Deep Blue vs. Kasparov: How Computers Finally Beat Humans
In 2025, IBM announced the 30th birthday of Deep Blue, the chess supercomputer that proved a worthy opponent to a world champion. Deep Blue famously played two matches against the reigning chess world champion, GM Garry Kasparov, in 1996 and 1997. How did silicon processors beat the world’s best chess player? To borrow a phrase from the computer-powered internet—it’s complicated.
Credit: Amin Zabardast / Unsplash
Development of Deep Blue by IBM
During the late 20th century, companies like IBM sought to find the limits of what computers could do. Chess was an ideal proving ground to test the capabilities of more powerful supercomputers. Could a computer “think” like a seasoned chess player and beat a grandmaster? Early chess programs relied on reviewing thousands of possible moves and computing a best move. It wasn’t elegant play, but it could be overwhelming for human players to counter.
In 1989, IBM hired Feng-hsiung Hsu and Murray Campbell, the computer scientists working on Deep Thought, the first computer to beat a grandmaster. Their mission? Improve Deep Thought (renamed Deep Blue in a nod to IBM) enough to beat the current chess world champion. The chess community was skeptical, and none more than GM Garry Kasparov, the world champ from 1985 to 2000.
One of the two towers of Deep Blue is on display at the Computer History Museum in California. Photo by Jim Gardner used under Wikimedia Commons license
The new supercomputer used 32 processors in coordinated parallel computing, an innovation that allowed it to evaluate 200 million chess positions per second. The processing speed of 11.4 billion floating point operations per second was blindingly fast for the time. (Source: IBM, IBM Heritage: Deep Blue) During matches, the computer took about three minutes to evaluate and choose a move, but those three minutes were very busy.
In order to become more than just a number-crunching behemoth, Deep Blue’s team included GM Joel Benjamin, embedding in the coding the thoughts of a great chess player. This was at least partially successful, but it was a mammoth undertaking to hard code this kind of behavior. Once they were satisfied with Deep Blue’s capabilities, it was time to test them against the best in the world.
1996: Ready Match One
Kasparov had beaten Deep Blue’s predecessor, Deep Thought, in two games in 1989. To make the match more interesting, IBM offered $400,000 to the winner and $100,000 to the loser, the equivalent of a ~$1M purse today. While the moves Deep Blue made were completely determined by the computer, its human team was on hand to make decisions regarding draws and resignations.
1. Deep Blue vs. Kasparov: The first game of the match was a win for Deep Blue. This was the first win for a chess-playing computer against a reigning world champion under standard classical time controls.
2. Kasparov vs. Deep Blue: This was a long game at 73 moves, with Deep Blue resigning. Kasparov played a blocking game, preventing Deep Blue from developing its positions. As an interesting note, the computer had a configuration error that prevented it from accessing its opening book, but it was able to recreate opening theory with the information it could access.
3. Deep Blue vs. Kasparov: This game was similar to Game One, with Kasparov playing the Sicilian Defence and Deep Blue pulling out the Alapin Variation. The game ended at 39 moves with a draw.
4. Kasparov vs. Deep Blue: The fourth game was ultimately another draw, despite the Deep Blue team declining a draw offered by Kasparov. This game lasted 50 moves, a variation of the Queen’s Gambit Declined.
5. Deep Blue vs. Kasparov: The score was tied moving into Game Five, but this was where the match turned. Kasparov successfully played black to a resignation by the Deep Blue team after the 47th move. Kasparov offered a draw at move 23, which the Deep Blue team declined. This error in judgment is a good example of the human side of the computer during this match.
6. Kasparov vs. Deep Blue: Playing white this time, Kasparov played to thwart the computer at every turn. This was a game of small, strategic moves that slowly advanced Kasparov’s position while preventing Deep Blue from doing much. Deep Blue resigned after 43 moves.
Kasparov did what great chess players do, countering the playing style of the machine and winning the match 4-2. Deep Blue could run through hundreds of millions of variations, but it could not play with the same ingenuity or aggression. The programming issue in Game Two highlighted how susceptible the computer could be to errors. A human player would be unlikely to forget everything he knew about openings!
1997: Deep Blue II: Electronic Boogaloo
IBMs engineers and scientists sought to make the rematch with Kasparov in 1997 more competitive by helping the machine “think” more like a grandmaster. Beyond making the endgame engine more robust, they brought in other grandmasters to advise the team. One of the key innovations was programming to help disguise the computer’s strategy.
1. Kasparov vs. Deep Blue: Kasparov won Game One in 45 moves using the Réti Opening and King’s Indian Attack. This game was marked by a sub-par move by Deep Blue at move 44, which, according to the team, was a fail-safe move triggered by a bug preventing the selection of a desirable move.
2. Deep Blue vs. Kasparov: Probably the most controversial of all the games played by Kasparov and Deep Blue, Game Two went to Deep Blue after Kasparov resigned. The controversy came toward the end of the game when the computer passed up material gains in an uncomputer-like move. Kasparov alleged that another grandmaster must have been involved. IBM maintained that this was the result of more sophisticated programming. This game was also notable as Kasparov missed a potential winning strategy after this. Safe to say, the world champ was not pleased with anything about this game.
3. Kasparov vs. Deep Blue: In Game Three, Kasparov tried to throw off the computer by playing an unusual opening, the Mieses Opening. Despite this tactic, the game ended in a draw after 48 moves.
4. Deep Blue vs. Kasparov: Game Four was another draw, this time likely due to Kasparov getting into time trouble and playing a series of rushed moves. The game was a Caro-Kann Defence that moved into the Pirc Defence.
5. Kasparov vs. Deep Blue: In this game, Deep Blue’s additional work on endgames paid off, with a turnabout that changed an expected win for Kasparov into a draw.
6. Deep Blue vs. Kasparov: Going into Game Six, the score was tied 2½-2½. Again, Kasparov assumed the machine would be playing conservatively to maximize material gains. Deep Blue made a surprising knight sacrifice that wrecked Kasparov’s strategy. The game was concluded in only 19 moves with a Deep Blue victory and match win at 3½ -2½.
Depending on your allegiances, the match was hailed as a triumph of machine over man or as an example of the world champion playing uncharacteristically poorly. Kasparov was dismissive of the machine’s victory at times, suggesting any creative moves were the result of human intervention. He demanded a rematch, which both IBM and Feng-hsiung Hsu declined. While unquestionably a triumph in the advancement of computers, the results received mixed applause in the chess community.
Was a Computer as Good at Chess as a World Champion?
Despite its victory in the rematch in 1997, it’s still questionable whether Deep Blue was as good as the world champion. Kasparov was not provided with previous games of the computer against other GMs, despite requesting them. Study of the opponent’s historical games is a normal part of preparation. The Deep Blue team certainly did have Kasparov’s games available when designing their algorithms. Additionally, the computer received updated programming between games. A human GM wouldn’t have the time to add completely new lines of attack or a library of positions to their play.
That said, the estimated ELO ratings of Deep Blue are comparable to Kasparov’s at the time, around 2800. Deep Blue is generally rated a little lower than Kasparov, but since we don’t have an objective measurement, this could be bias against the machine or fair due to the low number of games it played. IBM was more concerned with the marketing and technical opportunities afforded by the Deep Blue victory than its chess career.
The Legacy of Deep Blue
The ability of computer chips to run millions of scenarios in seconds is the strongest attribute of chess engines like Deep Blue. Early chess computers didn’t have much else, however. Modern engines like Stockfish will run about 40 million positions per second, but they have the advantage of neural network learning to evaluate positions without “brute forcing” moves. Stockfish has an ELO of 3600, substantially higher than Deep Blue (~2800) or GM Magnus Carlsen (peak 2882).
Deep Blue was a substantial leap forward for computing as a whole. The advancement of massively parallel processing machines enabled breakthroughs in financial modeling, healthcare, and the sometimes-controversial fields of artificial intelligence and data mining. As an experiment in expanding our capability to analyze very large datasets and build predictive models, Deep Blue was a smashing success.
A Historic Match for Chess Foreshadows Our Current Cultural Shift
If you’d like to learn more about the details of the match, with insight directly from Kasporov, IBM, and others, the documentary “GAME OVER: Kasparov and the Machine” has interviews and footage from the games. The Deep Blue vs. Kasparov series of games is fascinating both for the play of the games and the glimpse into the future of computing, predictive of the thorny questions we wrestle with today around artificial intelligence. A machine can win chess games, but can it be “better” than us?