Chips vs. the chess masters
Five years after a historic defeat, humans may be poised for a comeback
When IBM's Deep Blue supercomputer beat chess champion Garry Kasparov five years ago, the case seemed closed: The wetware of the human brain was simply no longer a match for the hardware of a chess-playing machine. So what, exactly, is the point of Man vs. Machine, Round II--two upcoming contests that pit today's best chess programs against Kasparov and the current world champion, Vladimir Kramnik?
Deep Blue's 1997 victory over Kasparov felt hollow to computer chess experts. They had dreamed for decades about beating the world's best human, but in the event, Kasparov fell apart psychologically. He later said he felt "ashamed" of the way he'd played, having made an obvious blunder in the final, deciding game. "I should have been exulting, but I was feeling empty inside," writes Deep Blue team member Feng-Hsiung Hsu in Behind Deep Blue, a memoir published this fall. "The game felt too easy."
As a result, Kramnik says, no one really knows how computer chess measures up. "The question was still open, is still open: Is it stronger than, let's say, the strongest human being?" Deep Blue isn't available for a rematch, so these new contests aim to answer that question. But this time it could be the computers that aren't at the top of their game. Some experts say that, compared with Deep Blue, they may actually be a step backward.
On the fritz? This week Kramnik will take on one of them, a German program named Deep Fritz, in Bahrain. Kasparov, not to be outdone by his rival Kramnik, will go to Jerusalem in December to compete against Deep Junior, the current world computer chess champion. Both software teams say their creations are much tougher opponents than Deep Blue. In a computer-on-computer match against the IBM group, "I would expect them to be killed by Junior. Maybe by others," says Amir Ban, cocreator of Deep Junior. "These programs are playing at least as good as Deep Blue," agrees Monty Newborn of McGill University, whose book Deep Blue: An Artificial Intelligence Milestone will come out in October.
Yet while the power of computer chips has marched forward over the past five years, that doesn't necessarily mean these new cyberchamps would outperform Deep Blue. A chess-playing machine rather than a mere program, Deep Blue drew its awesome power from chips designed by Hsu to do nothing but play chess. The IBM team put 256 of these processors into a supercomputer, allowing it to analyze at least 100 million chess positions a second. Fritz and Junior, by contrast, exist as off-the-shelf software for PCs, which anyone can buy to play at home. The "deep" versions run on multiple Pentium processors--essentially, a battery of PCs--but they'll consider only around 2.5 million positions per second.
Finesse in the software can help make up for a relative lack of brute force. As a program looks deeper and deeper into an opponent's possible future moves, the number of board positions explodes, overwhelming even the fastest computers. Techniques for choosing the most advantageous move from the fast-growing tree of possibilities become critical. The current top programs use strategies that Deep Blue didn't--for example, "pruning" away unpromising lines of play.
But can software alone make up for the power gap? Press releases touting this week's match boast that Fritz has beaten both Kasparov and Deep Blue. The win over Kasparov came, however, in a superfast kind of chess, where computers have a decided edge. And Fritz didn't really beat Deep Blue--it beat an early version of its software running on slower hardware.
Robert Hyatt at the University of Alabama doubts that Deep Blue has been bested. He wrote a program for a Cray supercomputer that, in the 1980s, defeated every other machine until Deep Blue's predecessor, Deep Thought, came along. And Deep Blue was up to 100 times faster. "Once you've been run over by a freight train, you know it's real painful," says Hyatt. "But until you do, you can only imagine."
Imagine is truly all anyone can do. Deep Blue played against Kasparov and then disappeared, dismantled after negotiations for a rematch with Kasparov fell apart. "At the end of the day, IBM's business is not in the chess computing software business," says My Luu, a public relations manager at IBM.
Singing the blues. But Deep Blue's untimely demise has left a hole at the center of this field. "Of all the things that have really hurt computer chess," says Hyatt, "that has hurt." Frederic Friedel, head of Chessbase, the company that created Deep Fritz, says it's as if scientists went to Mars and, instead of exploring, simply headed back home.
Victories by Deep Fritz and Deep Junior could dispel any doubts that computers have the upper hand. But most experts predict Kramnik and Kasparov will win easily if they play their best. Some critics say the software makers and the grandmasters are both hyping the programs' prowess and predict the match will prove more useful for selling software than advancing science. Kramnik, however, says he used Fritz to analyze the Deep Blue vs. Kasparov games and found Fritz suggested much better moves. "It's simply a stronger program," he says. "So I know it's going to be an even more serious challenge."
Yet a human victory will only postpone the inevitable. Already, the top humans can't beat a computer in checkers, and chess will go the same way. That's because chess, at its core, is a game of complex mathematics. While software gets better and runs faster every year, the human brain, well, doesn't. "We're at a balance point," says McGill's Newborn. "But the pendulum is flipping in the other direction, and it won't come back."
This story appears in the October 7, 2002 print edition of U.S. News & World Report.