AlphaGo is a DeepMind program that became the first to beat a professional human, and then a world champion, at the board game Go.
In plain words
Go was long seen as the hardest classic game for AI, because there are too many possible moves to search and positions are hard to judge. AlphaGo combined two neural networks, one to choose promising moves and one to judge who is winning, with a search method that plays out likely futures. In a paper published in 2016, it beat the European champion 5 games to 0, a feat thought to be a decade away. A later version, AlphaGo Zero, learned only by playing itself.
A quick example
In a game, AlphaGo's move network suggests a few sensible moves, its value network rates the positions they lead to, and the search explores the most promising lines before it places a stone.
Where you'll meet it
- A landmark in game-playing AI and reinforcement learning.
Related
Sources
- PaperMastering the game of Go with deep neural networks and tree search, Silver et al. (Nature) · read 29 Sept 2026
- DocsAlphaGo Zero: Starting from scratch, Google DeepMind · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.