NEXT week, the world’s top chess player, Garry Kasparov, will face a non-human opponent for the first time since IBM’s Deep Blue computer defeated him in 1997. But his latest challenger, Deep Junior, is a new kind of chess program. A hundred times slower than Deep Blue, it relies on strategy instead of prodigious number crunching, say its Israeli creators.
The World Chess Federation is sponsoring the six matches which will take place in New York, beginning on 23 January. A million dollar prize is at stake. The contest is also Kasparov’s chance to trump his rival, Vladimir Kramnik, ranked second in the world. Kramnik drew with the German chess computer Deep Fritz last year.
Deep Junior promises to be a formidable opponent. While Deep Blue relied on sheer computing power to calculate as many moves as possible, searching through 200 million positions per second, Deep Junior runs at least a hundred times slower, say its creators Shay Bushinsky, former chief executive of the Internet chess company ChessDev, and Amir Ban, co-founder of the flash disc company M-Systems.
Advertisement
Instead of sifting through all possible moves, Deep Junior decides on the most fruitful areas to search. The program assigns weight to each possible move depending on its complexity. This helps it decide whether to calculate moves further down the line. The approach is perhaps closer to the way a human would play, and allows the computer to come up with more interesting and powerful moves, say Bushinsky and Ban. They have also been teaching Deep Junior to learn from its mistakes in previous games.
Humans have also progressed since 1997. Man has been studying machine – Kasparov has been practising against various versions of Deep Junior for the past five years. But the latest incarnation still has a few new tricks, says Bushinsky.
While we wait for the results, the scientific community is divided over whether such matches can really tell us anything useful about intelligence. “Chess has outlived its usefulness,” says Jonathan Schaeffer of the Games Group at the University of Alberta’s Department of Computer Science. “It turns out to be easier than we thought.” Speaking at the “Man Versus Machine: The Experiment” symposium at Israel’s Haifa University in October last year, he suggested that the popular Chinese board game “Go” is more interesting than chess, as it cannot be won simply by searching through all possible moves. “If you want to understand intelligence, Go is much more demanding.”
But Fernand Gobet, an expert in intelligent systems at the University of Nottingham, does not agree that chess is no longer worth studying. He is trying to model how grandmasters think, based on a theory that they see the board as patterns of groups or “chunks”, rather than individual pieces. “Chess is a very rich domain of research for psychology, especially for studying complex decision-making,” he says.
Others are working on modelling opponents. Rather than trying to calculate the best move, the computer is taught to take into account its opponent’s weaknesses. “This turns out to be extremely difficult,” says Jaap van der Herik of Maastricht University’s Institute for Knowledge and Agent Technology. But he reckons it will be worth the effort – if you can model a player’s thought processes, he says, perhaps you could model any opponent, even a political one.
Future chess-playing computers will undoubtedly get wilier. If Kasparov cannot beat Deep Junior this time round, he may have missed his chance.