Mathematicians weigh AI’s place in research

OpenAI recently announced ten advances in mathematics and computer science made with its unreleased model Astra, spanning areas including geometry, cryptography and coding theory. The company also issued a statement on its “responsibility to the mathematical community”, acknowledging concerns over attribution, accountability and the changing nature of mathematical discovery.

LLMs such as ChatGPT and Claude are already disrupting research, teaching and publishing. University departments are weighing ethical questions, journals are facing AI-written submissions, arXiv has seen a sharp rise in mathematical papers, and funding agencies are developing rules for acceptable AI use. The debate now reaches beyond technical capability to questions about why mathematics is done, who receives credit and whether understanding matters as much as producing new theorems.

Recent collaborations show the range of responses. Saul Freedman and Melissa Lee solved the semiregularity problem about highly symmetric networks without using AI, after Freedman objected on environmental and social grounds. In another case, Aluna Rizzoli used an OpenAI model and a supercomputing cluster to find an object researchers had sought for over two years, completing the computation in 43 hours before inviting the existing team to co-author a paper.

No single consensus has emerged. The Leiden Declaration calls for AI to augment rather than replace human creativity, while emphasizing transparency, accountability and attribution. The central question is shifting from whether AI can contribute to mathematics to how it can be used without weakening collaboration, integrity and the culture of the discipline.