<b>Jenny Offill</b> looks at the projected learning curve for machines, builds her own bots, and takes a ride to the bleeding edge of Artificial Stupidity.
For nearly five decades, artificial-intelligence researchers have tried to build robots of human-level intelligence. In 1956, Allan Newell and Herbert Simon, creators of the first artificial-intelligence program, predicted that machines would soon be capable of understanding and translating spoken language, composing classical music, and inventing new mathematical theorems. This would happen, they said, by 1970.
Recent predictions about the trajectory of machine intelligence have been no less heady. Futurists speculate that robots will soon clean our houses and baby-sit our kids. They will run factories and fight in wars. There will be robot sonnets, robot paintings, robot symphonies. They will surpass us in intelligence and inherit the earth.
Canadian roboticist Mark Tilden calls this scenario "Attack of the Killer Robots" and dismisses it as pure fantasy. Unlike most AI researchers, Tilden isn't interested in creating a robot as intelligent as a human being. He just wants to build one that's as smart as your average bug.
Tilden, a researcher at the Los Alamos National Lab, creates small solar-powered robots based on the principles of evolutionary biology. His devices contain no computers and minimal electronics, yet they can run continuously for years. Whatever intelligence they have, they gain from interacting with the world.
Tilden's ideas build on the groundbreaking work of Rodney Brooks, the MIT roboticist featured in Errol Morris's documentary "Fast, Cheap and Out of Control." In the early 1980s, Brooks broke away from the dominant paradigm that had guided AI researchers for years. This paradigm stated that a robot must first perceive the world, then think about it, then translate this cognition into action. Brooks thought this was one step too many. Why not create a robot that could act as soon as it perceived something? Brooks wanted to build mobile robots that skipped over the complicated and painstaking "mapping" process that traditional robots used to navigate the world and instead build creatures that learned as animals in the wild did, through experience. Brooks's robots, modeled on insects, carried out a range of simple behaviors such as "move," "turn," and "stop," etc. based on the input they received from their sensors. The theory was that more complex behaviors would emerge as the robot was forced to learn and make adaptations to its changing environment. Eventually, by watching these insect-like robots evolve, Brooks hoped to discover how to mimic life-forms higher up the evolutionary chain.
In October of 1989, after hearing a talk by Brooks at MIT, Tilden was inspired to switch from the traditional robots he'd been working on to ones that could be built with minimalist technology. He also took Brooks's ideas one step further; instead of building creatures with very simple brains, Tilden proposed to build ones that contained no computer processors at all. Nothing would be preprogrammed into his robots. Any intelligence they developed would emerge as a result of their learning to adapt to the world around them.
Brooks's and Tilden's theories made news because they cut right to the heart of one of the biggest disappointments in AI research: Despite years of development and millions of dollars of funding, artificial-intelligence researchers still haven't built a robot that could survive for long outside a laboratory. Even the smartest ones don't have the instincts of an insect, much less the common sense of a human being. They tend to perform poorly when taken out of their structured environments because they can't adapt to an unprogrammed world. Even Deep Blue, the IBM supercomputer smart enough to defeat Gary Kasparov, the world's former chess champion, lacks this basic animal intelligence. As AI pioneer Marvin Minsky put it: "Deep Blue might be able to win at chess, but he wouldn't know to come in from the rain."
Mark Tilden has founded a branch of robotics aimed at developing just this sort of animal intelligence in machines. It is known by the acronym BEAM, which stands for Biology, Electronics, Aesthetics, and Mechanics. The tenets of BEAM robotics are simple: Use minimalist electronics, recycle and reuse technoscrap such as defunct Walkmans and dead watches, and whenever possible use solar power as a source of energy. The result is a kind of everyman's robot, simple to build and easy to use. Tilden likes to call his creations "feral machines" and sells kits to make them on the Web. Last week, I ordered two BEAM robotics kits and prepared to try my hand at a bit of homegrown evolution. What follows is my robot diary.
Day 1
The robots arrive in the mail, two small Ziploc bags full of electronic parts. I tell myself that they will be more impressive when assembled. Still, they are clearly not the domestic robots of my dreams. After reading the instruction manual, my worst suspicions are confirmed. These creatures will not be able to wash my windows or mix my drinks. I am not to leave small children in their care nor will they walk the dog. Instead, like some sort of loopy New Agers, they are guided by three imperatives: "Go to the light! Avoid the darkness! Let nothing stand in your way!" That is to say that the manual promises only that my completed robots will be light-seeking and obstacle-avoiding and won't require batteries.
Day 2
According to Mark Tilden's calculations, building a traditional robot typically takes about eighteen months of manpower and programming. BEAM robots can be built in considerably less time. Somehow this does not reassure me. The instruction manual is filled with blurry pictures and daunting explanations. ("When the voltage level matches a preset point at the 1381, it sends a pulse out to the 2N3904 transistor, turning it on, then some power splits off through the 2.2.k resistor and goes into the 2N3906 transistor.") Hmm…
I decide to assemble a crackerjack team of experts (my architect roommate and computer whiz boyfriend) to help me build my robots. They are excited, as is every male friend that I tell about the project. All boys love robots.
Day 3
There is one obvious hitch to Mark Tilden's theory that robotics should take a cue from Mother Nature and try to evolve greater life-forms from lesser ones. Such a plan presupposes self-reproducing robots, which are not yet possible. Unlike many of his colleagues, Tilden isn’t even sure they'll ever be. The solution to this problem, he says, is to view a human being as a robot's way of making another robot. It is a weighty responsibility, I think, to be in charge of another species' procreation. Still, before we begin building the robots, I consider various names for them. After much indecision, I decide on Rod and Todd. We clear off the kitchen table and get to work, or rather my friends get to work while I serve a variety of hot and cold beverages. I have ordered two of the most basic BEAM kits; one is rated at a beginning skill level, the other at intermediate. It soon becomes apparent that an extraordinary amount of painstaking and time-consuming soldering is required in BEAM robotics. My roommate, who has previously constructed buildings and, more importantly, radios, wields the iron while my boyfriend assists. Four hours later, they have completed a small robotic car, slightly larger than a postage stamp. This car is called the Solar Speeder and is used for races in various robot competitions. We place it under a lamp to charge. The manual advises beginning roboticists to be patient as they wait for the Solar Speeder to make its first mad dash across the room. This can take up to three minutes, it says. We wait patiently. Forty-five minutes later the car still hasn't moved. A certain surliness descends upon my team of experts. One goes home and the other pronounces the Solar Speeder a complete dud. In a fit of pique, he fiddles with it one more time and miraculously, it begins to work. We put it back on the table and wait for the magic. The Solar Speeder zooms about a foot and a half and then stops. Thirty minutes later, it does this again. I consult the manual. This indeed is all our first BEAM robot is designed to do. I decide to name him Todd.
Day 4
The next day we build the Photovore Photopopper robot, aka "Rod." Although the Photopopper is a much more complicated kit, my roommate's torturous apprenticeship with Todd serves him well and he puts this one together in short order. When assembled, Rod is about the size of my palm. He resembles a cross between a hermit crab and a waterbug. As soon as we hold him up to the light, his motors begin to pulse. This is an excellent sign, according to the manual. Rod should now make a beeline for the brightest light, skillfully maneuvering around any obstacles in his way. We set Rod on the table and wait. After a moment, he lurches forward like a drunken cockroach, then stops. A minute or so later, he does it again. "Beeline" turns out to be a bit optimistic. Ten minutes later, Rod reaches the light. Later, I put a grape in his path and he puzzles over it for some time before finally making his way around it. Still, I must admit he seems to be an intellectual giant compared to Todd.
Day 5
I start my schooling of Rod and Todd by reading them Isaac Asimov's famous three rules for robots, which are essentially protect humans, obey humans, and then take care of yourselves. Later, I discover that Mark Tilden has invented his own rules for BEAM robots, which are as follows:
A robot must protect its existence at all costs.
A robot must obtain and maintain access to a power source.
A robot must continually search for better power sources.
Otherwise known as: Protect thine ass; Feed thine ass; Look for better real estate.
Tilden claims these rules will be enough to guide a robot in the wild, though domesticated robots will require more sophisticated handling. Wild robots will have to be bribed, tricked, or evolved into household roles, he says, and this will take time, but the wild robot is the start of everything. I look at Rod and Todd. By Tilden's definition, they are feral machines. I note that they appear unmoved by the rules of robotics.
Day 6
Rod appears to have fallen in love with a shoe. Three times I have moved this shoe (an unremarkable Converse sneaker) and three times in the night he has returned to its side.
Todd is less smitten. Once or twice he has zipped past the sneaker, taking little note of it.
Day 7
In the afternoon, I watch TV with Rod and Todd. All day they have seemed listless perhaps because I have repeatedly moved their favorite lamp away from them. A special report comes on and Peter Jennings announces that the Florida Supreme Court has ruled for Gore in a surprise decision. Todd races across the room and moments later Rod lurches towards the TV. From this behavior, I surmise that they are Democrats.
Day 8
In the interest of fostering the robots' nascent spirituality, I read to them from a book on near-death experiences. My hope is that they will become intrigued by the mysterious tunnel of light and connect it to their own experience of the world. They remain impassive in the face of testimonials from people who claim to have visited the other side. Later, it occurs to me that they were only pretending to listen.
Day 9
In the interest of speeding Rod and Todd along the evolutionary path, I decide to subject them to a variety of novel and complex stimuli and see how they adapt. To this end, I bring out a fearsome toy known as the Ladybug Parade, which consists of a giant polka-dotted bug trailing three revolving bugs in its wake. I wind up the Ladybug Parade and let it proceed. It rattles along for awhile, then stops a few inches away from Rod and Todd who are clustered in a corner of the living room. On one side of the robots, I place a large plastic frog, and, on the other, a small Lego forest ranger. For good measure, I throw in a harmonica, a pack of gum, and Rod's beloved shoe. Surveying the scene, I find it hard to imagine a richer laboratory environment for my feral machines. I take a chair in the far corner of the room and observe. After awhile, Todd zips forward and rams into the gum. "Aggression," I write in my notebook. Some time later, Rod inches tentatively toward the Ladybug Parade, then stops short. This I characterize as "fear." I wait and watch, but nothing else happens. Finally, I give up and turn off the light. All the creatures wait quietly until I leave the room. Later, I can't sleep and stand outside the door to see if I can tell what they are up to. It is only a matter of time, I think, before my robots run amok.
Jenny Offill is the author of the novel Last Things. She lives in
Brooklyn, New York.
Other articles by Jenny Offill
Back to Article