Computers Move Closer To Checkmating Humans
Chess tourney provides latest test of mind over megabytes
BOSTON — THE score is: Chess players 5, Computers 0.
But in the sixth Harvard Cup Human vs. Computer Chess Challenge, which starts today in New York, the chess-playing machines may finally win one.
"It should be a close match," says Larry Kaufman, an international chess master who is also the coauthor of a computer program, called Socrates, which will compete in the tournament. "The consensus is still that the grandmasters could win. But ... each year the computers get faster and stronger."
The tournament - an annual event designed to measure the progress of computer chess - comes less than two months before world chess champion Garry Kasparov faces off against IBM's "Deep Blue" computer in a much anticipated six-match contest in Philadelphia.
These tournaments have become one indication of how far artificial intelligence has come - and how far it has to go.
"Our event is a team competition," says Christopher Chabris, co-producer of the Harvard Cup. The two-day, 36-game tournament pits six top American players against six computer programs.
Just how good is chess software? Although computers have scored individual wins against the very best chess players - called grandmasters - the edge still rests with the players when they have adequate time to consider their moves. Mr. Kasparov, who has lost individual speed-chess games to machine opponents, has never lost a match to one.
There are some differences between this week's tournament and Kasparov's upcoming contest. For one thing, the players can't be expected to play at Kasparov's level. Another disadvantage: games are limited to 25 minutes. This time constraint generally favors the computers. In the upcoming Kasparov match, each game can last up to eight hours.
The Harvard Cup players do have one advantage. If they're not quite world champions, neither are their machine opponents. One of the programs - Chessmaster 4000 for Windows 95 - is available on store shelves. The others - Junior, Virtual Chess, Mr. Kaufman's Socrates, WChess (last year's best software at the Harvard Cup with a 5-1 record), and M-Chess Pro (the current world microcomputer champion) - are expected to be incorporated in future versions of commercial products.
On the other hand, IBM's Deep Blue system, which will take on Kasparov, is a high-end research computer. It is capable of calculating 100 million moves a second and has won an impressive string of matches against other chess-playing computers.
So far, the Harvard Cup players have consistently beaten their machine opponents. Since the first Harvard Cup tournament in 1989, when the machines won a paltry 9 percent of the games, the programs have gotten faster and more sophisticated. Last year, the computers scored their best ever, winning 39 percent of the games.
"It's hard to say whether this is artificial intelligence or not," says Joel Benjamin, a grandmaster who is the tournament's defending champion. "It's pretty clear that the way computers play chess is very different from the way humans do."
Chess computers are very good at making calculations. Mr. Kaufman's Socrates program, for example, calculates all the possibilities for the next three moves, then selects the most promising possibilities to calculate even more moves ahead. As a player, Kaufman says he occasionally might work out 10 moves ahead but only considers two possible lines of play.
"Humans have a very general approach and think about strategy," adds Mr. Benjamin. Computers "are performing a task they've been instructed to do."
This marks a big change from the late 1940s and early '50s, when artificial-intelligence scientists began contemplating a chess-playing computer. Back then, the idea of beating a reasonably competent player seemed an incredible advance, says Mr. Chabris of the Harvard Cup.
But computer scientists expected to achieve it by building machines that could play the game as people do. Since then, the chess programs have gotten good enough to routinely beat people, even whose who play quite well. But their reliance on brute-force calculations, rather than strategy, would have surprised some early artificial-intelligence researchers, who believed machines could mimic the way people think.