IBM (Deep Blue Era)

Contribution to AI

Special-purpose silicon was the wager. Feng-hsiung Hsu's ChipTest and Deep Thought had already shown at Carnegie Mellon that putting move generation and position evaluation into custom chips bought search depth no general-purpose machine could reach, and IBM hired that group in 1989 to scale the idea until it reached the world champion. The machine that played in 1997 was a thirty-node RS/6000 SP driving 480 chess chips, each scoring positions against roughly eight thousand hand-specified patterns, with the shallow plies of the search run in software and the deep ones in hardware — an unusually complete case of an algorithm and its circuits designed together for a single problem, guided by grandmaster advice on what the evaluation should reward. The 1996 match in Philadelphia produced the first game a reigning champion had lost to a machine at tournament time controls, though Kasparov took the match four-two; the rematch a year later went to Deep Blue by three and a half to two and a half, after which IBM retired it rather than play again. What shifted was the standing of the question. Chess had served for four decades as the agreed proxy for machine intelligence, and once it fell the consensus was that it had never been one, which redirected attention to games where enumeration is hopeless. The staged public match against a named human, with corporate reputation on the outcome, has been the template for announcing machine capability ever since.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).