Big Blue wants to help science find life-saving drugs, forecast hurricanes, and study galaxies. It wants to divine the economy. First, though, it wants to whip Gary Kasparov's butt.
IBM is into chess. Its Deep Blue supercomputer, the fruit of eight years of research, will take on world chess champion Kasparov in a six-game match in Philadelphia starting on Feb. 10. The winner gets $400,000 from the sponsoring Association for Computing Machinery. No computer has ever beaten a world champion in a standard tournament of unhurried games--but this time, some chess fans are wagering on the machine.
No wonder. Deep Blue is a new sort of supercomputer, based on a standard parallel-processing system called the Risc System 6000 Scalable Power Parallel System. That machine--SP for short--is already pretty fast. But IBM's crew made it go perhaps 1,000 times faster on chess problems by surrounding each of its 32 processors with eight chips hard-wired to do nothing but race through the analysis of millions of chessboard positions. Deep Blue can sort through an awesome 100 million positions per second.
MIDGAME MANEUVERS. The hybrid design has commercial potential, IBM believes. The basic computer can be programmed to handle all kinds of tasks. And when it comes time for a specialized job--drug design, say--the extra chips kick in horsepower on rote tasks. While Deep Blue isn't for sale, similarly customized SP machines could come to market in the next couple of years at prices ranging from $150,000 to tens of millions of dollars.
Kasparov admits to being a bit spooked by Deep Blue, especially since the brand-new computer hasn't played any games for him to analyze. He thinks he will win, though, by playing as coolly as the computer itself: "Mistakes come when you're in a fury or nervous." And he believes he'll be able to detectDeep Blue's weak points. A computer, he says, is hamstrung by its inability to adjust its programmed-in priorities during a game. So if Deep Blue is instructed to give top priority to keeping the king safe, it continues to do so even when developments in the game dictate a shift in strategy.
IBM agrees that the inability to learn midgame is Deep Blue's Achilles' heel. So can it win? Chung-Jen Tan, leader of the Deep Blue engineering team, is a bit diffident: "We're pretty confident, but it's hard to tell," he says. Kasparov, typically, is bolder: "My general knowledge of the game and my ability to keep the game under control will eventually prevail." Deep Blue wasn't available for comment.