Deep Blue 1997: the day the machine beat the world champion

New York, 11 May 1997. The nineteenth move. Garry Kasparov looked at the board, tipped his king over and resigned. Less than an hour of play. The best chess player on the planet had just lost to a machine — and the evening news tried to explain what had happened without ever really understanding it.

Today it's 29 years since that moment. It's worth pausing to understand what was actually at stake in that Manhattan hotel.

Kasparov and the first battle — 1996

Garry Kasparov became world chess champion in 1985, at 22, and dominated the sport almost absolutely for more than a decade. When IBM proposed a match against Deep Blue in 1996, in Philadelphia, Kasparov accepted with confidence — and he had reason to.

He won. 4–2. Kasparov showed he could exploit the computer's weaknesses: he forced long-term strategic positions where raw calculation of variations wasn't enough. The 1996 Deep Blue was powerful, but predictable. An opponent that played well within a script.

IBM went back to work. The 1997 rematch was a different computer.

What the 1997 Deep Blue was

The IBM team — led by Feng-hsiung Hsu and Murray Campbell, with support from Grandmaster Joel Benjamin to fine-tune the position evaluations — spent the year rebuilding the system end to end. The 1997 version could evaluate around 200 million positions per second, running on dedicated chess chips developed specifically for the project.

It's important to say what Deep Blue wasn't: there was no machine learning, no neural network, nothing resembling what we call artificial intelligence today. It was extraordinarily well-directed brute force — a tree search over variations with an evaluation function meticulously calibrated by human specialists. Impressive, certainly. But different from intelligence.

Kasparov probably knew that. What he didn't expect was what happened in game two.

Game 2 and the move that shook Kasparov

The 1997 match ran to six games. Kasparov won the first with relative ease. In the second game, at a critical point in the middlegame, Deep Blue played a long-term positional move — one that brought no visible immediate gain, but created an advantage that would only materialise many moves later. It was the kind of calculation expected from a human Grandmaster, not from a computer of that era.

Kasparov lost game 2. And, visibly shaken, he asked for access to the computer's internal logs — accusing IBM of human intervention in the process. IBM denied it. The controversy was never fully resolved. What is certain is that the move cost more than a single game: it cost Kasparov his focus.

The scoreline continued: two draws, a Kasparov win in game 5, another draw. Everything came down to game 6.

Game 6 — defeat in 19 moves

11 May 1997. Kasparov had the black pieces. He needed a win to draw the match. What happened was the opposite: in 19 moves he made a serious opening error, fell into a hopeless position and resigned. The most important confrontation in the history of chess lasted less than an hour.

Final score: 3.5 to 2.5 for Deep Blue.

It was the first time a computer had beaten a reigning world chess champion under regulated tournament conditions. Kasparov demanded a rematch with full access to the game logs. IBM refused both demands. Months later, Deep Blue was decommissioned and dismantled.

What Deep Blue actually proved

The popular narrative of 1997 was: "the machine surpassed the human in intelligence". That reading is mistaken — and AI researchers of the day were the first to point it out.

Deep Blue didn't understand chess. It had no sense that a move was elegant, it didn't know what winning meant, it had no model of the world beyond the rules of the game. It was a specialised system operating in a tightly bounded search space. That's impressive — and it was an engineering milestone — but it's fundamentally different from intelligence.

What Deep Blue's victory really demonstrated was that problems with a well-defined search space — even enormous ones, like chess — can be conquered with fast enough hardware and well-tuned heuristics. A lesson that shaped decades of research in computing.

29 years on — the trail it left

Deep Blue opened a path. In 2011, IBM's Watson beat two Jeopardy! champions. In 2016, DeepMind's AlphaGo defeated Lee Sedol at Go — a game with a possibility space far larger than chess, solved by a completely different approach: reinforcement learning, not brute force. In 2017, AlphaZero learned chess from scratch, with no database of human games, in under 24 hours — and surpassed every system that had existed before.

Kasparov, curiously, became a serious thinker on AI. In his book Deep Thinking (2017), he revisits the 1997 defeat with more perspective than bitterness. His most interesting conclusion is the centaur model: a human plus an AI working together is stronger than either alone. But that window of human advantage is closing as the systems grow more capable.

On that 11 May 1997, nobody knew exactly what had happened. Twenty-nine years later, there's still debate about game 2. What isn't up for debate is the historical weight of that afternoon in New York — the moment the line between tool and opponent became, for the first time, genuinely blurred.

If you enjoy these milestones in computing history, it's worth stopping by the post about Claude Shannon — the man who invented the bit, whose information theory is the mathematical foundation on which systems like Deep Blue were built. And if you remember where you were when you heard Deep Blue had won — or if you simply weren't around yet — the comments are open.