Artificial mathematics

Matt Kerr, Professor and Director of Graduate Studies, Department of Mathematics at Washington University in St. Louis

The unofficial logo of my tech-magnet high school was a computerized version of Hokusai’s massive blue wave, and with the OpenAI slop drop1 it finally feels like that wave is poised to flatten the far more creative ecosystem to which I fled. Encountering Strogatz’s essay2 “A Mathematician Reads the AI Future” in its wake, I had hoped for a more vigorous defense of an intellectual tradition which, with philosophy, has always belonged at the nerve center of Arts & Sciences despite being emphatically neither. While he worries that something deeply human could be lost if, in the rush to ride that wave, we let verification replace understanding and hand the process of its development over to machines, the essay ends with a sense of communal resignation.

My initial thought, from the stories Strogatz tells, was that events are driving a subtle but significant wedge between pure and applied mathematics and mathematicians. Applied math is oriented toward well-defined external goals, like improving imaging or data compression algorithms, and has the most obvious and immediate value to society. To its practitioners, AI simply feels like another tool, like a microscope or telescope. While it may change the stage of reasoning at which the computer becomes involved in their work, this is nothing to be “ashamed” of, as I heard a department chair say at a recent AMS discussion of AI.

If that’s true, then perhaps applied mathematicians, and even some pure mathematicians who are oriented toward knowing rather than understanding, don’t share the sense of loss that pure mathematicians are facing right now. Strogatz avers that despite his sympathy with pure math, he believes that math is a science, and that there are external “truths” that exist without humans, and which we discover. (In fact, he gets a pure mathematician to say this.) Hardy also makes the latter error in his otherwise moving “Apology”.3 Aside from the fact that science doesn’t deal with truths, it deals with theories, there are two problems with externalizing math from humans in this way.

The first is that what mathematicians regard as logical reasoning is merely a form of grammar, or a grammar of form, constantly reshaped with definitions and conventions. It helps us to clarify and distill our ideas and frame them in communicable shapes, in much the same way that a poet or composer uses sonnets or sonata form or polytonal counterpoint as a clarifying fire and vessel for their own. There are no fewer possibilities for a work of mathematics than for a verse play or an opera; but just as with monkeys and Shakespeare, it’s humans who need to decide what to write and what is beautiful or important. In my mind, pure math is not about what correctly aligns with something which is “out there”, which I suppose was an acceptable view in the so-called Enlightenment when people didn’t know any better, but what constitutes a strong poem with the potential to move and influence humans.

In short, the representationalist view of mathematics is anti-humanistic, and also (in my view) incorrect. If we hand our subject over to computers, they will generate everything — and thus nothing, as in the “pleroma” of Jung’s Sermons to the Dead — according to whatever state of grammar was programmed into Lean as the Project 2025 of a misguided gang of mathematical technocrats. (In fact, it has a little bit in common with the religious anti-humanist fervor, reflected in the actual Project 2025, that the best thing about us is what exists before we are even born and become human, but replacing innocence with infallibility.)

The second error, corollary to the first, is to deny the centrality of human effort and cognition to pure math. The entire point of the subject is the human understanding (and communication) of ideas and the “struggle and joy of discovering them ourselves” to quote Strogatz, though again I’d dispute the use of the word “discovering” here even if that is what it can feel like. (Heard at a conference — speaker: “I feel like Columbus”; audience member: “Wow, that’s modest.”) A better term would be “creating” (so: God rather than Columbus; also not modest).

Each philosopher in history has a unique feeling, a sort of idea cluster, associated with them, which was distilled from a lifetime spent struggling in this way, simplifying and consolidating a worldview according to their very human discernment, experience, vision and taste. This is not about the small collection of final results, but a manner of contemplation, evident throughout voluminous work, which brings transcendent perception in a way that is recognizable by other humans experiencing that work. Likewise, the exact final results are so unimportant for us in mathematics that we emulate the philosophical naming tradition, attaching to famous mathematicians a theorem or theory or object which captures their life’s interests even if it was not proved or defined in that form by them.

It certainly isn’t impossible to imagine a coalescence of our community around more human-centric norms: that pursuing one’s interests without AI remains in the interest of mathematical progress; and that its use for computations and literature searches that increase our cognitive load is also acceptable. But if our community pressures itself, in the name of scientific progress, to hand the slowest and most inconvenient part over to AI, I expect we lose the whole point of the endeavor. With the combined legacies of past struggles frozen in place, and mathematical grammar with them, the subject will begin to resemble Middle Earth at the end of the Third Age, with its inhabitants sailing through the Gates of Argonath and unable to comprehend their construction or the civilization that could build such things. Our monuments will stand, but as ghosts; and while their fellow citizens look for a time machine at the local Costco, someone still gets the computers to churn out papers on Hexagon4 with “no human understanding”.

I’ve restricted myself here to the question of using AI to do math, not to do things I personally think are legitimate — like running administrative reports, latexing handwritten notes, “vibe-coding” (ugh) tikz pictures, experimenting with examples — and things I do not think are acceptable — like writing a paper or a letter of recommendation, which represents our final attempt to communicate to other humans about an idea or about another human. If we don’t have the heart or the energy to do it ourselves, maybe on that point we should remain silent. Perhaps a necessary reaction to ChatGPT’s keyboard monkey explosion is that we (at least, we who have tenure) stop writing as if Stakhanovite productivity is the point and only write what is essential. This isn’t a promise that every paper I write from this point on will be important — there is plenty of “socially necessary” writing — but maybe the community as a whole can try to discuss what counts as beautiful exposition and recenter its writing around that.

In the background of Hokusai’s painting stands a snow-capped mountain, for this is only one of his famous 36 views of Mt. Fuji. Likewise, this will be one episode in the ongoing development of mathematics, which while difficult, will ultimately prove that what we do is not a game which can be solved. At its best, our subject is deeply social but not competitive, rigorous but not scientific, full of endless variety and individuation achieved through scaling the mind’s alpine wilderness. It is essential enough for society and for the human spirit that it will endure the present wave, and students who love it should not give up their dream of becoming a mathematician.

A coffee mug decorated with Hokusai’s Great Wave, standing in front of tall stacks of papers beside a notebook page of handwritten mathematics.
Photo supplied by Matt Kerr.
  1. The unofficial name given to OpenAI’s release of over 700 manuscripts on GitHub, none of them peer-reviewed, most of them not formalized, and several retracted since the release. ↩︎
  2. S. Strogatz, The Atlantic, October 25, https://www.theatlantic.com/ideas/2026/10/ai-math-human-future/689133/. I have long admired Strogatz for his outreach on behalf of the mathematics community. ↩︎
  3. Hardy, G. H. A Mathematician’s Apology. Cambridge University Press, 2012. First published 1940. ↩︎
  4. A new human-moderated platform for the online publication of preprints, modeled on the arXiv but designed to collect and organize work produced in concert with LLMs. https://hexagonmath.org ↩︎

Received 9 November 2026.

More by →

One response to “Artificial mathematics”

  1. Scott Guthery Avatar
    Scott Guthery

    Cobble up a bunch of definitions and axioms, create an imaginary mathematical world, explore the world and make exhilarating discoveries. The stacks in the basement of your local research library are full of hundreds of bound issues of defunct journals recording peregrinations in these now-abandoned fantasy landscapes. Academic mathematicians have been acting like your “misguided gang of mathematical technocrats” for centuries. It’s all great fun, but it’s little more than finding the Easter eggs you hid yourself; the exhilaration is self-induced and, consequently, rather cheap. If a landscape dries up (string theory, descriptive geometry, nomography, lattice theory, whatever), we just move along, cobble up another set of definitions and axioms, start another journal, and carry on. Abstract mathematics has become just a giant series of Sudoku games whose only purpose, as far as I can tell, is to get people tenure by being good players of the latest version. AI has become a player, and a good one. It isn’t going to get tenure, but it is showing how silly the game was all along by mowing through the outstanding problems. We have the Olympics and the Paralympics. Maybe we should have two branches of mathematics: with and without AI.

Comments are moderated. Read our comment policy.

Add to the discussion

New posts by email.

Prefer a feed reader? Subscribe by RSS.

Also on Mathstodon.

Latest comments across the site.

  1. Cobble up a bunch of definitions and axioms, create an imaginary mathematical world, explore the world and make exhilarating discoveries....

  2. Statement of tool use: This comment is powered by Lavazza[tm] espresso and was spell-checked by ispell. Other comments have already...

  3. >”Embrace AI”This is a very questionable proposition for a whole range of reasons – if it weren’t this forum would...