Rich Sutton on AI Creativity and Discovery

In a Twitter thread, Rich Sutton (co-inventor of reinforcement learning) shares his thoughts on AI creativity and discovery. He argues that true artificial intelligence must be able to generate novel ideas and hypotheses, not just optimize known goals. Sutton praises recent work on open-ended learning algorithms, such as POET and GO-Explore, which create their own challenges. He suggests that future AI systems should be evaluated on their ability to discover new knowledge rather than just solve predefined tasks. He also warns against over-reliance on large language models, which are fundamentally pattern-matching rather than generating truly new ideas.