
Using AI to Accelerate Scientific Discovery
Demis Hassabis, co-founder and CEO of DeepMind, argues that artificial intelligence is entering a phase where it can accelerate scientific discovery itself rather than just excel at games. Speaking at Oxford's Sheldonian Theatre, he traces DeepMind's path from AlphaGo, the system that beat the world Go champion, through to AlphaFold, which solved the fifty-year protein structure prediction problem and produced a near-complete map of the human proteome. He explains how self-learning systems trained on games became a proving ground for general-purpose algorithms now being aimed at biology, chemistry, and physics problems. Hassabis discusses what made protein folding tractable for machine learning, what still resists it, and what a future where AI systems propose and test scientific hypotheses might look like. Drawing on his own background in neuroscience and games, he situates DeepMind's work within a longer argument about intelligence, learning, and discovery, running over an hour with audience questions.