Math Experts Push Back Against AI Hype, Warn of "Plausible but Unreliable" Solutions
Over 150 mathematics experts have signed a declaration warning governments and institutions not to overestimate AI's ability to solve long-standing mathematical problems, directly challenging recent claims from OpenAI and others about breakthroughs in the field.
The Leiden Declaration on AI and Mathematics emerged after a string of headlines claiming AI had solved problems that have stumped mathematicians for decades. Earlier this year, a 23-year-old used ChatGPT to claim he'd solved one of the ErdΕs problems. Last month, OpenAI announced its AI had disproven an 80-year-old conjecture, calling it "the first time AI has autonomously solved a prominent open problem."
The declaration's signatories, including International Mathematical Union vice president Ulrike Tillmann, argue that such claims reflect commercial incentive rather than scientific reality.
The Core Problem: Plausible Nonsense
The declaration identifies a critical flaw in using AI for mathematics: these systems can generate solutions that sound convincing but contain errors that are hard to spot.
"Current automated techniques can produce plausible but unreliable (or even incorrect) arguments which are difficult to distinguish from correct mathematical proofs," said Leslie Ann Goldberg, head of computer science at the University of Oxford and a declaration signee.
This matters because mathematical research builds on previous work. If unreliable results enter the literature, they contaminate future research. The broader scientific community has already seen this happen, with journals struggling to manage an influx of AI-generated papers that risk compromising peer review.
Commercial Pressure and Academic Precarity
The declaration also addresses why mathematicians might feel pressure to endorse AI despite reservations. The technology industry has offered lucrative positions, funding, and computing resources at a time when academic employment remains precarious and higher education funding has declined.
"There is currently a strong commercial incentive on the part of the technology industry to overstate the capabilities of their products," the declaration states, advising policymakers to consult mathematicians directly rather than relying on industry press releases.
Consent and Broader Concerns
A separate issue raised in the declaration: AI models train on published research without author permission. Mathematicians whose work appears in training datasets never agreed to contribute to AI development.
"Mathematicians who never intended to contribute to AI development are having their work used for this purpose without their consent," said Rodrigo Ochigame, an anthropologist of AI at Leiden University who helped draft the declaration.
The declaration also flags concerns beyond mathematics, citing the AI industry's involvement in military applications, mass surveillance, misinformation, and environmental costs.
The statement represents the strongest public pushback yet from the mathematics community against claims of AI-driven breakthroughs. It signals that experts in the field are willing to publicly dispute narratives that dominate mainstream coverage.
For researchers evaluating AI tools, the declaration offers a clear message: scrutinize results carefully, verify claims independently, and recognize that plausible-sounding output is not proof of correctness.
Learn more about how large language models work and their limitations, or explore AI's role in academic research.
Your membership also unlocks:
