As AI automates parts of pure mathematics research, a look at formalization challenges and why human imagination remains key to deciding which questions to ask
Headlines and History — The headlines keep coming: such and such an AI system has solved such and such a math problem.
Context & Ripple Effects
Mathematics had become a prominent proving ground for AI in 2026: researchers argued that newer reasoning models were making systems more useful in the field, while also treating math performance as a measure of AI progress. That shift toward mathematical evaluation gives formalization—not merely answer generation—greater importance.
The debate had already moved into research practice through tests using unpublished mathematical questions and reflections from mathematicians at the 2026 International Congress of Mathematicians. Wolfram’s essay sharpens the remaining division of labor: machine assistance may broaden the search, but mathematicians retain responsibility for selecting worthwhile questions.
First-order effects
- Mathematicians assessing AI tools must distinguish between generating or checking formalizable steps and choosing the research problems that justify sustained attention.
- AI developers building systems for mathematics face a harder product requirement: support for formalization workflows, not just demonstrations that return answers to isolated problems.
Second-order effects
- Projects such as First Proof gain a stronger rationale for testing AI against unpublished work, where useful assistance must be judged in a research context rather than against known solutions.
- Reasoning-model teams are pushed to compete on reliability within formal mathematical representations, because informal plausibility alone does not settle whether a result can enter mathematical practice.
Third-order effects
- If formalization becomes a larger share of mathematical work, research workflows may separate into machine-assisted verification and exploration on one side, and human-led agenda setting on the other.
- Mathematical benchmarks will increasingly need to measure contribution quality beyond solved-problem counts, including whether systems help researchers pursue meaningful questions.
The trend: AI in mathematics is moving from headline-grabbing problem solutions toward integration with the formal verification and problem-selection practices that define research work.