Mathematicians Sign Leiden Declaration to Establish AI Guidelines for Research
Key Takeaways
- ▸Over 60 researchers and policymakers created comprehensive guidelines to prevent AI from undermining mathematical research integrity while maintaining the field's open-science traditions
- ▸Tech companies frequently conceal AI methodology details for competitive advantage, contradicting mathematics' commitment to transparency and accessibility (e.g., Google DeepMind's AlphaProof took 18+ months to publish peer-reviewed results)
- ▸The declaration mandates AI disclosure in papers, commitment to peer review, equitable access to computational resources, and comparable legal/funding support between academic and corporate entities
Summary
Mathematicians, computer scientists, and math historians have released the "Leiden Declaration on Artificial Intelligence and Mathematics," establishing comprehensive guidelines for responsible AI use in mathematical research. The 11-page declaration was developed following a workshop at Leiden University in the Netherlands where approximately 60 researchers and policymakers gathered to address growing concerns about AI's rapid advancement in mathematics, catalyzed by breakthroughs like OpenAI's solution to the famous "unit distance" geometry problem and Google DeepMind's AlphaProof.
The mathematicians identified critical challenges: AI-generated proofs often lack proper attribution, tech companies withhold methodological details for commercial reasons (contradicting the field's open-science tradition), and the explosion of AI submissions is overwhelming journal editors' capacity for meaningful peer review. Among their key prescriptions are mandatory disclosure of AI use in research, ensuring rigorous peer review of all papers, and leveling the playing field between well-funded corporate AI labs and academic institutions through legal resources and public funding.
The declaration emphasizes that mathematics has long valued transparency and accessibility—nearly all modern mathematical papers are freely available on arXiv.org—but corporate AI developers often retreat "behind closed doors" due to commercial interests. For example, Google DeepMind's AlphaProof announcement in 2024 took more than a year before methods were published in a peer-reviewed journal, illustrating the divide the mathematicians seek to address.
- Core concerns include improper attribution in AI-generated work, overwhelming journal submission volume, and the risk of losing the openness and accessibility that mathematicians have long prioritized
