Well-VersedLibraryCoverageAbout
2026-10-31 · Games & Pastimes
Games · Chess

Deep Blue

The first machine to beat a world chess champion in a match won the deciding game in nineteen moves, and the champion left convinced a person had helped it.

A tall black two-door computer cabinet labelled IBM and RS/6000 SP, standing on a dark plinth in a museum gallery, with a small printed placard leaning against its base and an illuminated panel beside it showing an enlarged microchip.
One of the two cabinets that made up Deep Blue, on display at the Computer History Museum in Mountain View, California. The panel beside it shows one of Feng-hsiung Hsu’s chess chips. Source: Wikimedia Commons, where the author and licence terms for this file are recorded.

The last game lasted a little over an hour. On the afternoon of 11 May 1997, at the Equitable Center in New York, Garry Kasparov resigned after nineteen moves, stood up from the board and walked out. He was thirty-four, had been world champion since 1985, and was widely held to be the strongest player who had ever lived. Across the table from him sat a member of the IBM team, relaying moves from a supercomputer he could not see. The defeat settled a six-game match 3½ to 2½ and made Deep Blue the first computer to beat a reigning world champion in a match played at tournament time controls.

A colour photograph of a dark-haired man in a pinstriped suit, chin resting on his clasped hands, frowning over a set of wooden chess pieces in the foreground.
Garry Kasparov at the board in New York in 2003, six years after the match. Source: Wikimedia Commons, where the author and licence terms for this file are recorded.

The machine began as a graduate student’s chip. At Carnegie Mellon in the mid-1980s Feng-hsiung Hsu designed a piece of silicon that did nothing but generate and test chess moves, and built it into a machine called ChipTest. Its successor, Deep Thought, became in 1988 the first chess machine to beat a grandmaster in tournament play. IBM hired Hsu and his colleague Murray Campbell in 1989, added the programmer Joe Hoane and others under the manager C. J. Tan, and gave them one assignment. The version that met Kasparov in 1997 joined a thirty-node IBM RS/6000 SP supercomputer to 480 of Hsu’s chips, each searching two to two and a half million positions a second. In quiet positions the whole system ran close to 200 million.

Speed was most of the method. Deep Blue judged a position by adding up thousands of weighted features, which the team tuned with the help of grandmasters led by Joel Benjamin, and it looked further down the tree of possible replies than any person could. It held no idea of a plan. It simply calculated, and the bet was that enough calculation would pass for judgement.

The first meeting went to the man. In Philadelphia on 10 February 1996 Deep Blue won the opening game, the first time a reigning champion had lost to a computer at classical time controls; Kasparov then won three and drew two to take the match 4 to 2. IBM more than doubled the hardware for the rematch and raised the purse to $700,000 for the winner and $400,000 for the loser, with the machine’s share going back to IBM.

Kasparov won the first game of the rematch on 3 May. The second, a day later, is the one that broke him. On move 36 Deep Blue passed up a chance to win two pawns with its queen and instead made a quiet pawn exchange followed by a bishop move that tightened its grip on the board, the kind of long positional squeeze computers were supposed to be blind to. Kasparov resigned after White’s 45th move. The next morning his friends told him the position had been a draw: a queen move to e3 led to a perpetual check that Deep Blue could not escape. He accused IBM of letting a grandmaster intervene, and he played the rest of the match like a man looking over his shoulder. Games three, four and five were drawn.

With the score at 2½ each, Kasparov took Black in the final game and chose the Caro-Kann, a solid defence he seldom played, apparently to steer the machine out of its preparation. On move 7 he pushed his rook’s pawn to h6. In that line the move is a known error, because it lets White give up a knight on e6 for two pawns and a black king stranded in the centre. Kasparov later said he expected the computer, which counted material, to retreat. “To my horror,” he recalled, “Deep Blue immediately took on e6.” The sacrifice was sitting in the opening book its grandmaster advisers had loaded.

r . b q k b . r
p p . n . p p .
. . p . N n . p
. . . . . . . .
. . . P . . . .
. . . B . N . .
P P P . . P P P
R . B Q K . . R
Deep Blue–Kasparov, game six, after 8.Nxe6. The white knight has been given up on e6 for a pawn, and if Black takes it the e-file opens straight at the king, which has not castled. Diagram drawn for this lesson from the game score.

Kasparov took the knight, his king fled to d8 and then c8, and White’s rook, queen and bishops poured through the gaps. On move 19 the machine pushed a pawn to c4 to open one more line, and he resigned rather than watch the rest.

The scoreGame six · New York, 11 May 1997

1.e4 c6 2.d4 d5 3.Nc3 dxe4 4.Nxe4 Nd7 5.Ng5 Ngf6 6.Bd3 e6 7.N1f3 h6 8.Nxe6 Qe7 9.0-0 fxe6 10.Bg6+ Kd8 11.Bf4 b5 12.a4 Bb7 13.Re1 Nd5 14.Bg3 Kc8 15.axb5 cxb5 16.Qd3 Bc6 17.Bf5 exf5 18.Rxe7 Bxe7 19.c4, and Black resigned.

Each numbered pair is a move by White, then Black. A capital letter names the piece (N for knight, B bishop, R rook, Q queen, K king) and no letter means a pawn; the square is where it lands, “x” is a capture, “+” is check and 0-0 is castling. Step through the game with notes →

Deep Blue v. G. Kasparov, IBM rematch, game 6. Caro-Kann Defence, Steinitz Variation.

Kasparov wanted a rematch. IBM declined, closed the project and sent the hardware to museums, which left his suspicion of human help with nowhere to go. The team’s own technical account, published by Campbell, Hoane and Hsu in 2002, describes a machine that won by searching an average of 126 million positions a second through the match. Twenty years on, in his book Deep Thinking, Kasparov dropped the charge and wrote of the engineers with new respect.

· · ·

What Deep Blue proved is narrower than the headlines of 1997 suggested. Its strength came from people: Hsu’s chips, the grandmasters who tuned its sense of a good position, the advisers who wrote its openings. Nobody could hand-write that kind of judgement for Go, which is why the Asian board game held out for nineteen more years, until a program taught itself what a strong position looks like. The chess machine needed to be told, and in 1997 it had been told well enough to beat the best player alive.

Up next
Pizza Margherita
The pizza named for a queen in June 1889 is older than her visit, and the letter that ties them together looks made later.
Read →

Images on this page are reused from Wikimedia Commons. Each is linked to its file page, where the author and the licence terms for that file are recorded. Thumbnails in “Up next” are credited on the lesson they link to.