Deep Blue vs. Kasparov: the match that shook the world (1997)

On May 11, 1997, a machine defeated the best chess player alive. Deep Blue, an IBM supercomputer designed by Feng-hsiung Hsu and a team of engineers, beat world champion Garry Kasparov 3½–2½ in a six-game match in New York, the first time a computer had won a match against a reigning world champion under standard tournament conditions.

The event was broadcast worldwide. It made front pages everywhere. And it forced a question onto the public stage that researchers had been debating for decades: what does it mean when a machine outperforms a human at a task considered the pinnacle of intelligence?

The road to 1997

The project began in the 1980s at Carnegie Mellon University, where Hsu built ChipTest, a custom chip designed to evaluate chess positions at extraordinary speed. The approach was not subtle: evaluate millions of positions per second using brute-force search, guided by an evaluation function tuned with help from grandmaster consultants.

IBM recruited Hsu and the project evolved through Deep Thought into Deep Blue. By 1996, Deep Blue was fast enough to evaluate 200 million positions per second. In February 1996, it played Kasparov and lost the six-game match 4–2, but won the first game, the first time a computer had beaten a reigning world champion in any single game under standard time controls.

For the 1997 rematch, IBM doubled Deep Blue’s speed and, crucially, refined its evaluation function. The team studied Kasparov’s games extensively. Kasparov, meanwhile, had no access to Deep Blue’s preparatory games.

The match

The 1997 match unfolded dramatically:

• Game 1: Kasparov won. But a bug in Deep Blue’s code caused it to make a random move when it couldn’t find a good option, and Kasparov, not knowing this, interpreted the seemingly purposeless move as evidence of deep strategic thinking. It unsettled him.
• Game 2: Deep Blue played a move on turn 37 that stunned observers, a positional sacrifice that looked like intuition. Kasparov, shaken, resigned, only for post-game analysis to reveal he could have forced a draw. He publicly accused IBM of cheating, alleging that a human grandmaster had guided the computer’s moves.
• Games 3–5: Three draws. Kasparov played cautiously, visibly affected by Game 2.
• Game 6: Kasparov blundered in the opening and resigned after just 19 moves. Deep Blue won the match.

The controversy

Kasparov’s cheating allegations dominated the aftermath. He demanded to see Deep Blue’s log files; IBM initially refused. He called for a longer rematch; IBM declined. IBM then retired Deep Blue and dismantled the hardware (components later went to the Smithsonian), fuelling further suspicion.

IBM denied any improper human intervention, stating that only routine bug fixes occurred between games, as permitted by the match rules. No evidence of cheating was ever found. Kasparov eventually retracted his accusations in 2016.

The deeper question, though, was not about cheating but about what Deep Blue’s victory meant. Kasparov himself later described the machine as “as intelligent as your alarm clock.” Deep Blue did not understand chess. It did not learn. It searched, 200 million positions per second, guided by hand-tuned evaluation functions. Two grandmasters who played it described the experience as “a wall coming at you.”

Why it matters

Deep Blue proved that raw computation, combined with domain-specific engineering, could defeat the best human at a task long considered a benchmark for intelligence. But it did so without anything resembling general intelligence, learning, or understanding.

The victory marked the end of chess as the frontier challenge for AI. Researchers increasingly recognised that beating Kasparov had not brought them closer to building thinking machines, it had simply shown that search and hardware could brute-force a narrow problem.

The real successor to Deep Blue’s legacy was not another chess engine but AlphaGo, which, nearly twenty years later, beat the world champion at Go using deep learning and reinforcement learning rather than brute force. Where Deep Blue searched, AlphaGo learned. That shift, from hand-crafted evaluation to learned representation, is the defining transition of modern AI.