This episode commemorates the 10-year anniversary of Google DeepMind's AlphaGo defeating Go world champion Lee Sedol, marking a pivotal moment for AI. Guests Thore Graepel and Pushmeet Kohli discuss how AlphaGo's novel approach, combining intuition and calculation, laid the groundwork for modern AI breakthroughs like AlphaZero and AlphaFold, pushing the boundaries of human knowledge in complex scientific domains.
Summarized by Podsumo
AlphaGo's 2016 victory over Lee Sedol was a landmark event, demonstrating AI's ability to combine intuition and calculation, not just brute force, in a game of immense complexity.
Move 37 by AlphaGo in the second game was a "divine move" that baffled human professionals, expanding their understanding of Go strategy and proving AI could generate novel insights.
AlphaZero's self-play training, without any human game data, led it to not only rediscover human Go knowledge but also surpass it, finding "alien" yet superior strategies.
The core techniques of AlphaGo, particularly its search algorithms and policy networks, directly influenced projects like AlphaFold for protein folding and AlphaTensor for optimizing matrix multiplication, tackling scientific grand challenges.
The episode explores how AI systems can go beyond human knowledge, generate new insights, and the critical role of verifiers in distinguishing genuine discoveries from "hallucinations" in scientific applications.
"Not a single human player would have chosen move 37."
— Professional commentators
"If we can do this, then what else could we do?"
— Thore Graepel
"AlphaGo was that transition point where it became very, very clear that the moment of transition where we go beyond human level intelligence in particular areas is not a science fiction or many decades later it is happening now."
— Pushmeet Kohli