April 9, 2001
In the turn-of-the-millenium techno-visionary pantheon, Danny Hillis cuts a unique figure. He waxes philosophical with the best of them, holding forth eloquently about transhumanism and the end of the human race and the whole shebang. He’s building a clock intended to last ten thousand years. And yet he’s neck deep in practical work, having just resigned a plum job as a Disney exec to start a new company providing technology and consulting to the entertainment industry. His one big stab at true fame and fortune was Thinking Machines Inc., a firm that lasted 11 years, created the world’s best parallel computing hardware, yet failed to either create a thinking computer program or make Hillis fabulously wealthy. Yet, at five years short of fifty, Hillis seems relatively unruffled by the whole crazy rollercoaster ride. His visionary prognostications lack the alarmism of Bill Joy or Jaron Lanier, and also avoid the starry-eyed enthusiasm of Ray Kurzweil. He comes across, in person and in his writings, as a mild-mannered, curious and creative guy. Although he now talks tough about business like any other seasoned exec, in many ways he’s still an MIT hacker at heart, delighted with the task of building the next cool gadget or intricate algorithm, and looking forward to the gadgets and algorithms of the next millennia in a remarkably matter-of-fact way.
Hillis’s early life prepared him well for the tumultuousness of the technology industry. Born in Baltimore in 1956, his father was an Air Force epidemiologist, so the family moved frequently on the trail of hepatitis outbreaks, and he grew up with no fixed home. Moving from place to place in Rwanda, Burundi, Zaire, and Kenya, he avoided formal schooling and the pressures for social conformity that go along with it. As he says, “We were typically out in the middle of the jungle so I was just taught at home.“ His mother did most of the teaching and her interest in mathematics jibed well with his natural abilities. His father encouraged him to study biology, a pursuit that gave him an early appreciation for the complex machines that are living organisms. “My best biological experiment,” he says, “ was tissue culturing a frog heart and keeping the heart beating even while it was growing in the test tube. It was amazing to me that somehow they got together and did this coordinated activity even though they were just this homogenized mass of cells.” Much of his career was spent creating complex computer systems capable of displaying spontaneous coordinated activity, like the cells in a frog heart.
His education is about what you’d expect – an undergraduate math degree from MIT in 1978, followed by a MIT master’s in robotics three years later. Along the way he found time to pursue his avid interest in toys and to indulge his entrepreneurial streak --working at the MIT Logo Laboratory developing computer hardware and software for children, designing computer-oriented toys and games for the Milton Bradley Company, and co-founding Terrapin Inc., a producer of computer software for elementary schools. For his PhD work, Hillis began the endeavor that has been his greatest contribution to science and humanity so far – his work on the Connection Machine, a massively parallel computer going far beyond any other computer system of the time in terms of its potential for artificial intelligence, simulation of complex physical systems.
Ordinary computers are “serial” – they have only a single processor and hence they can carry out only one operation at a time. The trick is that they’re fast -- a single operation can be done very, very quickly. So a computer can give the illusion of doing many things at once – say, running a game while downloading e-mail while showing an animation – when in fact its processor’s time is swapping back and forth from one task to another rapid-fire. The brain, on the other hand, has around a hundred billion neurons, and in principle they’re all working in parallel, simultaneously. Each one of them acts much more slowly than a computer processor, but what they lack in speed they make up for in bulk and in parallelism. Hillis’s Connection Machine was an elegant compromise, the nature of which changed over time as computer hardware technology evolved. The idea was to make a computer whose processors were fast like those of ordinary computers, but also massively parallel like in a brain. In this way, one could have the best of both worlds, and one could build a really intelligent system with perhaps hundreds of thousands or millions of computer processors tightly linked together.
Today, engineering workstations– fancy, expensive machines – may have 2-4 processors, and the machines powering major Websites may have up to 128 processors. Hillis’s machines were nothing like this. The biggest Connection Machine ever built had 64,000 processors, and a 128,000 processor version was fully designed. Far short of the number of neurons in the brain, but still, pushing up toward the level of a workable compromise between traditional computing and brain-style information processing. Alternative parallel processing machines, like the Cray supercomputers, are specialized and inflexible, focused on doing the same exact operation on a large amount of data all at once. Hillis’s system, on the other hand, had the flexibility of the brain – each processor could do what it wanted when it wanted. Leading to the possibility of computational chaos,intelligent coordinated activity, or most intriguing, the combination of the two.
Thinking Machines Inc., founded in 1983 while Hillis was in the middle of his PhD work, was a remarkable organization. At its peak the research staff, about half the corporation, numbered in the hundreds. Despite the name of the company, there was not a coordinated company-wide R&D program aimed at making the Connection Machine think. Rather, there was a variety of research groups aimed at doing all sorts of different things with the Connection Machine, ranging from straightforward artificial intelligence research to simulation of fluid flow, computational immunology, experimental mathematics – you name it. Astrophysics, aircraft design, financial analysis, genetics, computer graphics, medical imaging, image understanding, neurobiology, material science, cryptography, subatomic physics…. Work on data mining – the automatic analysis of large and complex data sets – was particularly successful and later became a central part of the company’s business model.
The motivation underlying this diverse approach was simplistic but ambitious. “Clearly,” Hillis says, “the organizing principle of the brain is parallelism. It's using massive parallelism. The information is in the connection between a lot of very simple parallel units working together. So if we built a computer that was more along that system of organization, it would likely be able to do the same kinds of things the brain does.”
Of course, this approach to building AI presupposes that parallelism itself is something close to the chief ingredient of intelligence – that there is no further “secret sauce” required to make a mind come out of a distributed network of processors. Hillis believes that "intelligence is just a whole lot of little things, thousands of them. And what will happen is we'll learn about each one at a time, and as we do it, machines will be more and more like people. It will be a gradual process, and that's been happening." This is not so far off from Marvin Minsky’s Society of Mind theory, which holds that the mind is a collection of agents, each one taking care of a particular aspect of intelligence, and communicating with one another, exchanging information as required. Some AI theorists hold other views of course. Some maintain that it’s not the underlying computation mode that’s crucial, but rather that there are particular algorithms (of reasoning, memory, perception, etc.) that are really the key. Others argue that the right combination of “little things” is needed to give rise to the overall emergent patterns of coordinated activity that constitute real intelligence. But Hillis’s philosophy is a plausible one, and he had built a hardware platform and an organization well suited to validating or refuting his theory through ongoing engineering and research work. Most AI research is far less ambitious, consisting of small-scale, detailed work on one or another particular aspect of intelligence. In the history of AI, Hillis stands as one of a very small number of people who made a serious attempt to actually create a thinking machine.
And then the supercomputer industry died. Networks, it became clear, were the wave of the future. Networking large numbers of weak machines together, one had distributed computing, different from parallel computing in design, but somewhat similar in result. The last Connection Machine designed, the CM-5, was something like a computer network internally – it consisted of standard Sun Microsystems processors hard-wired together rather than traditionally networked. This was a big change from the earlier Connection Machines, which had been unique on the processor level as well as on the level of overall system architecture. In the end, Thinking Machines Inc. revised its business model, abandoning hardware altogether, focusing on selling their data mining software for use on distributed computing system composed of ordinary computers.
In 1994, the firm dispersed. The hardware side of Thinking Machines Inc. ended up at Sun Microsystems. Much of the data mining group ended up on Wall Street. Several Thinking Machines executives started TopicalNet, a company building text categorization software. And Hillis, after a stint working with the MIT Media Lab as an AI guru, abandoned the push for AI and went back to one of his earlier loves, toys and games. His new title: VP of R&D in the Imagineering Department of Walt Disney Corporation.
He entered this new phase of his career with wide-eyed optimism. "I've wanted to work at Disney ever since I was a child," he said. "I remember listening to Walt Disney on television describing the 'Imagineers' who designed Disneyland. I decided then that someday I would be an Imagineer. Later, I became interested in a different kind of magic--the magic of computers. Now I finally have the perfect job-- bringing computer magic into Disney."
Post Thinking Machines, his scientific work was becoming more practical in orientation – he was designing new technologies to underlie games and theme park rides rather than working directly toward digital intelligence. But at the same time, his philosophical side was hardly dormant. The far future came to occupy his thoughts more and more. In 1993, with Thinking Machines on its last legs, he wrote the following manifesto:
"When I was a child, people used to talk about what would happen by the year 2000. Now, thirty years later, they still talk about what will happen by the year 2000. The future has been shrinking by one year per year for my entire life. I think it is time for us to start a long-term project that gets people thinking past the mental barrier of the Millennium. I would like to propose a large (think Stonehenge) mechanical clock, powered by seasonal temperature changes. It ticks once a year, bongs once a century, and the cuckoo comes out every millennium."
The Clock of the Long Now. A clock built to last 10,000 years or more, powered by seasonal climactic fluctuations. The clock is not yet built, but a piece of land in rural Nevada has been purchased, the design is completed in detail, and construction of the components is underway. The Long Now Foundation (http://www.longnow.org/) is accepting donations online.
Hillis holds some 40 U.S. patents--for disk arrays, forgery prevention methods, a color camera and various software and mechanical devices. Among all his inventions, the clock is definitely one of the coolest – one that would make any MIT hacker proud. And it resonates with something deep and powerful in the human soul – the same aspect of human essence that finds the Cheops Pyramid more impressive than the Nintendo Gameboy, in spite of the incredible complexity of the hardware and software engineering involved in the latter. The Clock of the Long Now appeals to our embodiedness, to our embeddedness in space and time, whereas Hillis’s work on AI, like most AI work, ignored embodiedness and embeddedness and focused mainly on cognition, abstract thinking, on the most rarefied parts of the mind.
Abstractly, one could build a mind operating a thousand times faster than a human mind, or a thousand times slower. 10,000 years would mean something different to each of these differently time-scaled minds. But the mathematics and theory of AI would apply equally well to all of them, as would many of the same hardware engineering principles. The Clock of the Long Now is focused on palpable human reality, not the abstract mathematics of mind or the subtleties of hardware engineering. In fact it represents a step back from fancy modern electrical engineering. Modern technology provides few systems of 10,000-year durability, and so the design of the Clock of the Long Now required a number of purely engineering innovations.
One could easily portray Hillis’s interest in clock-building as a symptom of a mid-life crisis. After all, the older you get, the more interesting time seems to you. Perhaps, having failed to create real AI, the guy was reviewing his own life, and feeling his own death moving closer. Perhaps he found it comforting to remind himself how little it matters, from a 10,000 year viewpoint, whether any one human or any one company succeeds at doing any one thing. No doubt, there is an element of truth to this view. But this doesn’t seem to be a terribly large aspect of his motivation for pursuing the clock project – not as large, for example, as his sheer love of building cool stuff. And of course, both a thinking machine and a 10,000 year clock are Big Things – projects that appeal to the entrepreneurial, adventurous, overambitious soul.
The clock got all the media attention, but for Hillis personally it was never a full-time occupation. His new job at Disney was the bulk of his life. It was exciting – there was lots of money to build lots of great stuff, and he was involved in a lot of different projects. But if one reviews the time Hillis spent at Disney, one has a hard time finding any Disney project that really showcases his flair for large-scale innovative engineering. The details of his time at Disney aren’t open for public discussion, but, it’s not hard to reconstruct the story. Disney is a huge organization, and carrying exciting projects from concept to real-world implementation, without layers of bureaucracy getting in the way, probably wasn’t the easiest thing in the world. Anyway, last year Hillis left Disney, taking with him Bran Ferren, the head of the Imagineering group.
Ferren shares Hillis’s visionary streak, and also his interest in escaping from Internet Time into historical time. He, Hillis and Nathan Myrhvold (former CTO of Microsoft) have enjoyed hunting together for dinosaur bones. The conceptual clash between dinosaur bones and cutting-edge computer technology is just the kind of thing that leads brilliant minds in new directions. Ferren and Hillis are exploring these new directions via their new start-up Applied Minds, a company aimed at providing technology and consulting services to entertainment firms, presumably including Disney.
Having spent most of his career at the intersection between business and science, Hillis is acutely aware of the difficulties of balancing the different goals of these very different enterprises. There was a transition in the life of Thinking Machines, he observes, when it became less of an R&D shop and more of a real business – and at that point it became more and more difficult to move toward the goal of building real AI. When the firm became a real business, efficiency became important -- but creativity is exploratory, evolutionary, and fundamentally inefficient. Basically, in a company narrowly focused on making money, every minute of everyone’s day must be judged by whether it contributes to the bottom line. But the nature of the creative process is such that it just can’t be judged until it’s finished – there’s no way to tell which kooky train of thought or experimentation is going to lead to useful results.
What appealed to him about Disney, when he started out there, was the fact that it was a real business that was making real efforts to keep creativity alive within its walls. This was the express purpose of the Imagineering group. The defection of Hillis and Ferren, however, is an indication that Disney’s efforts in this regard have not been entirely successful. Applied Minds is a fascinating venture, which one suspects will do a better job of combining creativity with business focus than was possible inside Disney. But still, it’s worth noting how Hillis’s efforts have bifurcated: the Clock of the Long Now and Applied Minds each embody different aspects of his mind and soul, which were fused together in his earlier work with Thinking Machines.
Of course, if Applied Minds becomes an extremely profitable business, then it will be able to fund more and more interesting research over time. It will be interesting to see what happens in this regard. By remaining at Disney through the whole Internet bubble, Hillis missed out on his chance to cash in on the tech boom while it lasted. And given Disney stock’s poor performance in recent years, Disney stock options presumably weren’t a wonderful thing to own. The end-game of Thinking Machines Inc. did not result in making Hillis tremendously rich either. So, Hillis has a nice house, the back yard facing a beautiful lake, and is having an even nicer one built. But he isn’t currently in a financial position to build amazing new things on his own dollar. If Applied Minds puts him in this position, who knows what will emerge?
Perhaps something as fantastic as Thinking Machines Inc. – which remains the high point of Hillis’s story. It was a fascinating enterprise in many ways, but largely because of the way it fused science and business in the service of a single, immensely ambitious initiative. The Connection Machine was too big of a project to be initiated outside of industry, yet too innovative to be done without a large team of visionary scientists. The things Hillis is involved in now are less paradoxical and complex, and ultimately for that reason perhaps a little less intriguing. The Clock of the Long Now is a great work of conceptual art, with deep philosophical overtones and involving some neat engineering problems. Applied Minds is a real business through and through, using new science as required to provide customer solutions. These are both intriguing and sensible things, and yet they lack the Don Quixote –ish majesty of Thinking Machines Inc. and the Connection Machine, which to this day remain Hillis’s greatest creations.
A story like this reminds us that business, science, engineering and art are not fundamental divisions of the universe, any more so than earth, air, fire and water. Great innovations and enterprises stand outside these divisions, because they are crystallized around concepts that go beyond the temporary structures of any one human culture and society. The human race’s urge to create intelligence beyond itself – whether through building AI machines or through, as Hillis has recently discussed, putting “chips in the brain” – is a fundamental force that cuts across categories of human endeavor. Our need to understand our relationship with time is a similar fundamental force. And some human beings – like Danny Hillis – and some human organizations –like Thinking Machines and, to a lesser extent the Long Now Foundation -- reflect these fundamental forces in particularly elegant and powerful ways. To paraphrase what Hillis said about the frog’s heart he experimented with as a youth, it is remarkable that we can “do this coordinated activity… even though we are just a mass of cells.”