Richard Sutton, a prominent AI researcher, shared a perspective on generative AI's creative capabilities in a video posted on X. He argued that generative AI models trained by supervised learning are inherently incapable of making novel discoveries. This statement was part of a speech titled "AI Creativity and Discovery" shared on June 10, 2026, addressing AI's relationship to science and mathematics, according to his X post.
In the video and accompanying text, Sutton explained that while generative AI can produce new combinations of existing knowledge, it does not truly innovate or discover fundamentally new concepts. He emphasized that supervised learning models generate outputs based on patterns in their training data, limiting their ability to create original scientific or mathematical insights. Sutton's remarks were intended to provoke discussion on the limits of current AI approaches in advancing knowledge.
This perspective challenges some common narratives around AI's potential to revolutionize scientific discovery. Sutton's view highlights a distinction between generating novel-seeming content and genuine innovation, which requires more than pattern recognition. His comments contribute to ongoing debates about the role of AI in research and the expectations placed on generative models, especially in fields demanding original thought and discovery.
The full speech and video are available on Sutton's X account and YouTube channel, providing detailed reasoning behind his stance. His insights add to the discourse on AI's capabilities as of June 2026, underscoring the need for further research into AI methods that might enable true discovery beyond supervised learning.