The Silicon Immortal

In December 2017 Google DeepMind let its chess program AlphaZero loose on Stockfish, the strongest conventional engine of the day, in a hundred-game match played at one minute per move. AlphaZero had been given nothing but the rules of chess and several hours of self-play against itself; it emerged assessing positions the way strong humans do, favouring long-term binds and piece activity over raw material count, rather than searching brute-force the way Stockfish did.

This game, one of ten released alongside DeepMind's original paper and nicknamed the Silicon Immortal by chessgames.com, shows that style at its purest. AlphaZero, with White, built a small space advantage out of a Queen's Indian fianchetto, then spent the middlegame manoeuvring its rook and bishop around Black's king (21.Rd6, 23.Bh6) until Stockfish's pieces had no coordinated targets left. Material only started to fall on move 57, well after the position had already been squeezed past the point of defence.

The wider match finished 28 wins to Stockfish's zero, with 72 draws, and reshaped how top engines are built; both Stockfish and the open source Leela Chess Zero absorbed lessons from AlphaZero's neural network approach within a few years. The exact calendar day this particular game was played has not been confirmed in any public source; DeepMind's paper appeared on 5 December 2017 with the ten sample games attached.