How to Formalize All of Mathematics
xenaproject.wordpress.com
October 16, 2026
12 min read
47/100
Summary
A group of mathematicians and formalization researchers argues that AI can help formalize most of mathematics within three years, but only if the resulting Lean libraries are reusable, reviewed, documented, and governed by mathematical communities. They warn that rapidly generated code has limited value when nobody can understand, extend, or trust it. The authors cite an AI-generated formalization of Maryna Viazovska’s sphere-packing work that its original team discarded because cleanup was harder than rebuilding it, and say Anthropic’s 13-million-line, 11-day formal verification of Fermat’s Last Theorem is unlikely to replace a more modern formalization. They support coordinated efforts built around Mathlib, an open-source Lean library with 65 reviewers and 950 contributors that received the 2026 Jean-Pierre Demailly Prize for Open Science in Mathematics. Tau Ceti uses AI-written contributions and AI adversarial review under expert-curated roadmaps; it has grown beyond Mathlib’s 3 million lines of code and includes results such as Riemann–Roch and Chebotarev density. Formal Frontier uses AI-agent systems to turn source texts into interconnected libraries, with 41 released libraries totaling about 200,000 lines. The authors call for open-source interoperability, public review, avoidance of duplicated results, links to literature, sustained compatibility with Lean and Mathlib, and mathematician control over infrastructure and resources.
Key Takeaways
- The authors estimate that most mathematics could be formalized for practical use by mathematicians within three years, potentially sooner, if high-quality libraries and supporting infrastructure are developed.
- Mathlib is an open-source, community-curated Lean library with 65 reviewers and 950 contributors, and it received the 2026 Jean-Pierre Demailly Prize for Open Science in Mathematics.
- Tau Ceti uses AI systems to write and adversarially review Lean code under mathematician-curated roadmaps; it is slightly larger than Mathlib’s 3 million lines of code.
- Formal Frontier has released 41 AI-assisted formal libraries totaling roughly 200,000 lines of code and is designed to scale to thousands of textbooks.
- The authors argue that formalizations must be reusable, documented, publicly reviewed, interoperable, and connected to existing mathematical literature rather than treated solely as correctness certificates.
What the discussion said
The liveliest AI discussion was not about whether every theorem can be encoded, but about what AI changes once formal mathematics becomes a usable substrate. Several readers see a powerful loop: proof assistants and formal libraries make mathematical claims machine-checkable, while AI can lower the barrier for newcomers by explaining unfamiliar material, generating visual walkthroughs, and surfacing techniques from distant fields. That prospect feels less like replacing mathematicians than giving more people a route into serious mathematics. The caution was equally concrete. Readers who have tried learning difficult subjects with language models warn that confident errors are uniquely damaging in mathematics: a plausible but false explanation can seed a whole tower of misconceptions before the learner notices. Others report that AI is productive for drafts, exposition, and connecting known methods, yet still stalls precisely at the novel bottleneck where research needs an original idea. The preferred role is therefore a partner constrained by formal verification and human judgment, not an autonomous mathematician whose opaque proofs merely happen to check. A few commenters also worry that AI could encourage inscrutable, machine-oriented mathematical branches that humans cannot effectively build on. The thread broadly treats formalization as a major human achievement, but not as evidence that mathematical understanding has become trivial or that current AI can safely teach or discover without supervision.
Where opinion split
The sharp dispute is whether AI is a credible teacher and research collaborator in mathematics. Optimists argue that it can create tailored explanations, visualizations, and rapid connections to techniques a researcher may not know; skeptics say its fluent errors and weak model of a learner make it hazardous, while its output collapses at genuinely new research steps.
Community Sentiment
Positives
- Formal proof libraries paired with AI could turn advanced mathematics from a closed guild into something motivated learners can explore with interactive explanations and community support.
- AI-generated visual, step-by-step treatments of technical material can make concepts such as model-training pipelines far more approachable, provided the learner still does the intellectual work.
- For working mathematicians, AI can rapidly bring unfamiliar but established techniques into play, making it useful for drafting and escaping narrow disciplinary silos.
- Machine-checked formalization offers a valuable guardrail: AI-assisted mathematical work need not rest solely on persuasive prose when proofs can be verified mechanically.
Concerns
- A language model can deliver a crucial mathematical falsehood with total confidence, leaving learners to build elaborate mental models on a broken premise before discovering the damage.
- Current AI often writes opaque, poorly sequenced explanations that assume undefined concepts and fail to adapt to what the student actually understands.
- Researchers report that AI helps around well-mapped territory but freezes at the decisive novel insight, so high-volume drafts do not equal research progress.
- There is apprehension that AI may generate formally valid but humanly unreadable mathematics, creating isolated machine-facing knowledge that others cannot meaningfully extend.
