Young mathematicians doubt their future: AI hits first those who haven't yet started doing research

By Business Insider · November 1, 2026.

Three MIT-affiliated students published a survey this week of 268 mathematicians in training (PhD students and postdocs). According to the results, about 76% say they are increasingly unlikely to stay in academic mathematics after completing their degree. Another 76% say they feel less motivated to do research “as a result of AI advances.”

The survey was conducted using a Google Form, with multiple-choice and open-ended questions. The authors themselves caution that it is a snapshot of “a very small window of time.” The sample is limited, so it does not necessarily represent the entire profession. Moreover, the response period closed just after OpenAI published, on October 6, hundreds of solutions to long-standing open problems.

According to Business Insider, the earlier trigger came in September, when OpenAI published solutions to the Millennium Prize problem on Navier-Stokes (the equations that describe the motion of fluids). We have not been able to verify the status of those solutions here or their validation by the community; they should be treated as company claims until there is independent review.

The article also highlights a recurring complaint in the open-ended responses: a lack of academic leadership. Respondents describe institutions as erratic, silent or ineffective. “It would be great if institutes gave guidance instead of platitudes,” one wrote. According to the outlet, OpenAI and Anthropic have together contributed $15 million to a foundation supporting mathematicians, and OpenAI has created an advisory group to evaluate its mathematical work. The students say this has changed little.

**Our take.** What matters most is not the percentage but what it reveals about the transition. Mathematics is the field where AI first showed a clear edge, because proofs can be verified. Young researchers are the first link exposed: they invest years of training in a craft whose perceived value is eroding before they get to practice it. That is the harsh short-term reality, and it should not be sugarcoated.

But the most weighty phrase, “none of the individual work will matter,” conflates two different things: the usefulness of a result and the meaning of producing it. If a system proves a theorem first, the theorem is known; what remains contested is who understands it, who formulates the questions and who decides which ones are worth pursuing. In general, those functions (choosing problems, interpreting, teaching, connecting fields) are the most resilient, much as we see in other sectors where judgment outlasts execution.

There is also a management lesson. The $15 million in funding is a gesture, but the underlying complaint is about direction: students are asking for roadmaps, not donations. Universities and mathematical societies have a concrete task here: redesign doctoral training toward verification, formalization and problem formulation, and state clearly which careers still make sense.

In the long run, our thesis is optimistic: if AI accelerates science, the benefit (more theorems, more applications in medicine, energy and materials) will be enormous, and a generation freed from routine work could devote itself to what it is truly passionate about. But that future will not arrive on its own. In between, there is a generation paying the cost of the transition, and meeting it with platitudes instead of plans is an avoidable failure.