Deep Blue vs. Kasparov: 1997 Match, Hardware, Controversy

Custom chips, not learning

Deep Blue chose moves with brute-force tree search accelerated by 480 custom chess chips, not with anything resembling modern machine learning. Each chip carried a hardwired evaluation function recognizing about 8,000 positional features and could search 2 to 2.5 million positions per second alone. The 1997 machine ran as a 30-node IBM RS/6000 SP system holding those chips, with a guaranteed speed near one billion positions per second and a sustained rate around 200 million. [2]

The search split across two levels: workstation nodes ran an eight-ply software search on top of the chips' own hardware search, and opening and endgame databases handled the start and end of a game with little computation. Between games, engineers adjusted the weights the evaluation function gave positional features, using human judgment to patch weaknesses Kasparov had exposed. The system learned nothing during play itself. [2]

Nineteen moves and a resignation

The six-game rematch in New York finished 3.5 to 2.5 in Deep Blue's favor. Kasparov won game one, Deep Blue won game two, three draws followed, and on May 11, 1997, Kasparov resigned game six after nineteen moves, down a piece with no real compensation. It was the first time a computer had beaten a reigning world champion in a full match under normal tournament time controls, and the first resignation of Kasparov's competitive career. [1][2]

The hand of God, and the rule that let it happen

After losing game two, Kasparov accused the IBM team of taking covert human help, comparing Deep Blue's move to Diego Maradona's hand of God goal. The suspicion traced to one move, 36.axb5, which Kasparov found too composed for a machine that had otherwise played erratically. IBM's team, including Joel Benjamin and Murray Campbell, denied any intervention during play and said the only human input came between games, retuning evaluation weights within the agreed match rules. [3][4]

Atlas interpretation: That retuning was real and documented, which is what gave the accusation room to grow. A system that could not learn mid-game could still be reworked overnight, and Kasparov had no way to tell a better-tuned evaluation function apart from a hidden grandmaster's hand. Engines built decades later independently agree move 36 was simply correct, and Kasparov walked the allegation back in a 2016 interview, crediting Deep Blue's human designers rather than a rival behind the screen. [3][4]

What people meant by intelligence, before and after

Atlas interpretation: The match answered the question it was built to test: whether a machine could beat the best human player at chess under real conditions. The public reaction then changed the question, treating the win as proof chess had never really measured intelligence at all. Kasparov's later framing, that the victory belonged to Deep Blue's human designers rather than the machine, is a version of the same move: crediting the people behind the system rather than granting it anything resembling a mind. [4]