Math has only one application that I care about: To feed my family. Any other application has largely been to the detriment of us all. Look no further than Hiroshima and Nagasaki to see why matrices were not meant to be multiplied.
Now the one application I care about is being threatened. I am tenure track at a good institution where I teach amazing students and have fantastic coworkers who all fill me with joy. I love my job, and I want to keep it. To keep it, I need to get tenure, but what does research work for tenure mean now? I have asked several people this question. Many don’t understand why this question is being asked, which is a major concern to me. I will explain why I ask. But even those I know who agree that this question must be answered tell me it is not up to them. Instead, it is up to famous people, those blessed by the Gods old and new who can speak math into being and tap dance and smoke a cigar in one long drag.
However, when looking for famous people’s opinions, I find in its place much discussion about philosophical issues. Uniformly I hear the sentiment that mathematics is about understanding. I don’t understand why people are saying this en masse. It is true of course but unhelpful for the problem at hand. The appropriate reaction to the situation is not to dig down deep and think: Why do we really do mathematics research? It’s so we can feed our families. It’s not an interesting question.
This is the obvious flaw of the AHM. They forget that if you have to decide between using AI or letting your family starve, the choice is obvious.
I believe many mathematicians do not understand that this is not an abstract discussion of AI’s capabilities anymore. Let me be concrete. In the last month I have solved so many distinct open problems with GPT Pro personally that I do not know what to do with the material. Under our traditional metrics, I should be publishing as much mathematics at as high of a quality as I can produce as fast as possible. But if I were to do that, I would release 10 papers this semester, which is clearly wrong. What should I be doing instead?
AI is better at proving stuff than us now. There’s no question anymore. We like to joke about it, but it’s not actually bad at explaining stuff either. If you ask GPT even without the pro version to explain a proof step by step, it does a phenomenal job. I know because I’ve done this with many results personally for more than a year now. Focusing on us giving the best explanations is not a solution either. I’ve seen that proposed. I don’t think that will work.
That’s the reality, which is why I ask how research is to be evaluated. AI can produce high quality research infinitely, better than us, at rapid rates, and it is only getting better. We urgently need to make a radical policy decision to account for this. To be clear what I am looking for, here is a candidate policy solution:
The only thing I can imagine valuing mathematicians for is as human beings who work to understand what the AI is producing. The sole question is how to verify that. Writing won’t work as AI can do that better than us and infinitely. Any sort of manufacturing of a product will not work for that same reason. As a result, we should end the journal system completely. The only way is for us to communicate about it with each other live human to human. The structure of the community would then surround research workshops organized around understanding particular topics. Our corporate overlords could fund some of them to get us to parse their breakthroughs. Then participants earn credit by presenting various lemmas and so on. These workshops focus moreso on compute allocation and yield a final product that is a mix of human and AI generation, which is published to the math community at large with no external refereeing as fodder for the models. The actual most relevant skills for the job are interpersonal. Teaching, mentoring, working with colleagues in meetings, and service work are the core of the job and all that really matters now. We should just hire based on the skills relevant for the job like other professions. We can ask for materials that are evidence of that in applications such as letters of recommendation, workshop participation, teaching, organizing events, and outreach activities.
Maybe you dislike this proposal. My point is not that I have the perfect solution. It is that we need to stop wasting time and answer: How are we going to evaluate research output now that our previous model is defunct? I have heard many of us in denial. There is no way for journals to survive, because AI can easily produce submissions of high quality from anyone independent of their skillset. At the very least, I don’t want to write 10 papers this semester or worse yet referee proportionately many papers. We need an alternative. This is not a hypothetical. I need to know the answer, so I know how to get tenure. PhD students and postdocs need to know because they are going on the job market. Faculty need to know because we are hiring people right now. The publication process is broken. We can’t use it anymore in its current form. I gave you an alternative. You don’t like it? I can totally understand that. Propose a better idea, or better yet, implement it at scale. I don’t actually care what the solution is as long as it works. Just tell me what I need to do to feed my family.
Let me end on a positive note. We need to remember that as a community we decide collectively what constitutes mathematical scholarship. OpenAI and the National Science Foundation do not control it. It’s not up to my mom either. I asked her, and she said no. There is no threat to mathematics at all. That’s why this is not the mathocalypse. Math will be fine. This is the math-research-normsocalypse, which is less catchy but easier to address. The only things that need to change here are our evaluation practices. Honestly, I hate the current norms anyway. Publishing makes me want to perish. So let’s pretend we’re back in our first graduate course in real analysis and see what happens when we change the mathematical norms.
Received 12 October 2026.
Add to the discussion