Descriptions:
On the tenth anniversary of AlphaGo’s landmark 4-1 victory over 18-time world champion Lee Sedol, Google DeepMind’s podcast host Professor Hannah Fry sits down with two of the system’s core architects: distinguished research scientist Thore Graepel, who was present in Seoul and played early versions of AlphaGo himself, and Pushmeet Kohli, who now leads DeepMind’s science work and traces those early techniques into today’s AI breakthroughs.
The conversation covers the technical reasons Go was considered a uniquely hard challenge — orders of magnitude more complex than chess in both search breadth and game depth — and how combining supervised learning on professional games with reinforcement learning self-play produced capabilities that surprised even the team. Both Move 37 in Game 2 (AlphaGo’s alien-seeming offensive play) and Move 78 in Game 4 (Lee Sedol’s “divine move” that exposed the system’s fragility) are analyzed in detail, offering a window into how the system reasoned and where it broke down.
Beyond the historical narrative, Kohli maps the direct technical lineage from AlphaGo’s reinforcement learning core to protein folding (AlphaFold), large language models, and modern agentic AI. The episode is an unusually substantive primary-source account of a pivotal moment in AI history, making it valuable for anyone wanting to understand not just what happened in Seoul in 2016, but why it mattered for the architecture of systems being built today.
📺 Source: Google DeepMind · Published March 10, 2026
🏷️ Format: Podcast
