Deep Blue beats Kasparov
Era 3 · From the expert boom to the statistical turn (1980–2011) › Expert systems and the second winter
- Lineage
- Symbols and logic
- Kind
- Popularization
Who
IBM
What happened
In New York, Deep Blue, IBM's chess computer, wins a six-game match against the world champion, Garry Kasparov, by 3½ to 2½. The sixth game, on May 11, 1997, decides the result.
Why it mattered
For the first time, a machine beats a reigning world champion in a match under tournament time controls. It does so through massive search on specialized hardware, not through learning.
The full story
Since Turing and Shannon, chess had been AI's favorite test bed. Deep Blue was born at Carnegie Mellon, where a student, Feng-hsiung Hsu, developed ChipTest and then Deep Thought, which in 1988 beat a grandmaster, Bent Larsen, in a tournament game. IBM hired the team, with Murray Campbell and Joe Hoane, and built a machine with 30 processors and 480 chips designed for chess, which in 1997 examined an average of about 126 million positions per second and peaked at 330 million.
In February 1996, in Philadelphia, Deep Blue won the first game against Kasparov, but the champion took the match 4 to 2. Kasparov later wrote in Time of sensing "a new kind of intelligence" across the board. IBM improved the machine, and in the May 1997 rematch in New York, Deep Blue won 3½ to 2½. In the sixth game Kasparov resigned in fewer than twenty moves. After the second game Kasparov had accused IBM of human intervention, an accusation without evidence.
Deep Blue did not learn. It combined search through the game tree, with alpha-beta pruning, and an evaluation function tuned by hand by engineers and grandmasters, all running on chips designed for chess. It was the culmination of AI based on search and computing power.
The victory caused an enormous stir. Twenty years later, AlphaZero would reach a higher level by learning on its own, playing against itself.
Under the hood
A chess program explores the game tree with minimax: it assumes each side chooses the move that suits it best according to the evaluation function applied to the positions at the end of the search. Chess has on average about 35 legal moves per position, so looking d moves ahead means about 35^d nodes.
Alpha-beta pruning discards branches that can no longer change the decision. If moves are examined in a good order, the number of nodes falls to about 35^(d/2): for the same effort, the search goes almost twice as deep.
Deep Blue split the work between the software on its 30 processors, which explored the first moves, and its 480 chess chips, which searched the last levels and evaluated each position in hardware. It also extended the search along the most forcing lines.
Common misconceptions
What people often say: Deep Blue learned to play chess.
In fact: It did not learn. It searched the game tree with alpha-beta pruning and scored positions with a function tuned by hand by engineers and grandmasters. The leap to learning came twenty years later with AlphaZero, which learned by playing against itself.
Videos
Leads to
Watson wins Jeopardy! (February 14, 2011)
Documented direct influence
IBM framed Watson as the grand challenge that followed Deep Blue
Threads that pass through here
- Machines versus champions (step 4 of 7)
Sources
Original source
- Deep Blue. Artificial Intelligence, 134(1-2), pp. 57-83 (opens in a new tab). Murray Campbell, A. Joseph Hoane Jr., and Feng-hsiung Hsu, 2002
Scholarship and context
How to cite
“Deep Blue beats Kasparov.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/deep-blue-beats-kasparov
This entry was last reviewed on November 13, 2026.