Deep Blue

A machine that beat the world champion

In May 1997 IBM's Deep Blue became the first computer system to defeat a reigning world chess champion in a match played under standard tournament time controls ^[1]. Playing conditions mattered as much as the result: the machine won under the same clock and rules used in human competition, not in a specially arranged demonstration.

How it played

IBM says Deep Blue used 32 processors to perform a set of coordinated, high-speed computations in parallel ^[1]. Working together, those processors allowed it to evaluate 200 million chess positions per second ^[1]. IBM gives the machine's processing speed as 11.38 billion floating-point operations per second ^[1]. For scale, IBM's first supercomputer Stretch, introduced in 1961, had a processing speed of less than 500 flops ^[1].

A chess program works by assigning a value to a position — how favourable it looks for one side or the other — and then looking ahead through possible moves and replies. The more positions a machine can examine in the time available, the deeper it can look, and the less often it is caught out by a line it never considered. Deep Blue's advantage lay in doing that examining on many processors at once.

From ChipTest to Deep Blue

The project began in 1985, when Carnegie Mellon University graduate student Feng-hsiung Hsu helped develop ChipTest, which won the North American Computer Chess Championship in 1987 ^[1]. Hsu and Murray Campbell then built Deep Thought, which IBM describes as the first program to defeat a grandmaster ^[1]. In 1989 IBM Research hired Hsu and Campbell, and the chess computer was renamed Deep Blue as a nod to IBM's nickname Big Blue ^[1]. Among the other computer scientists on the team, IBM names Joe Hoane, Jerry Brody and C. J. Tan ^[1].

The work between the two matches

The team did not stand still between the 1996 and 1997 matches. It improved the endgame databases, which store known results for positions with few pieces left, and created a more powerful evaluation function — the part of the program that judges how good a position is. The team also hired additional chess grandmasters, so that human expertise could be built into the machine, and developed methods to disguise the computer's strategy, so that its choices would be harder to read from the outside ^[1].

What it left behind

IBM says the underlying technology advanced supercomputing for drug discovery, financial risk assessment, pattern discovery in large databases and gene research ^[1]. The company also says Deep Blue's architecture was later applied widely in financial modelling and in data mining ^[1]. After the 1997 victory, Deep Blue was retired to the Smithsonian Museum in Washington, D.C. ^[1].

Deep Blue is remembered less as a chess product than as a public demonstration of what a large parallel machine could do against a task long treated as a test of human intelligence. The hardware was soon overtaken, but the pattern it established — parallel computation applied to problems that can be split into many small pieces — became ordinary practice.