

Tuomi: Clearly, there have been huge qualitative changes in desktop computers during this time. The problem, then, is how to take them into account.

One approach is to create so-called matched-model price indexes. It is possible to measure price changes of a given computer type across several years and deduce from this the actual price change. So, instead of looking yearly changes in the list price of desktop computers, .. . we can look how the price for a given PC configuration has changed.

Kurzweil: This is not appropriate because a given model of computer does not necessarily remain the best value model.

Tuomi: A complete personal computer contains several different types of chips and other components, such as hard disk drives, CR-ROM drives, a keyboard, and a displays. Price changes in PC, therefore, reflect quality changes in several different types of technologies. The hedonic estimation models, however, tend to break down when new components and functionality are added to computers. When notebook computers started to become important towards the end of the 1980s, the different existing technical characteristics became revalued. For example, more memory perhaps meant progress for desktop users but for notebook users it implied shorter battery lifetime or the need to stay close to the power plug when using the computer.

Kurzweil: Battery life for a notebook computer has not decreased. Fuel cells are on the way, with dramatically improved battery (fuel cell) life. NEC is introducing fuel-cell-powered notebooks in 2004.

Tuomi: . . . .recent growth may reflect increase in cost-adjusted performance. For example, architectural changes in microprocessors during the second half of the 1990s moved much of the external and expensive cache memory onto the processor chip. On the other hand, the rapid improvements in average PCs sold might also reflect, for example, availability of consumer credit and funding for information technology intensive firms.

Kurzweil: This reflects the increase in the number of chips that have been produced, which represents another exponential trend.

Tuomi: Price changes, of course, reflect market competition, imbalances in supply and demand, technical change, and new markets that open as new uses are found for technology. These have very little to do with the original formulations of Moore’s Law. Yet, one may ask whether the current economic evidence supports the claim the cost of computing halves every 18 months.

Kurzweil: Tuomi is really stuck on beating this dead horse. The doubling time (of price-performance or other measurement) depends entirely on what is being measured. Some in fact are 18 months, others range from under 12 months to 36 months. The 36 month categories, however, do not reflect price-performance.

Tuomi: The change in the average values as well as the yearly fluctuations show that the price decreases have not followed an exponential trend. The great differences between the results of the different modeling exercises show that we do not know how quality-adjusted computer prices should be measured or how they have changed during the last decades.

Kurzweil: Tuomi’s statement, “the price decreases have not followed an exponential trend,” is a remarkable conclusion. His own paper provides myriads of exponential trends. This conclusion defies common sense and ignores reams of data that can be looked at in many different ways, all of which show exponential trends in price-performance, and other measures of capability.

Tuomi: In the previous sections we have reviewed the original formulations of Moore’s Law and its revisions. We found that Moore changed his interpretations of Moore’s Law during the 1960s and 1970s, and that its subsequent extensions have added qualitatively new and important aspects to it. Whereas the original formulations of Moore’s Law focused on counting components on integrated circuits, its extensions made claims of exponential increase in processing power and exponentially dropping quality-adjusted prices of computing. We reviewed the available empirical evidence for these different versions of Moore’s Law and found that they have little empirical support. Semiconductor technology has not developed according to Moore’s Law. The claims that future developments in semiconductors, computer technology, or information processing would be determined by the continuation of Moore’s Law are, therefore, obviously invalid.

Kurzweil: The above statements only make sense if one interprets Moore’s Law as strictly being the 18 month statement. Tuomi is correct that the 18 month version that is often quoted is not correct – but to imply that there is no exponential trend in computing and semiconductors has no validity.

Tuomi: Here, of course, the industry dynamics play an important role. For example, computers require software. One of the important drivers for buying increasingly powerful computing equipment has been that new versions of operating system and application software have typically demanded more processing power. Although it seems clear that today personal computers are much more functional than twenty years ago, it is not clear how much more functional they are.

Kurzweil: For many applications, linear improvement requires exponential gains. For example, we need exponential gains in computing power to get linear gains in chess scores. This assumes we hold the software constant. Recent progress in the software of terminal-leaf evaluation in the minimax algorithm shows that improvements can be gained from software alone. However, linear progress in what are inherently exponential problems is very powerful.

Tuomi: The regular doubling and exponential growth that underlies the different versions of Moore’s Law implies a very unique claim. It fundamentally says that the described phenomenon grows purely based on its internal characteristics. Exponential growth means that after the growth speed is set, the future unfolds based on nothing but the current state of the system. Contrary to what some commentators of Moore’s Law have claimed, exponential growth is not uncommon. When we put money on a fixed interest rate account, and reinvest the returns, it will grow exponentially. In this sense, a bank account is a prototype of self-determined endogenous growth. Exponential growth, however, is very uncommon in real world. It usually ends when it starts to matter.

Kurzweil: Tuomi provides no basis to conclude that exponential growth in computing has ended or is about to end. As mentioned earlier, the inherent limits to exponential growth of computation and communication are extremely high (that is, the minimum matter and energy required to compute a bit or transmit a bit is extremely low).

Tuomi: During its history, the semiconductor industry has several times hit the speed limit. First it was bailed out by the digital clock and calculator industry, then by mini and mainframe computer industry. In the mid-1980s, just when no one seemed to be able to make a profit, the IBM PC and Microsoft saved the day.

Kurzweil: Bailed out? This is the inherent process of innovation powering these exponential trends. The number of such “saves” (in Tuomi’s terminology) is increasing. We recently have cell phones, digital cameras, PDAs, portable game machines, MP3 players, pocket computers, and many other new categories.

Tuomi: It is . . . . no surprise that semiconductor industry has not actually followed an exponential growth trend.

Kurzweil: This is a remarkable conclusion that does not follow even from his own reasoning. His own paper is filled with exponential trends, even if most are not 18 month doubling times.

Tuomi: As the size and importance of computer and information processing technologies now is becoming more than a couple of percents of national economies, it can be predicted that the endogenous growth in this industry cluster is about to end. The imbalance between supply and demand shifts and the social basis of demand makes itself increasingly visible. The open source movement, for example, effectively disconnects the economics of operating systems from the economics of semiconductor manufacturing, thus splitting the industry cluster in half.

Kurzweil: This is a big leap, with no support to land on.

Tuomi: In reality, the belief in rapid development has often paid off. Discontinuous innovations have created new uses and markets for semiconductors and have produced an expanding market. Instead of filling a market need, the semiconductor industry has actively and aggressively created markets. At times the aggregate market has grown at a speed that has appeared to be almost infinite in relation to existing manufacturing capability.

Kurzweil: These references to “infinite” demand make no sense. Nothing is infinite in today’s world of technology.

Tuomi: The rapid growth of semiconductor industry, therefore, has not been driven simply by technical advance in semiconductor industry. Although the aggressive pricing policy has facilitated the wide use of semiconductors, the high demand for semiconductor technology has fundamentally reflected a continuous stream of innovations that have occurred outside the semiconductor industry. In other words, the apparent explosive big bang in semiconductor technology is also an illusion.

Kurzweil: Tuomi’s statement that “the apparent explosive big bang in semiconductor technology is . . . an illusion” is another remarkable conclusion.

Tuomi: . . . .many discussions on the future of Moore’s Law have focused on physical limits. In recent years economic considerations have gained legitimacy also in this context, partly because Moore himself has frequently predicted that the increases in chip complexity will not be limited by physics but by the exponentially increasing costs of manufacturing plants.

Kurzweil: Three-dimensional methodologies promise to reverse this – new approaches use self-organizing methods to allow many faulty components in a system.

Tuomi: As computing technology becomes increasingly pervasive, we eventually have to ask what benefits it actually brings. Fundamentally, this question can only be answered in a theoretical framework that is able to define development. In theory, there are many different ways to approach this question, both old and new. It should, however, be clear that development cannot be reduced to shrinking line-widths, maximum number of components on a chip, or minimal manufacturing costs.

Kurzweil: These remain, nonetheless, powerful driving factors, although it is correct to say that there are other factors of innovation. Moreover, exponential growth is not limited to memory or MIPS, but includes essentially all information-based technologies. Other examples include magnetic disk density (a completely independent phenomenon), telecommunication speeds and price-performance, DNA sequencing, brain reverse engineering, human knowledge, and many others.

Responses to Ilkka Tuomi’s “Kurzweil, Moore, and Accelerating Change” ³⁸

Tuomi: As Moore noted, on average integrated circuit component counts grew rapidly during the 1960s, almost with a one-year doubling time. During the first decade of microprocessors, the transistor counts grew at about 22 months doubling time, when measured using a least squares trend, which slowed to about 33 months during the following decade. During the 1990s, transistor counts grew at varying speeds. In the Intel’s Pentium chips, the transistor counts grew with around 54 month doubling time. After that the transistor counts grew very rapidly, partly because large amounts of memory were added onto microprocessor chips (Tuomi, 2002b).

Kurzweil: As noted earlier, Tuomi’s 54 month doubling time is not consistent with the historical data, as indicated in the graph on transistors in Intel microprocessors provided earlier. There are always less capable processors offered than the current standard, so including these on the chart only serves to skew the results.

More importantly, measurements related to processors, whether transistor counts, or MIPS ratings, are not the most meaningful items to measure. We are more concerned with performance per unit cost. Also note that MIPS measure does not take into consideration word size. Since word sizes have increased over time, there has been even greater progress than the MIPS ratings alone would suggest. See my graph above on processor performance (MIPS).

Tuomi: Alternative growth rate estimates can be based simply on transistor counts on representative microprocessors introduced at two points of time. If we use the first microprocessor, Intel 4004, as a starting point, the exponential growth time for the 1971-82 period is 21 months, for the 1971-91 period 26 months, and for the 1971-98 period 27 months.This calculation indicates a slowing down in the component growth rate. If we separately calculate the growth rate for the 1982-91 period, it is about 35 months, and for the 1991-98 period somewhat over 30 months. It therefore appears that during the first decade of microprocessors component counts increased much more rapidly than during the 80s. In the 1990s the growth rate was faster, 30 months for both the first half and the whole 1990-1998 period, but considerably slower than during the first decade of the microprocessor history.

Kurzweil: These numbers are also dubious, but even less relevant than MIPS ratings. There is an optimal number of transistors for the functionality of a processor for a given word length and instruction set, so measuring the number of transistors is not the most appropriate measure of the exponential growth of price-performance. Nonetheless, there has been exponential progress on this measure. See my graph above on transistors in Intel processors.

Tuomi: If one studies the share of inputs that are used computer manufacturing in the US, one can see that the biggest input cost is associated with wholesale trade (about 14 percent of total output) and semiconductor devices (also about 14 percent). This is followed by payments for other electronic component manufacturing, software publishers (about 9.5 percent), computer storage device

manufacturing, computer peripheral manufacturing, internal sales in the computer industry, and management consulting services. A more detailed study reveals that computer manufacturing requires such inputs as air transportation, computer terminals, sheet metal work, and food services and drinking places. The average price changes in the 1990s are greatly influenced by the extremely rapid drop of prices in the second half of the decade. For example, the semiconductor input prices in the computer industry dropped over 40 percent annually during 1995-99. In the first half of the decade, however, they declined only 11 percent annually.

Kurzweil: This is all irrelevant. We are concerned with what the computer industry has been able to accomplish in terms of price-performance, not in measuring what they pay for their “inputs,” such as “food services and drinking places.” We see exponential improvement in all facets of computer price-performance: MIPS per dollar, RAM capacity, hard disk capacity, and other features.

Tuomi: A reasonable estimate for the average annual decline in quality-adjusted computer prices is probably about 18-30 percent during the last couple of decades, which corresponds to 2.6 to 4.2 year “doubling times.”

Kurzweil: As I cited above, Tuomi makes this mistake repeatedly. A 30 percent decline in price means a price of 70% (of the original price) after one year, 49% after two years. Thus the price has fallen to less than half in 2 years, so the doubling time is less than 2 years, not 2.6 years. This consistent error is significant since Tuomi is citing “high” doubling times as evidence that estimates of 2 year doubling times are not accurate. Of course, the doubling time depends entirely on what one is measuring.

Tuomi: Clouds are continuously changing their form. The ripples on a stormy sea encode huge amounts of information. Any argument about speed of change therefore has to neglect most sources of change. The selection is obviously made by relevance. When we say that the evolution is progressing at an accelerating pace, we have to abstract away all change that doesn’t matter. 20 Which sources of change are left out of the equation depends on our present interests.

Kurzweil: Tuomi is missing the entire point of the law of accelerating returns, which pertains to exponential growth of the capacity and price-performance of information-based technologies. It is not my position that all exponential trends go on indefinitely.

Indeed, exponential trends do hit a natural limit, but the key point is that the known limits for computation and communication are extremely high, vastly exceeding current technology. When I refer to a paradigm shift (such as integrated circuits, or the Internet), I am not referring to any type of change, such as “clouds . . . . changing their form,” but rather technological methods that provide for the continuation of the exponential growth of the capacity and price-performance of an information related technology.

Notes

name="_edn1">¹ Tuomi, Ilkka, "The Lives and Death of Moore’s Law." First Monday, volume 7, number 11 (November 2002). http://firstmonday.org/issues/issue7_11/tuomi/index.html.

name="_edn2">² Ibid.

name="_edn3">³ Ibid.

name="_edn4">⁴ Ibid.

⁵Intel Corp. and The International Technology Roadmap For Semiconductors: 2002 Update, International SEMATECH, 2002. http://public.itrs.net.

⁶ Microprocessor Quick Reference Guide. Intel Research. http://www.intel.com/pressroom/kits/quickrefyr.htm. Gregory Allen has also addressed similar issues in a mathematical analysis of microprocessor performance growth ("The Bit Flip Rate, Frequency and Transistor Density Equations," private communication, 2003). He agrees that the "bit flip rate" of microprocessors and transistor density are growing exponentially.

⁷ Data from Dataquest and Intel reports.

⁸ Cullen, Steve. "Semiconductor Industry Outlook." Instat/MDR. 2003.

name="_edn9">⁹ Tuomi, Figure 4.

name="_edn10">¹⁰ 1976-1999: E.R. Berndt, E.R. Dulberger, and N.J. Rappaport, 2000. "Price and quality of desktop and mobile personal computers: a quarter century of history," 17 July 2000. http://www.nber.org/~confer/2000/si2000/berndt.pdf. 2001-2016: The International Technology Roadmap for Semiconductors: 2002 Update. On-Chip Local Clock: Table 4c Performance and Package Chips: Frequency On-Chip Wiring Levels-Near-term Years (Update), page 167

.

name="_edn11">¹¹ Average transistor price: Dataquest/Intel. Microprocessor speeds: 1976-1999: E.R. Berndt, E.R. Dulberger, and N.J. Rappaport, 2000. "Price and quality of desktop and mobile personal computers: a quarter century of history," 17 July 2000, http://www.nber.org/~confer/2000/si2000/berndt.pdf. 2001-2016: The International Technology Roadmap for Semiconductors: 2002 Update. On-Chip Local Clock: Table 4c Performance and Package Chips: Frequency On-Chip Wiring Levels-Near-term Years (Update), page 167.

name="_edn12">¹² http://www.ii.uni.wroc.pl/~jja/ASK/HISZCOMP.HTM, http://www.siliconspirits.com/scomputer.html, http://www.seagate.com/cda/products/discsales/index, Byte magazine advertisements, 1977-1998, PC Computing magazine advertisements, 3/1999, Understanding Computers: Memory and Storage Time Life Editors. Time Life, 1990.

name="_edn13">¹³ Internet Software Consortium, http://www.isc.org/ds/host-count-history.html

name="_edn14">¹⁴ Trajtenberg, Manuel. Economic analysis of product innovation: the case of CT scanners. Cambridge, MA : Harvard University Press, 1990; Michael H. Friebe, Ph.D., President, CEO NEUROMED GmbH (email)

¹⁵ "GenBank Statistics," GenBank, National Library of Medicine, August 15 2003, http://www.ncbi.nlm.nih.gov/Genbank/genbankstats.html

name="_edn16">¹⁶ See discussion in my essay “The Law of Accelerating Returns”

name="_edn17">¹⁷ Tuomi, Ilkka, “Kurzweil, Moore, and Accelerating Change.” Ilkka Tuomi

Curriculum Vitae. European Commission Joint Research Centre. http://www.jrc.es/~tuomiil/moreinfo.html.

name="_edn18">¹⁸ Tuomi, Ilkka, "The Lives and Death of Moore’s Law." First Monday, volume 7, number 11 (November 2002). http://firstmonday.org/issues/issue7_11/tuomi/index.html.

¹⁹ The International Technology Roadmap For Semiconductors: 2002 Update, International SEMATECH, 2002. http://public.itrs.net.

name="_edn20">²⁰ Fredkin, Edward, “A Physicist’s Model of Computation,” Proceedings of the 26^th Recontre de Moriond, 1991. http://www.digitalphilosophy.org/physicists_model.htm.

name="_edn21">²¹ Tuomi, Ilkka, "The Lives and Death of Moore’s Law." First Monday, volume 7, number 11 (November 2002). http://firstmonday.org/issues/issue7_11/tuomi/index.html.