I don’t think they fret and reason endlessly about mental states, as we do. They simply dream a different dream, probably much like the one we used to dream, before we crocheted into our neural circuitry the ability to have ideas about everything. Other animals may know you know something, but they don’t know you know they know. Other mammals may think, but we think about having thoughts. Linnaeus categorized us in the subspecies of Homo sapiens sapiens, adding the extra sapiens because we don’t just know, we know that we know. Our infants respond to their surroundings and other people, and start evolving a sense of self during their first year. Like orangutans, elephants, and even European magpies, they can identify themselves in a mirror, and they gather that others have a personal point of view that differs from their own.
So when people talk about robots being conscious and self-aware, they mean a range of knowing. Some robots may be smarter than humans, more rational, more skillful in designing objects, and better at anything that requires memory and computational skills. I reckon they can be deeply curious (though not exactly the way we are), and will grow even more so. They can already do an equivalent of what we think of as ruminating and obsessing, though in fewer dimensions. Engineers are designing robots with the ability to attach basic feelings to sensory experience, just as we do, by interacting with the world, filing the memory, and using it later to predict the safety of a situation or the actions of others.
Lipson wants his robots to make assumptions and deductions based on past experiences, a skill underlying our much-prized autobiographical memory, and an essential component of learning. Robots will learn through experience not to burn a hand on a hot stove, and to look both ways when crossing the street. There are also subtle, interpersonal clues to decipher. For instance, Lipson uses the British “learnt” instead of the American “learned,” but the American “while” instead of the British “whilst.” So, from past experience, I deduce that he learned English as a child from a British speaker, and assume he has lived in the United States just long enough to rinse away most of the British traces.
Yet however many senses robots may come to possess—and there’s no reason why they shouldn’t have many more than we, including sharper eyesight and the ability to see in the dark—they’ll never be embodied exactly like us, with a thick imperfect sediment of memories, and maybe a handful of diaphanous dreams. Who can say what unconscious obbligato prompts a composer to choose this rhythm or that—an irregular pounding heart, tinnitus in the ears, a lover who speaks a foreign language, fond memories evoked by the crackle of ice in winter, or an all too human twist of fate? There would be no Speak, Memory from Nabokov, or The Gulag Archipelago from Solzhenitsyn, without the sentimental longings of exile. I don’t know if robots will be able to do the sort of elaborate thought experiments that led Einstein to discoveries and Dostoevsky to fiction.
Yet robots may well create art, from who knows what motive, and enjoy it based on their own brand of aesthetics, satire (if they enjoy satire), or humor. We might enjoy it, too, especially if it’s evocative of work by human artists, if it appeals to our senses. Would we judge it differently? For one of its gallery shows, Yale’s art museum accepted paintings inspired by Robert Motherwell, only to change its mind when it learned they’d been painted by a robot in Lipson’s Creative Machines Lab. It would be fun to discover robots’ talents and sensibility. Futurologists like Ray Kurzweil believe, as Lipson does, that a race of conscious robots, far smarter than we, will inhabit Earth’s near-future days, taking over everything from industry, education, and transportation to engineering, medicine, and sales. They already have a foot in the door.
At the 2013 Living Machines Conference, in London, the European RobotCub Consortium introduced their iCub, a robot that has naturally evolved a theory of mind, an important milestone that develops in children at around the age of three or four. Standing about three feet tall, with a bulbous head and pearly white face, programmed to walk and crawl like a child, it engages the world with humanlike limbs and joints, sensitive fingertips, stereo vision, sharp ears, and an autobiographical memory that’s split like ours into the episodic memory of, say, skating on a frozen pond as a child and the semantic memory of how to tilt the skate blades on edge for a skidding stop. Through countless interactions between body and world it codifies knowledge about both. None of that is new. Nor is being able to distinguish between self and other, and intuit the other’s mental state. Engineers like Lipson have programmed that discernment into robots before. But this was the first time a robot evolved the ability all by itself. iCub is just teething on consciousness, to be sure, but it’s intriguing that the bedrock of empathy, deception, and other traits that we regard as conscious can accidentally emerge during a robot’s self-propelled Darwinian evolution.
It happened like this. iCub was created with a double sense of self. If he wanted to lift a cup, his first self told his arm what to do, while predicting the outcome and adjusting his knowledge based on whatever happened. His second—we can call it “interior”—self received exactly the same feedback, but, instead of acting on the instructions, it could only try to predict what would happen in the future. If the real outcome differed from a prediction, the interior self updated its cavernous memory. That gave iCub two versions of itself, an active one and an interior “mental” one. When the researchers exposed iCub’s mental self to another robot’s actions, iCub began intuiting what the other robot might do, based on personal experience. It saw the world through another’s eyes.
As for our much-prized feats of scientific reasoning and insight, Lipson’s lab has created a Eureqa machine, a computer scientist able to make a hypothesis, design an experiment, contemplate the results, and derive laws of nature from them. Plumbing the bottomless depths of chaos, it divines meaning. Assigned a problem in Newtonian physics (how a double pendulum works), “the machine took only a couple of hours to come up with the basic laws of motion,” Lipson says, “a task that occupied Newton for years after he was inspired by an apple falling from a tree.”
Eureqa takes its name from a legendary moment in the annals of science, two thousand years ago, when Archimedes—already a renowned mathematician and inventor with formidable mastery in his field—was soaking in his bathtub, his senses temporarily numbed by warm water and weightlessness, and the solution to a problem came to him in a flash of insight. Leaping from the tub, he supposedly ran naked through the streets of Athens yelling, “Eureka!” (“I have found it!”)
For two thousand years, that’s how traditional science has run: solid learning and mastery, then the kindling of observation and a spark of insight. The Eureqa machine marks a turning point in the future of how science is done. Once upon a time, Galileo studied the movement of the heavenly bodies, Newton watched an apple fall in his garden. Today science is no longer that simple because we wade through oceans of information, generate vast amounts of additional data, and analyze it on an unprecedented scale. Virtuoso number-crunchers, our computers can extract data without bias, boredom, vanity, selfishness, or greed, quickly doing the work that used to take one human a lifetime.
In 1972, when I was writing my first book, The Planets: A Cosmic Pastoral, a suite of scientifically accurate poems based on the planets, I used to hang out in the Space Sciences Building at Cornell. The astronomer Carl Sagan was on my doctoral committee, and he kindly gave me access to NASA photographs and reports. At that time, it was possible in months to learn nearly everything humans knew about the other planets, and the best NASA photos of the outermost planets were only arrows pointing to balls of light. Over the decades, I attended flybys at the Jet Propulsion Laboratory in Pasadena, California, and watched the first exhilarating images roll in from distant worlds as Viking and Voyager reached Mars, Jupiter, Saturn, Neptune, and an entourage of moons. In the 1980s, it was still possible for an amateur to learn everything humans knew about the planets. Today that’s no longer so. The Alps of raw data would take more than one lifetime to summit, passing co
untless PhD dissertations at campsites along the trail.
But all that changes with a tribe of Eureqa-like machines. A team of scientists at the University of Aberystwyth, led by Professor Ross King, has revealed the first machine able to deduce new scientific knowledge about nature on its own. Named Adam, the two-armed robot designed and performed experiments to investigate the genetics of baker’s yeast. Carrying out every stage of the scientific process by itself without human intervention, it can perform a thousand experiments a day and make discoveries.
More efficient science will solve modern society’s problems faster, King believes, and automation is the key. He points out that “automation was the driving force behind much of the nineteenth- and twentieth-century progress.” In that spirit, King’s second-generation laboratory robot, named Eve, is even faster and nimbler than Adam. It’s easy to become mesmerized watching a webcam of Eve testing drugs, her automated arms and stout squarish body shuffling trays, potions, and tubes with tireless precision, as she peers through ageless nonblinking eyes, while saving the sanity of countless graduate students, spared sleepless nights in the lab tending repetitive experiments.
How extraordinary that we’ve created peripheral brains to discover the truths about nature that we seek. We’re teaching them how to work together calmly as a society, share data at lightning speed, and cooperate so much better than we do, rubbing brains together in the invisible drawing room we sometimes call the “cloud.” Undaunted, despite our physical and mental limitations, we design robots to continue the quest we began long ago: making sense of nature. Some call it Science, but it’s so much larger than one discipline, method, or perspective.
I find it touchingly poetic to think that as our technology grows more advanced, we may grow more human. When labor, science, manufacturing, sales, transportation, and powerful new technologies are mainly handled by savvy machines, humans really won’t be able to compete in those sectors of the economy. Instead we may dominate an economy of interpersonal or imaginative services, in which our human skills shine.
Smart robots are being nurtured and carefully schooled in laboratories all over the world. Thus far, Lipson’s lab has programmed machines to learn things unassisted, teaching themselves the basic skills of how to walk, eat, metabolize, repair wounds, grow, and design others of their kind. At the moment, no one robot can do everything; each pursues its own special destiny. But one day, all the lab machines will merge into a single stouthearted . . . being—what else would we call it?
One of Lipson’s robots knows the difference between self and other, the shape of its physique, and whether it can fit into odd spaces. If it loses a limb, it revises its self-image. It senses, recollects, keeps updating its data, just as we do, so that it can predict future scenarios. That’s a simple form of self-awareness. He’s also created a machine that can picture itself in various situations—very basic thought experiments—and plan what to do next. It’s starting to think about thinking.
“Can I meet it?” I ask.
His eyes say: If only.
Leading me across the hall, into his lab, he stops in front of a humdrum-looking computer on a desk, one of many scattered around the lab.
“All I can show you is this ordinary-looking computer,” he says. “I know it doesn’t look exciting because the drama is unfolding in the software inside the machine. There’s another robot,” he says, gesturing to a laptop, “that can look at a second robot and try to infer what that other robot is thinking, what the other robot is going to do, what the other robot might do in a new situation, based on what it did in a previous situation. It’s learning another’s personality. These are very simple steps, but they’re essential tools as we develop this technology. And with this will come emotions, because emotions, at the end of the day, have to do with the ability to project yourself into different situations—fear, various needs—and anticipate the rewards and pain in many future dramas. I hope that, as the machines learn, eventually they’ll produce the same level of emotions as in humans. They might not be the same type of emotions, but they will be as complex and rich as in humans. But it will be different, it will be alien.”
I’m fascinated by the notion of “other types of emotions.” What would a synthetic species be like without all the lavish commotion of sexual ardor, wooing, jealousy, longing, affectionate bonds, shared experiences? Just as I long to know about the inner (and outer) lives of life forms on distant planets, I long to know about the obsessions, introspections, and emotional muscles that future species of robots might wrestle with. A powerful source of existential grief comes from accepting that I won’t live long enough to find out.
“Emotional robots . . . I’ve got a hunch this isn’t going to happen in my lifetime.” I’m a bit crestfallen.
“Well, it will probably take a century, but that’s a blip in human history, right?” he says in a reassuring tone. “What’s a century? It’s nothing. If you look at the curve of humans on Earth,” he says, curving one hand a few inches off the table, “we’re right there. That’s a hundred years.”
“So much has happened in just the last two hundred years,” I say, shaking my head. “It’s been quite an express ride.”
“Exactly. And the field is accelerating. But there’s good and bad, right? If you say ‘emotions,’ then you have depression, you have deception, you have creativity and curiosity—creativity and curiosity we’re already seeing here in various machines.
“My lab is called the Creative Machines Lab because I want to make machines which are creative, and that’s a very very controversial topic in engineering, because most engineers—close the door, speak quietly—are stuck in the Intelligent Design way of thinking, where the engineer is the intelligent person and the machines are being created, they just do the menial stuff. There’s a very clear division. The idea that a machine can create things—possibly more creatively than the engineer that designed the machine—well, it’s very troubling to some people, it questions a lot of fundamentals.”
Will they grow attached to others, play games, feel empathy, crave mental rest, evolve an aesthetics, value fairness, seek diversion, have fickle palates and restless minds? We humans are so far beyond the Greek myth of Icarus, and its warning about overambition (father-and-son inventors and wax wings suddenly melting in the sun). We’re now strangers in a strange world of our own devising, where becoming a creator, even the Creator, of other species is the ultimate intellectual challenge. Will our future robots also design new species, bionts whose form and mental outlook we can’t yet imagine?
“What’s this?” I ask, momentarily distracted by a wad of plastic nestled on a shelf.
He hands me the strange entanglement of limbs and joints, a small robot with eight stiff black legs that end in white ball feet. The body is filamental, like a child’s game of cat’s cradle gone terribly wrong, and it has no head or tentacles, no bulging eyes, no seedlike brain. It wasn’t designed as an insect. Or designed by humans, for that matter.
Way back in our own evolution, we came from fish that left the ocean and flopped from one puddle to another. In time they evolved legs, a much better way to get around on land. When Lipson’s team asked a computer to invent something that could get from point A to point B—without programming it how to walk—at first it created robots reminiscent of that fish, with multihinged legs, flopping forward awkwardly. A video, posted on YouTube, records its first steps, with Lipson looking on like a proud parent, one who appreciates how remarkable such untutored trials really are. Bits of plastic were urged to find a way to combine, think as one, and move themselves, and they did.
In another video, a critter trembles and skitters, rocks and slides. But gradually it learns to coordinate its legs and steady its torso, inching forward like a worm, and then walking insectlike—except that it wasn’t told to model an insect. It dreamt up that design by itself, as a more fluent way forward. Awkward, but effective. Baby steps were fine. Lipson didn’t expect grace. He could mak
e a spider robot that would run faster, look better, and be more reliable, but that’s not the point. Other robots are bending, flexing, and running, using replica tendons and muscles. DARPA’s “cheetah” was recently clocked at a tireless 30 mph sprint. But that cheetah was programmed; it would be a four-legged junkpile without a human telling it what to do. Lipson wants the robot to do everything on its own, eclipsing what any human can design, unfettered by the paltry ideas of its programmers.
It’s a touching goal. Surpassing human limits is so human a quest, maybe the most ancient one of all, from an age when dreams were omens dipped in moonlight, and godlike voices raged inside one’s head. A time of potent magic in the landscape. Mountains attracted rain clouds and hid sacred herbs, malevolent spirits spat earthquakes or drought, tyrants ruled certain trees or brooks, offended waterholes could ankle off in the night, and most animals parleyed with at least one god or demon. What was human agency compared to that?
ROBOTS ON A DATE
Looking around Lipson’s quiet lab, I sense something missing. “You have real students sitting at the computer benches. I don’t see any chatbots.”
Lipson smiles indulgently. His chatbots have been a YouTube craze. “That was just an afternoon hack. It went viral in twenty-four hours and took us completely by surprise.”
He doesn’t mean “hack” in its usual sense of breaking into a computer with malicious intent, but as highwire digital artistry. The Urban Dictionary defines its slang use like this: “v. To program a computer in a clever, virtuosic, and wizardly manner. Ordinary computer jockeys merely write programs; hacking is the domain of digital poets. Hacking is a subtle and arguably mystical art, equal parts wit and technical ability, that is rarely appreciated by non-hackers.”