Deep Blue was a chess-playing computer developed by IBM, notable for becoming the first computer system to defeat a reigning world chess champion in a match played under standard tournament time controls. Deep Blue's defeat of Garry Kasparov in May 1997 was a landmark event in the history of Artificial intelligence, widely covered internationally as a symbolic milestone in machines surpassing human intellectual capability at a task long regarded as a hallmark of strategic and analytical thought.
Development and the 1996 match
Deep Blue grew out of a chess-computer research project called ChipTest and its successor Deep Thought, developed by Feng-hsiung Hsu and colleagues at Carnegie Mellon University before the team moved to IBM to continue the project. IBM first pitted Deep Blue against Kasparov in a six-game match in February 1996, which Kasparov won 4 games to 2, including a loss to Deep Blue in the first game, marking the first time a computer had defeated a world champion under tournament conditions in a single game, though Kasparov won the match overall.
The 1997 rematch
IBM substantially upgraded the system, roughly doubling its search capability, and arranged a rematch held in New York City in May 1997. The upgraded Deep Blue won the six-game rematch 3.5 to 2.5, including a decisive final-game win, becoming the first computer to defeat a reigning world champion in a full match under standard chess tournament conditions. Kasparov and some commentators raised suspicions during and after the match that IBM engineers had made illegal human interventions between games, given a move in game two that Kasparov found uncharacteristically strong and difficult for a computer of that era to find; IBM denied the accusation and declined Kasparov's request for a further rematch, retiring Deep Blue from competitive play shortly afterward.
Architecture
Unlike later game-playing systems such as deep-learning-based AlphaGo and AlphaZero, Deep Blue relied on brute-force search combined with hand-engineered chess evaluation functions rather than learned pattern recognition, an expert-system-style approach that encoded human chess knowledge directly into rules rather than learning it from data. It used specialized parallel-processing hardware, including custom VLSI chess chips, to evaluate up to roughly 200 million chess positions per second, searching many moves deep using the minimax algorithm with alpha-beta pruning, and its evaluation function was tuned in part with input from grandmaster chess consultants, including Joel Benjamin, to encode positional chess knowledge.
Significance
Deep Blue's victory is frequently cited in histories of AI as demonstrating the power of exhaustive search and specialized hardware for well-defined, fully observable games, an approach fundamentally different from the pattern-recognition and self-play learning methods that later powered AlphaGo's 2016 defeat of Lee Sedol in the far larger search space of Go. Commentators often contrast the two milestones to illustrate a shift in AI research strategy: Deep Blue succeeded primarily through computational brute force applied to a domain small enough for exhaustive-style search, while AlphaGo and AlphaZero succeeded through learned intuition in a domain too vast for brute-force methods to work. IBM did not commercialize Deep Blue directly but the project's engineering talent and techniques informed IBM's subsequent AI initiatives, including IBM Watson.