← Curiosities

Chess & History · 2026-04-21

From Deep Blue to AlphaZero: How AI Reinvented Chess Strategy

In 1997 a machine out-calculated us. Twenty years later, one taught itself to play — and changed how chess is understood.

When Deep Blue beat Kasparov in 1997, it won by brute force, calculating further than any human could. Two decades later a very different machine arrived — and it didn't just beat us at chess. It taught us new things about the game itself.

A machine that taught itself

In 2017, DeepMind unveiled AlphaZero. Where Deep Blue relied on human-tuned evaluation and raw speed, AlphaZero was given only the rules and learned entirely by playing itself — millions of games, with no opening books and no human games to study. In about four hours of self-play it surpassed the strongest traditional engine, and it went on to beat Stockfish in a 100-game match with 28 wins, 72 draws and zero losses.

A different kind of mind

The most striking part was how it played. AlphaZero examined only about 80,000 positions per second — against Stockfish's 70 million — yet won by 'understanding' rather than counting, using a neural network to focus on the most promising ideas. Its style stunned grandmasters: bold, long-term piece sacrifices and pawn pushes that looked almost human, even romantic, reviving ideas masters had long dismissed.

Engines built on these ideas now shape opening theory and train every top player. The machine that learned chess by itself ended up teaching it back to us.

From rival to teacher

The arc from Deep Blue to AlphaZero is the story of AI in miniature: first matching human skill by force, then surpassing it by learning. For chess, the result isn't the death of the game many feared in 1997 — it is a richer one, with human creativity and machine insight feeding each other.

In short: In 2017 DeepMind's AlphaZero learned chess from scratch by self-play in about four hours and beat Stockfish with 28 wins, 72 draws and 0 losses — examining far fewer positions but 'understanding' more, reshaping modern strategy.

Frequently asked questions

What was AlphaZero?

A 2017 program by DeepMind that taught itself chess from only the rules, through self-play, with no opening books or human games. It quickly surpassed the strongest traditional engines.

How did AlphaZero do against Stockfish?

In a 100-game match it scored 28 wins, 72 draws and no losses, despite examining only about 80,000 positions per second versus Stockfish's 70 million, relying on a neural network.

How is AlphaZero different from Deep Blue?

Deep Blue (1997) won by brute-force calculation using human-designed evaluation. AlphaZero (2017) learned chess by playing itself and won through pattern-based 'understanding', influencing modern strategy.

See also

Play History's Gambit →More curiosities

A curiosity from History's Gambit, where chess meets history. You may cite or describe it with attribution to historysgambit.com.