The Alarm Bell of the Era of Large Models: Richard Sutton Calls for a Rekindling of Scientific Exploration into Intelligence Understanding

At the recent RL China 2025 opening ceremony, Professor Jun Wang from University College London had an in-depth conversation with "the father of reinforcement learning," Richard Sutton, discussing the essence of intelligence and its future direction. As an authority in the field of intelligent information systems, Professor Wang reviewed the academic foundations of reinforcement learning (RL) with Sutton, while also focusing on the impact of the rapid expansion of the artificial intelligence industry on fundamental scientific research.

Sutton pointed out that current artificial intelligence technologies, especially large-scale language models (LLMs), although demonstrating strong capabilities in practical applications, are not the way to truly understand intelligence. He emphasized that LLMs lose their ability to learn after training because they lack clear goals and reward mechanisms. In contrast, reinforcement learning emphasizes achieving goals through interaction with the environment, which better understands and simulates the operation process of intelligence.

Sutton proposed that true intelligence requires clear definition of goals and learning driven by "reward" signals. He believes that learning from experience is the core of reinforcement learning, rather than relying solely on methods such as gradient descent. To promote the development of intelligence, researchers need to combine search with gradient descent to explore more possibilities. He also emphasized that although the influx of industry funding has promoted application development, this should not cause scientific research to deviate from long-term goals.

In his message to young researchers, Sutton encouraged them to focus on basic science, emphasizing that the exploration of artificial intelligence is a long-term process, not a problem that can be solved in the short term. He mentioned that although the industry focuses more on achievable technologies at present, the scientific community must be committed to exploring unresolved fundamental issues to promote a true understanding of intelligence.

This conversation reminds us that while pursuing technological advancement, we should not neglect the in-depth understanding of the essence of intelligence.