

Deep Blue vs. Kasparov: The Day a Machine Beat a World Champion
In May 1997, a machine without intuition beat the best chess player alive, and proved that brute-force search can look exactly like genius from the outside.
On May 11, 1997, in a conference room on the 35th floor of a Manhattan office tower, IBM's Deep Blue defeated reigning world chess champion Garry Kasparov in the sixth and final game of their rematch, winning the match 3.5 to 2.5. Kasparov had beaten an earlier version of Deep Blue in 1996. For the rematch, IBM's engineers had roughly doubled the machine's evaluation speed, giving it the ability to assess up to 200 million chess positions per second across 480 custom-built chess chips.
Deep Blue was not learning anything. It had no neural network and no self-play. It was alpha-beta search with a handcrafted evaluation function, tuned with help from grandmaster Joel Benjamin, backed by a large opening book and endgame tablebases, and a mountain of purpose-built silicon that let it search far deeper than any general-purpose computer of the era. The intelligence, such as it was, lived in the depth of the search, not in anything resembling understanding.
The match ended a narrative stretching back to Claude Shannon's 1950 paper on programming a computer to play chess and Alan Turing's own hand-simulated chess algorithm: that mastering chess was a meaningful proxy for general intelligence. Within a few years, chess engines running on ordinary consumer hardware could beat any grandmaster alive, and chess stopped being treated as an AI benchmark at all. It became a tool.
Kasparov, convinced that an unusually quiet, almost "too human" move in Game 2 meant a grandmaster had intervened in real time, accused IBM of cheating. IBM refused a rematch, dismantled Deep Blue shortly afterward, and declined to release the full game logs for independent analysis. It later emerged that the suspicious move was most likely the product of a software bug that caused the machine to fall back to a random legal move, a glitch that looked, to a world champion under pressure, exactly like insight.
That pattern keeps recurring. Every time a narrow system clears a benchmark long assumed to require general intelligence, observers reach for understanding as the explanation when the real answer is optimised search over a well-defined space. AlphaGo's Move 37 two decades later, and the debates over whether today's reasoning models truly "think," restart the same argument Deep Blue first forced the world to have.
What it actually meant
Deep Blue proved that a system can completely master a narrow, well-defined domain through depth of search alone, with none of the general reasoning spectators assumed the feat required. Chess mastery and general intelligence turned out to be far more separable than decades of AI research had assumed.
What enterprises get wrong today
Enterprises see a model clear a benchmark, a certification exam, a coding test, a scripted customer conversation, and conclude it has acquired the general competence the benchmark was built to signal. The benchmark measured one narrow capability under narrow conditions; deployment exposes everything the test never covered.
Related capabilities
More from Moments
Ask about this story
In 1997, IBM's Deep Blue beat world chess champion Garry Kasparov using raw search, not learning. Want to know how it actually worked, or why Kasparov accused IBM of cheating?