The Bayesian Conspiracy 37 – AlphaGo Returns
10 snips
Jun 21, 2017 Patrick Chapin, pro Magic: The Gathering player, game designer, and writer, returns to talk AlphaGo's evolution and its strange, alien play. He covers AlphaGo’s anonymous online dominance, human-plus-AI collaborations, and efforts to extend Go-style systems to complex games like StarCraft. They also digress into sports tech, tool-assisted play, and how mods spawn new genres.
AI Snips
Chapters
Transcript
Episode notes
AlphaGo Learned By Playing Itself
- AlphaGo shifted from learning human games to self-play and built its own novel strategies.
- Patrick Chapin explains Master trained mostly by playing millions of games against itself, producing moves humans hadn't seen before.
AlphaGo's Master Disguised Win Streak Online
- Master impersonated a mysterious strong player on an elite Go server and went on a win streak before being revealed as AlphaGo.
- Patrick recounts Master playing extremely fast and making multiple 'legendary' moves per game that raised suspicion.
AlphaGo Became Alien And Hard To Interpret
- AlphaGo's later play became alien and hard for humans to interpret despite being stronger.
- Patrick says its strategy no longer adhered to human heuristics, making games less instructive to watch.