Skip to content
AI Digested
History1985 to 19973 min read

Deep Blue beats Kasparov: the day a computer became chess champion

In May 1997, IBM's Deep Blue defeated world chess champion Garry Kasparov. The story of the match, how Deep Blue worked, the controversy, and what it did and didn't prove about AI.

First published · AI Digested

On 11 May 1997, in a television studio in New York, Garry Kasparov, the world chess champion and by many measures the greatest player who had ever lived, resigned after just 19 moves. He had lost the match to a computer. For many people, it was the moment they realised machines might outthink us.

Why chess?

Chess has been called the "drosophila of AI", the fruit fly that geneticists study, because it's a perfect test bed. The rules are simple and fixed. The game is fiendishly deep. And it's widely seen as a measure of intelligence.

Alan Turing wrote a chess program in 1948, before there was a computer powerful enough to run it, and simulated it by hand. In 1958, AI pioneers predicted a computer would be world champion within ten years. It took nearly forty.

Building Deep Blue

Deep Blue began as a student project. In the mid-1980s at Carnegie Mellon University, Feng-hsiung Hsu built a chess-playing chip called ChipTest, which evolved into a machine called Deep Thought. IBM hired the team in 1989, and the project became Deep Blue.

By 1997, Deep Blue was a powerful IBM supercomputer fitted with 480 custom chess chips. It could evaluate about 200 million positions per second, looking many moves ahead in the most important lines.

Its chess knowledge, how to judge whether a position is good, was largely written and tuned by its programmers working with grandmasters such as Joel Benjamin. It also had a huge library of opening moves.

The 1996 match: humans win

The first match was in Philadelphia in February 1996. Deep Blue won game one, the first time a computer had beaten a reigning world champion in a game under standard tournament conditions. But Kasparov adapted, and won the match 4 to 2.

The 1997 rematch: machine wins

IBM upgraded Deep Blue, roughly doubling its speed. The rematch began in New York on 3 May 1997, with enormous media coverage.

  • Game 1: Kasparov won.
  • Game 2: Deep Blue won, with a quiet, positional move that unsettled Kasparov. He felt it was too "human" for a computer and suspected interference.
  • Games 3, 4 and 5: draws.
  • Game 6: Kasparov, rattled, made an error in the opening and resigned after 19 moves.

Final score: Deep Blue 3.5, Kasparov 2.5.

The controversy

Kasparov accused IBM of unfair play and demanded a rematch and access to the machine's logs. IBM declined a rematch and retired Deep Blue. Kasparov's suspicions were never supported by evidence.

In the 2003 documentary Game Over: Kasparov and the Machine, and in his 2017 book Deep Thinking, Kasparov reflected on the match. One of the developers later suggested that a strange move in game 1, which unsettled Kasparov, had been caused by a bug.

What it proved, and what it didn't

Deep Blue showed that in a well-defined game, enough computing power plus expert knowledge could beat the best human.

But it wasn't "intelligent" in a general sense. It couldn't play noughts and crosses unless reprogrammed. It didn't learn from its games. It didn't understand chess the way Kasparov did; it calculated.

Many people concluded that games requiring intuition, rather than calculation, would be safe from computers for a long time. The ancient game of Go, with vastly more possible positions than chess, was the obvious example. That belief lasted until AlphaGo in 2016, which learned to play using neural networks rather than brute force alone.

What happened next

Chess engines kept improving and became tools that every serious player uses to train. In 2017, DeepMind's AlphaZero taught itself chess in hours by playing against itself and beat a top engine of the day.

Kasparov became a champion of "centaur" chess, where humans and computers play as a team, and of the idea that AI should augment human intelligence rather than replace it.

Read on

Common questions

Did Deep Blue use machine learning?

Not in the modern sense. Its evaluation of positions was largely designed and tuned by programmers and chess grandmasters. Its strength came from custom chips that searched an enormous number of possible moves.

Did IBM cheat?

Kasparov suspected human help during the games, particularly after a surprisingly subtle move in game 2. No evidence of cheating was ever found. IBM released the logs later, and one of the developers has suggested a bug may have caused a puzzling move in game 1.

Can anyone beat a chess computer today?

No. Free chess programs running on a phone are now far stronger than any human player.

Sources and further reading

  1. IBM: Deep Blue
  2. Campbell, Hoane & Hsu (2002). Deep Blue. Artificial Intelligence, 134

Spotted a mistake or something out of date? Tell us and we will fix it. Read how we research and write our articles.