On the impact of AI on the mathematical practice

Manuel Rivera, Associate Professor at Purdue University

I’ve been recently following some interesting debates about the future of mathematics in the age of AI. I think many discussions about how AI is “disrupting” or “changing” (pure) mathematics tend to conflate, or at least fail to distinguish between, three interlocked interpretations of mathematics:

(1) Mathematics itself, as a body or domain of ideas and concepts living somewhere at the intersection of knowledge and art.

(2) Mathematics as a vocation, as the human practice of acquiring knowledge, understanding, and clarity about structure, guided by a sense of aesthetics.

(3) Mathematics as a profession, as the academic practice that exists in institutions and systems, and includes universities, publications, teaching, recognition, competition, etc.

The first is very hard to understand or define, and I lack the proper philosophical foundations to discuss it eloquently. I suspect it is also importantly different from science and perhaps close to something like theology.

The second is the human practice through which we access the first. By “mathematics as a vocation” I mean doing mathematics as a calling or way of life, rather than as an occupation. It exists individually and/or in community and, like other artistic endeavors, is often in tension with political and societal structures. This kind of practice is utopian: there is a sense of authorship without ownership, a particular sharing economy, and it is not bound to market forces. For many of us, participating in this vocational practice is what being a mathematician means; it is at the core of our identity. Figures like Poincaré, Grothendieck, Erdős, Gödel, and Thurston, to name a few, exemplify this practice.

The third is easier to explain. It depends on politics, systems, markets, biases, trends, diplomas, funding, prizes, etc. Teaching mathematics as a service to science and engineering students, as well as training PhD students, also belongs to this domain. It is what many of us choose as a job because, so far, it is the most comfortable way we can get to (2) without renouncing the benefits of societal structures.

I don’t think it makes sense to discuss AI “disrupting” (1), because technology can only disrupt the practice of mathematics, not mathematics itself.

AI will certainly disrupt (3), and I have some concerns about the direction the professional practice, already quite imperfect, may take. For instance, I think the AI industry might push (3) further away from (2), because the dependence of (3) on the market may become heavier and take a different form. Even if the technology itself is a useful tool (modulo its ethical implications) that, if used correctly, can enhance communication and writing, find counterexamples, and help navigate the literature, there is also a high price to pay: enormous amounts of nonsense and slop, together with widespread misuse by students and early-career mathematicians. But again, the economic and institutional forces surrounding AI – the industry – may alter the profession (and society in general) in ways that are much more consequential.

While I do worry about how (3) is going to change, and about the broader societal and ethical implications of AI, I don’t think (2), the practice of understanding structure through creative means guided by aesthetics rather than market forces, will fundamentally change. Some people might argue that (2) is healthier without tools like AI; others suggest we should somehow “own” the technology and incorporate it into our vocational practice. I don’t know. In any case, for many of us, the incentives for devoting our lives to mathematics do not come from (3), the domain where most changes are expected. This vocational practice is where many of us are centered, and there is a community of friends, not colleagues, built around it that I believe no industry can destroy.


Received 23 August 2026.

4 responses to “On the impact of AI on the mathematical practice”

  1. a Avatar
    a

    One of the incentives (not discussed in the article) for doing math is to attach our names to our results. It seems this will not be possible in this new era. This may destroy practice (2) as well.

    1. Manuel Rivera Avatar
      Manuel Rivera

      I don’t consider that as a major incentive in the vocational practice and community only in the profesional context… and it is briefly mentioned in the article when alluding to a sense of authorship without ownership.

      1. a Avatar
        a

        I think “attaching one’s name to a theorem” carries the meaning of “being permanently connected” (so gives some kind of immortality to the mathematician). I am not sure that “authorship” reflects this meaning well.

  2. Novum Organum Avatar
    Novum Organum

    I think this is the start of a golden era for AI-amplified math, and I’d push a little further on your (3): I think the risk you name — the industry pulling (3) away from (2) — is real, but I’d argue the disruption is more likely to realign (3) toward (2) than to push it away. Look at your own list of what the tools usefully do: “enhance communication and writing, find counterexamples, help navigate the literature.” Add verification, formalization, and routine lemmas — things the models already do superhumanly — and that union is, uncomfortably, most of the daily work most working mathematicians actually do. The part of the job that isn’t yet superhuman is shrinking, and it’s precisely the part (2) has always valued most: problem selection, judgment, understanding, teaching. So the restructuring of (3) is less a drift away from (2) than a forced return to what (2) already prizes. The profession just hasn’t admitted the arithmetic.

    The Mathathon is the clearest instance of this. It isn’t the industry colonizing mathematics; it’s the youngest cohort running a cheap experiment on exactly the transition you’re describing. The open letter asked for its suspension; the outcome was OpenAI pulling its sponsorship and the event still running on October 30. The profession scared off the money but not the experiment — and the norms they were genuinely worried about (disclosed logs, verification before publication, judging by understanding rather than by solved-problem count) remain uncodified. That’s the difference between blocking and steering.

    The template is well-worn. Writing, art, and software engineering all went through the same sequence: block, misuse, mastery, codification. Your “widespread misuse by students and early-career mathematicians” is the second stage, not the terminal one — it’s the tuition every practitioner of those fields paid. They got through it by accepting that digital intelligence is already superhuman over what used to be their work — the routine, the medium — and redefining the craft around the remainder. Mathematicians are at stage one. The acceptance phase is coming, and it is cheaper to lead it than to be caught in it: codify the new practices now, so (3) reorganizes around selection, understanding, and judgment instead of around the industry’s narrative of who proved what faster.

    I suspect you’re right that (2) itself is safe — the person who sits with a problem for a year is still doing (2), and a community of friends rather than colleagues is hard for an industry to take. The risk is (3): that the profession doesn’t steer in time, and (2) ends up as something people do on the side, outside a profession that no longer funds it, teaches it, or certifies it. Codification is how (3) keeps that mandate.

Add to the discussion

New posts by email.

Prefer a feed reader? Subscribe by RSS.

Discover more from Proofs and Prompts

Subscribe now to keep reading and get access to the full archive.

Continue reading