David Silver on the 'Era of Experience' and Beyond Human Data
In this episode of Google DeepMind: The Podcast, host Hannah Fry speaks with David Silver about his position paper proposing a new 'era of experience' for AI—one that moves beyond the current paradigm of training on human data. Silver argues that while large language models have extracted enormous value from human-generated text, this approach has a hard ceiling: it can only reproduce and recombine human knowledge, never exceed it. Drawing on AlphaZero—which famously learned Go, chess, and Shogi from zero human data via self-play reinforcement learning—Silver contends that agents generating their own experience from interaction with the world can break past that ceiling indefinitely, much like AlphaZero's Move 37 revealed strategies no human had conceived.
Original sources
Podcast Google DeepMind: The Podcast
YouTube Google DeepMind: The Podcast
AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.