Chess Match Tests Limits Of Computer 'Intelligence'
Champ Garry Kasparov takes on IBM's Deep Blue tomorrow in six-game contest.
In the realm of game-playing, computers and humanity are running neck-and-neck.Skip to next paragraph
Subscribe Today to the Monitor
Computers are better at checkers. People hold the edge in bridge, Scrabble, and backgammon. And in chess, where the most artificial-intelligence work has been done, a computer has never beaten the reigning world champion in a full regulation match.
That is why tomorrow, when chess champion Garry Kasparov and IBM's Deep Blue computer meet in New York to play the first game of their six-game rematch, the world will be watching. The games computers play tell us not only about the state of artificial intelligence. They also reveal much about ourselves and the complexity of human intelligence.
The message is a comforting one. Even though these machines are beginning to beat us at our own games, their "smarts" and mankind's intelligence are fundamentally different. Thus, the foreseeable future will not entail some apocalyptic vision of mankind versus machine. Overall, computers cannot match human wits. Instead, artificial intelligence will complement real intelligence.
"I'm more and more impressed at how people work - that they can do what they do," says Paul Rosenbloom, an artificial-intelligence researcher at the University of Southern California in Los Angeles. "The more we understand about how they work ... the more amazing we think they are."
Knowledge researchers are gaining from game-playing computers is being applied to a variety of fields, from calculating investment decisions to scheduling transit routes to war games.
Dr. Rosenbloom's institute, for example, is working on computers that simulate the behavior of helicopter pilots for the Pentagon's simulated battles. The military needs the computerized "agents" because it doesn't have the resources to conduct real war games with tens of thousands of participants. At times, these computer "agents" have worked on par with the tactics of a third-world pilot; at other times, they've been incredibly stupid, as when the company commander returned to base alone after a mission, not realizing the need to bring back the rest of the team.
That's the problem with artificial intelligence. It doesn't come with any built-in common sense. "One of the interesting things we've learned through artificial intelligence ... is that intelligence is very complicated," says Itiel Dror, a cognitive neuroscientist at Miami University of Ohio in Oxford. Despite 40 years of work in the field, scientists cannot agree on a definition of intelligence.
Is IBM's Deep Blue chess computer intelligent? Its creators say no, calling it a "turbo-charged expert system" instead. The machine doesn't even know it's playing chess, they point out. But if the definition of intelligence is processing data and using knowledge to accomplish a goal, then Deep Blue is intelligent, other researchers argue.
In fact, the computer employs the same skills that people use when playing chess, but in different proportions. Deep Blue's main strength is that it can search through 200 million chess positions a second to come up with the best move. (That's twice as fast as last year's machine, which Mr. Kasparov beat 4-2.) By contrast, Kasparov, widely acknowledged as the best player in chess history, can calculate only about three positions a second.
His advantage is that he can recognize and evaluate chess positions - how the pieces interrelate in an overall structure - and develop a strategy. Deep Blue has chess knowledge but is poor at developing a long-term strategy. The point is that by emphasizing different methods, Deep Blue and Kasparov are pretty evenly matched (although chess experts expect Kasparov to win).
"I think the most important thing that we are going to learn from all this is that we and machines are completely different," says Matt Ginsberg, founder of the Computational Intelligence Research Lab at the University of Oregon in Eugene.
Dr. Ginsberg is organizing an international conference on artificial intelligence in July, where some of the world's top players will square off against computer opponents in backgammon, bridge, checkers, Go, poker, and Scrabble, as well as chess. The sides are fairly evenly matched in most of these games. The exception is Go, an ancient Oriental strategy game, which has so many potential moves that even computers choke on calculating the possibilities.
Researchers say that their game research has many real-world applications, from data mining (where computers search data to find meaningful patterns) to financial modeling (where investors need to digest data rapidly to make split-second trades). But will computers ever be as intelligent as people?
To achieve that, researchers need a much deeper understanding about what makes up intelligence - and they are decades away from understanding emotion and consciousness. The question boils down to this: "What is really a human being?" says John Yen of the Center for Fuzzy Logic, Robotics and Intelligent Systems at Texas A&M University at College Station.
He adds: "I personally don't believe all aspects of human intelligence can be duplicated by a computer. I just don't believe that we can do what God does."