Quick moves claim computer-chess title.

Quick moves claim computer-chess title

After losing decisively last fall to world chess champion Gary Kasparov, chess computer Deep Thought returned to the digital world last week and successfully defended its title as the North American computer-chess champion. But it wasn't easy. Deep Thought lost one game to Hitech--only its third loss to a machine--and had to share the title with Mephisto, a strong contender from Germany, which also lost just one game.

This year's championship, sponsored by the Association for Computing Machinery and held at the Supercomputing '90 meeting in New York City, featured 10 chess machines and computer programs. Belle, world champion in 1980, came out of retirement to participate in the tournament but managed to win only one game.

"That shows how much computer-chess programs have changed and improved," says David Levy of Intelligent Software Ltd. in London, England.

Mephisto ranks as the top commercially available computer-chess player. Last April, a Mephisto computer became the first machine to defeat a former holder of the human chess title when it beat Anatoly Karpov. Even though Karpov was simultaneously playing 23 other opponents, the computer's success remains significant, Levy says.

Last year, Mephisto beat Deep Thought in the final round of the computer-chess championship. This time, Deep Thought won the rematch. "Mephisto played a horrible move in the opening and never recovered," Levy says. "It was a typical computer move, which computers make very often in certain positions, and most people haven't yet worked out how to get it out of their programs."

Deep Thought has remained fundamentally unchanged over the last two years and has a number of weaknesses, says Feng-hsiung Hsu, one of its creators. Now working at the IBM Thomas J. Watson Research Center in Yorktown Heights, N.Y., Hsu and his colleagues are developing a more sophisticated version that can also respond to patterns.

To defeat Deep Thought, Hans Berliner of Carnegie Mellon University in Pittsburgh developed a strategy for giving Hitech a significant advantage at the beginning of the game. That proved enough to overcome Deep Thought's greater speed, which normally provides a clear advantage at the end of a game.

Zerker, a promising newcomer developed at the University of California, Berkeley, can search roughly three times faster than Deep Thought, evaluating up to 7 million moves in 1 second. But damage to the machine during shipment from California forced its withdrawal.