IBM’s supercomputer beat chess champion Garry Kasparov in 1997. What terrifyingly 'genius' move broke his confidence?

The quiet bishop move that unsettled Kasparov

The move was 37.Ba6, played by IBM’s Deep Blue in the second game of its 1997 match against Garry Kasparov. The notation means that Deep Blue moved its bishop to the square a6. It was not a flashy checkmate or a sudden attack. Its force came from being a patient positional move: it improved the bishop’s placement and helped tighten the pressure on Kasparov’s position.

That was part of what made it so jarring. Chess players expect computers to excel at calculating concrete variations—working through chains of possible moves. But 37.Ba6 looked like the sort of long-range judgment a strong human player might make, even though its payoff was not immediately obvious.

Why it seemed almost too human

Deep Blue did not think about chess as a person does. It searched many possible positions, evaluated them using chess knowledge built into the program, and chose a move from that analysis. Its hardware let it examine enormous numbers of possibilities, while its evaluation rules helped it judge features such as king safety and piece activity.

Those methods could produce moves that looked intuitive without relying on human intuition. A move might be selected because its consequences held up across the computer’s search, not because the machine grasped a strategic idea in the human sense. But to Kasparov, a quiet move with a subtle plan was harder to interpret than a brute-force tactical sequence. It raised the unsettling possibility that Deep Blue was seeing something he had missed.

A move, not the whole match

Kasparov resigned the game after Deep Blue built a winning position. The bishop move became a focal point in his later suspicion that the computer’s play might have involved human intervention; IBM denied that claim. The match’s result, however, did not hinge on one mysterious stroke. Deep Blue won the six-game match 3½–2½, using speed, chess-specific evaluation, and carefully prepared opening knowledge.

The move’s lasting power is that it made computation look like judgment: a machine chose a restrained plan, and its opponent had to decide whether that plan was brilliant—or whether he simply could not see its point.