He may be a humble researcher dedicated to the pursuit of knowledge, but in May, C. J. Tan might just cut loose and play to win. That’s when his new and improved chess-playing supercomputer, Deep Blue, is scheduled to meet world chess champion Garry Kasparov in a rematch of last February’s heartbreaker, which Deep Blue, despite surprising almost everybody by taking the first game, lost, 4-2.
This time, Tan and his colleagues at ibm think Deep Blue has a better shot at being the first machine to beat the reigning human in a championship match. For one thing, they’ve learned the hard way that raw computing speed--Deep Blue crunches 200 million moves a second, to Kasparov’s two--is simply not enough to beat the formidable Russian. Tan says he’s not even going to bother making the machine any faster for the rematch. What we lacked in the system was chess-specific knowledge and strategy, and those are the areas we are concentrating on, he says. Since August, grandmaster Joel Benjamin has been playing Deep Blue almost every day, pointing out weaknesses in the computer’s game.
Tan expects new programming tools to eliminate another of Deep Blue’s weaknesses: its inability to adapt to its opponent’s strategy and change its own playing style. This disadvantage became painfully apparent last February when, midway through the match, Kasparov shifted to a less confrontational style, making Deep Blue flounder like an amateur. Perhaps the biggest change this time, however, lies in Tan’s own attitude toward the game. Last February his scientific curiosity got the better of him in the crucial fifth game, when he refused Kasparov’s offer of a draw. Kasparov went on to win the game. Last time we were really just doing an experiment, says Tan. This time I have very high confidence that we will do very well.