Wikinventia — Atlas of discoveries and inventions · Digital Age

Mastering the game of Go via reinforcement learning — DeepMind

2016 AD · Transmission: Global

AIInventionBritish

In March 2016 AlphaGo, developed by DeepMind under Demis Hassabis, defeats world champion Lee Sedol 4-1 in a globally televised match. Go's state space of 10^170 had made it a paradigmatic example of a domain unreachable by classical AI. AlphaGo combines deep convolutional networks trained on human games, reinforcement learning via self-play, and Monte Carlo Tree Search. The later AlphaGo Zero (2017) learns exclusively via self-play, beating AlphaGo 100-0.

InstitutionDeepMind — London, United Kingdom

Historical regionUnited Kingdom

Primary sourceSilver, D. et al. — "Mastering the game of Go with deep neural networks and tree search" (Nature, 529, 484-489, 2016). DOI: 10.1038/nature16961

Secondary sourceSilver, D. et al. — "Mastering the game of Go without human knowledge" (2017)

Original languageEnglish

View this entry in the interactive atlas → View in graph →