Deep Blue vs Kasparov, 1997

Two matches, one famous result

The first meeting came in February 1996 in Philadelphia. Deep Blue won the opening game, the first time a computer had beaten a reigning world champion in a game under normal tournament time controls, but Kasparov recovered and took the match 4 to 2. IBM's team went back to work, upgraded the hardware and tuned the program with help from grandmasters.

The rematch in New York in May 1997 was a media event. Kasparov won game one, Deep Blue struck back in game two, and three draws followed. In the sixth and final game Kasparov collapsed in just 19 moves. Deep Blue took the match 3.5 to 2.5.

How Deep Blue played

Deep Blue did not learn chess the way modern AI systems do. It was a supercomputer running a carefully hand-built program. Its heart was a set of custom chess chips that could generate and score moves very quickly, letting the system examine about 200 million positions per second in favorable conditions.

Strategy came from an evaluation function, a long list of weighted features such as material, king safety and pawn structure that human experts helped design. The machine also drew on an opening book and endgame databases. In short, it won by searching deeper and more precisely than any person could, guided by human chess knowledge.

Controversy at the board

Kasparov was shaken by a quiet, human-looking move in game two and questioned whether people had intervened during play. IBM denied it, and later analysis found no evidence of cheating. Kasparov asked for a rematch, but IBM declined and retired the machine. Years later he wrote about the match in a more reflective tone and became a well-known voice on how people and machines can work together.

Why it still matters

Chess had long been seen as a test of the highest kind of human thinking, so the result felt like a cultural turning point. Yet Deep Blue could do nothing but chess. That gap between narrow skill and general intelligence is still a central theme in AI.

The match also hints at what came next. Two decades later, systems such as DeepMind's AlphaZero learned chess largely by playing against themselves rather than relying on hand-written rules. Today one of Deep Blue's two cabinets is at the Computer History Museum in California and the other is at the Smithsonian in Washington.

← The Turing TestThe T in GPT: The 2017 Transformer Paper →