Mathematics: Call to Boycott OpenAI After 700 AI Proofs

After OpenAI's release of over 700 AI-generated proof manuscripts, the Association for Human Mathematics is calling for a boycott. The AGMAI warns against a future of research dictated by AI laboratories and emphasizes the autonomy of researchers. The community is divided, with some supporting the boycott and others defending open publication, amid examples of texts deemed unreadable without machine assistance.

AGMAI Warns Against Research Guided by AI Labs

The Assembly for the Governance of Mathematics and AI (AGMAI) takes a more nuanced approach than the boycott call, believing that the publication represents an initial step while the mathematical assimilation of the results is just beginning. However, AGMAI warns against a future of mathematical research that would be limited to exploiting results from AI laboratories. For the organization, it is essential that mathematicians are able to pose their own questions, design their methods, and investigate areas that have not been selected as examples illustrating the capabilities of an AI system. AGMAI emphasizes that equitable access to powerful research tools and adequate computational resources is crucial for this freedom. Reactions published on Terence Tao's blog show a community divided on how to respond.

Boycott: AHM's Grievances Against OpenAI

The Association for Human Mathematics (AHM) advocates for a boycott of OpenAI, accusing the company of violating fundamental principles of scientific research by disseminating AI-generated proofs. AHM states that OpenAI did not adhere to a key recommendation from AGMAI, which stipulates that complex mathematical problems should not be experimented on proprietary models. It notes the simultaneous release of over 700 files, which it views more as a display of power than an act of scholarship. AHM's position also mentions the numerous lawsuits for copyright infringement targeting OpenAI and implies, without explicitly stating, that the company may have obtained these results by using the work of mathematicians for training, whether legally or not. AHM calls on mathematicians to cease all collaboration with OpenAI.

A Concrete Case: The Unique Games Conjecture Put to the Test

The series of documents published by OpenAI includes a purported proof of the Unique Games Conjecture, an open problem in theoretical computer science. Dana Moshkovitz, who has dedicated her career to this conjecture, described the text as resembling something written by someone under the influence of psychotropics and deemed large passages vague. She points out that the article cites numerous works without explaining their relevance, while prior results should have excluded them. According to her, the article is so poorly written that it is impossible to read without the aid of AI. She adds that the proof relies on an entirely new and strange construction, which she describes as "extraterrestrial madness."

Validation Standards: Open Disagreements and Massive Publication

A controversy is emerging regarding the legitimacy of declaring a problem solved before a proof has been independently verified and made accessible for human understanding. Some researchers report that articles are perceived as particularly compact, sometimes only accessible with the help of AI. OpenAI's dissemination of several hundred manuscripts, including over 700 files published all at once, fuels these debates. Within the discussions, some participants support the boycott, recommend ceasing the use of OpenAI products, and suggest strengthening the protection of preprint platforms like arXiv against massive data scraping for training purposes. Others view AHM's position as unrealistic, believing that OpenAI will continue its research and that publicly released results are better than private ones. Finally, some question the possibility of considering a problem solved when neither the authors nor others fully understand the proof.

Tao and Aaronson Frame the Public Debate

Terence Tao, president of AHM, shared the association's statement in a guest post on his blog and mentioned on Mastodon an era of "Math 2.0," where problem-solving would no longer be the primary goal of the discipline. Meanwhile, complexity theorist Scott Aaronson referred to the situation as a "Mathocalypse" on his blog, helping to publicly frame the terms of a debate that goes beyond the initial publication.