Did you know that a system learning through its own games beat a leading chess engine? Results published in 2018 compare AlphaZero with Stockfish. This is not a human facing a machine: the game pits two computer methods against each other. AlphaZero combines learning and search to choose its moves, training through self-play. The film is an original reconstruction of the opening moves of an AlphaZero–Stockfish 8 game from 18 January 2018, using the PGN published by DeepMind. It is not a recording of the match.
Out of a thousand games in the test cited, AlphaZero has one hundred and fifty-five victories, six defeats and eight hundred and thirty-nine draws. The vast majority is zero. To say that he won all the games would erase the most important part of these figures actually published.
The result is the versions and conditions of this comparison. It doesn’t prove that AlphaZero beats every modern engine today. The story remains fascinating: learning to play can produce a strong strategy, but a result does not automatically translate to other test conditions.
What we know
The test cited gives 155 wins, 6 defeats and 839 draws over 1,000 games. It compares historical versions and does not rank all modern engines.
To go to the source
References consulted on October 12, 2026.
Signaler une erreur ou apporter une précision
