A.I. Could Solve Some of Humanity’s Hardest Problems. It Already Has.
About this episode
Ezra Klein asks Google DeepMind CEO Demis Hassabis how AI went from games to solving grand scientific challenges. Hassabis traces his path from video-game design through neuroscience to founding DeepMind, and describes AlphaFold's solution to protein-structure prediction — now 200 million structures freely available — as AI's first great scientific triumph, accelerating drug discovery and enzyme design. They discuss applying AI to materials science, fusion research, and climate modeling. On risks, Hassabis acknowledges the dangers of advanced AI, including misuse and misalignment, but argues the benefits are enormous and that overly restrictive regulation could concentrate power in a few hands. Klein presses him on racing dynamics between labs and nations and the case for international coordination. They disagree mildly on timelines and governance but share an optimistic view of AI's scientific promise.