Turing Award winner Sutton: pure generative AI cannot do real science

Richard Sutton identifies a core weakness of generative AI: inability to evaluate its own results, preventing genuine scientific discovery

Published 10sem1 sourceNotable

≈ 23sSpeed

The fact

Only built-in evaluation loops (as in AlphaGo or AlphaProof) enable genuine creativity and sustained discoveries

Without evaluation mechanisms, novel results remain fleeting and cannot build cumulative scientific progress

Click the link to read an article on the topic:

Explore this topic

Auto-synthesis from 1 media source · identified on June 1, 2026