In early October, OpenAI posted more than 700 AI-generated mathematical manuscripts in an online public repository, prompting both excitement and concern across the mathematics community. If the arguments are tested and found correct, they could represent a surprising amount of new mathematical knowledge. But researchers are also worried about the effect on their careers, the prospects of younger mathematicians, and the lack of transparency surrounding how AI-generated research is published.
As a co-author of the Leiden Declaration on Artificial Intelligence and Mathematics, published in June, I believe many of the concerns about OpenAI’s approach were foreseeable. The declaration does not oppose the use of AI. Instead, it calls for practical safeguards, including clear citation policies, accurate disclosure of how AI is used, and a renewed commitment to the long-standing principle that mathematical proofs should be independently verifiable and that human authors should be able to explain and take responsibility for their arguments. Some AI companies have agreed to follow the declaration. OpenAI has not.
As mathematicians have begun examining the AI-assisted proofs, it has become clear that many of the 700 preprints are written in a rough, sketch-like style that makes them difficult to understand. The inner workings of OpenAI’s closed model are unknown, and no human author has been identified as responsible for explaining the work or correcting possible errors. Checking the manuscripts and rewriting them in a professional form suitable for peer review requires substantial time and effort. Understandably, many mathematicians are reluctant to undertake that work, regardless of how interesting the underlying mathematics may be.
Who should shape mathematics in the age of AI?
The future of mathematics will require creative rethinking and action at every level. Leadership from mathematicians such as Terence Tao and grassroots initiatives such as the Leiden Declaration should be complemented by engagement from organizations representing the wider field, including the International Mathematical Union. It would be a disservice to mathematics if AI companies were allowed to establish standards for conducting and publishing mathematical research without broader input from academics.
Source: www.nature.com
