Paul Allen: the Singularity isn’t near
October 13, 2011 | Source: Technology Review
The Microsoft cofounder and a colleague say the Singularity is a long way off.
Their main issue: the “complexity brake. As we go deeper and deeper in our understanding of natural systems, we typically find that we require more and more specialized knowledge to characterize them, and we are forced to continuously expand our scientific theories in more and more complex ways…. Without having a scientifically deep understanding of cognition, we can’t create the software that could spark the singularity.”
Comments (23)
by xd
Couple of points. I don’t know if it’s required that the singularity means *no* human can understand the breakthroughs that are taking place or just that the average human cannot understand it. If the latter I think we are close to already being there as the rate of change to my naive sample of one eyes seems to have sped up by a factor of somewhere between 2 and 50 times. I’ve been reading science journals since undergraduate days in the late 1980s and it seems to me that answers are being found to posed questions in many areas every couple of months whereas it seemed to take 2-5 years before. Also IBM in it’s battery 500 used some kind of simulation to find a chemistry for an advanced battery in two years, something that might have taken 50 years if the 2 million potential candidate chemistries had to be tested individually in the lab.
Just the internet itself, the ability to cross reference words and the very fact that much of the published research is already on the internet increases collaboration to an enormous extent. There is now no longer the need to reinvent the wheel because you don’t know other teams are working on the same thing. Everyone in the field can read everything and the “groupmind” is up to speed everywhere.
I would not be surprised if we’re not already in the early stages of the singularity quite frankly.
by Onthebeach
Paul may think that it will take until 2045 for Windows to work properly. He could be right. Of course by then Windows will be so massive that it may well collapse into a gravitational singularity. This would certainly be a singular moment for windows users.
Uploading a human consciousness is probably theoretcally possible but it would not be ‘you’ but a copy. When you switched it on it would think it was ‘you’ but you yourself, the original, would still eventually die. The only way you can achieve true ‘immortality’ is with a gradual system of implants over time so that ‘you’ remain ‘you’ throughout the process. In other words, ‘There can only be One’.
by Micheal Karg
Sorry, i don’t see the difference.
by eldras
It isn’t Ray’s predictions…it’s his methods that are decisive.
He trends development then cross references their trends.
Complexity explosion was the argument starting the A.I. winter in the 70′s (The 1973 Lighthill Report concluded A.I. was impossible and funding dried up. Back then it was called the ‘combinatorial explosion’).
The successful rebuttal is from trend observation.
1. Machines [i]are[/i] getting smarter.
2. Systems [i]are[/i] able to do tasks once thought uniquely human.
3. Top down & bottom up brain approaches [i]are trending as Kurzweil predicted.
by tim333
It’s interesting how little the “complexity brake” seems to slow things in practice. Like say self driving cars, people have been mucking about with them for years and the first ones that seem about ready for prime time, the Google cars come along about when the hardware is ready.
I expect the kind of engineers who are building self driving cars today are already figuring out how to extend the algorithms to human level AI and it’ll be pretty much ready to go when the hardware arrives.
Personally I think Kurzweil is being conservative with 2045, understandably so as if he gave an earlier date and it came about with no singularity he’d look kinda silly. For me the singularity is the day when you can tell a robot to go build a better robot and they can go do it without human assistance. 2035 perhaps?
by Maxwell
I think that if we wait for a singularity to happen among machines running a Paul Allen’s Microsoft OS, we might indeed see that in another 2’000 years (or never…)
by MysticMonkeyGuru
I agree with everything that Paul Allen and Mark Greaves have said in their article. The Complexity Brake is very real, and will slow down technological progress by a considerable amount.
by Jod
and you fail to mention their primary reason which is the understanding of cognuity. all of it is hypothetical to me but it would be nice to see our species progress and return some favors to our earth
by PeterKinnon
Trans-humanism suffers from the fundamental misconception that an “intelligence” greater than ours will be created in an intentional way by our species.
Kurzweil completely misses the reality that species, including the inorganic one which is soon to be born of technology, are not created but evolve.
Unfortunately we are saddled with a deeply entrenched mind-set that we create things, that we are designers. It is the very same idea that, by projection, has led to development of religions. That the cosmos must have a creator, a designer.
In actuality, if we make the effort to step away from such self-centered viewpoints and look objectively at the development of science and technology we see that it is truly an evolutionary process for which the gross outcomes are independent of individual efforts.
We would have geometry without Euclid, calculus without Newton or Liebnitz, the camera without Johann Zahn, the cathode ray tube without JJ Thomson, relativity (and quantum mechanics) without Einstein, the digital computer without Turin, the Internet without Vinton Cerf. (The last mentioned development representing the embryonic new phase of life. I suspect that for this phase transition, for which the inappropriate buzz-word “singularity” is used, Kurzweil’s timing at least may not be far wrong.)
Technology is seen to inevitably evolve within the medium of collective imagination of our species, moderated, as are biological evolutionary processes, by selection pressures of the prevailing environment.
This is very informally expanded upon in “The Goldilocks Effect: What Has Serendipity Ever Done For Us?” (free download in e-book formats from the “Unusual Perspectives” website)
by thinkahol
the thing is we actually do have a good over-arching theory of cognition and consciousness, everyone just hasn’t realized it yet. Thomas Metzinger describes it quite well in his talk “Being No One”: http://www.youtube.com/watch?v=mthDxnFXs9k
by funkervogt
I wouldn’t be so dismissive to Allen’s ideas. He’s a brilliant man, and, as has become especially clear since 2009, Kurzweil’s predictions frequently don’t come to pass until years later than he forecast.
Allen isn’t saying the Singularity isn’t going to happen, he’s just saying that it probably won’t happen until sometime after 2045. Such a delay would be consistent with how Kurzweil’s predictions have been faring so far: Most of them end up happening, it just takes longer than Kurzweil predicted.
It’s also significant to note that Allen believes the Singularity hinges on the successful creation of AI, not radical life extension, genetic engineering or advanced nanomachines. I agree with that. The other technologies are very important, but only AI will cause the pace of change on this planet to explode beyond human levels of comprehension.
by PacRim Jim
Is it not obvious that the human brain can be reverse-engineered. Once it’s up and running, it would be auto-didactic, teaching itself more and more.
That would be faster than developing enormously complex strong AI, and would be more robust, also.
by tbrehmer
Who is Paul Allen?
I can’t help it—Ray Kurzweil’s work just makes all kinds of sense to me. For people with little imagination, its perspectives on the future are just unimaginable.
by Luke25
I should add: that’s what Paul Allen seems to be thinking: if it’s going to take 30 years to develop an hardware powerful enough for AGI, we’ll need another 30 years to develop the software. Nonsense. Hardware and software improve together. We’ll already beginning to see something related to AGI (speech recognition is beginning to take pace) on current personal hardware.
by Luke25
Paul Allen… Brilliant! You can hardly believe he’s the Microsoft cofounder. You can even more hardly believe that he really wasted his time to speak such nonsense. Complaining for the lack of good A.I. routines today is exactly like complaining the lack of augmented reality software on a Commodore 64. Is it truly possible that he can’t see that some applications begin to be possible only when an adequate hardware is available? Kurzweil predictions are based on processing power because without such powerful hardware some software models can’t simply be implemented. Of course an A.G.I. software will need a lot of refinements when we’ll have enough processing power to let it run, since it will be the first one, but if you don’t have the hardware, it doesn’t matter what brilliant ideas you have: you can’t simply make them a reality. Perhaps Paul Allen thinks that even IBM is a fool with his ‘Blue Matter’ project. Electronic engineers and neuroscientists have begun working on true brain’s simulation. Of course, with the processing power they have now they can simulate a mouse’s cerebral cortex at best, but this doesn’t mean these simulations won’t improve later on.
by rencomp
This sort of article strikes me as having a sort of “whistling in the dark” quality. So many of us want so very badly for the brain to remain an unyielding frontier of mystery that will always foil our scientists’ efforts to understand it. I sympathize. Reducing consciousness to a set of physiological principles that can be replicated in another substrate is a frightening proposition, and that fear can lead to a state of denial. But each new generation of technological advancement gives us the tools to create the next level of advancement, etc. And while the human brain doubtless possesses greater complexity than we currently are aware of, the tools we have to manage complexity are becoming exponentially more powerful, so I wouldn’t bet on this ‘brake’ fending off the singularity for very long. It’s true, that predictions about future technological abilities based on observing past trends are not exactly rock solid reliable eventualities, but I personally wouldn’t place any bets. The Cray Titan’s 50 petaflop supercomputer for 2012 is actually ahead of Kurzweil’s predictions.
by Spikosauropod
Paul Allen is a perfect example of “intuitive linear” thinking. Nearly everything he says is suggestive of this mindset.
After seeing this set of videos, I almost wonder if the singularity might come in the next couple of years:
http://www.kurzweilai.net/henry-markram-simulating-the-brain-next-decisive-years
by Logic
I was so excited to read a well-informed rebuttal to Kurzweil’s ideas. As I read the full article, however, I realized that Allen has made the same basic intellectual error as all detractors before him: He makes a false correlation between a specific technology’s limit (in this case, our current limited understanding of the human brain) and the Law of Accelerating Returns.
Any limits to our current knowledge are not in themselves a valid rebuttal to the Singularity prediction, unless they pinpoint where the LAR (or Kurzweil’s use of it) is flawed. Allen’s argument doesn’t make that link because it ignores everything outside its sphere of focus.
The “Complexity Brake”, however, is an excellent concept, and could certainly stymie Kurzweil’s Singularity timeline predictions. I think *as a concept* it’s a wonderful insight we can use to analyze progress and trends. (It would be great to analyze the data and determine a “complexity brake” metric on the scale of Kurweil’s LAR). Even if Allen’s use of the concept is flawed by his misunderstanding of the context of Kurzweil’s arguments (which it is), the idea isn’t necessarily wrong because he’s used it poorly. The “complexity brake” is a potentially real and serious stumbling block worthy of thought. Personally, I think it’s factored into the LAR, but if we could determine a metric, it might help overcome forward progress hurdles in a real and serious way.
The problem is that Allen ignores the cross-pollination of ideas and technologies. He makes the same mistake most people make: he fails to recognize the impact of one industry/technology upon unrelated industries/technologies. So he sees linear progression (or at least slower-than-exponential progress) because he’s blindsided by *current* limitations he can’t imagine solutions for.
Just because you can’t imagine a solution doesn’t mean the breakthrough required to solve the problem must be epic. On an exponential scale, small steps forward tomorrow lead to massive steps forward by the end of the year.
I believe opposing points of view like this are extremely valuable, as any scientific advance requires challenging our most deeply held assertions. The more we challenge Kurzweil’s ideas, the better our clarity of direction. Allen’s argument, while flawed, is a wonderful contribution to the discussion.
by BePo
Essentially, I think Paul means that “HE” won’t take us there. (‘CAN’T’ never will.) There are 10-dozen ‘CAN-DO’ types already alive and busy doing it.. Move over Mr. Allen; we need people with a little more ‘CAN DO’ vision than that. It’s a simple matter of ‘WHEN’.
No one can pick the exact date; but half of what Paul claims won’t happen soon, has already happened in science-labs around the world. He simply isn’t keeping up. (Maybe he should subscribe to Ray’s newsletter?)
by RobinSongs
More understanding of the complexity of technical innovation is important. But I don’t see that as much of a problem since tech and its future beneficial concepts is more popular now than it’s ever been.
Back in the 80s, Kurzweil predicted the ARPANet which was a precursor to the internet and his prediction came true. But that was a time when most of us didn’t really care. Back then, we couldn’t conceptualize the social benefits (in terms of facilitating tech innovation) of things like search engines, open source, YouTube, blogs, etc.
This is another reason why the rate of tech grows exponentially. Its popularity slowly expands along with the innovations and it spreads just as fast as tech progress grows. They work with each other in order to keep the growth going.
A good example are video games since many of them today incorporate a level of AI and the gamers now expect the AI to get better with future titles; it’s now something to look for when reading the description of a game newly released on Amazon, much like the quality of the latest TV.
So the popularity of the concept of AI and its beneifts will spread like crazy, especially when the gamers can eventually use open source to develop it further (eg Goertzel’s OpenCog project).
I’m starting to agree with Steve Jobs when he said that Microsoft doesn’t have much of an imagination.
by eldras
Allen’s opinion is utterly ill-informed. Like great minds microsoft have blocks on certain things. AGI is definitely one of them. Project Halo which Allen ste up is ill conceived.A necessary requirement for intelligence is generalism: one simply cannot do that from getting good in one discipline, or 2 or 3 or 4 etc
Whoe many specialised approaches eg linguistics or artificial neural networks will get there singly they would take 100 years.
An SAI /AGI attempt has to incorporate as many sub-disciplines as a Los Alamos project to build the A Bomb.
It has to be run by a generalist and CEO’s by a specialist…probably an engineer in my view.
It would also be dangerous not to run it as a military project.
by jmlvu
I think Paul’s experience at Microsoft warps his opinion. The Window operating system certainly suffers from the “complexity brake”, (Excel runs slower today than it did in Windows 95) but this is from feature creep and poor programing choices not complexity. An operating system written from scratch would run 1000 times faster.
Once we distill the fundamentals of human learning, transferring it to computers will make the singularity inevitable.
by Singme
It is kind of funny that Apple’s Co-founder, Steve Woz. is on board with Kurzweil’s Singularity. Time will tell!