The Age of Robots
We're close to making humanlike machines. It's time to reckon with the promises and perils
The millennium was still a half century off in the future when Isaac Asimov penned his sci-fi classic, I, Robot. So it must have seemed plausible to imagine a world populated by big, strong, intelligent humanoid robots. The mechanical replicas he conjured may have had shiny metal bodies and glowing red eyes, but they otherwise resembled people, thought like people, and--most important of all--devoted themselves to taking care of the human race.
Contrary to Asimov's genre-defining tale, humankind is still operating pretty much on its own. Indeed, of all the great science-fiction predictions to go bust at the end of the millennium--no time machines, no intergalactic space travel--surely the most galling is the absence of a single decent robotic maid. Or butler, take your pick. Oh sure, the new Robomower will trim your lawn while you recline in the hammock, and the Dyson DC06 robotic vacuum cleaner will soon be available to suck the lint from your carpets. But if you want something from the fridge, you're still going to have to fetch it yourself.
If visionaries like Asimov may have been wrong about the timing, they were right to predict a bright robotic future. Indeed, robots of various stripes seem to be popping, whirring, buzzing, and gliding up just about everywhere. Very practical-minded bots are at work in the real world right now, exploring distant planets, assisting with precision surgery, and locating deadly land mines.
The toy market, already home to Aibos and Poo-Chis, is just months away from a massive invasion by robotic cats, dogs, and mice as well as mechatronic aliens and babies and dinosaurs. Steven Spielberg will set the stage for the fall shopping frenzy this summer when his movie A.I. does for supersmart, supersensitive robots what his Jurassic Park did for supersmart, supervicious dinosaurs. And labs around the world are busily working on the robotic parts--feet and knees for walking, hands for grasping, versions of eyes and ears--that will someday be stitched together into a fully functional humanoid robot.
Early lessons. The dream of such a humanoid compatriot--machine enough to boss around but human enough to be a good sidekick--stretches back at least to the early 20th century. But the early decades of research proved mainly this: that humans are a lot more complex than originally considered--and it's really, really hard to build one. In the past few years, however, important advances in computer science, artificial intelligence, biomechanics, and material science have once again raised hopes of reaching the holy grail of robotics. In fact, progress toward a fully autonomous, intelligent robot has been so convincing that any number of technofuturists are worrying publicly about the perils of robotics. At least one highly regarded scientist, Bill Joy, a co-founder of Sun Microsystems, has predicted that our own robotic creations might one day replicate themselves and contribute to humankind's demise.
Whether humans are in a hopeful or precarious place, the journey here has been an intellectual challenge. Don't even contemplate the brain for a moment. How about something simple like walking on two legs. Humans do it naturally, and our ancestors have for millions of years, but it took one of the largest industrial companies in the world 10 years and untold millions to build a machine that could master a workable form of bipedalism. That was the Honda P3, a 5-foot, 3-inch, 290-pound astronaut look-alike unveiled by the Japanese car company in 1996. Widely hailed as a triumph of pure engineering willpower, the P3 does walk convincingly and can even go up and down stairs. But the Honda spokesbot--P3 has been succeeded by the more diminuitive Asimo--can manage only a sluggish 1 mile per hour on the straightaway. It's going to take a lot more hustle than that to make it in the domestic-service racket.
The trick to making an athletic robot is simulating the finely tuned orchestra of muscle, bone, and nerve that has evolved over countless millennia. All robots make use of the same basic components to do this. A jointed metal or plastic frame serves for a skeleton, and a variety of actuators (motors, pulleys, gears, hydraulics, and so forth) provide muscle power. But the new humanoids are not just bodies; they're also sophisticated sensing machines, packed with cameras, microphones, even "haptic" sensors that mimic the sense of touch. Significant engineering challenges still remain--one of the most fundamental is finding a way to power the energy-hungry machines--but most researchers are confident that they'll get the physical side of things worked out in the near future.
And then there are the brains. At MIT's humanoid robotics lab, the cartoonish, head-only robot Kismet is just slightly larger than a normal human noggin. And yet the contraption relies on a bank of 15 external computers to control its social abilities and impressive array of facial expressions. Asimo and P3 are downright doltish by comparison, depending on remote-control operators and pre-scripted programs to tell them what to do. Other advanced humanoid bodies leave all the thinking to humans. NASA's prototype space worker Robonaut, for example, mimics the movements of a human operator in a sensor-laden "tele-presence" suit. The operator bends his elbow, and the robot bends its elbow in response, like a mechanical mirror image.
Despite decades of intensive research in artificial intelligence, the brains are lagging behind the knees and wrists. During the field's early days, says Rodney Brooks, director of MIT's AI lab and its humanoid robotics group, researchers tried to make machines smart by writing elaborate, computerized problem-solving programs. They assumed the sequences of facts, physical laws, and logical relationships would add up to thinking. The results could be impressive--just ask chess grandmaster Garry Kasparov, who was humbled by IBM's Deep Blue in 1997--but making a thinking machine turned out to be much harder than the scientists imagined. Chess, for all its challenges to human brain power, turns out to be a simpler pursuit than, say, making soup. A chess-bot needs only information and logic, but a chef-bot without a dash of creativity, intuition, and flair would be little more than an expensive, programmable Cuisinart. Take that pot of soup. You cut some vegetables, boil some bones, throw in a bay leaf or two, maybe some other spices. But what vegetables and how many? How to tell a turnip from a turkey leg? And, ahem, whose bones? And how on Earth could a robot add salt to taste with no sense of what "saltiness" means--and, for that matter, no sense of taste?
The high marks of Enlightenment thinking--logic and problem solving--turn out to be much easier to simulate than the perceptual and intuitive things that any kid can do, Brooks says. Stuffing a computer full of facts (chicken bones good, dog bones bad) and equations (salt tolerances between 1 and 3 teaspoons per quart, say) works well for number-crunching tasks. But for real-world smarts--remembering to grab an umbrella if it looks like rain--logic alone just doesn't cut it.
So how then to proceed? Increasingly, AI researchers are looking to children for the answer. Kids are essentially learning machines, and while no one is quite sure how they do it, there's clearly a lot of imitation and interaction, and plenty of room for trial and error. If robots are ever going to have humanlike intelligence, the new thinking goes, perhaps they'll have to develop it the way babies do. And that, says Cynthia Breazeal, the MIT roboticist behind Kismet's licorice-whip lips and Groucho Marx eyebrows, requires social interaction.
Kismet is programmed to seek sensory stimulation--voices, movement, brightness, and color--which it attracts with beguiling expressions and a sort of babbling baby talk. If an expression works, and a passing human comes up to play, Kismet's internal "social drive" is satiated. If not, the levels sink and Kismet tries a new strategy to connect. But there are balancing desires: Get too close and Kismet will let you know you've invaded its space with an exaggerated look of annoyance. Play too rough and the usually docile head may assume an alarming grimace or turn away.
"The whole point," says Breazeal, "is that the robot is trying to get you to interact with it in ways that can benefit its ability to learn." Basic movements are programmed into Kismet's behavior, but its handlers hope human feedback will help it learn new gestures and vocalizations by imitating people and storing successful attempts in its memory. "It helps the robot learn the social meaning of its actions," Breazeal says. The goal is for Kismet to learn not just to "think" for itself but also, as every child must, to understand that its actions have consequences.
What's good for the mind is good for the muscles. Maja Mataric, a computer scientist and roboticist at the University of Southern California, is trying to solve the problem of motor control. The entire range of human--or robot--mobility, she says, "can be collapsed down to a reasonably sized set of movements." Called "primitives," these basic motions can be combined or modified to produce novel activity. Take reaching. Whether you're stretching your racket arm out to volley a tennis ball or grabbing the lid off a boiling pot, says Mataric, "you use a standard way of reaching, the same basic movement." Once a robot has the basic moves down, mimicking people is much easier. All it takes is a little practice, plus a few learning and adaptation algorithms to help the machine capitalize on its mistakes. The result: a bot that can learn to dance the Macarena just by watching. In theory anyway--a good body is hard to come by, so Mataric works mostly with computer simulations with realistic physical properties such as gravity and mass.
Practical applications. Not all researchers buy the developmental theory of robot building. Kazuhiko Kawamura of Vanderbilt University, for example, programs his humanoid ISAC to perform practical tasks, such as feeding disabled patients. Machines that learn from the ground up might make for interesting interactions, Kawamura says, but "a humanoid robot needs to have more than just fascinating behaviors. To me, that approach only gets you a robotic baby. And we don't need a robotic baby."
While many roboticists are focused on developing useful machines, a few like Mataric and MIT graduate student Brian Scassellati are more interested in what humanoids can tell us about humans. "Humanoids give us a platform for research," says Mataric. There's nothing like trying to build a simulation of a baby, for example, to show you just how much you don't know about how babies are built.
Scientists are also starting to use the machines to test theories and notions about how brains work. "There are some really nice models of how children learn basic social skills" but few ways to test them, says Scassellati, who has a special interest in autism. The robot he works on, a rugged-looking head and torso unit named Cog, is the product of almost 10 years of evolutionary development. Finally, says Scassellati, "we're starting to be able to look at these models [with Cog] and say something intelligent about them." Testing behavioral theories with a robot, he says, may provide a major advantage over computer simulations, the only other method presently available. "In building something I have to deal with the real social scenarios," he says, "not my idea of how social scenarios should work. In building something, I'm forced to get it right."
And not just right--approachable, too. These are social robots after all, so they won't be much use if they give people the willies. "We try to build something that looks enough like a person so that you could treat it like a person and you don't feel too weird about it," says Scassellati. Stephen Jacobsen, a University of Utah professor whose robotics company Sarcos makes amusement park automata as well as terrifically sophisticated humanoid robot bodies, says giving a machine the look of life is really quite simple. "First it's the movement, then it's the eyes," he says. Jerkiness is a sure turnoff. And as with the body, it's not always how real eyes look; rather, it's how they move. "If the eyes are at all awkward, people just don't like it," says Jacobsen. "But if they're graceful, people are intrigued."
If robots are ever going to live and work with us, they've got to look good, too. Some designers prefer the stylized, impersonal look of Robonaut or the sleekly modernist Japanese humanoid SIG. For the more truly social machines, designers have two options; they can go for the disarmingly goofy look of MIT's Kismet, or they can shoot for realism. At the Science University of Tokyo, Fumio Hara and his team study robot-human interactions using intricately constructed "face robots," mechatronic skulls complete with lifelike dentures and eyeballs. Silicone masks can be pulled over the underlying mechanism--transforming it into a famous athlete, for example. Nineteen different actuators pinch, push, and stretch the rubbery skin into myriad expressions, some of them human, some of them most decidedly not. Too much realism, it turns out, can be just as much of a social obstacle as too little. "As you start looking more and more like a person," says Scassellati, "you pass a certain point where it becomes scary and off-putting because it looks enough like a person and yet it's wrong."
Bad PR. If it seems like the robot builders are trying to put a kinder, gentler face on their creations, they are. In North America, at least, robots have had terrible PR. We associate robots with herky-jerky movements, brutish strength, and a personality limited to grim, remorseless logic. That, says Mataric, amounts to nothing less than antirobot prejudice. "There are all sorts of misconceptions about robots as halting, mechanical things. It's a stereotype. What people know is what they saw in the old movies. It has nothing to do with reality."
Robots' image is much better in Japan. "I've always taken it for granted that robots are friendly and not something to compete with," says Japanese roboticist Hiroaki Kitano, whose futuristic SIG could win an automaton's beauty contest. "It's very curious for me why the Americans nearly always portray robots [as evil]." In fact, many Japanese researchers credit their childhood love of fictional robots--especially the peppy and resourceful Astro Boy, who delighted postwar Japan--with inspiring the national drive to develop helper robots. By legend, Astro Boy came into being in 2003--a date as significant to Japanese sci-fi aficionados as 2001 is here.
There's more to Japan's domination of the emerging humanoid robot world than an old cartoon, of course. With their traditional markets saturated, says Takeo Kanade of Carnegie Mellon University, corporate giants like Honda and Sony are casting about for new products. "Humanoid robotics is one of many things they're looking at as a potential new industry. They have the money, the technology, and the long-range vision to move into new areas." The products might be toys now, but these and other robotic pioneers are serious about developing humanoids to work as office assistants, caretakers for the elderly, and other human aides.
Living robots. If learning, memory, and creative intelligence really all are possible, then can machine consciousness be far behind? That would depend on what exactly consciousness is, of course, and to date there is no agreed-upon definition. There's no evidence that consciousness exists anywhere outside of a biological brain, notes philosopher Colin McGinn in his book The Mysterious Flame. But neither can anyone point to a reason why it couldn't, short of invoking the religious or mystical. A growing number of experts are beginning to accept that conscious robots are all but inevitable sometime in the future.
The vision they conjure up looks pretty bright for intelligent machines, but our own prospects may be decidedly more grim. In his bracingly ominous Wired essay, "Why the Future Doesn't Need Us," Sun Microsystems' Bill Joy all but sounded the death knell for the human species last year. Advances in robotics, genetic engineering, and nanotechnology, he wrote, could lead to a world populated by super-organisms, both biological and mechanical. By building machines that are like us, only smarter, stronger, and more easily produced, Joy suggests, we could in fact be creating our own worst enemy in an evolutionary battle for survival. James Martin, a technology and business consultant, warns of a coming alien intelligence in his book of the same name. As machines become ever more intelligent, he argues, they will not only outpace our cognitive abilities but will develop new forms of thinking that will be beyond our comprehension. If we can't understand what we've built, we may not be able to control it. Ray Kurzweil, an artificial-intelligence pioneer, gives us about 20 more years of intellectual superiority over computers. By that time, he argues in The Age of Spiritual Machines, computers won't just be intelligent, they'll be conscious, feeling beings deserving of the same rights, privileges, and considerations we give each other.
Beyond a morass of ethical issues, what exactly might all of this mean for humanity? The speculations range from the catastrophic to the merely creepy. Most unpleasant is the "what goes around comes around" scenario, where the machines turn the tables and enslave us for a change. There is actually historical precedent for a robot rebellion; the word robot comes from robota, the Czech for an annual debt of forced labor. In 1848, the serfs rose up against their Austro-Hungarian landlords in protest. A different story, to be sure, but the term and the concept of mechanized serfs entered the Western consciousness with the grim baggage of class warfare.
Alternatively, the robots of the future could simply ignore us, leaving us to pursue our archaic organic mode of life, irrelevant but hardly dangerous. Finally, there is the "if you can't beat 'em, join 'em" scenario. Hans Moravec, a roboticist at Carnegie Mellon University, proposes that humanity may be able to survive, and even achieve a level of immortality, by digitally uploading our own consciousness into advanced robots.
We're probably decades away from having to worry about anything more than running out of batteries. Still, it seems clear that big changes are coming, and while humans--the flesh and blood type--usually manage to adapt to technological change, the period of adjustment can sometimes get pretty uncomfortable. As with any new technology, there will certainly be some unintended, and quite possibly unpleasant, consequences as robots begin to play a regular role in our day-to-day lives, USC's Mataric notes. But she's confident that the potential benefits outweigh the risks. "I hope society is strong and wise enough to stop abuses without stopping science," she says, "but I think all of that is still a long way off." Before anyone has to start really worrying about our place in the future, the techies have a heck of a lot more work to do.
Man-made Man
Many of the skills we take for granted--running, listening, recognizing a friend's face--are still beyond even the most advanced robots. But scientists are trying to fine-tune the pieces, one by one. Here's a look at some of the important parts of being human.
Head
Robot heads are packed with sensors and communication gear. To date, the computerized "brains" are usually too big to fit inside a human-size head, so many models are linked to an external computer bank; others rely on human controllers.
Face
Even a robot likes to look its best. Some designers favor a rugged mechanical feel while others shoot for realism. This silicone mask is fitted with metal shape-memory actuators to mimic human expressions.
Eyes
Our two eyes deliver front and peripheral vision, in stereo. Robots like Cog need four cameras and plenty of computer power to imitate the trick. The next challenge is actually recognizing objects and faces.
Hands
Not all robots need four fingers and an opposable thumb. A hand is a terribly complex thing, but to work with human tools Robonaut requires the full complement of human digits, its hands can grasp, hold, and twist, just like the real thing.
Arms and muscles
Robots don't always know their own strength. ISAC is strong, but stretchy artificial muscles called "rubbertuators" let it deliver a firm, but not crushing, handshake. Many robots have position, force, and tactile sensors to control their limbs.
Legs
Walking on two legs is basically the art of falling forward without falling over. It takes babies months, and lots of tumbles, to learn. Gravity sensors and solid-state gyroscopes called gyrometers help robots to keep their balance.
Hips
On humans they tilt, sway, and twist, but even Honda's multimillion-dollar robots--note the lack of a pelvis--are a little stiff. Macarena robot, yes; Elvis robot? That'll take some work.
Ears
No, the little pink cones and other earlike protrusions don't rally help with hearing--microphones do that. And right now, Kismet can't tell when it's being called. Sound localization technology is in the works.
GRAPHIC BY DOUG STERN--USN&WR
With Peter Hadfield
This story appears in the April 23, 2001 print edition of U.S. News & World Report.