Rich Sutton의 AI 창의성과 발견

Sutton argues that generative AI, while useful, is fundamentally limited to mimicry and cannot achieve genuine discovery without reinforcement learning and systematic evaluation. The breakthrough systems like AlphaGo, AlphaFold, and Claude-Code succeed not through generation alone, but through iterative cycles of variation, evaluation, and selective preservation. This insight is crucial for understanding AI's true capabilities and limitations in scientific and creative domains.