OpenAI VP Responds to Fields Medalists: Future AI Could Be a Research Partner in Math

On September 14, 2026, Liam Fedus, OpenAI's vice president, responded to a joint statement by Terence Tao and 25 Fields Medalists about the role of AI in mathematics. He agreed that conceptual understanding matters, but argued that the current limits of AI should not constrain its future possibilities.

Tao published the statement, "Severe Misalignment of AI in Mathematics," on his blog on September 11. The co-signatories span generations of Fields Medalists, from Pierre Deligne (1978) to Yu Deng (2026). The statement warned that AI companies are using "solving famous math problems" as a central benchmark, which may be misaligned with the mathematical community's goals of conceptual understanding, methodology, and knowledge transmission.

Fedus said future AI systems should not only solve problems but also generate valuable insights and establish new conceptual frameworks for mathematicians to build on. If training objectives expand beyond getting correct answers to explaining why and helping humans understand, AI could become a research partner that gives answers and opens new directions.

He also agreed with an analogy to AlphaGo: although AlphaGo could not explain its moves, human players learned from its play, and Noam Brown had previously observed that human Go players improved after AlphaGo. Fedus suggested mathematics could see a similar shift. The main difference is that mathematicians worry about damage to accumulated knowledge, while Fedus is more optimistic that AI will eventually take part in creating it.