Deep Blue–Kasparov 1996: the first computer game win

In Philadelphia in 1996, IBM’s Deep Blue defeated Garry Kasparov in the first game of their match. It was the first time a computer beat a reigning world champion in a game under standard tournament conditions.

Complete game score

37 moves. Result: 1-0.

  1. 1.e4 c5 2.c3 d5 3.exd5 Qxd5 4.d4 Nf6 5.Nf3 Bg4 6.Be2 e6.
  2. 7.h3 Bh5 8.O-O Nc6 9.Be3 cxd4 10.cxd4 Bb4 11.a3 Ba5 12.Nc3 Qd6.
  3. 13.Nb5 Qe7 14.Ne5 Bxe2 15.Qxe2 O-O 16.Rac1 Rac8 17.Bg5 Bb6 18.Bxf6 gxf6.
  4. 19.Nc4 Rfd8 20.Nxb6 axb6 21.Rfd1 f5 22.Qe3 Qf6 23.d5 Rxd5 24.Rxd5 exd5.
  5. 25.b3 Kh8 26.Qxb6 Rg8 27.Qc5 d4 28.Nd6 f4 29.Nxb7 Ne5 30.Qd5 f3.
  6. 31.g3 Nd3 32.Rc7 Re8 33.Nd6 Re1+ 34.Kh2 Nxf2 35.Nxf7+ Kg7 36.Ng5+ Kh6.
  7. 37.Rxh7+

What happened in this game?

Deep Blue chooses the Alapin Sicilian and reaches an isolated-queen’s-pawn structure. The computer exchanges Black’s f6-knight and then occupies c4, using concrete calculation to maintain pressure against the queenside and center.

The tactical sequence 23.d5 Rxd5 24.Rxd5 exd5 leaves Black with active-looking pieces but exposed dark squares. White’s queen enters on b6 and c5 while the knight reaches d6 and b7.

Kasparov creates threats with ...f4, ...f3 and ...Nd3, but 32.Rc7 and the later knight checks keep control. The final 37.Rxh7+ is decisive. The result was historic, though Kasparov recovered to win the six-game match.

A first for a computer

On 10 February 1996 in Philadelphia, IBM’s Deep Blue beat Garry Kasparov in the first game of a six-game match. It was the first time a computer had beaten a reigning world champion in a game under normal tournament conditions, with a time control of 40 moves in two hours.

Kasparov recovered with three wins and two draws in the remaining games and won the match 4–2. In the 1997 rematch in New York, Deep Blue won 3½–2½, the first match defeat of a reigning champion by a computer.

Opening and middlegame

The IBM team chose Deep Blue’s opening moves. Against the Alapin Sicilian, Kasparov tried the new 10...Bb4, whose merit is still debated; Stockfish 18 rates it only slightly behind the usual 10...Be7. After 18.Bxf6 gxf6 his king was more exposed, and 19.Nc4 stopped him taking the d4-pawn because of Qg4+.

After 10...Bb4 Deep Blue left its opening book and calculated from there. Had it played 11.Nc3 Qa5 12.Qb3, the game would have transposed to one Kasparov had already played against Kramnik.

Kasparov said he might have played 23.d5 himself, a pawn sacrifice that damaged Black’s structure and opened the board. Stockfish rates it about level: not the engine’s first choice, but sound.

Where the game turned

The annotators criticised 25...Kh8, with Burgess suggesting 25...Ne7 and Keene 25...Rd8, but Stockfish rates all three about level. The engine’s turning points come later: 27...d4 cost more than half a pawn, 30...f3 more than a pawn, and 32...Re8, the move the notes call a huge blunder, about two pawns.

In between, Deep Blue took a pawn with 29.Nxb7, a materialistic choice typical of computers of the time, made because it saw no mating threat. Kasparov later said his attack would have intimidated many players into a mistake or two, but not this opponent.

The machine also defended precisely. 27.Qc5 stopped 27...Qg5, which would have forked g2 and the rook, and 30.Qd5 avoided 30.Qxd4??, which loses to 30...Nf3+.

The finish

After 33.Nd6 Re1+ 34.Kh2 Nxf2 35.Nxf7+ Kg7 36.Ng5+ Kh6 37.Rxh7+ Kasparov resigned. After 37...Kg6 38.Qg8+ Kf5 39.Nxf3, Black cannot meet the threats of 40.Nxe1, 40.Rf7 and 40.Qd5+ at once, and Stockfish rates the final position about eight pawns in White’s favour.

What else was playable?

Stockfish 18 at the moments the notes discuss, from White’s point of view:

  • 10...Be7 (instead of 10...Bb4): About level, the usual move; Kasparov’s new idea is only slightly worse.
  • 21.Qg4+ (instead of 21.Rfd1): About +0.4, the move the notes preferred; Stockfish rates the two almost the same.
  • 32...Qe5 (instead of 32...Re8): About +2.5 against +4.4 after the game move.

What to take away

  • Taking material is safe only when the counterattack has been calculated; Deep Blue’s 29.Nxb7 held because Black’s threats were not real.
  • An exposed king is a long-term weakness: the king opened up by 18...gxf6 was the target of the final attack nearly twenty moves later.
  • Annotators’ marks are worth checking. Stockfish finds 25...Kh8, criticised at the time, about as good as the suggested improvements.
  • A counterattack that does not break through leaves your own weaknesses behind; Kasparov’s play on the g-file never got there.

Want to turn one of your own games into a video? Try the ChessMotion Chess Video Maker.

Sources