arxiv: 2603.03684 · v3 · submitted 2026-03-04 · 🧮 math.HO · cs.AI

Recognition: no theorem link

Mathematicians in the age of AI

Jeremy Avigad

Authors on Pith no claims yet

Pith reviewed 2026-05-15 17:07 UTC · model grok-4.3

classification 🧮 math.HO cs.AI

keywords AI theorem provingmathematical practiceformal mathematicsinformal reasoningAI in researchproof generationmathematicians adaptation

0 comments

The pith

AI can prove research-level mathematical theorems formally and informally.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that recent AI progress enables proving theorems at the level of current mathematical research. It calls on mathematicians to learn about these tools and prepare for changes in how proofs are discovered and verified. Readers should care because this could speed up mathematical progress while changing the skills that define expertise in the field. The essay reviews existing examples and suggests ways to engage constructively with the technology.

Core claim

Recent developments show that AI can prove research-level theorems in mathematics, both formally and informally. This essay urges mathematicians to stay up-to-date with the technology, to consider the ways it will disrupt mathematical practice, and to respond appropriately to the challenges and opportunities we now face.

What carries the argument

The capability of AI to generate proofs at research level, which drives the argument for updating mathematical practice.

If this is right

Mathematicians should monitor and adopt AI tools for theorem proving.
Mathematical practice will likely incorporate more machine assistance in both formal and informal reasoning.
New challenges will emerge around verifying AI-generated results and maintaining standards of rigor.
Opportunities will open for tackling harder problems through combined human and AI efforts.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

AI assistance could lead to faster resolution of long-standing conjectures.
Mathematical education might shift to emphasize interaction with AI systems.
Questions about the nature of mathematical understanding may arise as machines handle more formal work.

Load-bearing premise

That AI has advanced to a stage where it can meaningfully affect how mathematicians carry out their research.

What would settle it

A controlled comparison showing that mathematicians using only traditional methods match or exceed the output of those using AI tools in solving open problems.

read the original abstract

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Avigad's essay assumes AI already proves research-level theorems and urges adaptation, but the claim sits without examples or citations.

read the letter

The central message is that AI systems can now handle research-level theorems both formally and informally, so mathematicians should learn the tools, anticipate shifts in daily practice, and decide how to respond. That premise drives the whole piece, yet the text supplies no named theorem, no specific system, and no reference to back it up. The stress-test note is right on this point: the factual starting point is asserted rather than shown. Without at least one concrete case, the rest of the advice floats on an unverified foundation. This is an opinion essay, not a technical report, so it does not need to prove new results, but it does need to ground its key observation. On the credit side, the recommendations themselves are sensible and measured. Avigad draws on his own work in formal verification to suggest practical steps: keep up with current capabilities, think through effects on proof exploration and checking, and weigh both opportunities and risks to the culture of mathematics. The tone stays calm and avoids hype, which makes the suggestions easier to take seriously. The soft spot is limited to that opening gap; once past the unsupported claim, the discussion of practice changes is straightforward and proportionate. The essay is aimed at working mathematicians who have not yet spent much time with AI tools but want a short, accessible overview of the issues. It would suit a department seminar or a reading group for people new to the topic, though specialists already following the literature will not find new data or arguments. It deserves a serious referee for an opinion or essay section in a venue such as the Notices or Bulletin of the AMS. A referee could ask for one or two concrete examples to anchor the premise without altering the overall message, and the questions raised are worth public discussion even if the evidence needs tightening.

Referee Report

2 major / 1 minor

Summary. The manuscript asserts that recent AI developments enable proving research-level mathematical theorems both formally and informally, and urges mathematicians to remain current with the technology, anticipate disruptions to traditional practice, and address resulting challenges and opportunities.

Significance. If substantiated, the essay addresses a timely intersection of AI and mathematics by advocating adaptation in research and education practices. Its value as an advisory piece is reduced by the absence of concrete evidence, limiting its ability to ground calls for change in verifiable developments.

major comments (2)

[Abstract] Abstract: The central premise that 'Recent developments show that AI can prove research-level theorems in mathematics, both formally and informally' is stated without any specific theorem examples, AI system names, proof methods, or citations. This unsupported assertion is load-bearing for the subsequent recommendations on practice disruption.
[Main text] Main text (opening paragraphs): The argument for urgent adaptation relies on the claim of AI achieving research-level results, yet no concrete cases (e.g., formal proofs in Lean or informal theorem generation) are referenced, leaving the weakest assumption—that current capabilities meaningfully disrupt practice—unexamined and unverifiable.

minor comments (1)

[General] The essay would benefit from explicit section headings (e.g., 'Challenges' and 'Opportunities') to organize the advisory content and improve readability for a broad mathematical audience.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their review and for highlighting the need for greater specificity in our essay. We agree that the central claims about AI capabilities would be strengthened by concrete examples and will revise the manuscript to incorporate them while preserving the piece's essay format and forward-looking focus.

read point-by-point responses

Referee: [Abstract] Abstract: The central premise that 'Recent developments show that AI can prove research-level theorems in mathematics, both formally and informally' is stated without any specific theorem examples, AI system names, proof methods, or citations. This unsupported assertion is load-bearing for the subsequent recommendations on practice disruption.

Authors: We accept the point and will revise the abstract to name specific developments. The revised version will reference, for example, DeepMind's AlphaProof achieving silver-medal performance on IMO problems and recent Lean formalizations of research-level results in algebraic geometry and homotopy theory. These additions will anchor the premise without changing the essay's advisory character. revision: yes
Referee: [Main text] Main text (opening paragraphs): The argument for urgent adaptation relies on the claim of AI achieving research-level results, yet no concrete cases (e.g., formal proofs in Lean or informal theorem generation) are referenced, leaving the weakest assumption—that current capabilities meaningfully disrupt practice—unexamined and unverifiable.

Authors: We agree that the opening paragraphs require concrete cases to make the disruption argument verifiable. In revision we will insert brief, cited illustrations of both formal (Lean) and informal (transformer-based) theorem generation at research level, together with a short acknowledgment of current limitations. This will allow readers to evaluate the practical implications directly. revision: yes

Circularity Check

0 steps flagged

No circularity: advisory essay with no derivations or fitted claims

full rationale

The paper is a short opinion essay containing no equations, derivations, predictions, fitted parameters, or mathematical claims that reduce to prior inputs. The opening assertion about AI proving theorems is presented as an observation without any self-referential definition, self-citation chain, or renaming of results. No load-bearing step exists that could be circular by construction. The text is self-contained as advisory commentary and does not invoke uniqueness theorems or ansatzes from prior work by the author.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a non-technical essay containing no formal derivations, fitted parameters, background axioms, or newly postulated entities.

pith-pipeline@v0.9.0 · 5315 in / 936 out tokens · 42579 ms · 2026-05-15T17:07:08.261991+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

A Milestone in Formalization: The Sphere Packing Problem in Dimension 8
math.MG 2026-04 unverdicted novelty 5.0

The sphere packing problem in dimension 8 has been formally verified in the Lean theorem prover, with the final stages completed by the AI model Gauss.