Leiden Declaration Aims to Keep AI from Overpowering Math

OpenAI’s AI solved the unit distance problem, sparking a wave of AI‑generated proofs that threaten mathematical integrity.

AI Proofs Flood Peer Review

AI‑generated proofs are now clogging journal editors’ inboxes, making it hard to vet each submission. Large language models regurgitate human ideas without attribution, creating a “mess” that undermines transparency and accessibility—values long prized in mathematics.

The 2024 AlphaProof release illustrates the problem. Google DeepMind announced that its model solved three difficult math‑competition problems, yet it took more than a year before the methods appeared in a peer‑reviewed journal. When commercial interest mounts, researchers often retreat behind closed doors, leaving the broader community in the dark.

Leiden Declaration’s Core Tenets

The 11‑page Leiden Declaration on Artificial Intelligence and Mathematics, drafted by 60 researchers at Leiden University’s Lorentz Center, lays out concrete safeguards:

• Disclosure – Authors must state when AI contributed to a proof.
• Attribution – Prior work must be properly cited, even if the AI regurgitated it.
• Extra Scrutiny – AI proofs should undergo additional verification to catch subtle, hard‑to‑spot errors.
• Leveling the Field – Public funding and legal resources should counterbalance corporate power, ensuring independent mathematicians can influence the direction of research.

Rodrigo Ochigame, an anthropologist of AI, notes that the declaration “calls for commercial AI companies to adhere to these principles.” He worries that mathematicians “never intended to contribute to AI development” yet find their work used without consent.

Industry Response and Future Steps

The International Mathematical Union plans to endorse the declaration, and Jim Portegies will speak about it at the IMU’s upcoming conference this summer. If the declaration gains traction, it could set a precedent for how AI tools are vetted in any rigorous scientific field.

The declaration’s success hinges on widespread adoption by both academia and industry. Until then, mathematicians will continue to navigate a landscape where AI can both accelerate discovery and erode the very foundations of the discipline.

Source: Mathematicians sign declaration to rein in AI use
Domain: scientificamerican.com