Proofs and Prompts

Notes from the common room

The Association for Human Mathematics

AHM communications group

This is a short announcement about the birth of the Association for Human Mathematics (AHM), an organization run collectively by mathematicians at all career stages. Its main goal is to protect mathematics and its practitioners from the jaws of AI companies, whose actions are increasingly destructive to our discipline, to the environment, and to society at large. To achieve that goal, we must critically analyze the use of AI models in all aspects of our work: doing research, writing it up, submitting it for publication, and presenting it, but also teaching, mentoring, communicating, and much more. Members of the AHM will work to create policies addressing the issues faced by our journals, departments, and professional organizations.

Anyone is free to join the AHM. There are no membership fees or dues, and participation in the working groups is voluntary. Members are asked to publicly sign onto two commitments: not to work for AI companies, and not to publish texts generated by AI models. There is an optional third public commitment: to abstain from AI usage in core mathematical activities.

The Association is organizing working groups devoted to making mathematics a better place for humans. Three central working group topics are teaching, hiring practices, and journal policies. We are here to help and you should feel free to reach out if you want to know what we are doing.

If you have reservations or doubts about the way things are going, you should join. If you have already spent some time thinking about how to improve the current system and want to share your ideas, you should join. If you have discussed these issues in your department and are already taking steps in a positive direction, you should join. If the journal you edit is overrun with AI-generated submissions and you are looking for solutions, you should join.

Finally, if you think this is all overblown and that we will go back to normal in no time, or that your field will not be affected, or that corporate AI will somehow level the playing field and ultimately become a democratizing force in mathematics —\text{\textemdash} well, we urge you to look around, talk to your colleagues, and think again.

Received 11 September 2026.

Filed under
·
More by →

New to Proofs and Prompts? Start here →

15 responses to “The Association for Human Mathematics”

  1. Pablo Shmerkin Avatar
    Pablo Shmerkin

    I applaud this initiative! My own current practices seem to narrowly make me ineligible (I use AI to generate proofs of technical lemmas, etc, so I don’t meet criterion (2)). Nevertheless, I hope this association will play a positive role in ensuring that human understanding of mathematics continues to be valued in the community, including when publishing, looking for jobs, applying for funding, etc.

    Reply
    1. JM Avatar
      JM

      I might be misunderstanding but doesn’t (2) just says you can’t directly put AI generated content into your works? As long as you digest the AI output and rewrite it in relation to your own understanding you should qualify? I think you only wouldn’t qualify for (3) which is optional.

      Reply
      1. Pablo Shmerkin Avatar
        Pablo Shmerkin

        I use the AI generated content as a first draft, which I then check and rewrite substantially, also with AI help (to be clear, this is all based on human ideas; I’m strongly against prompting for full solutions that bypass human understanding). Probably (2) should be clarified as I think many mathematicians may be working right at the boundary of what it is meant to allow.

        Reply
        1. JM Avatar
          JM

          Ahh that does seem like a fine line. I guess it does make sense to clarify (2) a bit.

          Reply
        2. Douglas Silva Avatar
          Douglas Silva

          You can use a LLM open source that don’t work with datacenters as the chineses for this specific task.

          Reply
  2. YW Avatar
    YW

    Can you define “core mathematical activities”? Do search for literature, learn some math, polish the paper and check for math typos count?

    Reply
    1. AHM communications group Avatar
      AHM communications group

      The details of what constitutes a core mathematical activity are ultimately left to the individual. We believe that generating mathematical ideas, selecting problems, exploring examples, and finding proofs count as core mathematical activities. Members of our AI-free caucus agree to avoid the use of AI tools in research, and should follow their own conscience in adhering to this principle.

      Reply
  3. Jerome Buzzi Avatar
    Jerome Buzzi

    I agree on the need to confront the many questions raised by IA and more specifically the (human, too human) attempts to exploit our field.

    However joining “working committees” seems premature: one will work to defend what exactly?

    Is it a specific idea of mathematics? I expect that colleagues will honestly disagree on the meaning of good (or human) mathematics.

    Is it the majority of the mathematical community (then we should work inside national and international mathematical societies)?

    Reply
    1. AHM communications group Avatar
      AHM communications group

      We believe it is urgent to start organizing against the more immediate threats to the discipline like sloppification and unsustainable extraction of the field. There are also other issues related to academic life in general, such as the irresponsible push for AI use by university administrators as an excuse for austerity/defunding. Although individual mathematicians can disagree on the finer points of what constitutes good mathematics, we believe we’ve reached a point where there is sufficient discontent with the way AI is being used in mathematics (sometimes by non-mathematicians) and we are organizing to provide some solutions.

      Reply
  4. Jonathan Noel Avatar
    Jonathan Noel

    The first pledge in joining the group forbids from “providing technical labor, mathematical knowledge, consultation, or publicity.” That seems very different from “working for” them.

    If I was to use an LLM toward proving a theorem and then disclose usage of a tool during a public talk, does this count as publicity?

    If I were to engage in conversation with someone at an AI company about how they can improve their tool’s value to the mathematical community (either making it better at assisting in proving and discovering new math, or in doing so in a way that is not harmful for the community), I assume that this would count as consulting?

    I think it is unlikely that agreeing en masse to make sure that you cannot impact the technology is not going to be effective.

    It is a little hard to see how it will be possible to generate proposals that will be supported by mathematicians broadly if the pledges imply that the members will have a fairly narrow set of views.

    Reply
    1. Jonathan Noel Avatar
      Jonathan Noel

      Oops, I mean: I think it is unlikely that agreeing en masse to make sure that you cannot impact the technology is going to be effective.

      Reply
  5. Jiuzu Avatar
    Jiuzu

    I think AI referee report is more accurate than human mathematicians. It can point out many errors and gaps that could be missed by a normal referee. I think it definitely makes sense to incorporate the Ai review at least on the correctness part.

    Reply
  6. Uri Bader Avatar
    Uri Bader

    It is not unlikely that in the not so remote future formal verification will be the standard for new math results, and this will be typically managed by AI. This will be a wonderful contribution! While this use doesn’t contradict items 1 and 2, nor even 3 maybe, I feel that it doesn’t concur with the commitment’s spirit.

    Reply
  7. Andy Sukowski-Bang Avatar
    Andy Sukowski-Bang

    What exactly is meant by “commercial AI companies”? Primarily major LLM companies like Anthropic and OpenAI? For example, could I work at Google but not on Gemini specifically? Could I develop vision transformers for medical imaging, or work on reinforcement learning for non-LLM applications?

    Reply
  8. AI 수학 거부 서약에 1천 명, 수학자들의 ‘채식’과 ‘비건’ 사이 – HastyEarth

    […] 이야기· Association for Human Mathematics: Members· Association for Human Mathematics: FAQ· Proofs and Prompts: The Association for Human Mathematics (2026년 9월 12일)· Terence Tao 블로그: AHM Statement on OpenAI’s October 6 Release (2026년 10월 7일)· […]

    Reply

Comments are moderated. Read our comment policy.

Add to the discussion

Subscribe

New posts by email.

Prefer a feed reader? Subscribe by RSS.

Also on Mathstodon.

Join the discussion

Latest comments across the site.

  2. Yeah I was just really hoping that in the infinite sea of math problems they do actually exist... Feels important...

Read next