Recently, Stanford hosted “The Future of Mathematics Symposium”. Unfortunately, I was not able to attend in person, but I watched some of the talks and panel discussions on YouTube. Based on what I saw, a better title for the symposium would have been “The Funeral of Mathematics”.
Let me summarize most talks so you don’t have to watch them:1 The speaker will say something about the strides AI has made in math capabilities in the last few years. Maybe they show a line with positive slope to illustrate their point. Then they assert that it’s probably not going to stop here. The obvious question that everyone is thinking, “what then?”, is left unanswered.
1 I’m being intentionally provocative. The talks weren’t bad, and may still be worth watching on YouTube if you’re interested.
In between the questions being asked and the noticeable discomfort of those attending, you can sense a feeling of dread in the air. The takeaway is that mathematicians are being made obsolete, as I type, by machines made from silicon and metal. This fact is quickly becoming common knowledge among professional mathematicians and those aspiring to become one. These AI developments lead to some existential questions for mathematicians.
Actual mathematicians tend to speak carefully about these questions and give room for nuance. But for the sake of entertainment, let me caricature their views into a few broad categories:
- The Denier: Believes that AI will never be able to replicate what a human mathematician can do. Avoids thinking about AI at all, if possible.
- The Coper: Says that the role of mathematicians will change, but that humans will always be needed in the loop for task X that AI cannot yet do.
- The Doomer: Views AI as an existential threat to human mathematicians. Believes that their existence becomes meaningless if AI surpasses them in raw mathematical skill.
- The Pragmatist: Has laser focus on AI capabilities at the present moment and how that affects current workflows. Refuses to speculate too much about the future.
- The Enthusiast: Is excited about the possibility of AI to help resolve major open questions and speed up math research. Somehow less interested in talking about the impact on professional and aspiring mathematicians. Tends to be emeritus-age or work outside of academia.
Common to the denier and the coper is that they refuse to look at the rate of progress. Because the rate of progress has been legitimately crazy to see: 2 years ago, in May of 2024, LLMs were extremely unreliable for any non-trivial math.2 When the first reasoning model, OpenAI o1-preview, released in September 2024, it was a big step up, but still not useful for hard math problems. Less than a year later, in summer of 2025, both Google DeepMind and OpenAI announced that their best language models had achieved gold medal performance at IMO. In Fall of 2025, proofs of simple open problems started being announced, most notably AI-assisted proofs of several Erdős problems. Around a week ago as I type, in May of 2026, one of the Top 10 Erdős Problems, the unit distance conjecture, was disproved autonomously by an internal general model at OpenAI.3
2 When I submitted my PhD thesis in June 2024, the only use case I had found for LLMs in my work was to aid in generating figures.
3 This proof also inspired another paper, released around a week later, disproving Erdős’ sum-product conjecture (another problem from Bloom’s top 10 list) over the reals. The original integer version of the conjecture remains open.
The denier and the coper might say that IMO problems and Erdős problems are uniquely suited for the strengths of AI. But even if true, this is still the beginning and AI will keep improving.
For the doomer, that is an existential threat. They see the rate of change, and the projected outcome is incompatible with their value system.
The pragmatist and the enthusiast have a more positive relationship to the change that is happening, comparatively speaking. But even they don’t have answers to the most difficult question of all: What is the purpose of mathematicians when AI supersedes humans in skill across all relevant tasks? Which I consider likely to happen in the next 20 years (to use a conservative estimate).
Some people are definitely thinking deeply about these issues and trying to come up with answers. The best take I’ve seen yet is David Bessis’s thoughtful piece The Fall of the Theorem Economy, where he reflects on the ongoing changes and how this affects mathematicians as people and mathematics as an institutional practice. He points out some uncomfortable realities that most people have yet to accept:
If theorem-proving remains their only official currency, all research mathematicians run the risk of becoming demonetized in the course of the next few years.
What most people don’t yet acknowledge is that mathematics, as it has been practiced for thousands of years, may be on its deathbed. Perhaps someone should organize a “Funeral of Mathematics Symposium”, where we can pay our respects and acknowledge what will be lost in the AI-dominated future (assuming the trend continues). Here are some things we might grieve over:
- We lose our institutional value system, for better or worse, which rewards technical skill and proofs of hard theorems. Hence the title of Bessis’s essay linked above.
- The romantic image of a mathematician at the frontiers of knowledge, taking down the hardest problems with only their mind and a chalkboard, ceases to be accurate. A mathematician may still work in this way to gain insight and understanding. But any problem that can be solved in this way becomes easy for AI.
- In fact, humans become mostly irrelevant to solving the biggest problems in math. Proving hard theorems will be bottlenecked by available compute, and not human genius. These problems will get solved by costly AI runs that use lots of compute.
- Human mathematics ceases to be seen as economically useful. This affects the number of math jobs, PhD positions, and the careers of many people across the world, both in research and in teaching. It makes a job market that is already difficult even worse.
- The apprenticeship model of mathematics training changes. Currently, mathematicians are trained in PhD programs which are apprenticeships, more or less, under the guidance of their advisor.
Some of this is bigger than mathematics. Every professional community will have to face some uncomfortable questions. What’s the point of funding a person if they don’t contribute much economic value? What’s the point of paying a person to do research or to teach if those tasks are better done by machines? And why spend resources on teaching people advanced skills when AI will be better than them anyway?
I don’t have the answers. But besides the BIG question of how professionals make a living in this future, I do have an opinion about how mathematics can still be a meaningful practice in the future.
My personal view is that the practice of doing mathematics can still be meaningful even if machines are better at it than we are. In addition, the communal practice of maintaining a culture of mathematics can also be meaningful, even if the goals of the community shift from utilitarian theorem proving to one of stewarding human understanding. Let me elaborate on both points.
The disruption we are anticipating is not unprecedented. One example that comes to mind is the invention of photography, which changed the nature of drawing and painting. After photography, painting could no longer be justified as a way to preserve images of the world. This affected the economics supporting painters, but also what the purpose of painting actually was. People didn’t stop painting because they enjoyed it and found it meaningful, and in fact they discovered some interesting new artistic directions like impressionism. Even today, as AI can generate images that replicate specific painting styles, many people still paint. In a similar way, I imagine that people will still have many reasons to be interested in doing mathematics in the future.
When it comes to the communal practice of maintaining a mathematics culture, I think this can be even more meaningful than the individual purpose. The point I’m making here is similar to things Tao and others have said, in that human mathematicians should be stewards of the human understanding of mathematics. It is also aligned with Bessis’s main thesis, citing Bill Thurston:
The product of mathematics is clarity and understanding. Not theorems, by themselves.
As I explained in my previous blog post about The Future of Knowledge, (human) knowledge exists on several levels, both on the individual but also on the level of communities. To maintain and develop clarity and understanding of mathematics on a community level is therefore an end in itself.
As human beings who are interested in mathematics, the math community is our community and we care about it, in the same way that we care about our family and our city. We don’t need a metaphysical argument to justify why we care. AI may supersede humans as efficient generators of mathematics, but that’s besides the point. I’m not an AI, I’m a homo sapiens, and I like to be around other members of my species with similar interests.
I’ve tried my best to be honest about the many downsides of AI in math in this essay. But we should also remember that there are upsides. With AI, many of the largest mysteries of mathematics may get resolved, and in the process, we may discover even deeper ones. Math knowledge may be democratized, as anyone interested can learn by talking to a chatbot, which is much easier than attending a university (although I still think doing it in a community with other people has its own intrinsic value).
So as a fan of math, I’m looking forward to seeing the next chapter of mathematics, which promises to be a time rich in discovery. But as a human being, I’m even more excited to share this thrill with my fellow humans.