Mathematicians release guidelines to govern AI use in research
A group of mathematicians, computer scientists, and historians have published rules to prevent artificial intelligence from reshaping their field without oversight. The "Leiden Declaration on Artificial Intelligence and Mathematics" emerged from a workshop last fall where about 60 researchers and policymakers gathered at Leiden University in the Netherlands to discuss how AI will affect mathematics.
The declaration calls for three core practices: disclosure of AI use in research, mandatory peer review of all papers, and legal support and public funding to help academic researchers compete with for-profit companies.
What prompted the guidelines
OpenAI's announcement last month that AI had solved geometry's "unit distance" problem crystallized concerns among researchers. While some welcomed the achievement, many worry that unchecked AI development will damage the field's core values.
Journal editors now receive more AI-generated proofs than they can verify. Large language models reproduce human ideas without attribution. Researchers fear the integrity of mathematics itself is at risk.
The discipline has long prioritized transparency and open access. Nearly every modern mathematics paper is freely available on arXiv.org, and the American Mathematical Society maintains a public repository of papers, books, and reviews. This openness allows anyone to read and build on new work.
Tech companies operate differently. When Google DeepMind announced in 2024 that its AI model AlphaProof had solved three difficult math competition problems, more than a year passed before the methods appeared in a peer-reviewed journal. "We retreat behind closed doors because there is now a lot of commercial interest," said Jim Portegies, a mathematician at Eindhoven University of Technology.
The core concerns
AI proofs contain subtle errors that are difficult to spot, unlike human-written proofs that experts can verify directly. The declaration recommends extra scrutiny for AI-generated work to catch these mistakes.
The goals of mathematicians and tech companies diverge. Mathematicians pursue questions that may yield new techniques and ideas. Companies often focus on problems that showcase their AI systems but have limited impact on mathematics itself. Independent funding can ensure mathematicians retain influence over their field's direction.
Rodrigo Ochigame, an anthropologist of AI at Leiden University, flagged another issue: "Mathematicians who never intended to contribute to AI development are having their work used for this purpose without their consent." Commercial AI companies train on published mathematics without permission or compensation.
What comes next
Some recommendations rest on individual action-disclosing AI use and properly crediting prior research. Others require broader intervention, such as regulating the AI industry or changing how companies access research.
The International Mathematical Union plans to endorse the declaration. Ilka Agricola, chair of the IMU's Committee on Publishing, said the document provides "a starting point for decision making, for discussion."
Portegies will present the declaration at the IMU's conference this summer.
For researchers interested in understanding how these issues affect their work, AI Research Courses and Generative AI and LLM Courses offer relevant training on responsible AI practices in academic settings.
Your membership also unlocks:
