APS News

This Month in Physics History

February 1996: Kasparov vs. Deep Blue

Kasparov vs. Deep Blue

Ever since the introduction of the sentient computer HAL in Stanley Kubrick's 2001: A Space Odyssey, the explosive advances in computing technology have begged the question: Can truly intelligent computers be constructed? Can a man-made machine ultimately out think its creators? In February of 1996, a computer known as Deep Blue, developed by IBM researchers, made history when it took on the reigning world champion of chess, Gary Kasparov, for a series of six games.

The origins of Deep Blue date back to the dawn of modern computing. The first chess program was written by Alex Bernstein of MIT in the late 1950s. When Kasparov first became World Champion in 1985 at the age of 22, solid chess playing machines were already being constructed. In the 1970s, the Machack IV computer became the first to play in a human chess tournament, and with the introduction of integrated circuits, the first chess playing computers went on the market in 1976. It wasn't until 1983 that a computer managed to triumph over a chess master in any tournament, and the Deep Thought project launched a few years later lost miserably to Kasparov in 1986.

In 1995, a Carnegie Mellon doctoral student named Feng hsiung Hsu began developing a chess playing computer called "Chiptest." After earning his PhD, Hsu joined the research staff at IBM and he and his colleague (and former classmate) Murray Campbell adapted his work on Chiptest as part of an effort to explore how to use parallel processing to solve complex computer problems. The Deep Blue project was born.

In order to build a chess playing computer capable of testing the best chess players in the world, Hsu's team sought to design a chess specific processor chip. Deep Blue had 128 processor chips running in parallel, enabling Deep Blue to calculate one billion positions per second. Yet 97% of the computer was constructed from components that could be purchased by the average consumer.

Kasparov was confident going into the match, but Deep Blue stunned the experts by winning the first game. It accomplished this by offering a pawn sacrifice early in the game to gain a lead in position: a common strategy among chess players, but risky, since the outcome is uncertain. The computer went on to recover the sacrificed pawn, ultimately winning the match. Kasparov later told TIME magazine that he was "stunned" by the computer's decision to sacrifice a pawn. "I had played a lot of computers, but had never experienced anything like this," he said. "I could feel a new kind of intelligence across the table."

Kasparov recovered his equanimity and ended up winning the match, winning three games and playing two to a draw to collect the $400,000 prize. He later said that he eventually defeated the computer by switching strategies mid game, since the computer did not so much think, as react to its opponent's moves. "My overall thrust was to avoid giving the computer any concrete goal to calculate toward," he said. "So although I did see some signs of intelligence, it's a weird kind, an inefficient, inflexible kind that makes me think I have a few years left."

As it happened, he had about one year left. In May 1997, Kasparov faced the latest, improved iteration of Deep Blue in a rematch that made history. Early in Game 6, Kasparov made a disastrous mistake, allowing Deep Blue to sacrifice a knight and obtain an overwhelming positional advantage, going on to take Kasparov's queen in exchange for a rook and a bishop. Kasparov resigned the match after only 19 moves. It was the first time a current world champion had lost a match to a computer opponent under tournament conditions.

The biggest improvement made to Deep Blue in the year following the first match with Kasparov was speed, thanks to faster processors that gave the computer to evaluate 200,000,000 positions per second. [For comparison, Kasparov can examine approximately three positions per second.] In addition, Deep Blue's general knowledge of chess was significantly enhanced through the efforts of IBM consultant and international grandmaster Joel Benjamin, so that it could draw on vast resources of stored information, such as a database of opening games played by grandmasters over the last 100 years. The increase in computing power also allowed Deep Blue to adapt to new strategies as the game progresses a weakness cleverly exploited by Kasparov to win the first match.

Campbell and his IBM cohorts took the lessons learned from building the system and applied them to other complex and difficult problems that required a tremendous amount of computational power.

Today, massively parallel computers are being applied to finance, medicine, education, and nearly every other major sector, not just in the US, but worldwide. Scientists have not yet created artificial intelligence, but systems like Deep Blue and its descendants have helped us make better use of the real thing.

APS encourages the redistribution of the materials included in this newspaper provided that attribution to the source is noted and the materials are not truncated or changed.

Editor: Alan Chodos
Associate Editor: Jennifer Ouellette