Mathematicians issue warning as AI rapidly gains ground
Mathematicians are circulating the “Leiden Declaration” to warn that rapidly advancing AI could reshape their field, from how proofs are generated and verified to how credit, careers, and training pathways work. Commenters debate whether math is primarily about producing correct answers or about human understanding, creativity, and a shared research culture, and worry that AI‑generated proofs and papers could overwhelm peer review and hollow out early‑career training. Others argue these fears are overstated, seeing AI as a powerful assistant that will democratize advanced math and accelerate progress, much as engines changed chess and automation changed other professions.
Scope of AI in Mathematics
- Some mathematicians in the thread think the “warning” is overblown; they see AI as an optional tool, at least for now.
- Others argue that if AI becomes a strong productivity multiplier, non‑use will become career‑limiting (publications, hiring, grants).
- There is debate whether academia is currently pushing AI use: some say no pressure yet; others see long‑term structural pressure via “publish or perish” and compute access.
What Mathematics Is For
- One camp emphasizes math as a machine for producing correct answers and practical tools; funding is justified by societal ROI, not as a “jobs program for nerds.”
- Another camp stresses that the key value is understanding, asking good questions, and building conceptual frameworks, not just truth values.
- Ongoing tension between “pure” curiosity‑driven work vs applied, commercially relevant math; some argue pure math is culturally like art, others dispute that picture.
Capabilities and Limits of Current AI
- Participants note recent AI proofs of nontrivial problems as evidence of genuine capability, not just “slop.”
- Others caution that many results look like extremely deep literature search and recombination, not “new theory.”
- Concerns: AI may generate vast amounts of technically correct but esoteric or incomprehensible math, potentially beyond human understanding or utility.
- Counterpoint: current LLMs excel near their training data, not at radically novel abstractions; they still show a “long tail” of glaring errors.
Impact on Careers, Training, and Access
- Widespread worry about the “junior problem”: low‑hanging problems that used to train PhDs may be solved by machines in hours, undermining apprenticeship paths.
- Some see parallels with other fields where AI removes entry‑level work and thus the ladder to senior expertise.
- Access to compute and proprietary models may further stratify research groups and countries.
Quality, Peer Review, and Verification
- Fear that AI‑generated papers will flood journals, overwhelming already under‑rewarded peer review.
- In math, formal proof systems could mechanically verify correctness, but most work is not yet fully formalized; importance and context still require human judgment.
Education and Public Understanding
- Many see AI as a powerful tutor that makes advanced math more accessible, especially for those without strong local mentorship.
- Others warn that easy answers can blunt the hard work needed to build deep intuition; “use it or lose it” concerns about human reasoning are raised.
Ethical, Economic, and Cultural Tensions
- Some participants worry that commercial and military incentives driving AI development conflict with traditional academic values.
- Others respond that math has long been influenced by commercial utility, and human mathematicians must adapt rather than treat the field as exclusively human.