Turing Award winner Richard Sutton says pure generative AI can't do real science

Richard Sutton argues that generative AI can make novel-looking outputs, but it cannot reliably tell which ideas are actually good.

He says real scientific discovery needs variation, evaluation, and selective retention, not just generation.

Sutton points to systems like AlphaGo, AlphaFold, AlphaProof, and Claude Code as examples that combine generation with feedback or verification.

He still sees generative AI as useful for summaries, assistants, and entertainment when it is faster, cheaper, or more customizable.