Skip to main content

AlphaGo: Judgment Grown

This video presents the same text shown beside it, spoken and on screen. It adds nothing the text does not say.

State

AlphaGo beat the world Go champion by combining tree search with learned judgment: networks trained on human games, then on play against itself.

Show

By its maker's account: a policy network proposing moves, a value network judging positions, sharpened by self-play.

Watch for

Where a written judge can be read and argued with, a grown judge cannot — a cost with its own unit later.