Terence Tao says AI could trigger math's biggest crisis since Gödel
Terence Tao warns AI could trigger a foundational crisis in mathematics comparable to the early 20th-century upheaval caused by Russell's paradox and Gödel's incompleteness theorems The First-Proof Project demonstrated that AI systems solved 7 out of 10 research-level math problems with passing-quality solutions at costs of tens to hundreds of dollars per problem Tao's core concern is not mathematical truth but the implicit framework of mathematical values: what counts as a contribution, what ge
Analysis
TL;DR
- Terence Tao warns AI could trigger a foundational crisis in mathematics comparable to the early 20th-century upheaval caused by Russell's paradox and Gödel's incompleteness theorems
- The First-Proof Project demonstrated that AI systems solved 7 out of 10 research-level math problems with passing-quality solutions at costs of tens to hundreds of dollars per problem
- Tao's core concern is not mathematical truth but the implicit framework of mathematical values: what counts as a contribution, what gets rewarded, and whether machines can truly "do" mathematics
- He proposes a practical rule: if authors cannot give an expert-level talk on their results, the work should not be published, regardless of formal verification
- The Leiden Declaration on AI and Mathematics (June 2026), backed by the International Mathematical Union, offers concrete guidance for the field's adaptation
Why It Matters
This essay forces the AI and mathematics communities to confront uncomfortable questions about the purpose of research in an era of automated proof generation. For AI practitioners, it highlights how Goodhart's law applies when benchmarkable outputs become the primary goal, risking a flood of polished but shallow results. The broader implication is that the mathematics field must redefine what constitutes genuine understanding and contribution before AI-driven incentives permanently distort the ecosystem.
Technical Details
- The First-Proof Project tested four AI systems against ten never-published research-level math problems under controlled conditions, with seven problems receiving at least one passing grade judged as essentially flawless or needing only minor revisions
- Tao's working hypothesis states AI tools will soon perform a reasonable fraction of research-level mathematical tasks with reasonable levels of success, quality, supervision, and cost
- The Leiden Declaration on Artificial Intelligence and Mathematics was published in June 2026 and is backed by the International Mathematical Union as an institutional response framework
- Tao's proposed publication standard requires authors to convincingly demonstrate the ability to give a clear, expert-level talk on their results that is correct and properly attributed
- Tao personally uses AI for literature search, diagram creation, text completion, and slide-to-paper conversion, while prominent mathematicians Gowers and Sarnak acknowledge LLMs' mathematical abilities but identify limits in generating genuinely new ideas
Industry Insight
The mathematics community should proactively establish clear AI-use guidelines before market incentives and benchmark-chasing reshape research culture irreversibly, drawing on the Leiden Declaration as a starting framework. Institutions funding and publishing mathematical research must prioritize depth of understanding over proof abundance, potentially requiring oral defenses or expert-level presentations as publication prerequisites. AI tool developers working in mathematical domains should consider building verification and attribution layers that preserve the "human friction" in proofs—such as intermediate lemmas and revision traces—that makes mathematical exposition genuinely learnable.
Disclaimer: The above content is generated by AI and is for reference only.