Computerworld - She may be only a robot baby orangutan, but someday Lucy may tell us about how the cerebral cortex of the brain works.
And when she does, she'll be able to help people develop and build new computational architectures inspired by biological systems, as well as applications based on those systems that are more adaptable, intelligent and robust, according to Steve Grand, Lucy's creator.
Grand, a recognized authority on artificial life and the founder of Cyberlife Research Ltd., an artificial intelligence research company in Somerset, England, has written a book about Lucy, Growing Up With Lucy: How to Build an Android in Twenty Easy Steps (Weidenfeld & Nicolson), due out in January.
Grand says he hopes Lucy can tell him the basic operating principles of the brain—engineering that evolution discovered when it came to making nervous systems but that mankind hasn't yet unraveled. He is looking to use the neural building blocks of the brain as a map for creating AI.
Grand says there are two kinds of AI: "soft" AI, which tries to create high-level reasoning by explicitly programming rules for it into a computer, and "hard" AI (his preference), which involves making machines that are genuinely intelligent and can teach themselves. Enter Lucy.
"What I'm interested in are . . . the principles that enable a brain to organize itself into a set of machines that enable it to do all the things that brains do," Grand says. He wants to find the basic principles that enable the cerebral cortex to wire itself up in response to experience, until it becomes a very complex and specialized set of computing machines.
"How does that happen? It's completely unlike any technology we've ever made. It's as if you could take 50 million transistors and stick them in a heap on the carpet and show them Microsoft Office, and half an hour later, they'll spontaneously assemble themselves into a computer."
Grand says he wants to replicate that in Lucy with neural networks simulated on PCs. Lucy's intelligence will be a consequence of the interactions between thousands of simulated neurons. His goal is to develop a machine that can supplement or even supercede the digital computer—a machine that can think and learn.
On Her Own
Grand says Lucy is developing the ability to learn by herself. So far, she has learned to point to a banana—any banana: a green banana, a yellow banana, a big banana, a small banana. If you show her an apple and a banana, she points to the banana.
Grand says he hasn't programmed Lucy to do that; instead, he has given her a model of the bit of cerebral cortex that knows how to do it. "It doesn't sound like a huge achievement. Why not just program a computer to recognize yellow?" he asks.
But, he says, Lucy solved a whole series of problems by herself like detecting the lines that form the edges of the banana regardless of what position it's in or how far away it is, and she figured out how to point at it.
Thanks to a $68,000 grant from The National Endowment for Science, Technology and the Arts in London, Grand purchased 15 new computers and is building an improved Lucy.
When Lucy's new body is complete, she should have a voice as well as better eyesight and hearing, and she should be able to move her arms and legs, Grand says. He hopes Lucy will soon be learning to crawl and ultimately walk. He also hopes she will be able to repeat simple sounds, like toddlers do.
"Lucy won't be very smart, but it won't be far from the truth to say she'll have a mind of her own, albeit a very, very stupid one," Grand says.
Think of the robots that build cars, Grand says. "They don't adapt. If they go to pick up a part and it's not there, they pick up air. We have to build them so that one day they will be able to adapt to [different situations]."
But the question remains: Will he be able to do it?
Limits of Knowledge
Well, maybe, says Larry Yaeger, an expert on AI and a distinguished scientist at Apple Computer Inc. who lives in Bean Blossom, Ind. "With Lucy, [Grand] appears to be taking no shortcuts with sensory inputs or motor outputs, as he is striving to integrate real vision and audition, as well as voice, arms and legs," Yaeger says.
That said, Yaeger claims that the greatest difficulty Grand faces is mankind's limited knowledge of human brains. "The wiring diagrams, the details of the different kinds of neural and synaptic mechanisms and the almost unexplored influence of the baths of chemicals our brains are awash in, [are] still very much in its infancy," he says. "But it's possible that what we already know is enough for Steve to succeed."
But, Yaeger cautions, "I believe that . . . the complexities of those wiring diagrams are more likely to yield to evolution than to engineering. [Grand] believes design is the answer. I think evolution, and a willingness to evolve and learn from very primitive organisms first, before we hope to obtain simian or human-level intelligences in the computer, may be the better approach. But if anyone on the face of the earth can engineer intelligence from scratch, I believe it would be Steve Grand."
Free Insider Guide