The rapidly evolving AI sector seems to be on the cusp of a critical turning point.
On September 11th, at the 2025 Inclusion·The Bund Conference, Richard Sutton, the “father of reinforcement learning,” pointed out that the AI industry is currently in the “Age of Human Data.” Most machine learning efforts today aim to transfer existing human knowledge to AI systems that are static and lack autonomous learning capabilities.
The problem, however, lies in the fact that under this paradigm, the dividends of human data are approaching their limits—yet continuous learning is crucial to the effectiveness of intelligence. Sutton argued that AI is entering an “Age of Experience” centered on continuous learning, and its potential will far surpass that of previous eras.
In his speech, Sutton stated that we are stepping into the “Age of Experience,” which demands a new type of data source generated through direct interaction between intelligent agents and the world. This is precisely how humans and other animals learn—and it is also the path that enabled AlphaProof to win a silver medal at the International Mathematics Olympiad (IMO) recently.
Sutton explained that “experience” refers to three types of signals—observation, action, and reward—that flow back and forth between an intelligent agent and the world. “Knowledge stems from experience, and learning occurs through experience. The intelligence of an agent depends on its ability to predict and control the input signals it receives. Experience is the core and foundation of all intelligence.”He also noted that while reinforcement learning has ushered us into this new Age of Experience, unlocking its full potential requires two currently immature technologies: continual learning and meta-learning.
Addressing widespread concerns that AI could bring about bias, unemployment, or even human extinction, Sutton argued that such fears about AI are overblown—and stoked by certain organizations and individuals who stand to benefit from them. He pointed out that economies function best when individuals have diverse goals and capabilities; similarly, intelligent agents with different objectives can achieve win-win outcomes through decentralized collaboration.
“Humanity’s most remarkable superpower is its ability to collaborate better than any other animal. Our greatest successes lie in collaboration itself—economies, markets, and governments are all products of successful collaboration,” Sutton said.
He emphasized that AI and human prosperity will stem from decentralized collaboration. “Collaboration is not always achievable, but it is the source of all good things in the world. We must pursue collaboration, support it, and work to institutionalize it.”
Looking ahead to AI’s future, Sutton proposed four practical “principles of prediction”: First, there is no universal consensus on how the world should operate, and no single perspective should take precedence over others; Second, humans will truly understand intelligence and use technology to create it; Third, the current level of human intelligence will soon be far surpassed by super AI—or by humans enhanced with super intelligence; Fourth, power and resources will flow to the most intelligent agents.Based on these principles, he believes that replacement of the latest AI will be inevitable in the course of human development.
Taking a broader view of cosmic history, Sutton divided it into four eras: the Age of Particles, the Age of Stars, the Age of Replicators, and the Age of Design. He argued that humanity’s uniqueness lies in “pushing design to its limits”—creating things that can design themselves, which is exactly the goal pursued through AI today. Humans are at the very least a catalyst, and more importantly, pioneers in ushering in the universe’s fourth great era: the “Age of Design.”
“AI is the inevitable next step in cosmic evolution, and we must embrace it with courage, pride, and a spirit of adventure,” Sutton concluded.
