Wolfram Is Building Tools So Humans Can Check AI-Written Math
Stephen Wolfram says work is underway to encode pure mathematics concepts into the Wolfram Language, with an emphasis on letting people verify that an AI-generated proof actually answers the researcher's question.
On September 28, Stephen Wolfram published an essay on where pure mathematics research is headed in the age of AI. Wolfram said work is underway to extend the Wolfram Language, his computational mathematics programming language, so that it can express a wider range of pure mathematics concepts. The idea is to have AI turn a mathematical request into something computable, and then let a human read and check the result.
How Do You Verify Math Written by AI?
Wolfram proposed a workflow in which humans inspect the mathematical expression that the AI produces at an intermediate stage. The AI translates a mathematical request into Wolfram Language, a person reviews that translation and corrects it where needed, and only then is it used for computation or proof. The point is to move beyond researchers simply receiving an AI’s final answer, and to let them examine how the AI rendered the question in the first place.
The core of the extension work lies in expressing abstract mathematical concepts precisely. Wolfram cited concepts such as sheaves, Lie groups, and Clifford algebras as targets. The goal is not just to make each concept computable, but to make its representation readable and checkable by a human. He also wants these representations used when AI translates mathematics into strictly rule-bound form.
Wolfram pointed out that even when an automatically generated proof passes verification, a separate check is still necessary. Whether a proof follows the rules and whether it proves the statement originally intended are two different questions, he argued. If AI-handled mathematical work is to be trusted and used, he stressed, a human has to confirm that the question it started from was captured accurately.
An Idea Floated in 2014 — Where Does It Stand Now?
The notion of expressing pure mathematics broadly in Wolfram Language is not new. Wolfram put forward a plan for an extension covering pure mathematics as a whole back in 2014. In this latest essay, he connects that effort to AI’s mathematical representations and to the human review process. A new point of discussion has been added to an old idea: how to use AI.
Mathematica was first released in 1988, and version 15 of Wolfram Language and Mathematica arrived in June 2026. Wolfram said the pure mathematics extension he described is still a work in progress. The essay is not an announcement of finished features, but a statement of development direction and of his views on how research should be conducted.
So What Is Left for the Researcher?
Wolfram argued that deciding which questions to explore and what goals to pursue is a human role. He said AI can find connections among papers and assist with mathematical work. In doing so, he addressed both the scope of AI assistance and the responsibility for setting research direction.
For readers thinking about using AI in mathematical research, the heart of this proposal is the review step. The goal is to produce representations that a person can read and correct, showing how the AI interpreted the request. What to watch next is how much the extension features now in development actually help with that kind of checking.