Lingua franca
Lingua franca Book Review     Fall 1998     pp. 12:13

Dr. Moravec's Cabinet of Wonders

BY JIM HOLT

ROBOT: MERE MACHINE TO TRANSCENDENT MIND
BY HANS MORAVEC * OXFORD UNIVERSITY PRESS * 224 PP $25 * NOVEMBER 1998




Robots with an inner life?
Hans Moravec, robot visionary
extraordinaire, says they're
just around the corner. And
you might even be one of them.

For some people, mere mention of the word "robot" is enough to set the pulse racing. I am not one of them. For me, robots are things that dwell in the banal literary culture of science fiction and in the lonely fantasies of computer hackers who long to be able to wake up in the morning and utter the words, "Well, what shall we do today, my mechanical friend?"

As a little boy growing up in his native Austria and then in Canada, Hans Moravec tinkered with toy robots and was haunted by the thought that he himself might be a robot. But Moravec is no ordinary robotics freak. Director of the Mobile Robot Lab at Carnegie-Mellon, he is a formidably clever fellow with a philosophical ax to grind--several axes, in fact. For one thing, he is an avowed Platonist. For another, he is a pan-psychist, a believer in the doctrine that consciousness infiltrates all matter. And, on the evidence of his writings, I think he might also be an undiagnosed Orphic.

In the reviewer's cliche, Moravec has written two books in Robot. One of them is a sober account of what has been been going on over the last couple of decades in robotics and artificial intelligence (AI)--fields in which the author ranks among the most accomplished and controversial figures. The other is a lotus-eating exercise in futurology and speculative metaphysics that sometimes left me breathless and sometimes made my flesh creep.

Let's take the sober book first. It is informative and also comes as something of a relief: If you haven't been paying attention to Al and robotics in recent years, you haven't missed a great deal. From the beginning, both endeavors have been long on hyperbole and short on results, though Moravec is not always candid in acknowledging this. In the 1950s, the future Nobel laureate Herbert Simon bragged that by getting a computer to deduce some theorems in elementary geometry, he had solved the mind-body problem. In 1970 a robot called Shakey was dubbed "the first electronic person" by Life magazine. Assembled by Stanford scientists, Shakey was a square cabinet on wheels with a TV camera and range finder mounted on top. On a good day the contraption could haltingly make its way down a straight corridor. "In from three to eight years we will have a machine with the general intelligence of an average human being," blustered Marvin Minsky, the father of Al.

Nothing of the sort has happened.despite the best efforts of scientists like Moravec. Arriving at the Stanford Al lab in 1971, he started working on a successor to Shakey. By the end of the decade the new robot was able to negotiate the rolling adobe outside the lab for about fifteen meters before becoming paralyzed with confusion. "Only in the 1990s" Moravec writes, "did complex mobile robots really begin to show some promise. Today there are dozens of wastebasket-sized machines wandering the halls of universities and other research institutions all over the world" Toss in a hundred thousand or so industrial robots performing repetitive assembly-line work like spot welding--but lacking sufficient consciousness to feel alienated from the product of their labor--and that's about it.

Why such pitiable progress? There are three reasons. First, the hardware. To get a computer with as much raw power as the human brain, you'd have to wire together a few dozen versions of the IBM Deep Blue machine that defeated Garry Kasparov in chess last year, and that would make for a rather gargantuan automaton. Admittedly, this may only be a temporary problem: Moravec predicts that brain-sized computers of brain-sized power will be available within two decades.

The second and more formidable obstacle is software. A robot worth the name must be able to perceive its environment, manipulate it, and reason about it. It took natural selection a billion years or so to engineer human vision. If you think mimicking that faculty in computer code is easy, Moravec's account of his own frustrations will disabuse you. Deep Blue could beat Kasparov, but no computer today, however festooned with sensors and servomechanisms, can do what any child can: move the chess pieces from square to square. And things are no better when it comes to reasoning, thanks to the notorious "frame problem." The most trivial inferences we make in everyday life involve background knowledge that potentially extends to the whole of human culture. The programmer who tries to represent this formally is faced with what Moravec calls a "combinatorial explosion" of facts interacting with facts. The context of the computer's reasoning must therefore be artificially limited to a "frame" of manageable size.That is why Al software shows no common sense-- why, for example, the best medical expert program will prescribe penicillin for a broken bicycle.

Finally, there is the matter of economic incentive. What profit is there in a robot that can do everything a human can? We already have complex, easily programmable, nonlinear servomechanisms that can be cheaply and easily reproduced by unskilled labor: us!

This isn't good enough for Moravec. He is unhappy having to inhabit an imperfectly designed body compounded of lime and jelly. "I resent the fact that I have these very insistent drives which take an enormous amount of effort to satisfy and are never completely appeased." he once told a reporter. Ten years ago. he wrote a book called Mind Children that described how we might shuffle off this mortal coil by downloading our minds into computers. A reviewer in The Washington Post called it "the most lurid book ever published by Harvard University Press." But this idea is not really so lurid. It is just the latest manifestation of the ancient cult of Orpheus. Our true life is among the stars, Orphism holds, but we are tied to the earth by our bodies; only by purifying and renouncing these bodies can we attain to the highest ecstasies. Though Moravec does not call himself an Orphic, this is precisely his philosophy, right down to his talk of our souls "transmigrating" into immortal robots.

In the latter chapters of Robot--the lotus-eating part--Moravec elaborates on his vision in bold, and surprisingly compelling, detail. He begins by sketching out the four generations of robots that will supposedly evolve between now and the middle of the next century.They will have the cognitive powers, respectively, of lizards, mice, monkeys, and, finally, men. By the year 2040, he claims, robots will not only be able to reason about the world by simulating it, they will also be able to reason about their simulations. In short, they will have an inner life.

I have the gravest doubts about these projections, but I am willing to crush them for the sake of entertaining what comes next: the end of capitalism (as the marginal value of human labor drops to zero), the transformation of physical space into cyberspace (as all matter is pressed into computational service), and the blurring of personal identities in a permanent orgy of experiential simulation. Now, Moravec's scenario of the future is unabashedly conjectural and even verges on science fiction.Yet it is all rooted in his sophisticated knowledge of physics and engineering.

And Moravec does not stint on the engineering details. He is already famous fur his conception of a robot with repeatedly branching arms that terminate in trillions of microscopic fingers capable of manipulating matter at the atomic level: a sort of fractally structured, animated bush. Here he generalizes the idea to robots with physically discontinuous parts smaller than dust motes that are powered and controlled by light beams. As is frequently the case with bravura engineering, a certain visual poetry emerges: Such a being, he writes, would appear to be "surrounded by an illuminated cloud that does its bidding as if by magic."

It is only when Moravec ventures into metaphysics that his logic gets wonky."But what is consciousness?" he asks on one page. "What is reality , anyway?" he asks on the next. He has already told us that he leans toward "physical fundamentalism"-- the dogma that only physical science deserves the title of true knowledge and that "other belief systems may have social utility for the groups that practice them,but ultimately they are just made-up stories." Presumably, then, the answers he proffers to these philosophical questions are to be taken in the spirit of "made-up stories."

Still, they won't have much social utility for philosophers. If minds can be extracted from brains and put into computers, as Moravec believes, then consciousness must be no more than the running of a very complicated software program. What the hardware happens to be--squishy neurons,hard silicon--doesn't really matter.

This theory of mind, called functionalism, used to be fashionable among philosophers and philosophically inclined cognitive scientists, but it has fallen into disrepute in the last decade. One of the problems with it was pointed out by the philosopher Hilary Putnam, who showed that any system with nonrepeating states--a waterfall, say--can be interpreted as a computer running any program you please. (Think of the physical states of the waterfall as an unending series of random numbers; depending on how this series is interpreted, It can be seen as encoding anything from the consecutive positions of the planets, to the complete works of Shakespeare, to the functioning of your next-door neighbor's mind.) So unless we are willing to attribute consciousness to waterfalls, we had better drop the notion that computation alone creates mind.

Moravec would rather attribute consciousness to waterfalls. He is, he says, a Platonist, and "Platonism holds that the soul is in the abstract relationships represented, not the mechanics of how they are encoded." Since the soul is just a mathematical object, a program, that anything sufficiently complex can be interpreted as encoding, it must follow that "anything can be interpreted as possessing ... consciousness and intelligence"--even, Moravec cheerfully concedes, a rock. It gets better. Since Interpretations too are mathematical entities, Platonically out there, the rock's consciousness is just as real as ours. Moravec, in short, turns out to be a pan-psychist.

And who's to say he is wrong? I, for one, wouldn't know how to go about proving that every object in the universe doesn't have a mind. But if this is true, I fear that the heroic labors of our robotic "mind children" in Moravec's futurological vision will be in vain. Why make the universe into mind, or at least a cosmic version of Bill Gates's house, if it already is mind?

Still, no reader will want to miss out on Moravec's clairvoyant perception of tomorrow's universe "teeming with unhuman superminds, engaged in affairs that are to human concerns as ours are to those of bacteria." Contemplating the Great Cosmic Revel to come, I was gloomily reminded that I would probably be dead before my mind could be downloaded into a vintage 2050 robot, and hence, unlike some of my younger friends, I would miss out on the eternal cyberparty promised by Moravec.

Born too soon, born too soon.