By Anna Pulley
By Erin Sherbert
By Chris Roberts
By Erin Sherbert
By Rachel Swan
By Joe Eskenazi
By Erin Sherbert
By Erin Sherbert
On Jan. 12, 1992, Dr. David G. Stork hosted a birthday party for ... well, you might say a friend. Stork is a professor of electrical engineering at Stanford, as well as a huge fan of Stanley Kubrick's 1968 film 2001: A Space Odyssey; by his count, he's seen the movie 43 times. Because he works in the field of artificial intelligence, he takes particular interest in any scene that involves the HAL 9000, the remarkably lifelike computer that operates the spaceship on its trip to Jupiter. Stork regularly lectures on HAL at conferences and universities. Over time, he has met nearly everyone intimately involved in the making of 2001 except Kubrick, and friends keep passing along HAL-related memorabilia. He owns two props from the film. One is the computer's faceplate. The other is the "key" astronaut Dave Bowman uses to lobotomize HAL toward the end of the movie.
As HAL gasps in that scene, "he" became fully operational on Jan. 12, 1992, in Urbana, Ill. Just for fun, Stork decided to celebrate the occasion by inviting over some friends and colleagues. The Associated Press even showed up, and photographed the revelers as they stood around a HAL-styled birthday cake and sang "Daisy," HAL's dying words, in honor of the miraculous computer that was, in HAL's own words, "foolproof and incapable of error," a gifted lip-reader, a kick-ass chess player, an elegant conversationalist -- and a coldblooded killer of four men.
Dr. Stork's big idea has parallels to HAL, but he's loath to suggest that what he's working on in Menlo Park is a HAL per se -- it's not that simple, and besides, it makes him look a bit like a mad scientist. In truth, he's a mild-mannered, middle-aged man with a Ph.D. in physics, a 20-page CV, and a job as chief scientist at Ricoh Silicon Valley, where he mainly researches new technologies for the Japanese imaging giant. His office is cramped and modest, and his work seems modest as well, when you consider that most of his office neighbors on this stretch of Sand Hill Road are the world's leading venture capital firms.
For nearly three years, with Ricoh's blessing, Stork has been quietly working on and promoting a project called the Open Mind Initiative. To define it roughly, it's a way of advancing research on artificial intelligence (AI) by realizing that scientists alone can't create a useful model for human intelligence. Instead, the Open Mind Initiative plans to harness the collective brain power of millions of Web-enabled regular people to contribute their everyday experiences, observations, knowledge, and old-fashioned common sense.
This marks something of a radical break in the field. Thus far, AI research has been performed mainly in labs with computers that are fed cold mathematical models and huge sets of data of a specific type. The most famous result of that approach is IBM's Deep Blue computer program, which beat chess grandmaster Garry Kasparov in 1997 and led many to think that we were now in the era of computers that could outstrip human intelligence. But Deep Blue was physically incapable of savoring the victory -- or even comprehending the concept of "victory" itself. And, as Stork points out, Deep Blue hasn't the slightest idea how to play checkers.
"The hardest problems in AI have been the ones that 2-year-olds do all the time, effortlessly," says Stork. "They can tell the difference between a cat and dog just by looking. No computer can do that." So what the Open Mind Initiative is focusing on, for starters, are the aspects of humanity that aren't so easy to codify into algorithms -- our capacity to understand speech, to recognize handwriting, and to comprehend very basic, common-sense ideas. The Initiative wants to tackle "simple, basic facts," says Stork. "For example: "Bill was hungry, so he went to the store, saw an apple on the shelf, and bought it.' How do we know he bought the apple and not the shelf? I don't even have to tell you it was the apple. It's that kind of knowledge that undergirds so much of our language and knowledge about the world. Getting a machine to be intelligent requires that. There's really no way around it."
Investigating those ideas requires gathering much more data -- and more "humane" data -- than has previously been available. But, watching the steady growth of Internet users over the last five years, Stork realized that the whole wired world can be the world's largest data set -- and who better understands common-sense issues than common people? Anybody online can voluntarily contribute information on (theoretically) anything. For example, users might be asked to write a short paragraph to describe a simple picture of a chair in a room. In a case like handwriting recognition, a typical Internet user might be presented with a written number and would be asked whether it was a four or a nine. The Initiative has already developed a version of the children's computer game Animals, in which a computer tries to guess what animal someone is thinking of through a battery of yes-or-no questions. Some of the "common sense" statements being tackled are "you can use lies to confuse," "ice can be formed into cubes," and "the first thing you do when you take a shower is get undressed" -- no-brainers for people, but stunning feats of cognition for a computer. The plan is to pool and then process the results of those responses to create a computer that can make intuitive leaps only humans are capable of now. Users would get incentives to participate -- play the handwriting-recognition game for a while, for example, and you'd receive a certain number of frequent-flyer miles.