Richard Sutton: Experience-driven is the core path for the next generation of AI.

According to the Securities Times, on July 17, 2026 Turing Award winner and founder of reinforcement learning Richard Sutton stated at the main forum of the 2026 World Artificial Intelligence Conference that current large-scale models and multi-modal generation only complete the pattern fitting of human historical data, and can only replicate existing knowledge without the ability to independently discover new knowledge. Issues such as factual inaccuracies and logical flaws are difficult to eliminate. There are two extreme perspectives on AI in the market over-hype and excessive panic and the global industry should adhere to multilateral cooperation and mutual benefit. In his view, the static annotated data model has reached a ceiling, and experience-driven is the core route for the next generation of AI. Currently, artificial annotation resources are continuously declining, unable to support model long-term iterations, and experience-based intelligent agents can rely on environmental feedback to continuously correct cognition. Autonomous evolution is an irreversible technological trend.