AlphaZero

One algorithm with no human games learned chess, shogi and go from scratch and beat the strongest program in each.

Why it matters

Human knowledge turned out to be not merely unnecessary but a brake: without it the system surpassed decades of work in hours.

AlphaGo learned from human games; AlphaZero was given only the rules. Within four hours it surpassed Stockfish at chess. It gave the cleanest evidence for the claim that for closed tasks with a clear objective human experience is not required.

Event record

Event date

December 2017

Timeline date

Event date

Verification

Sources gathered automatically · September 17, 2026

Lines

ID

evt-0235

Preprint of 5 December 2017.

Sources

primary Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm
arXiv · Published December 2017
primary A general reinforcement learning algorithm that masters chess, shogi and Go through self-play (Silver et al.), author preprint
DeepMind authors; preprint hosted at starcraft.ai (not the publisher)

Related events

Earlier

Extends AlphaGo versus Lee Sedol 2016
The last human component is removed: no human games in training.
- Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm · arXiv
Related The HEARSAY-II blackboard 1980
Self-play carried from backgammon to chess and go by one algorithm.

Later

Builds on AlphaStar at StarCraft II 2018links to this record
DeepMind places AlphaStar in the self-play line in which AlphaGo and AlphaZero reached superhuman play at Go, chess and shogi.
- AlphaStar: Grandmaster level in StarCraft II using multi-agent reinforcement learning (DeepMind blog) · Google DeepMind
Extends MuZero: planning without the rules 2019links to this record
MuZero matches AlphaZero at Go, chess and shogi without being given the rules AlphaZero was supplied with.
- Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model (Schrittwieser et al., arXiv:1911.08265v1) · arXiv (DeepMind)
Related AlphaDev 2023links to this record
The same self-play approach, carried over to search in the space of processor instructions.
Related AlphaGeometry 2024links to this record
The same approach of search guided by a trained model, carried from a game into formal mathematics.
Builds on Silver at the International Mathematical Olympiad 2024links to this record
AlphaProof learns by self-play in a space of formal proofs.
- AI achieves silver-medal standard solving International Mathematical Olympiad problems · Google DeepMind