Breakthrough or Hype? Over 2,000 Scholars Sign AI‑Mathematics Leiden Declaration Warning AI’s Limits in Fundamental Science

AI advances and concerns in mathematics

AI for Science has produced breakthroughs in biology, chemistry, and physics, but the impact of AI on mathematics has raised significant concerns.

Leiden Declaration on AI and Mathematics

Authored by 16 mathematicians from leading universities and research institutes; the declaration is 11 pages long.

More than 2,000 scholars—including mathematicians, philosophers, and computer‑science researchers—have signed the declaration.

The declaration warns that uncontrolled automation threatens both the way mathematics is conducted and the meaning of the discipline.

High‑profile AI claims

In the “First Proof” competition, an AI model solved several research‑level mathematical problems.

OpenAI announced that a large model had solved the 80‑year‑old Erdős Unit Distance Problem.

The declaration attributes these claims to strong commercial incentives that can lead to hype.

Market‑driven publicity versus peer‑reviewed science

Michael Harris, Columbia University professor and one of the declaration’s initiators, said laboratories are engaged in a “life‑or‑death” battle, using mathematics to attract investment. AI announcements follow market timelines rather than the peer‑reviewed scientific cycle, and specific mathematical tasks may be used misleadingly as proxies for general reasoning ability.

Technical risks identified

AI systems can generate proofs that appear plausible but contain errors that are difficult for humans to verify.

Systems may appropriate human researchers’ results to make speculative inferences, reducing recognition of genuine contributions.

Potential cultural impact

Widespread AI use could encourage trend‑driven research, shorten peer review, and turn mathematicians into service providers for AI developers, undermining independent mathematical inquiry.

Community responses

Ulrike Tillmann, Vice‑President of the International Mathematical Union, acknowledges new opportunities but calls for transparent, fair practices guided by global mathematical values.

Fields Medalist Terence Tao fully endorses the declaration and notes that these issues should have been discussed systematically years ago.

Historical over‑promising examples

Physicist Nick McGreivy reported disappointment after attempting to accelerate physics research with AI, citing survivor bias and overly polished results.

DeepMind’s GNoME model claimed discovery of 2.2 million crystal structures; materials scientists later found most generated compounds unusable.

Overall perspective

AI continues to achieve task‑specific breakthroughs and assist research, but the community is urged to move beyond the binary question of “can we use it?” toward deeper methodological, ethical, and philosophical reflection on AI’s role in fundamental science.

Reference links: https://leidendeclaration.ai/ https://phys.org/news/2026-06-mathematicians-dont-hype-ai-capabilities.html https://www.science.org/content/article/mathematicians-issue-warning-ai-rapidly-gains-ground

Code example

来源：ScienceAI
本文
约2000字
，建议阅读
5
分钟
尊重技术的边界，尊重科学本身不可替代的根基。