Deep Blue's Move 37 — The Move That Broke Kasparov, 1997

The Rematch Nobody Could Afford to Lose

Kasparov had beaten Deep Blue 4-2 in Philadelphia in 1996, and he arrived at the Equitable Center in Manhattan expecting to do it again. He was the strongest player alive, the dominant figure of the World Chess Championship era that began with his victory over Karpov in 1985, and he understood computers better than almost any of his peers among famous chess players. His plan for 1997 was explicitly anti-computer: avoid sharp tactics, keep the position closed, and exploit the machine's notorious greed for material. Computers of that generation snatched pawns because their evaluation functions counted wood. Kasparov intended to feed Deep Blue poisoned material and strangle it positionally.

It worked in Game 1, which he won convincingly. Game 2, with Deep Blue playing White, would expose the flaw in the plan. IBM's team, which included grandmaster Joel Benjamin as a full-time consultant, had spent the intervening year teaching the machine exactly the judgement Kasparov assumed it could never have.

The Position: A Slow Spanish Squeeze

Game 2 was a closed Ruy Lopez, one of the oldest battlegrounds in chess history. Deep Blue handled the opening with surprising patience, advancing on the queenside, fixing weaknesses, and steadily improving its pieces while Kasparov sat cramped behind his pawn chain. This was precisely the kind of slow manoeuvring game the champion had wanted, except that the machine was playing it better than he was.

By move 36 the critical moment had arrived. Deep Blue's queen had infiltrated to b6's doorstep, and the obvious continuation was the materialist one: send the queen in to win a pawn. Kasparov was counting on it. The pawn grab would have cost White time, and after 36...Qe7 Black gets ...Rb8 and the freeing break ...d5, when his passive pieces suddenly flood into the game. Every commercial engine in the press room wanted to take the pawn. Kasparov's entire match strategy was, in that moment, a bet on what Deep Blue would do next.

37.Be4: The Move No Computer Was Supposed to Find

Deep Blue played 36.axb5 axb5 and then 37.Be4. No capture, no check, no threat in the crude sense. The bishop simply settled on the central light squares, clamping down on the d5 break, eyeing the kingside, and leaving Black's counterplay stillborn. It is the kind of move textbooks on chess strategy call prophylaxis: a move whose purpose is to prevent the opponent's ideas rather than to execute your own. Strong human positional players make such moves by feel. Machines, in 1997, were not supposed to make them at all.

Kasparov was visibly stunned. He later wrote that the move was "brilliantly subtle" and that it shook his understanding of what he was facing. The refusal to grab material meant the machine was weighing piece activity, pawn structure and long-term pressure against a concrete pawn, and concluding the pawn was not worth it. With Black's counterplay dead, Deep Blue squeezed methodically, opened lines at the moment of its choosing, and ground Kasparov down without ever needing the tactical fireworks computers were famous for.

A Resignation in a Drawn Position

The collapse that followed was as much psychological as it was technical. Facing 45.Ra6, Kasparov resigned. Within days, analysts demonstrated that 45...Qe3 would have saved the game: after 46.Qxd6 the black queen gives perpetual check, and White has no way to escape the checks without ruinous concessions. The greatest player of his generation, a man celebrated for his defensive resourcefulness as well as his attacks, had resigned a drawn position. Nothing like it had happened to him in a serious game.

Kasparov's reaction revealed how deep the wound went. He insinuated that a human hand might have guided the machine, invoking Maradona's "hand of God", and demanded Deep Blue's logs. IBM refused to hand them over during the match, releasing them only later, and no evidence of intervention was ever found. The damage, though, was done in both directions: Kasparov never recovered his equilibrium, drawing the next three games from positions where he pressed nervously, and then losing Game 6 in 19 moves to a known Caro-Kann piece sacrifice. Deep Blue took the match 3.5-2.5.

What the Move Actually Proved

The popular story says a computer learned to play like a human. The truth is subtler. Deep Blue examined up to 200 million positions per second, and its choice of 37.Be4 emerged from brute search guided by an evaluation function painstakingly tuned with grandmaster input. The machine did not feel the position; it calculated deeply enough that positional truth became visible as numbers. The lesson, which the chess world absorbed slowly over the following decade, was that there is no mystical wall between calculation and judgement. Go deep enough and the quiet move is the best move, whether the player choosing it is made of neurons or silicon.

For Kasparov the move was a trap of his own making. His anti-computer strategy depended on the machine's greed, and when the greed was not there, neither was the strategy. He played the rest of the match against an opponent he no longer understood, and the games show it.

Legacy: From One Move 37 to Another

Deep Blue's quiet bishop move now stands alongside the great combinations of the famous chess games of the romantic era as a turning point in the game's story, though it could hardly be more different in character. It announced the era in which machines would first match and then exceed human judgement, reshaping opening theory, training and even endgame knowledge through tablebases and engines.

The echo arrived nineteen years later. In 2016, AlphaGo played its own legendary move 37 against Lee Sedol in Game 2 of their Go match, a shoulder hit so unconventional that professionals initially assumed it was a mistake. The numerical coincidence is accidental, but the symmetry is not: in both cases a machine made a quiet, creative-seeming choice that a world champion believed impossible, and in both cases the champion was never quite the same afterwards. The line from the Equitable Center in 1997 to modern artificial intelligence runs directly through Deep Blue's 37th move.

Frequently Asked Questions

Q: What exactly was Deep Blue's move 37 in Game 2?

After 36.axb5 axb5, Deep Blue played 37.Be4, a quiet bishop retreat to the centre of the board. The move looks unremarkable on the surface, but its meaning is profound: instead of sending its queen to grab the loose pawn on b6, which almost any chess engine of the era would have done instantly, Deep Blue chose to restrain Kasparov's only source of counterplay, the d5 pawn break, and keep the black pieces passive. The pawn grab would have allowed 36...Qe7 followed by ...Rb8 and ...d5, when Black's pieces suddenly come alive. By declining material in favour of long-term positional pressure, Deep Blue played the kind of move chess teachers describe as prophylaxis, and it slowly suffocated the world champion.

Q: Why did Kasparov resign Game 2 when the final position was drawn?

Kasparov resigned after 45.Ra6 because he believed his position was hopeless, but analysis published within days showed that 45...Qe3 would have forced a draw. After 46.Qxd6 the black queen delivers perpetual check, since White's king cannot escape the checks on e3, e2 and related squares without allowing material loss. Kasparov, by his own admission, never seriously considered the resource because he had stopped believing he could save the game. He later said that he was so shaken by the machine's human-like play earlier in the game that he assumed it had calculated everything. The resignation in a drawn position is widely regarded as the clearest evidence of how completely the game had broken his composure.

Q: Did a human help Deep Blue choose the move?

No credible evidence of human intervention has ever emerged. Kasparov publicly suggested that a grandmaster might have guided Deep Blue's choices during Game 2, alluding to Maradona's infamous "hand of God" remark, and he demanded the machine's logs. IBM declined to release them during the match, which deepened his suspicion, but the logs were made available afterwards and showed the machine reaching its decisions through search. Members of the team, including grandmaster consultant Joel Benjamin, had tuned Deep Blue's evaluation function between 1996 and 1997 precisely so it could weigh positional factors like piece activity and pawn structure against material. The move was not human; it was evidence that human positional judgement had been successfully encoded.

Q: Who won the 1997 Deep Blue versus Kasparov match?

Deep Blue won the six-game rematch in New York by a score of 3.5 to 2.5. Kasparov won Game 1, lost Game 2, and then drew Games 3, 4 and 5, with the match level going into the final game. In Game 6, played on 11 May 1997, a visibly exhausted Kasparov allowed a well-known piece sacrifice in the Caro-Kann Defence, 8.Nxe6, and resigned after only 19 moves. It was the first time a computer had defeated a reigning world champion in a classical match under standard tournament conditions. Kasparov requested another rematch, but IBM retired the machine, and Deep Blue never played a competitive game of chess again.