WHEN THINGS START TO THINK | Chapter 5: The Personal Fabricator

May 15, 2003

Originally published by Henry Holt and Company 1999. Published on KurzweilAI.net May 15, 2003.

Thomas Watson, the chairman of IBM, observed in 1943 that “I think there is a world market for maybe five computers.” In 1997 there were 80 million personal computers sold. To understand his impressive lack of vision, remember that early computers were

  • large machines
  • housed in specialized rooms
  • used by skilled operators
  • for fixed industrial operations
  • with a limited market

From there it was too hard to conceive of a computer that could fit on a desk without crushing it, much less on a lap. Unseen beyond that horizon in packaging lay the revolutionary implications of personalization.

Once computers became small enough and cheap enough for individuals to own them, their application became a matter of personal preference rather than corporate policy. Big companies have an unerring knack for doing dumb things because so many people are involved in specifying and evaluating what someone else does that it’s all too easy to forget to think. Once it became possible for individuals to write programs or configure packages to reflect their individual needs, then instead of marveling at someone else’s stupidity they could do something about it. This represented a loss of control for the computer companies that were accustomed to prescribing what hardware and software their customers would use; in return, the market for computers grew to something more than five machines.

Now consider machine tools. These are the large mills and lathes and drills that are used in factories for fabrication. For the ordinary person, they are about as exciting as mainframe computers. In fact, they really are quite similar. Machine tools are

  • large machines
  • housed in specialized rooms
  • used by skilled operators
  • for fixed industrial operations
  • with a limited market

Sound familiar? Big companies use big machines to make things we may not really want. Personal computing has not gone far enough: it lets us shape our digital environment, but not our physical environment. By giving computers the means to manipulate atoms as easily as they manipulate bits, we can bring the same kind of personalization to the rest of our lives. With the benefit of hindsight, plus a peek into the laboratory to see what’s coming, this time around we can do better than Thomas Watson and recognize the impending arrival of the Personal Fabricator.

One of the eeriest experiences in my life came when I first opened the door of a 3D printer and took out a part that I had just seen on the screen of a computer. It violated the neat boundary between what is inside the computer and what is outside. In a strange way, holding the part felt almost like touching the soul of the machine.

A 3D printer is a computer peripheral like any other, but instead of putting ink on paper, or data on a disk, it puts materials together to make objects. Working with a 3D printer engages our visceral connection to the physical world, which has been off-limits for so long whenever a computer is involved. When I set up a 3D printer in my lab, people would come by just to touch the machine and the parts that it produced. Watching them, I could almost hear the mental gears grinding as they began to think about how they could make what they wanted, instead of purchasing what someone else thought they wanted. If a static shape can have that kind of impact, then I’m not sure how people will react when the output from the printer is able to walk out of it. Because we’re also learning how to print sensors, motors, and logic.

To appreciate just how inevitable and important personal fabrication is, I had to retrace the history of computing in this new domain. When I arrived at MIT in the early ’90s, you couldn’t tell that it had been a pioneer in manufacturing technology. The few remaining machine shops on campus were in a sorry state, holdovers from the Iron Age in the midst of the Information Age. The campus had long since gone digital. Forlorn machine tools were left neglected by the throngs of students clustering around the latest computers.

As I set out to create a lab that could free computing from its confining boxes, I knew that we would need to be equally adept at shaping things as well as ideas. This meant that we would need to start with a good machine shop. A traditional machine shop is a wonderful place in which a skilled machinist can make almost anything. Since this one would be in the Media Lab, where neither the people nor the computers are traditional, this shop would need to be much more easily accessible.

The most versatile tool to be added was a milling machine. These come in two flavors. Manual ones are designed to be used by hand; a part to be machined is fixed to a bed that is moved under the rotating cutting tool by long lead screws turned by hand. Numerically controlled (NC) mills are designed to be run by a computer; the bed and sometimes the head are moved by motors under software control. The operator of an NC mill stands at a control panel that can be some distance from the workpiece, usually with a cryptic interface that makes it difficult to do much more than start and stop the machine.

I wanted both less and more than a conventional NC mill. Whatever we bought had to have a direct mechanical linkage that could be used manually. The feeling of the torque and vibration through the handles provides essential insight into how well a tool is cutting, which is invaluable for beginners and important also for experts working with unfamiliar materials. This is lost when the mill is controlled from a remote console.

Next, the mill needed to have a graphical interface that made it simple to specify shapes right on the machine, because the only things more unfriendly than typical NC controllers are the programs that engineers use to design parts. Finally, the mill had to be on speaking terms with the building’s network, to make it easy to transfer designs from the many types of computers in use.

Machine tool distributors laughed when I described what I was looking for. I knew I had a problem when I got the same reaction from machine tool manufacturers. Over and over, I was told that I had to choose between the immediacy of a manual interface and the power of a numerical one.

As I traveled around the world, I started to make detours to visit manufacturers. The turning point in my quest came on a trip to see one of the largest European makers of machine tools. The day started with them proudly showing me their products—huge mills that cost $100,000, required skilled technicians to operate and maintain them, and were difficult to use for anything but repetitive operations. I then sat down with their eminent director of development to discuss my requirements. As I spoke, he became increasingly agitated. Soon he was pounding on the table and shouting that I must not ask for such a machine.

At first I wasn’t sure whether to laugh or cry. As he continued to hold forth, I felt like I was being transported back a few decades to reenact a familiar scene. I wanted to buy a PC; he wanted to sell me a mainframe. He was making all of the same arguments for why small computers could never replace big computers: machining/computing requires expert skills, NC mills/mainframes have narrow markets, personal systems are toys that can’t do serious work. We parted agreeing that we lived on different planets.

Fortunately, I came back and found what I was looking for in my backyard. A company on Boston’s Route 128, Compumachine, started with a nice manual milling machine, added motors so that a computer as well as a person could control it, then put on a keyboard and screen to run it. Instead of the usual impenetrable industrial controller, the computer was a familiar PC.

In the machine tool industry this was viewed as dangerous lunacy, because the inevitable system crashes could cause real crashes of the mill, destroying it if not its operator. What makes the mill safe is an illusion: the computer only appears to be in charge. There’s a layer of circuitry between it and the mill that actually issues the commands, monitoring everything the computer requests in order to prevent anything unsafe from happening.

An NC mill is such a specialized piece of equipment that it usually has to earn its keep with manufacturing operations. Putting one in the hands of graphic designers, and programmers, and musicians, led to all sorts of clever things getting made in a way that would never occur to a traditional engineer. The mill was used to build a parallel computer with processors embedded in triangular tiles that could be snapped together in 2­D or 3­D sculptures, exchanging data and power through the connections. Miniature architectural models were created to serve as tangible icons for a computer interface. The success of these kinds of projects led me to wonder if physical fabrication could be made still simpler to reach still more people.

An alternative to making something by machining away parts that you don’t want is to assemble it by adding parts that you do. Lego blocks are a familiar example of additive fabrication that make it possible to quickly and easily build impressive structures. There’s a surprising amount of wisdom in this apparently simple system; Lego has spent decades getting the bricks just right, experimenting with the size of the posts, the angle of the faces, the hardness and finish of the plastic. The earliest bricks look almost like the current bricks, but these ongoing incremental improvements are what make them so satisfying to play with.

Anyone who has been around kids building with Legos understands the power of a child’s drive to create. This is on display at the Legoland near Lego’s headquarters in Denmark, an amusement park filled with spectacular Lego creations. There are Lego cities with buildings bigger than the kids admiring them, giant Lego animals, Lego mountain ranges. What’s so amazing about this place is the kids’ reaction. There are none of the things we’ve come to expect that children need to be entertained, no whizzy rides or flashy graphics or omnipresent soundtracks. Just great structures. And the kids are more engaged than any group I’ve ever seen, spending awed hours professionally appraising the marvels on display. The people who work for Lego, and who play with Lego, share a deep aesthetic sense of the pleasure of a nicely crafted structure.

In the Media Lab, Professor Mitch Resnick has one of the world’s most extensive Lego collections. For years his group has been working to extend the domain of Lego from form to function, embedding sensors and actuators as well as devices for computing and communications, so that the bricks can act and react. At first, conventional computers were externally interfaced to the Lego set; now they’ve been shrunken down to fit into a brick. These let children become mechanical engineers, and programmers, with the same comfortable interface they’ve known for years. Kids have used the system to make creatures that can dance and interact, to animate their fantasy worlds, and to automate their homes. This system has been so successful that it has now left the lab and become a new line of products for Lego. The only surprise, which isn’t really a surprise, is that grown-ups are buying the sets for themselves as well as for their kids.

At the opposite end of the spectrum is Festo. Festo is to industrial engineers as Lego is to kids; they make the actuators and controllers that are used to build factories. In fact, Lego’s factories are full of Festo parts making the Lego parts. Festo also has a system for prototyping new factories and teaching people to design and run them. This is Lego for grown-ups. When I first brought its components into my lab there was a rush of ooh’s and aah’s, because it could do what Lego couldn’t: make large precise structures. The hefty, shiny metal parts spoke to their serious purpose.

Following detailed instructions, with some effort, we used this system to put together an assembly line to make pencil sharpeners. And that’s all we did. People left as quickly as they appeared, because it soon became clear that it was too hard to play with something this exact. Too many specialized parts are needed to do any one thing, assembling them is too much work, and interactively programming the industrial controllers is all but hopeless.

Between Lego and Festo the needs of children and industrial engineers are covered. Mitch and I began to wonder about everyone else. Most people don’t make most of the things that they use, instead choosing products from a menu selected for them by other people. Might it be possible to create a system that would let ordinary people build things they cared about? We called this “Any Thing” and set up a project to develop it. Drawing on the lessons from Lego and Festo, this kit would build electrical connections for data and power into every mechanical joint, so that these capabilities did not need to be added on later. The parts would snap together in precise configurations, without requiring tools. And it would be made out of new composite materials, to be light, strong, and cheap. Armed with these guiding principles, we first tried out computer models of the parts, then prepared to start building it. At this point we called a design review meeting. We filled a room with as many alternative construction kits as we could find for comparison, and then added as many people and as much pizza as we could fit in. The night took many surprising turns.

Unfortunately, Any Thing had been developed entirely by a bunch of geeky guys (like me). This was the first time any women had seen it. One of the first questions that one of my female colleagues asked was what it was going to feel like; our jaws dropped when we realized that we had designed bones without thinking about skin. As the night proceeded and people played with the various commercial construction systems, they were uniformly frustrated by the nuisance of chasing around to find all of the fiddly little bits they needed to complete whatever they were working on. The Any Thing vision of personal fabrication necessarily entailed a personal warehouse to store an inventory of all the required components.

The more we talked, the more we realized that the most interesting part of Any Thing was not our proposed design, it was the 3D printer we were using to prototype it. Three-dimensional printing takes additive fabrication to its logical conclusion. Machine tools make parts by milling, drilling, cutting, turning, grinding, sanding away unwanted material, requiring many separate operations and a lot of waste. Just as a Lego set is used to construct something by adding bricks, a 3D printer also builds parts by adding material where it is needed instead of subtracting it where it is not. But where Lego molds the raw materials to mass-produce bricks in the factory, in a 3D printer the raw materials are formed on demand right in the machine.

There are many ways to do this. One technology aims a laser into a vat of epoxy to cure it just where it is needed, another spreads layers of powder and squirts a binder where it should stick together, and our Stratasys 3D printer extrudes out a bead of material like a computer-controlled glue gun that leaves a three­dimensional trail. These techniques can all go from a computer model to something that you can hold, without requiring any special operator training or setup. This convenience is significantly decreasing the time it takes companies to make product prototypes. The machines are currently expensive and slow, but like any promising machinist they are steadily becoming more efficient.

As interesting as 3D printing is, it’s still like using a PC to execute a mainframe program. The end result is not very different from what can be made with conventional machine tools, it’s just the path to get there that is simpler. The real question posed by our Any Thing design review was whether 3D printing could be extended as Lego had been to incorporate sensors and actuators, computing and communications. If we could do this, then instead of forcing people to use their infinitely flexible and personal computers to browse through catalogs reflecting someone else’s guesses at what they want, they could directly output it. Appliances with left-handed controls, or ones with easy-to-read large type, would become matters of personal expression rather than idle wishes.

This is the dream of the personal fabricator, the PF, the missing mate to the PC. It would be the one machine that could make all your others, a practical embodiment of the perennial science-fiction staple of a universal matter output device. As we began to dare to dream of developing such a thing, we realized that we had been working on it all along.

Joe Jacobson’s printer for the electronic book project was already laying down the conductors, insulators, and semiconductors needed to make circuits on paper. Just as an ink-jet printer has cartridges with different-colored inks, it is possible to provide a printer with more types of input materials so that it can also deposit structural shapes and active elements. One of the first successes was a printed motor that used carefully timed electric fields to move a piece of paper. If your desk was covered by such a structure then it could file your work away for you at the end of the day.

The semiconductor industry is trying to reach this goal of integrating sensors, circuits, and actuators, by cutting out tiny little silicon machines with the fabrication processes that are currently used to make integrated circuits. This idea is called MEMS, Micro-Electro-Mechanical Systems. Although the industry sees this as an area with great growth prospects, MEMS fabrication has the same problem as mainframes and machine tools. MEMS devices are designed by specialists and then mass-produced using expensive equipment. The desktop version could be called PEMS, Printed Electro-Mechanical Systems. Unlike MEMS, it can make people-size objects rather than microscopic machines, and the things are made where and when they are needed rather than being distributed from a factory.

A PEMS printer requires two kinds of inputs, atoms and bits. Along with the raw materials must come the instructions for how they are to be placed. One possibility is for these to be distributed over the Net. Instead of downloading an applet to run in a Web browser, a “fabblet” could be downloaded to the printer to specify an object. This would significantly lower the threshold for a company or an individual to sell a new product, since all that gets sent is the information about how to make it. But it stills leaves people selecting rather than designing what they want.

To bring engineering design into the home, the CAD software that engineers currently use is a valuable guide for what not to do. These packages generally betray their mainframe legacy with opaque user interfaces and awkward installation procedures, and isolate the design of mechanical structures, electrical circuits, and computer programs into modules that are barely on speaking terms.

A design environment to be used in the home should be so simple that even a child could understand it. Because then there’s some hope that their parents might be able to also. There’s already a pretty good precedent that’s routinely used by children without any formal training: Lego. It provides simple standard parts that can be used to assemble complex systems.

That was the original inspiration for the Any Thing project, which in retrospect was the right idea for the wrong medium. It was a mistake to assume that the physical fabrication process had to mirror the logical design process. By moving the system from a bin of parts to a software version that exists only inside a computer program, the intuitive attraction of building with reusable components can be retained without needing to worry about ever running out of anything. Computer programmers use “object-oriented programming” to write software this way; with a PEMS printer the objects can become tangible.

It’s important that the reusability be extended to the physical as well as the virtual objects. The arrival of fast high-resolution printers has created the paperfull office of the future because printing is now so convenient. This is a cautionary tale for PEMS, which threatens to do the same for physical things. Two-dimensional sheets of paper get thrown away in three-dimensional recycling bins; three-dimensional objects would need a four-dimensional recycling bin.

A slightly more practical alternative is to let the PEMS printer separate discarded objects back into raw materials. This is what a state-of-the-art waste facility does on a large scale; it’s a bit easier on the desktop because the task is restricted to a much smaller group of materials that can be selected in advance to simplify their separation. Just as coins get sorted by weight, differences in the materials’ density, melting temperature, electrical conductivity, and so forth can be used to separate them back into the printer’s input bins.

Complementing a PEMS printer with intuitive design tools and the means to recycle parts provides all of the ingredients needed to personalize fabrication. This just might eliminate some of the many bad ideas that should never have been turned into products. I have a theory for why so many companies full of smart people persist in doing so many dumb things. Each person has some external bandwidth for communicating with other people, and some internal processing power for thinking. Since these are finite resources, doing more of one ultimately has to come at the expense of the other. As an organization expands, the volume of people inside the company grows faster than the surface area exposed to the outside world. This means that more and more of people’s time gets tied up in internal message passing, eventually crossing a threshold beyond which no one is able to think, or look around, because they have to answer their e-mail, or write a progress report, or attend a meeting, or review a proposal. Just like a black hole that traps light inside, the company traps ideas inside organizational boundaries. Stephen Hawking showed that some light can sneak out of a black hole by being created right at the boundary with the rest of the world; common sense is left to do something similar in big companies.

So many people are needed in a company because making a product increasingly requires a strategy group to decide what to do, electrical engineers to design circuits that get programmed by computer scientists, mechanical engineers to package the thing, industrial engineers to figure out how to produce it, marketers to sell it, and finally a legal team to protect everyone else from what they’ve just done. It’s exceedingly difficult for one person’s vision to carry through all that. Focus groups help companies figure out when they’ve done something dumb, but they can’t substitute for a personal vision since people can’t ask for things they can’t conceive of.

Companies are trying to flatten their organizational hierarchies and move more decision making out from the center by deploying personal computing to help connect and enable employees. But the impact of information technology will always be bounded if the means of production are still locked away like the mainframes used to be. For companies looking to foster innovation, for people looking to create rather than just consume the things around them, it’s not enough to stop with a Web browser and an on-line catalog. Fabrication as well as computing must come to the desktop.

The parallels between the promise and the problems of mainframes and machine tools are too great to ignore. The personal fabricator now looks to be as inevitable as the personal computer was a few decades ago. Adding an extra dimension to a computer’s output, from 2­D to 3­D, will open up a new dimension in how we live.

This revolutionary idea is really just the ultimate expression of the initial impetus behind multimedia. Multimedia has come to mean adding audio and video to a computer’s output, but that’s an awfully narrow notion of what constitutes “media.” The images and motion don’t have to be confined to a computer screen; they can come out here where we are.

WHEN THINGS START TO THINK by Neil Gershenfeld. ©1998 by Neil A. Gershenfeld. Reprinted by arrangement with Henry Holt and Company, LLC.