Mathematicians React to OpenAI's Publication of Hundreds of Proofs

OpenAI has released hundreds of mathematical proofs covering number theory, theoretical computer science and logic, many of which address problems that have remained unsolved for decades or longer. The response from the mathematics community has been varied. Hugo Duminil-Copin, a 2022 Fields Medal recipient, said the release "annihilated my field of research" and described an emotional reaction. Steven Strogatz, a Cornell University mathematician, noted that the AI output has been described as "strip-mining" the discipline and said he feels both disoriented and exhilarated.

Strogatz compared the situation to a shift from a traditional, decorous environment to a more disruptive one, likening the contrast to the difference between Wimbledon and the US Open. He highlighted that AI can democratize access to mathematical results, citing a case where a Chinese medical resident used ChatGPT to solve Crouzeix’s conjecture, a problem that had resisted professional mathematicians for years.

Both scholars raised concerns about the nature of AI-generated proofs. Strogatz described a "Twilight Zone" aspect, noting that while the models produce correct results, their reasoning can be opaque to experts. He drew a parallel with Google Translate, which functions effectively without providing insight into linguistic theory.

The discussion also touched on broader implications for scientific understanding. Strogatz referenced historical examples where deep comprehension of underlying principles enabled further breakthroughs, such as Maxwell’s work on electromagnetism leading to Einstein’s theory of gravity. He warned that reliance on AI that offers knowledge without accompanying understanding could limit future innovation.

Regina Barzilay, an MIT computer scientist, was mentioned as an example of a researcher who acknowledges that she does not fully understand how her AI models detect breast cancer, yet emphasizes the potential life-saving benefits. The conversation underscored a tension between the desire for advanced problem-solving tools and the need to preserve human insight and the collaborative nature of scientific discovery.