The robot wars

How soon are we going to see thinking, feeling robots? The answer may lie inside a mechanical head at MIT

by Chris Wright

WHEN NOVA LABORATORIES launched its Strategic Artificial Nuclear Transport (SAINT) line of robots in the 1980s, the company knew it was making more than headlines.

Touted as "the most sophisticated robot on planet earth," SAINT was developed for military purposes but also possessed some decidedly non-military skills, such as mixing the perfect gin and tonic. On the day of SAINT's unveiling, as five prototypes trundled out before a cheering audience, a sense of history crackled in the air. What the Nova people couldn't have imagined, though, was that a few days later, one of the robots, Number 5, would start to develop thoughts and feelings of its own, and would declare, to everyone's astonishment, "I'm alive!"

Or not.

What we're describing here is the plot of the 1986 movie Short Circuit. In the real world, robots are a long, long way from achieving the self-awareness of Number 5. And they're even further from being compassionate, ambitious robo-quipsters like Andrew, the Robin Williams character in the new film Bicentennial Man. In fact, not only has robotics technology failed to produce a thinking, feeling machine, but it has struggled to come up with a robot that can make it across a room without a programmer to hold its hand -- if it has a hand at all.

As we stand on the threshold of the 21st century, the chilling words of HAL, the malevolent artificial intelligence of 2001: A Space Odyssey -- "I'm sorry, Dave, I'm afraid I can't do that" -- have taken on a literal ring. "No, Dave, I'm serious. I really can't do that."

"Robots are much less sophisticated than people think," says Jeanne Dietsch, of the New Hampshire-based robotics company ActivMedia. "People think in Star Wars terms. But we're nowhere near that."

Surely it's not such a stretch, though, to be thinking in Star Wars terms more than 20 years after Star Wars was made. Back then, the Internet was unimaginable; today it is almost mundane. In the same time period, genetic engineering has advanced to a point we once envisioned only in dreams. We live in an age of unbounded technological innovation. So what happened to HAL, C3PO, Commander Data? What the hell happened to Rosie the Robot?

It's a good time to be asking this question, and not only because the year 2000 is just around the corner. After a period of robo-boom and robo-bust, the industry is regrouping -- indeed, there is a sense of nervous excitement in the robotics community right now. Many believe the field is on the brink of a technological revolution similar to the one the computer industry has undergone in the past two decades.

"The explosion is pending," says Henry Thorne, inventor of a new household robot named Cye. "This is just the beginning."

"I think we're finally starting to see robots that are breaking the mold," says William Townsend, CEO of Cambridge robot manufacturers Barret Technology. "As far as the technology goes, the sky's the limit."

AMERICANS ARE actually surrounded by robots -- 98,000 of them at the last count. The number rises every year. This year, through September, a billion dollars' worth of robots -- of various shapes, sizes, and applications -- have been put to work in the US. That's up from $300 million worth in 1987.

Most of these are industrial robots, as has been the case since the 1950s, when General Motors installed George Devol and Joe Engelberger's "Unimate" -- the first robotic worker -- in its auto plants.

Since then, robots have proven to be vastly efficient at repetitive tasks such as spot-welding and mail-sorting. But they've been lousy at anything that requires adapting to unforeseen conditions. There are a handful of advanced robots endowed with the skills necessary to leave the factory walls and go out into the real world, but until recently these have been prohibitively expensive for everyone but a few government agencies.

One area where mobile robots have excelled is in exploring territory where humans can't, or won't, go. NASA's Mars-exploring Pathfinder robot is one high-profile example. Another is the Navy's Magnum robot, which recently rummaged through the wreckage of the ill-fated EgyptAir Flight 990 and helped retrieve its black boxes. On a more mundane level, buildings across the country are patrolled by security robots -- basically trolleys equipped with cameras and sensors. And a few robots have recently emerged that are able to perform menial household tasks, such as Electrolux's robotic vacuum cleaners and lawn mowers.

A company named Probotics, for example, has invented the aforementioned Cye, a mini robot that can vacuum floors while its owner uses a point-and-click computer program to steer it. Then there's ActivMedia's PeopleBot, hailed as "the first affordable intelligent mobile robot," which can "listen for phrases or sounds it recognizes" and "navigate without running over toes or into furniture." But PeopleBot is only as smart as the person who programs it. If it could speak on its own behalf, it would probably say something like "Duh."

These robots aren't labor-saving devices so much as they are novelty items, says Cye's inventor, Probotics CEO Henry Thorne. "They have to have some utility," he says, "but they've also got to be fun. We wouldn't have made Cye if it was just about saving labor. That would have been boring. Robots are fun."

Until robots get their act together, entertainment is likely to be one of the industry's biggest money spinners, keeping companies afloat while they come up with more-advanced technology. The most renowned funbot currently on the market is Sony's lap-dog-size electronic pooch Aibo, which has a starting price of $2500. Aibo's main achievement is that it can recognize pink, which enables it to chase a pink plastic ball around. It can also bark, wag its tail, and . . . well, that's pretty much it.

Nonetheless, Sony is banking on Aibo's being a hit; other companies are predicting similar success for their own robo-critters. Among these are Mitsubishi's robotic version of a fish called a coelacanth, and a super-smart robot kitten named Robokoneko, in the works from a company named Genobyte.

WHEN WE think of "robots," however, we don't just think of an automatic welding arm, or a mechanical kitten. We think of androids. Man-machines. The fictional robots that have most captured the popular imagination are the ones that are like humans (see "Notable Robots"). The quest to create a genuinely humanlike being -- in our own image -- is perhaps the grandest project ever undertaken by humanity.

Right now the race to build the first genuinely humanlike robot has reached the intensity, as one researcher puts it, of an "arms race." Faustex Systems has developed a martial-arts robot called the Hyperkinetic Humanoid, a lifelike sparring partner, which will soon, we are promised, "do human feats ranging from common tasks to performance arts such as music, dance, and theater." Florida Robotics has a robo-gogo girl named Ursula, a life-size robot that "walks, talks, dances, plays music, and more." Looking a bit like a tinny Deborah Harry, Ursula is actually quite sexy, for a machine. Don't laugh: many predict that sex (see "Sweet Things" ) will be one of the early moneymakers in the robotics industry. "Porn is always a technology leader," says Jeanne Dietsch. "Robotic sex toys will probably be among the first profitable robots."

But other humanoid-robot projects have more ambitious aims. Japanese researchers in particular have something of a fetish for humanoid robots, partly because the Japanese see a real need for them. Faced with a combination of falling birthrates, a rapidly aging population, and a disinclination for hiring foreign labor, many in Japan view household androids as a vital part of their future.

Robotics experts in the US agree, though for different reasons. As William Townsend says: "Robots make great companions for the elderly. If you're 80 years old and you need help going to the bathroom, you'd be a lot happier having a robot help you than a person. It's less degrading."

Of the humanoid robots in development, Honda's P3 is one of the most promising. Standing five feet three inches tall and weighing in at 286 pounds, P3 is an impressive sight: it looks like an astronaut from the future, or an alien. But the most striking aspect of P3's design is the simple fact that it has working legs and is thus, unlike most wheeled robots, able to climb stairs. In the race to create viable household robots, legs have been -- quite literally -- one of the biggest stumbling blocks. Honda literature claims that the P3 will eventually "be used in daily life" and "result in added value to society."

The key word here is "eventually." Despite the 13 years and more than $100 million that have gone into its development, P3 is nowhere near ready to go. Integrating bipedal movement, perception, and a coherent world-view is still way, way beyond current technology. P3 may look like science fiction, but it's still hampered by the same old limitations that have haunted the robotics industry for years.

THE WORD "robot" was coined in 1921 by the Czech playwright Karel Capek, who envisioned in his play R.U.R. (Rossum's Universal Robots) a world peopled with indentured artificial humans. Then came the 1926 film Metropolis, which posited a city filled with mechanized inhabitants. In 1950, sci-fi writer Isaac Asimov published his massively popular collection of stories I, Robot. By then, the concept of robots was so familiar that it seemed they already existed.

And, in some sense, they already did. In 1948, a British psychologist named W. Grey Walter developed light-seeking robotic "turtles." Though crude by our standards (radio tubes for brains, photo tubes for eyes), Walter's automata displayed apparently sophisticated behavior, avoiding obstacles and simulating a level of social interaction somewhere around that seen in junior-high football players.

In the '60s, a group of researchers at Johns Hopkins University applied emerging transistor technology to robotics, creating a robot named the Beast. Like Walter's turtles, it did little more than roll about trying not to bump into things.

By the '70s, with the advent of computer technology, thinking, feeling robots began to enter the realm of real possibility. As artificial intelligence (AI) blossomed, researchers tackled the problems of reliable mobility. Stanford University introduced a mobile robot named Shakey, and another called Cart. These projects consistently ran into trouble, though. In fact, if left to their own devices, they ran into pretty much anything and everything in their paths.

Still, researchers and investors soldiered on into the '80s, buoyed by the explosive advances being made in AI, waiting for the time when computer technology would unlock the secrets to perception and -- more tricky -- intention. Indeed, in these heady days great strides were made in robots' mobility, visual and auditory reception, speech, and basic comprehension.

"There was a lot of hope in those years," says Donald Vincent. "There were indications that this would be a boom industry in the '80s. In reality, the industry hit a slump in this period." For those who had invested their time and money in robotics, the '80s were less a slump than an out-and-out throw-yourself-off-the-ledge crash.

"It's been boom and bust," says Hans Moravec. "Nobody's made any serious money. Most people lost a lot."

Moravec is one of the field's true pioneers. He invented Stanford's Cart robot and founded the world's largest robotics program, at Carnegie Mellon University. And Moravec was one of those whose optimistic predictions led to the ill-fated robot spurt of the '80s. "We've been very disappointed," he says. "The problems we faced were much harder than we thought. We made a huge, huge miscalculation."

The hugest miscalculation was a basic one: it turned out to be much trickier than anyone expected to synthesize a brain that could not only amass knowledge of the world, but also grasp it, and thus apply it to everyday life.

In some sense, robotics researchers didn't so much overestimate the available technology as underestimate human physiology: simple tasks like reaching for the remote control and a bag of Cheetos turned out to involve processes that are a lot more complex than they seem. Despite the heroic efforts of the AI community, the physical world consistently proved to be more than even a huge computer brain could handle.

The clearest example of this is IBM's Deep Blue, a supercomputer that in 1997 took on Garry Kasparov -- considered to be the greatest human chess player ever -- and beat him two games to one (they tied twice). And yet, as Moravec points out in his recent book Robot: Mere Machine to Transcendent Mind (Oxford University Press), "Deep Blue needed a human assistant to see and physically manipulate the pieces. No robot existing could have done it in the wide range of circumstances Kasparov, or any child, finds trivial." In other words, Deep Blue was masterful in approaching chess as a computational problem, but totally incapable of approaching it as a board game.

This is a problem Moravec and others hammered away at through the late '80s and early '90s. As computer power grew, however, the bridge between "thought" and action remained elusive. As William Townsend puts it, this was the period when the "bubble burst."

IN THE midst of this slump, a schism erupted in the robotics community. In the mid '80s, a young MIT professor named Rodney Brooks put forward a simple theory: "The world is its own best model."

This seemingly innocuous statement was revolutionary. The reigning approach in robotics was to build a machine with a very powerful brain, and then pre-program that brain with a model of the world. Brooks suggested starting at the other end: endowing the robot with a set of behaviors and letting it figure the world out for itself. "This was a reaction against Hans's approach," says Brooks, who was an office mate of Moravec's at the Stanford AI lab in the late '70s.

In 1985, Brooks presented his ideas at an international robotics convention, and immediately became both celebrity and pariah in his field. "Only many years later," he has written, "did I learn that in the back row senior robotics people were shaking their heads asking why I was throwing my career away."

In those days, Brooks was a wild-haired interloper. Today he has the look of a successful middle-aged technocrat: rumpled shirt, slacks, hair just this side of long. And far from throwing his career away, Brooks has landed himself a successful robotics firm and a spot as director of MIT's world-famous Artificial Intelligence Laboratory -- a position he has held since 1997.

Sitting in his office at MIT, Brooks still shows vestiges of the defiance that once turned the AI world on its head. He uses refreshingly simple language to describe his approach: "I was led to it by watching insects fly around, chase each other, mate with each other, destroy each other," he says. "There's a tiny little brain there. There's not enough computation for them to be building a complete world model. They're doing things in a very different sort of way. They see stimulus and they react to that, without knowing all the clutter."

Brooks's first robots not only acted like insects, they looked like them. His little six-legged beastie Genghis became something of a star in the robotics world. Endowed with minuscule intelligence compared to its supercomputer-driven brethren, Genghis couldn't begin to know what an "obstacle" was, much less what obstacles were in its path. But when it encountered an obstacle, Genghis could adjust itself to deal with it, without outside help -- something its brainier counterparts had long struggled with.

"My idea was to let computation out into the world, rather than trying to pull the world in," says Brooks. "The world's there in front of you, and hell, that's the best place for it." Appropriately, Brooks didn't use complex theorems and mathematical formulas as the basis for his approach. Instead, he used what he saw: the world.

In a 1987 paper titled "Intelligence Without Representation," recently republished in a collection of his work called Cambrian Intelligence (MIT Press), Brooks argued that the large part of human evolution has consisted of mastering basic interaction with the world. Intelligent behavior developed relatively recently, and relatively quickly. "This suggests," he wrote, "that problem-solving behavior, language, expert knowledge and application, and reason are all pretty simple once the essence of being and reacting are available."

If Brooks was right, then the way to develop a robot with humanlike intelligence was to build an essentially ignorant machine that behaved like a human and send it out into the world. His plan was to do this by initially creating basic drives, then applying layers of increasingly sophisticated behaviors -- working, as it were, from the bottom up. For Brooks, always, it was existence precedes essence. Or: get the ass in gear, and the mind will follow.

These may not sound much like fighting words, but what Brooks was saying, in effect, was that robotics experts had been wasting their time. Not surprisingly, this raised a few hackles in the AI community -- particularly when Brooks's approach seemed to be working. Fifteen years later, animosity toward Brooks and contempt for his ideas are still going strong.

"Rod just made an enormous amount of noise by taking an extreme position," says Hans Moravec, calling Brooks's challenge to the robotics community "theater." "He's controversial on purpose. He's attracted a lot of attention, and he's raised money to start his company." Some people, says Moravec, go so far as to call Brooks "flaky."

Brooks attributes this criticism to what he calls "physics envy."

"That gets leveled at me, and I say, `Well, my 90-person company makes money every year. That's not theater. We're doing real stuff.' This upsets me a little, because it's a way to put me down. But no one else has a company with 90 people that is actually making money on this stuff."

THE COMPANY to which Brooks refers is his Somerville, Massachusetts-based IS Robotics, "the coolest robot place around," as company president Helen Greiner puts it. It's certainly the coolest place at the Twin City Mall in Somerville, which is otherwise a mess of discount stores and megamarkets. The interior of the IS office is as unexceptional as its setting -- but for the fact that there's a little robot sitting on the floor of the reception area, and another squatting on the president's desk, and another crawling around the main office area.

The bulk of the "real stuff" IS does is divided between Defense Department projects and toys being developed for Hasbro. (Ironically, the Hasbro research is off-limits, while the military projects are open to public scrutiny.) The star of IS's serious projects seems to be Bilbo, an "urban robot" that looks a bit like an oversized VCR on caterpillar tracks. Bilbo's main job is to go into buildings where armed adversaries may lurk. The robot, which is able to climb stairs by way of pivoting tracks, has "the most advanced mobility in the country," says Greiner.

There's also a six-legged critter called Ariel, which scuttles around sideways like a crab. Ariel's job is mine detection. Then there are IS's swarming robots, which look like overturned salad bowls on wheels. These robots are meant to be deployed in large numbers, exploring nuclear power plants, or "anywhere it's just really nasty to send humans." The swarming robots display a type of social interaction, and today a guy named James is attempting to get them to form a conga line, only to find they prefer forming clusters. "It's loads of fun," James says.

But it's not just fun. Greiner won't divulge IS's annual profits ("a 90-person staff should give you an idea"), but its competitors say it is a force to be reckoned with, right up there alongside industry giants such as Sony.

In terms of research, Brooks's crowning achievement to date is a six-year-old robot that goes by the name of Cog. Sitting in the corner of a large room at MIT's AI lab, Cog looks like a diagram of itself: a scribble of steel bones, plastic cartilage, and hydraulic muscles. Its wiry little head is dominated by a pair of camera eyes, like joke Coke-bottle glasses. In spite of the fact that it has no legs and appears to be bolted to the base of a drill press, Cog is a marvel of engineering -- one of the only robots in the world that has even begun to interact with the world in a humanlike way.

Most of the time, Cog sits quietly, its head slumped on its sinuous chest. But bring it to life, and Cog seems to be just that: alive. Look Cog in the eye, and it will look right back at you. If two people are talking, Cog will appear to follow the conversation by looking from one person to the other. It has learned to coordinate vision and reach, so it can pinpoint an object and pick it up. Give it a slinky, and Cog will shuffle it from hand to hand. Unlike the herky-jerky incremental movements we usually associate with robots, Cog's motion is fluid, eerily organic.

Having sat in as a drummer for the rock band They Might Be Giants -- listening to the rhythm and banging along -- Cog might be the funkiest robot in the world. And this, says Brooks, is a better place to start than by building massive databases.

Only by applying layers of behavior -- starting with playing and drumming, moving on to thinking and acting -- does Brooks believe a truly humanlike being will emerge, one that not only interacts with the world in a lifelike way, but compels us to think of it as a living being. "I've always said that whenever my graduate students start to feel genuine concern or remorse when they switch it off," Brooks explains, "then I will feel that we have succeeded."

OUR FASCINATION with robots is a fascination with life," says Henry Thorne. "When something imitates life, it shines a light on what life is. It helps us answer the question of what we are."

If Brooks's research is anything to go by, the answer to that question is: considerably less than we think we are. Brooks's theories force us to rethink not only artificial intelligence, but human intelligence as well. His bottom-up approach turns our Cartesian self-image on its head: I am, therefore I think.

"People say that we over-anthropomorphize robots," he says. "I say we over

-anthropomorphize people. People are actually a lot more like robots, but we tend to attribute a lot more intention to them. Most of the time we're pretty primitive creatures without much high-level stuff; all that stuff on top is a pretty thin veneer."

But if Brooks's robots are simply bundles of impulses, doesn't that make them little more than yahoos? "You're a yahoo," Brooks says in answer to this question. "We all are, in some sense. My view of you is that you're a machine. You're a bag of skin full of these little molecules that interact according to rules. I don't think there's any essence to you. I don't think there's a soul."

And he still doesn't believe there is any core of intelligence that can eventually be mimicked by a giant computer. In his book, he mockingly poses a question asked by traditional robotics researchers -- "[A]ren't our computers too small yet?" -- and answers, "Heck no. Part of my thesis is that it [inventing autonomous mobile robots] takes very little computational power; we've just been organizing it all wrong up until now."

All this, of course, is anathema to Hans Moravec, whose AI philosophy is still to keep plugging away until the advent of computers powerful enough to tackle the physical world. "I think Cog's a great overshoot," he says. "I think Rod's too impatient. I'm quite happy to plod along for 10 years, as long as I'm at peace with myself. I'll take as long as it takes."

Brooks, on the other hand, scoffs at Moravec's willingness to play the waiting game. "He hasn't done anything new in 20 years," he says.

The hard truth, however, is that time may be running out for both Moravec and Brooks, who are in their 50s. After years of toil and conflict, hope and disappointment, they may never actually get to see the fruits of their labor. "That doesn't matter," says Moravec. "I know this thing is bigger than I am. I'll just do what I can, even if I don't see the promised land."

For all their squabbles, it seems the two old rivals can agree on this one thing.

"Maybe we'll need an Einstein to come along," Brooks says. "Maybe we need some smart kid to come along and bash us old guys over the head and say, `Idiots!' " Then, getting up to dash off to another meeting, he adds, "I don't know. I feel bad that I'm not going faster, but maybe I'll get there."

He stops midway to the door and says, half smiling, "Or not."

Chris Wright can be reached at cwright[a]phx.com.