A Chinese forum post analyzes Richard Sutton's 2026 position paper 'Toward Enactive Artificial Intelligence' (arXiv:2605.24238), which argues that AI should adopt enactive cognition—treating perception as active, embodied interaction rather than passive input processing. The author identifies two internal contradictions: first, enactive AI demands that normativity arise from an agent's own organization, yet Sutton's own Reward Hypothesis (2004) defines all goals as maximization of externally defined scalar rewards—Sutton's paper itself admits RL evaluation criteria 'remain externally defined.' Second, the paper hard-codes philosophical cognitive theory (Merleau-Ponty, Varela's autopoiesis) into AI architecture, violating the principle from Sutton's 2019 essay 'The Bitter Lesson' that building in how we think we think fails long-term. The post also cites the scaling-up problem and the coupling-constitution fallacy as unresolved objections, compares the route to Rodney Brooks' defeated anti-representation program, and contrasts it with the LLM+VLA engineering mainstream. It notes David Silver's Ineffable Intelligence raised a $1.1B seed round at a $5.1B valuation betting on the experience-based RL路线, with a verdict expected by 2028-2030.
Richard Sutton's Enactive AI Paper Contradicts His Own Reward Hypothesis and the Bitter Lesson
Summary
A Chinese forum post analyzes Richard Sutton's 2026 position paper 'Toward Enactive Artificial Intelligence' (arXiv:2605.24238), which argues that AI should adopt enactive cognition—treating perception as active, embodied interaction rather than passive input processing. The author identifies two internal contradictions: first, enactive AI demands that normativity arise from an agent's own organization, yet Sutton's own Reward Hypothesis (2004) defines all goals as maximization of externally defined scalar rewards—Sutton's paper itself admits RL evaluation criteria 'remain externally defined.' Second, the paper hard-codes philosophical cognitive theory (Merleau-Ponty, Varela's autopoiesis) into AI architecture, violating the principle from Sutton's 2019 essay 'The Bitter Lesson' that building in how we think we think fails long-term. The post also cites the scaling-up problem and the coupling-constitution fallacy as unresolved objections, compares the route to Rodney Brooks' defeated anti-representation program, and contrasts it with the LLM+VLA engineering mainstream. It notes David Silver's Ineffable Intelligence raised a $1.1B seed round at a $5.1B valuation betting on the experience-based RL路线, with a verdict expected by 2028-2030.
This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177980942