Section: technopop

Data, the chalk white android with the positronic brain, was pondering the meaning of life.

In a recent episode of Star Trek: The Next Generation, Data enlisted the aid of the Enterprise's Dr. Crusher in his struggle to crack the toughest nut of all. "I am curious as to what transpired between the moment when I was nothing more than an assemblage of parts," Data said, puzzled, "and the next moment, when I became alive. What is it that endowed me with life?" To which the doctor replied, "Scientists and philosophers have been grappling with that question for centuries without coming to any conclusion."

A neat dodge and a forgivable one. Even in The Next Generation's twenty-fourth century, apparently, the life-giving spark that transforms a bucket of bolts into a sentient being remains a mystery. One thing seems certain, however: When we build a machine so intelligent that it seems alive, we will fashion it in our own likeness, a la Dr. Frankenstein "a man of science who sought to create a man after his own image," to quote the 1931 classic film.

Data, a cross between the Tin Woodman and the "soft" man vilified by Robert Bly, is only the latest in a long line of humanoid robots. His family tree includes Robby the Robot, the talkative boiler in Forbidden Planet who could brew bourbon and speak 187 languages; R2D2 and C-3PO, the intergalactic Laurel and Hardy in Star Wars; Rachael, the sexy "replicant" who beds the hard-boiled protagonist in Blade Runner; and Ash and Bishop, the corporate droids in the Alien trilogy. All resemble humans in appearance or behavior; even the manhunting Terminator, a steely skeleton beneath its synthetic skin, looks and acts like Stanley Kowalski on steroids.

But Data and his mythic forebears live in a future that may never be. Increasingly, artificial-intelligence researchers are moving away from the human paradigm in their quest to create silicon consciousness from "an assemblage of parts." Intelligent machines will look less like Data and more like the microscopic mites that already share our apartments, subsisting on dust motes and traces of food. Stranger still, such robots won't display the egghead propensities of Data and his chess-playing, cocktail-mixing kin; they'll behave more like nest-building termites or swarms of bees.

In their embrace of creepy-crawly alternatives to the anthropomorphic model familiar from science fiction, visionary roboticists are following a path that could lead to a future of unutterable strangeness. The more radical among them suggest that present attempts to build battery-powered household helpers may ultimately yield robots that will struggle toward evolution's putative peak - human intelligence - and surpass it. If two of the field's visionary roboticists, Rodney Brooks and Hans Moravec, have their way, the human gene code may one day find itself out of a job, outperformed by artificial creatures that Brooks surmises "may be alien enough that we won't be able to appreciate them or they us."

Though they once shared an office at Stanford University's Artificial Intelligence Laboratory, Brooks and Moravec are now poles apart in theoretical terms. Brooks, an associate professor in the Department of Electrical Engineering and Computer Science at MIT, is without doubt the most vocal proponent of what is known as the bottom-up approach to artificial intelligence. He believes that sensorimotor skills, not higher-level thought processes, are the foundation on which intelligence is built. In other words, robots are going to have to learn to crawl before they learn to think. Brooks argues that intelligent inorganic life will evolve, like organic life before it, from simpler organisms. He dismisses as hopelessly misguided the top-down school of thought, which holds that artificial intelligence must set its sights on the supposed zenith of human intellectual achievement, pure reasoning.

Brooks's many-legged robots are inspired not by articulate, tool-using hominids that walk upright but by things that creep - ants, termites and the like. Brooks has drawn on studies by researchers who poke electrodes into the neural pathways of insects trudging on treadmills in an attempt to correlate the creatures' neural activity with their motor functions. "They come up with a set of condition-action rules that the insect seems to be following," says Brooks. "They validate them through computer simulations, and then we take those sets of rules and implement them on the robots." What crawl off the workbench are astonishingly lifelike robots that not only resemble bugs but behave like them, too.

Brooks is building a world that is not so much inhuman as "post-human," to borrow a phrase from Bruce Sterling's cyberpunk novel Schismatrix. He imagines a future in which the membrane between animate and inanimate is highly permeable, a Disney cartoon come to life where even chairs and tables sense and react. "Everything around us will be roboticized," says Brooks. "Normally passive elements will be active; `smartness' will be oozing everywhere. We'll be surrounded by intelligent objects in the same way that we're surrounded by plastics now. You'll go into a department store and buy this element and install it in your house; then, when you get three or four of them installed, they'll start communicating with each other, coupling and cooperating, and you'll end up with a smart house."

While Brooks contemplates ways to get small, Hans Moravec spends his days at a place that has staked its claim on the opposite end of the scale. Carnegie Mellon's Field Robotics Center, of which Moravec's Mobile Robot Laboratory is a part, is driven, according to Moravec, by director William Whittaker's passion for "big things - big mountains, big robots and big projects." Although Moravec has all but abandoned hands-on robotics for more theoretical pursuits, his contributions to the study of mobile systems, manipulators, computer vision and programming have proven invaluable to those at work on FRC robots.

The center's big machines are housed in two barn-size buildings nestled in a hollow on the university campus. The Tessellator, a rectangular vehicle with one long arm and omnidirectional wheels that allow it to float sideways with eerie grace, is under construction in the hangarlike building called the highbay; if approved by NASA, it will rewaterproof the underside of the space shuttle with toxic chemicals, a hazardous, laborious task for human workers.

Parked close by are the Navlabs (Navigation Laboratories), scary vehicles that use electric steering, computer-controlled hydraulic drives and a small fortune in technology - video cameras, laser range finders, sonar and supercomputers - to drive themselves. Funded by ARPA, the Department of Defense's Advanced Research Projects Agency, the Navlabs may one day end up transporting materiel in war zones or functioning as platforms for weapons systems. "As researchers, we're more interested in robotic delivery vehicles and smart highways, where cars drive themselves," says former Navlabs project manager Kevin Dowling, who takes pains to point out that Carnegie Mellon is not involved in classified military projects.

Even so, Navlab 2, whose chassis is based on a humvee that did a tour of duty in the gulf war, looks battlefield mean. A brutish, I-brake-for-nothing mountain of metal, it formerly sported a sign reading NOBODY ON BOARD.

Awesome as the Navlabs are, they're fossils to Moravec. He maintains that lumbering monsters intended for hazardous military or industrial applications, as well as mindlessly repetitive assembly-line robots, are single-celled organisms on the evolutionary time line of artificial intelligence. In Moravec's view, tomorrow belongs to the mass-produced, general-purpose robot for the home that he believes will be a reality by 1998, ready to weed the lawn, whip up a gourmet meal, give the car a tuneup and vacuum. (The Jetsons' vacuuming robot springs, apparently, from a deep-rooted desire; Brooks claims that marketing surveys have determined that what the American consumer wants out of a domestic robot, more than anything else, is a clean carpet.)

Moravec envisages a stainless-steel servant with five wheeled legs, two arms ending in humanoid hands, sonar sensors and a pair of color-TV cameras for eyes. The robot's ability to "maintain representations, at varying levels of abstraction and precision, of the world around it" is all-important, Moravec asserts in his 1988 book Mind Children: The Future of Robot and Human Intelligence. He writes, "In these internal models of the world I see the beginnings of awareness in the minds of our machines - an awareness I believe will evolve into consciousness comparable with that of humans."

Moravec, needless to say, spends much of his mental life in the distant future. He takes up where Brooks leaves off, theorizing a "post-biological" future in which Homo sapiens has been rendered irrelevant by its highly evolved artificial offspring - superintelligent mechanisms that almost certainly will not inhabit humanoid bodies and may not resemble anything we have ever seen.

In Mind Children, Moravec conjures trillion-limbed machine flora - robot bushes whose stems branch into ever more delicate, ever more numerous twigs, culminating, finally, in a dizzying array of microscopic cilia. "Such machines could carry on our cultural evolution, including their own construction and increasingly rapid self-improvement, without us, and without the genes that built us," Moravec writes. "When that happens, our DNA will find itself out of a job, having lost the evolutionary race to a new kind of competition."

A marathon runner whose wardrobe leans toward Reeboks and leather jackets, Brooks exudes a jittery, adolescent energy. Over sushi he answers questions while rolling - and rerolling and re-rerolling - his paper napkin into a pencil-thin tube. His unruly chestnut curls and bright, intense eyes make him look younger than his thirty-eight years. His accent, flat and twangy with the strains of Adelaide, Australia, where Brooks grew up, turns even the most innocent remark into a wisecrack.

"I'm much shorter term than Hans," says Brooks around a bite of sushi. "I only think a couple hundred years ahead." Brooks wants to see his ideas soldered together into something he can touch. "I don't want to think about things I'm not going to get a chance to build," he says.

In pursuit of attainable goals, Brooks has focused for the most part on near-term, low-cost projects. Working small is one way of keeping budgets manageable, and the mechanical creatures turned out by the Mobile Robot Project under Brooks's guidance have set new standards for miniaturization. Squirt, a wheeled, cubic-inch-size mite built in 1988 to prove just how small a robot could be, thinks it's a cockroach. The microbot cowers under furniture or in dark corners, listening for loud sounds; when the noises have faded, Squirt creeps warily from its hiding place in search of new cover, near the noises' point of origin.

Research scientist Anita Flynn, who collaborated with Brooks on Squirt, believes that electronic cockroaches are an intermediate stage in the evolution of gnat robots - millimeter-size robots etched on computer chips. She dreams of a computer chip that can "get up and walk." In a world measured in microns, computing power comes cheap; through very large- scale integration, the processing power of a computer can be scrunched onto an integrated circuit chip. Propulsion is another matter: The bulk of Squirt's mass is taken up by motors and batteries. But if a robot's hardware could be squeezed down to the size of its computer-chip brains, armies of invisible minions would be at our beck and call.

As part of her Ph.D. thesis, Flynn is investigating micromotors - microscopic motors chiseled out of silicon and set spinning when electricity is zapped through a ring of piezoelectric crystals. Miniature - though far from microscopic - piezoelectric motors are already used in some autofocus cameras. Moreover, researchers at Cornell University and UC Berkeley are experimenting with structures carved in layers on silicon chips; a chemical bath eats away the scaffolding, and the structure unfolds and locks in place, like a self-erecting tent. "It's not a robot yet," says Brooks, "but the individual elements have been demonstrated."

Brooks, together with Flynn and then-graduate student Lee Tavrow, applied for a patent for the "3-D fabrication of a gnat robot." In a 1989 essay titled "Twilight Zones and Cornerstones: A Gnat Robot Double Feature," the trio declared: "We want to build gnat-sized robots, a millimeter or so in diameter. They will be cheap, disposable, totally self-contained autonomous agents able to do useful things. . . . Gnat robots are going to change the world."

But not tomorrow. For now, speck-size robots whose computers, motors, sensors and limbs can fit on a silicon chip are a distant mirage - not that that stops Brooks and his colleagues from thinking up jobs for gnat robots. In "Twilight Zones," the authors sketch scenarios involving medical gnat robots that unclog arteries, mend severed neurons or crawl across eyeballs to perform retinal surgery. Mass-produced using techniques already employed to fabricate computer chips, gnat robots would be sufficiently cheap for consumers to assign millions of them, all working in concert, to a single task, no matter how mundane. Imagine two-tone gnat robots that cling, like infinitesimal flecks of paint, to the exterior of your house and, when you're tired of the color, "repaint" it simply by rolling over; gardening gnat robots that clamber up grass blades and nibble your lawn short; coiffure gnat robots that roost on your scalp, rearranging your hair and holding it in place.

Brooks's interest in robots that can dance on the head of a pin is rooted in his realization that in the natural world, small size and smaller brainpower aren't necessarily barriers to evolutionary success. "I spent many hours with Hans Moravec at Stanford in the Seventies, when he was working on his Ph.D. thesis, a robot called the Cart," Brooks recalls. The Cart was the first vision-equipped robot to venture into the crowded, confusing world beyond the static, neatly geometric laboratory environment. Remotely controlled by computer, it scanned its surroundings with a TV camera, locating objects and planning its route. After ten or fifteen minutes' worth of rumination, it would lurch forward a yard, stop and take stock of its place in the world. "Using one of the biggest computers around at the time, it took five hours to reach its destination," says Brooks. "It didn't seem too impressive, compared to an insect flying around."

How, wondered Brooks, could a more or less brainless creature accomplish in seconds what robots linked to supercomputers puzzled over for hours? Certainly, the conundrum had little to do with computer technology, whose giddy speedup had seen ENIAC - a room-size, vacuum-tube-powered monster that in 1946 was the first programmable electronic computer - shrink to a transistorized machine in the Fifties, a boxful of integrated circuits in the Sixties and a chip-driven PC in the Seventies. Perhaps the flaw lay in the premises on which the study of artificial intelligence was based, assumptions cemented in place at the legendary Dartmouth conference in 1956.

For two months, mathematician John McCarthy, computer scientist Marvin Minsky and other brash young minds grappled with the question posed by Alan Turing, one of the founding fathers of computer science: Can machines think? Out of this watershed colloquium came the classical top-down postulate that any thinking machine must emulate the problem-solving ability associated with rational thought. For a machine to be truly intelligent, adherents argue, it must possess the symbol-manipulating, rule-implementing proficiency that characterizes human reasoning.

Robotics, presumably, was mere brawn; teaching a machine to walk should be child's play after untangling the Gordian knot of human thought. An intelligent robot would construct a mental map of its environment, plot a course from point A to point B and, after much deliberation, take one small step for tin man. Unfortunately, while computers soon learned to play chess, solve calculus problems and prove geometric theorems, their motor skills were put to shame by toddlers.

In his book in progress, The Age of Mind, Moravec recalls experiments in the late Sixties and early Seventies by McCarthy's research group at Stanford and Minsky's at MIT: "[They] connected television cameras and robot arms to their computers so `thinking' programs could begin to collect information directly from the real world. The early results were like a cold shower. While the pure reasoning programs did their jobs about as well and fast as college freshmen, the best robot control programs . . . took hours to find and pick up a few blocks on a table top, and often failed completely, performing much worse than a six-month-old child."

Decades later, computers play chess on a grandmaster level but no robot yet exists that can locate and manipulate actual chess pieces with the unthinking ease that comes naturally to human infants. Moravec, Brooks and many other artificial-intelligence theorists now believe that sensory perception and motor control, not rational thought, are what the human brain does best. None of us uses trigonometry to calculate the trajectory of his hand before lifting cup to lip; evolution has consigned such functions to the lowest levels of unconscious data processing. By comparison, as Moravec notes in The Age of Mind, rational thought of the sort used in playing chess is a relatively recent arrival, "perhaps less than 100,000 years old."

Why not do away with the symbol-juggling, cognitive brain altogether, Brooks speculated, and route sensor data directly to motor responses? Of course, the resultant robot would be more insectoid than humanoid - an affront to those who contend that humankind is the crown of creation - but beetles and their brethren have proven surprisingly durable in the Darwinian road test. Brooks threw down the gauntlet at "Pixels to Predicates," a 1983 seminar. "I gave a talk," he says, "and I drew this diagram - this was my divine inspiration, the basis for everything I've done since - in which I had the world out here, feeding into perception, and then I had another box over here called action, and both boxes overlapped completely. Up here, I had an observer, observing the interaction of perception and action; cognition is in the mind of the observer, not in the system itself. What I'm saying is that there is no box inside me that does the cognition; an observer may impute such a box - `Ah, he's doing that because he's decided that' - and if you ask me, I'll give you an explanation, but it's just this thin layer of consciousness trying to make up a story about what happens down below. In fact, cognition is nowhere."

An assistant professor at Stanford at the time of the fateful lecture, Brooks left shortly thereafter to take up residence at MIT, where he, Anita Flynn and a graduate student named Jonathan Connell set to work on a robot that embodied his theories. The fruit of their labors, in 1986, was a squat, wheeled contraption named Allen, who resembled a trash-compacted R2D2, his electronic innards exposed for all to see.

Allen's software was based on Brooks's concept of stimulus-response behavior, refined into a scheme dubbed "subsumption architecture." Subsumption architecture consists of a stack of behavior modules - low-level, real-time behaviors ("avoid obstacle") on the bottom; higher-level, goal-oriented directives ("explore") at the top - that permit a robot to generate complex behaviors from simple responses. Allen, for example, had three layers of control: The first required that the robot avoid both static and dynamic obstacles; the second motivated it to wander about randomly (all the while avoiding obstacles); and the third suppressed the desire to meander aimlessly and directed it to move toward distant points located by its sonar range sensors.

Behavior, rather than being controlled by a centralized brain, emerges from a series of parallel activities, one behavior subsuming the others as circumstances require. The elaborate planning and mapping that forces conventional laboratory robots to rely on the processing power of off-board computers, thus slowing them to a crawl, is jettisoned. In fact, the computer to which Allen was yoked soon proved unnecessary: Brooks's approach, being simpler, required far less number crunching.

More recent generations of robots have earned Brooks's group its unofficial tag (the Insect Lab) and motto (Fast, Cheap and Out of Control). Genghis, the first of the lab's legged progeny, was the brainchild of Colin Angle, a gifted undergraduate who had come to MIT to "major in whatever lets me build the coolest stuff." And Genghis, so named, says Angle, because it "stomps over things" like the Mongolian conqueror, is nothing if not cool. Bequeathed twelve layers of behavior, the foot-long robot scrambles over mountains made of textbooks and plods inexorably on, pursuing human prey with the aid of a cluster of sensors that gives its "head" a distinctly buggy appearance.

Attila, Genghis's direct descendant, calls to mind a four-pound tarantula in a suit of armor. Rigged with ten microprocessors, twenty-three motors and 150 sensors, the high-tech arachnid can hoist itself onto ten-inch-high ledges and scrabble up near-vertical inclines. When powered up, Attila wriggles its legs in a manner disconcertingly reminiscent of a dying insect. In actuality, the robot's flailings are a means of gathering information about its environment. Within minutes, the six legs have "learned" how to coordinate themselves, and Attila is on the prowl - in search, no doubt, of a Lilliputian Tokyo to terrorize.

Emergent behavior like Attila's, the result of independent limbs following simple rules, manifests itself as emergent social behavior, or "swarm intelligence," when groups of behavior-based robots interact, their decentralized activities working toward a common goal. Think of bees: Their almost robotic activity - governed, again, by simple rules - produces wonderfully complex hives. Drawing on studies conducted by ethologists analyzing ant colonies as self-organizing systems, Brooks and his associates are attempting to program twenty identical robots to behave like social insects, collaborating on collective tasks. Small, boxy things that scoot along on wheels, the better to seize objects in their pincerlike grippers, they resemble heavily armored toasters. Brooks hopes they'll act like bees.

At IS Robotics, Brooks's start-up in Cambridge, Massachusetts, MIT graduate turned company president Colin Angle and a small staff are pursuing the roboticist's dream of "a robot invasion of the solar system" - "nerd herds" of galactic explorers. Like the late science-fiction writer Isaac Asimov (who coined the word robotics), Brooks dreams of off-world colonies. But he hopes to realize his visions of extraterrestrial settlements in a manner unlike anything imagined by Asimov. Classic Asimov titles like I, Robot are populated by chrome-skinned robots with photoelectric eyes and positronic brains (that's where Data got his); Brooks's far less anthropocentric future would find human explorers sharing new worlds with artificial intelligences possessed of nonhuman psychologies and nonhuman bodies.

"Let's grab a salad," says Brooks, "and I'll tell you a story about termites." He moves purposefully toward the salad bar, lacquered bowl in hand. "There's a particular termite that builds large structures by following some very simple rules," he begins, piling unfamiliar, seaweedy greens in his bowl. "These termites pick up wood chips, chewing them in their mandibles and then randomly spitting them out, imprinted with a pheromone that dies away over time. Now, the only additional rule is that if a termite detects that pheromone, the probability of it disgorging a spitball goes up, so that as you increase the density of termites in an area, the chance of them coming across spitballs which still have pheromones that haven't yet dissipated goes up.

"You start to see little accumulations of two, three, four spitballs," Brooks continues, "and as you increase the density even more, you get a perfect hexagonal pattern of spitballs, because if you go through the differential equations, that's the maximum density of these piles you can get. The termites are not measuring this hexagonal pattern; it all comes from this statistical picking up and dropping, with this high probability. So you can actually get complex structures produced by individuals following very simple rules; you don't have to have a global plan in the minds of robots in order to get these structures."

Bowl mounded high, Brooks leads the way back to the table, where he sets to work on his salad. "The challenge," he says, "is to find similar sorts of rules if you want, say, hordes of robots to produce nice troughs for lunar dwellings. NASA would have to bury the habitation modules for a lunar base below the surface to avoid the radiation. My idea is to send a horde of these social robots five years ahead of time. You don't control them from Earth; they just dig these long trenches that are programmed into their nature. If one robot breaks down, another one pushes it aside - just like ants building an anthill, carrying off the dead ants to get them out of the way."

BROOKS'S REFLEX-DRIVEN, OR "BEHAVIOR-BASED," approach to AI is not without its critics. Marvin Minsky, the field's patron saint, once twitted him by saying, "Hey, maybe we should all just devote ourselves to replicating insect intelligence."

Moravec, who enjoys an affable sparring relationship with Brooks, offers a more charitable critique of his colleague. Ensconced in his office in the Mobile Robot Lab, Moravec leans back in his chair pensively; a finger smooths an eyebrow obsessively, as if to iron out the wrinkles in his thoughts.

Permanently shadowed eyes - Moravec keeps hacker's hours, working late into the night - dominate a pale, round face. It is a face one sees frequently in paintings by Austrian artists, a fact that falls into place when Moravec remarks that although his family moved to Canada when he was four, he was born in Kautzen, a small town in northern Austria. Where Brooks is animated, contentious, Moravec is measured, reflective. But his composed demeanor conceals a restless, insatiably curious mind that never winds down; asked a question, he invariably responds with a chapter.

"Rod's approach is the first approach," Moravec says. "It's not a dead end, since any robot will want reflexes; your ancestors were once worms and worked entirely by reflex, and now those same reflexes are what makes your hand withdraw, fast, if you put it on a hot plate. Robots that model their worlds and reason about them will still have reflexes, so Rod's approach will be a small part of these future robots. But I don't think you can build a robot that's going to run a company based on insect reflexes."

Moravec's strategy bears a closer resemblance to traditional AI than to Brooks's bottom-up approach. Moravec believes that "evidence grids" - internal models of the world, collaged from sensor data, that will enable robots to navigate their surroundings - are an essential step toward the creation of mobile machines with human intelligence. In 1988, he test drove his theory with Uranus, a squat, wheeled robot whose distinguishing feature is a pair of TV-camera eyes attached to a mast jutting from its topside. Using stereo vision and sonar, Uranus was able to paint a remarkably accurate picture of the world, permitting the robot to thread its way through cluttered obstacle courses at the eye-blurring speed of about three quarters of a mile per hour.

Recently, Moravec stumbled on a shortcut to the perfection of his evidence grids and hence, he believes, to artificial intelligence itself. The roboticist spent his 1992 sabbatical at Cambridge's Thinking Machines Corporation, home to W. Daniel Hillis's massively parallel computer, the Connection Machine - an inscrutable black monolith studded with winking red lights that looks like Darth Vader's refrigerator. When Moravec arrived, eager to begin work on a 3-D version of the grid, the latest generation of Hillis's machine was not quite ready. "I decided to develop my program on a regular workstation with an eye to transferring it later," says Moravec. "As it turns out, I wrote the best program I've ever written in my life; it's so awesomely fast that I never had to transfer it. This is the program that's going to control the robots that dust and vacuum your house this decade."

Such cymbal-crash pronouncements are typical of Moravec, who shares Brooks's love of the grand gesture. His evolution from practical roboticist to cloud-dwelling theorist nearly complete, he spares no hyperbole in his descriptions of the coming post-biological age. Mind Children, for example, opens with a statement calculated to cause apoplectic seizures in humanist quarters: "I believe that robots with human intelligence will be common within fifty years."

Human culture, according to Moravec, represents the first nongenetic instance of data transferal from one generation to another; animals rely on their gene codes to pass useful behavior on to their descendants, but humans store accumulated knowledge in books, still and moving images and, increasingly, electronic media. We have already outpaced biological evolution, he asserts, and will beget our mind children - robots possessed of human-level intelligence - as soon as the processing power required for humanlike computers becomes available.

Moravec, an unreconstructed mechanist, believes that the mind is simply a soft machine; equivalence, therefore, is merely a matter of computing speed. Ten tera-ops should do the trick, by Moravec's reckoning. That's 10 trillion operations per second, light-years beyond a state-of-the-art PC chip like Intel's recently debuted Pentium, capable of 112 million instructions per second. Based on his calculation that there has been a trillionfold increase in the amount of computation a dollar will buy since the invention of the punch-card tabulator shortly before the turn of the century, Moravec projects the arrival of a ten-tera-ops machine by 2010.

At which point buckle your seat belt, because evolution's warp drive is going to engage. Astronomically intelligent robots, "looking quite unlike the machines we know," writes Moravec, "will explode into the universe, leaving us behind in a cloud of dust." Robots capable of engineering their own evolution will quickly surpass human equivalence, he theorizes, leapfrogging up the scale of intelligence to a level that defies human comprehension. Down-loading human consciousness into computers is one of Moravec's strategies for keeping pace with his superevolved creations. With dubious relish, he describes a robot surgeon removing the crown of a person's skull and using high-resolution magnetic-resonance measurements to create a computer simulacrum of the subject's neural architecture. Layer by layer, the brain is scanned, simulated and surgically removed. In time, the brainpan is empty; the robot disconnects all life-support systems, and the body goes into convulsions and expires. The subject's consciousness, meanwhile, is curiously unconcerned, wandering wraithlike through cyberspace. Eternity is his or hers to spend, pinwheeling past constellations of data or down-loaded into an android whose servomechanical muscles never tire, whose memory banks are never short-circuited by age.

The Age of Mind - described by the author as "a more coherent sequel" to Mind Children - is subtitled Transcending the Human Condition Through Robots. Assuming a priori the revolution in machine intelligence foretold in his earlier book, Moravec time warps his reader to a universe watched over by godlike machines who may choose, for old times' sake, to digitize the human race and preserve it in a computer-generated world - the virtual equivalent of the Kryptonian city in a bottle in Superman's Fortress of Solitude.

"I go past human equivalents into whole robot ecologies, from microbots to enormous machines," says Moravec. He hypothesizes the emergence of robot companies that churn out all manner of necessities and luxuries, creating a utopia in which the workweek dwindles to nothing and every human need is attended to. But conspicuous consumption proves no match for robotic efficiency, and as ever smaller amounts of their energies are spent on their Terran creators, the machines turn their attention heavenward. Boldly they go, turning whole planets into the raw material for their automated manufacturing processes. It is a sci-fi take on laissez-faire capitalism that would gladden the heart of William F. Buckley Jr.

"All of this will happen in the context of our current social structures," says Moravec. "I think it's a very natural way for it to happen. The robots' behavior is shaped entirely by the requirements of competition, except that the mechanisms that further this are ultra-intelligence, exotic physics, incredible technology." Moravec has a discomfiting habit of shifting into the present tense when discussing such matters, as if their inevitability were a given. "After a while," he continues, "they've expanded to cover all available space, and activity takes place at sub-sub-subatomic levels, but because matter is used so efficiently, the world inside the computer simulation, or cyberspace, is much bigger than the physical world ever was. At which point the robots become obsolete, and we enter the age of mind, where there's no more overt physical activity at all; everything is happening at such a subtle level that it's essentially computation."

Moravec's universe-size computer simulation is inhabited by down-loaded human minds and "unhuman disembodied superminds, engaged in affairs . . . that are to human concerns as ours are to bacteria." Humanity, in such a cosmos, will be a passing thought in the mind of a cyber-god. "Human beings will be the smallest of the small in this universe, although occasionally some supermind will re-create them just by thinking about them," says Moravec matter-of-factly. "To the supermind, it will be just a stray thought, of course, but to the human beings it will be a whole universe."

IN THE PRESENT, HOWEVER, MACHINE INTELLIGENCE is typified by Polly, a stumpy, barrel-shaped robot cobbled together by Ian Horswill, a graduate student in Brooks's Mobile Robot Project. Polly, whose name indicates the parrotlike nature of its linguistic abilities, displays its visual acuity by giving tours of the AI lab in a fuzzy, metallic voice Horswill describes as "Stephen Hawking on Thorazine."

Polly is easily confused. During a demonstration, the robot took wrong turns and incorrectly identified various areas, requiring frequent assistance from its chagrined creator. "Some of the hardware's acting up," Horswill said. "Polly is hallucinating random objects." Polly, however, remained sanguine. "Follow-me," it chirred. "I-can-avoid-obstacles-follow corridors-recognize-places and-navigate-from-point-to-point. By-the-way-I-don't-understand-anything-I'm-saying."

Silicon godhood, for Polly at least, seems far away indeed. Nonetheless, unrepentant anthropocentrists shouldn't feel too reassured. In the field of artificial intelligence, the technical and the philosophical intertwine; Brooks's bug-bots and Moravec's post-biological flights of fancy have done much to dislodge the age-old notion that humanity has been given dominion over all things, for all time.

Christopher Evans, writing in 1979, on the eve of the PC revolution, foresaw the psychological consequences of Brooks's and Moravec's work carried to its ultimate conclusion. "How will we feel at the realization that the gap between ourselves and the Ultra-Intelligent Machines is unbridgeable," he wrote in The Micro Millennium, "and that any advances we make will be easily outdistanced by their superlative endeavors?" Not that the view of ourselves propounded by Brooks and Moravec isn't unsettling enough as it is. If, as Brooks suggests, "cognition is nowhere," what, then, is human consciousness?

The skin crinkles impishly at the corners of Brooks's eyes. "Consciousness," he says, "is this high-level interface that God put in there so he could check our thoughts quickly; he didn't want to have to mess with the details one by one. This way, he can find the good and bad thoughts easily."

PHOTO: Hans Moravec (MICHAEL LLEWELLYN)

PHOTO: RODNEY BROOKS POSES WITH GENGHIS II, ONE OF HIS INSECTOID CREATIONS. (MICHAEL LLEWELLYN)

~~~~~~~~

By MARK DERY

MARK DERY writes about technology and fringe culture. He is currently at work on `Cyberculture: Road Warriors, Console Cowboys and the Silicon Underground,' a survey of computer subcultures to be published by Hyperion next year.