AI in Mathematics: What Happens When Machines Become Better Than Mathematicians?
AI in mathematics has reached a point where the discipline can no longer pretend that the rules remain unchanged, because machines are beginning to solve problems that once defined entire human careers.
Philadelphia was supposed to be a place where mathematics celebrated itself. On July 24, at the International Congress of Mathematicians, some of the world’s most accomplished mathematicians gathered to discuss ideas that had taken decades to develop, problems that had resisted generations of researchers, and the strange beauty of proving something that nobody had been able to prove before. Yet beneath the celebration was a growing sense that the rules of the discipline were beginning to change.
When Terence Tao, one of the most influential mathematicians of his generation, took the stage for his keynote lecture, “Mathematics in the Age of AI,” he did not describe artificial intelligence as a distant technology that mathematicians could simply observe. He called the situation a “very confusing mess” and warned that mathematics was facing a crisis in its values and practices.
The reason was becoming impossible to ignore. Only days before Tao’s lecture, Anthropic’s Claude Fable AI had reportedly produced a counterexample to the Jacobian conjecture, one of the major unresolved questions in mathematics. The result appeared in an almost absurdly compact form, an equation short enough to fit into a tweet.
It was not an isolated achievement. ChatGPT had already been credited with solving two other important open problems, and mathematicians were beginning to confront a reality that would have seemed extraordinary only months earlier. A problem that once might have launched a young mathematician toward a major publication, a prestigious academic position, or even a career-defining reputation could now be attacked by a machine.
The disruption goes deeper than speed. According to the developments described by the mathematical community, researchers have used an opaque combination of systems from Anthropic and OpenAI to solve the first Millennium Prize Problem in twenty years. These achievements raise a question that mathematics has rarely had to confront: what happens when producing a correct answer becomes easier than understanding why the answer is important?

AI in Mathematics and the Moment the Discipline Lost Its Distance from Machines
For Tao, that is precisely where the real crisis begins. The question is not simply whether artificial intelligence can prove increasingly complicated statements. The deeper question is what mathematicians believe mathematics is for.
If machines begin producing proofs faster than humans can read them, mathematicians could spend their time checking machine-generated arguments, translating them into human language, or searching for the ideas hidden inside them. But what happens if the machine eventually discovers mathematical truths that no human being can fully understand? Would mathematics still be the same discipline?
Andrew Sutherland of MIT has suggested that mathematicians may be “the canary in the coal mine” for other professions. Mathematics offers an unusually clear view of what happens when AI enters intellectual work because the objective seems so precise. A proof is either valid or it is not. A conjecture is either true or false. Yet the more powerful the systems become, the more complicated the distinction becomes between obtaining an answer and understanding it.
The artificial intelligence industry has noticed this potential. Mathematics has become one of the most important arenas in the race toward increasingly capable AI systems, partly because mathematical reasoning is widely regarded as a possible pathway toward broader machine intelligence. Some companies are explicitly treating mathematics as a test of how close their systems may be getting to superintelligence. Math Inc. has expressed the ambition with a deliberately simple slogan: “Solve math, solve everything.”
For some mathematicians, the transformation has already become personal. At the press conference surrounding the Fields Medal, Jacob Tsimerman, recognized for his work at the intersection of number theory and geometry, announced that he was leaving mathematics to work on AI safety at OpenAI.
His decision was striking not only because of the prestige of the career he was abandoning, but because of what he said about the future he believed was coming. Tsimerman has developed a taxonomy of AI risks that includes the possibility of human extinction, and he has argued that artificial intelligence will become better than mathematicians at research mathematics very shortly.
“People just haven’t been listening,” he said.
His position immediately became controversial. Some mathematicians interpreted his decision as a betrayal of the discipline, particularly because a Fields Medal represents one of the highest recognitions available in mathematics. Others considered his public warnings irresponsible, fearing that they could discourage younger researchers from entering the field. Some simply struggled to understand why someone who had reached the summit of mathematical research would choose to walk away.
Tsimerman sees the situation differently. He believes the taboo surrounding catastrophic AI scenarios needs to be broken because the consequences are too important to leave unspoken. Yet he does not describe the future only in apocalyptic terms. If machines become capable of generating mathematical results on an enormous scale, mathematicians could inherit something resembling an infinite library filled with new theorems and discoveries.
The problem would no longer be finding enough mathematics to explore. It would be deciding what deserves to be understood.
Javier Gómez-Serrano, a mathematician at Brown University who works with Google AI on questions related to fluid dynamics and problems resembling the Navier-Stokes equations, has described the experience through the language of grief. His reference to the “five stages of AI grief” captures the psychological adjustment required when a profession discovers that a technology is beginning to perform tasks once considered uniquely human.
Gómez-Serrano does not expect mathematics to remain untouched. He believes there will be fewer jobs, different jobs, and a profound restructuring of the profession. The mathematics itself may survive, but the system surrounding it could look completely different. “The whole system is going to collapse,” he has warned.
Terence Tao has compared the current moment with another period of uncertainty in mathematical history. In 1900, David Hilbert presented his famous list of problems at a time when mathematics was struggling with paradoxes and questions about its foundations. The response was not to abandon mathematics, but to rebuild its foundations through a more rigorous axiomatic approach.
Tao sees a similar possibility today. Artificial intelligence may force mathematicians to reconsider not the truth of mathematics itself, but the culture surrounding it: how results are discovered, verified and communicated, who receives credit, and what kind of work is worth doing.
He is not calling for mathematicians to reject AI. His concern is that the rules should be established by mathematicians rather than imposed by companies whose interests may be very different from those of the mathematical community.
That question has already reached professional ethics. Tao has supported the Leiden Declaration, an attempt to establish norms around the use of AI in mathematical research. Among its proposals is the principle that artificial intelligence should not receive authorship of mathematical proofs. Even that seemingly simple rule has already become controversial, with OpenAI and some mathematicians challenging the idea that traditional concepts of authorship can survive unchanged once machines participate directly in research.
AI in Mathematics and the Mathematicians Trying to Understand the Machine
Ravi Vakil, president of the American Mathematical Society, offers a very different interpretation of the Jacobian episode. In his view, it is misleading to say that Claude Fable simply did the mathematicians’ work for them. The machine produced an important counterexample, but the mathematical community still had to understand what it meant.
Leading mathematicians gathered around a blog and began examining why the conjecture was false. What could have been a moment of professional humiliation instead became an intellectually fascinating problem. The machine had found something important, but human mathematicians were still trying to understand the structure behind it.
For Vakil, that may point toward a more optimistic future. Artificial intelligence could solve problems that humans cannot solve efficiently while mathematicians devote more of their energy to understanding those solutions. The machine might produce the result, but the human contribution would be interpretation.
A mathematical discovery, in this view, becomes valuable not simply because it is correct, but because mathematicians can place it inside the larger structure of mathematical knowledge and explain why it matters.
There is another reason Vakil is concerned about the current panic. Mathematics already faces pressure from funding cuts, uncertainty surrounding academic careers, and a shrinking number of opportunities for graduate students. Universities are accepting fewer students into some programs, while changes in immigration and visa policies have made international academic careers more difficult.
If AI becomes a convenient argument for reducing investment in mathematics, the technological revolution could accelerate a decline that had already begun for economic reasons.
The danger is also cultural. Young mathematicians are being asked to develop expertise in a world where refusing to use AI can itself feel like falling behind. A poll among PhD students presented a strikingly divided picture: when asked whether they would press a hypothetical button that would stop AI from doing mathematics, the community was roughly split down the middle.
Yet beneath that division was a remarkable degree of agreement. Students widely expected AI systems to become substantially better, and many believed that becoming a professional mathematician would consequently become more difficult and tenure-track positions more scarce.
The anonymity surrounding some of these opinions is revealing. The debate has become sufficiently sensitive that students do not always feel comfortable attaching their names to their views. One student described paying for Claude Max simply to keep up with developments, while suggesting that not subscribing could itself become a disadvantage.
The technology is creating a strange new form of professional pressure: researchers may disagree with the direction of AI while simultaneously feeling that they cannot afford to ignore it.
Denisse Escobar Parra, a PhD student at the University of California, Santa Barbara, represents a more hopeful position. She wants to know what is true, and if AI allows mathematicians to reach that truth more quickly, she does not automatically see that as a loss. She accepts that the number of jobs may decline and is willing to compete in a more demanding environment.
For some young mathematicians, the higher bar created by AI is not necessarily a reason to leave the profession. It may simply be a reason to become better.
Michael Harris of Columbia University takes almost the opposite position. He has argued against the mechanization of mathematics and compared the situation to what has happened in creative professions such as graphic art and screenwriting. Rather than simply adapting to whatever systems technology companies develop, he has advocated collective resistance and, where possible, restrictions on the use of AI in mathematical work.
For Harris, mathematics is not primarily an industrial process designed to maximize useful outputs. It is part of human flourishing. People do mathematics because they are fascinated by it, because they enjoy the struggle of solving difficult problems, and because mathematical beauty can matter even when there is no obvious practical application.
In a world organized around utility, doing something simply because you love doing it can become an act of defiance.
Harris has previously described mathematics as his “problematic vocation,” a phrase that captures something important about the current crisis. If the purpose of mathematics were simply to obtain correct answers, AI would seem like an extraordinary gift. But if the process itself has value, replacing the journey with an automated shortcut changes the meaning of the activity even when the final answer is better.
That distinction also complicates the argument about education. Geordie Williamson, who has written papers with AI companies and remains enthusiastic about the technology, is skeptical that AI will necessarily destroy mathematical teaching. Smartphones and Google changed the way people access information, but they did not make mathematical education irrelevant.
The more immediate danger, he argues, may be that the research side of mathematics becomes economically vulnerable because much of academia is ultimately supported by teaching.
There is also a question of expertise that cannot yet be answered. Sutherland believes that an expert human in the loop will remain valuable for some time, but nobody knows how long that advantage will last. Even more uncertain is how the next generation of mathematicians will acquire the expertise necessary to supervise machines if they no longer spend years learning mathematics through the traditional process of struggling with difficult problems themselves.
That is one of Tao’s deepest concerns. Imagine a machine produces a revolutionary mathematical result, and every test confirms that the proof is correct. What happens if there is no human being capable of explaining it?
Mathematics would possess a truth that its own community could verify but not understand. The discipline would still contain knowledge, but the relationship between knowledge and human comprehension would have changed.
AI in Mathematics and the Fight Over the Future of Mathematical Work
Naomi Sweeting, a newly appointed tenure-track assistant professor at MIT, believes the conversation itself is changing. Critical opinions about AI that once remained behind closed doors are increasingly being expressed publicly. Sweeting has adopted a personal policy of not using AI for her mathematical work, partly because she wants to make clear to graduate students that they are allowed to do mathematics for themselves.
In an environment where AI use can feel inevitable, simply refusing to use it becomes a statement about what kind of intellectual life mathematics should preserve.
That resistance has begun to develop its own literature. On August 3, Max Weinreich published an essay making the case for total opposition to AI in mathematics, arguing that mathematicians should refuse to use these systems and oppose their development. On August 9, Caltech PhD student Tasmin Chu published “The AI Dissenter Viewpoint,” arguing that mathematicians have moral obligations to avoid collaborating with AI companies.
A blog called “Proofs and Prompts,” launched on August 7, began collecting critical writings, including contributions from Fields Medalists. By August 17, a paper was circulating with a footnote warning that mathematicians were being instrumentalized as part of a large-scale advertising campaign by AI companies competing for market dominance.
The anger is not directed only at the machines. It is increasingly directed at the industry building them. Sweeting has argued that mathematics is being used as material for corporate press releases and technological agendas that have little connection with the goals of mathematicians themselves.
Srikanth Iyengar expressed the frustration more bluntly: “They don’t care about beauty; they don’t care about truth. They only care about money.”
The debate also exposes an uncomfortable class divide within mathematics. Emily Riehl has pointed out that access to frontier AI systems and high-profile industry collaborations tends to flow toward mathematicians who are already exceptionally well connected, disproportionately within established and mostly male networks.
If AI becomes a central source of mathematical opportunity, it could unintentionally revive an old structure in which a small inner circle controls access to the most valuable resources.
Even Geordie Williamson, one of the mathematicians most excited by AI, has expressed disappointment at Tsimerman’s decision to leave mathematical research. He described it as a possible “harbinger of things to come,” a warning that the technology may not merely change mathematics from within but persuade some of its most talented people to abandon it.
Williamson has even said that, given the choice, he would press the button that made AI disappear, not because he believes AI is inherently evil, but because he believes everything is happening too fast. He has quoted mathematician Zhiwei Yun to explain why: “As a mathematician, I care not just about the end results, but also the journey. AI is destroying my journey.”
That may be the simplest way to understand why the argument has become so emotional. The conflict is not really about whether machines can calculate faster or whether they can discover proofs. Those questions are increasingly being answered in front of everyone. The harder question is whether mathematics is defined by the destination or by the path taken to reach it.
A machine capable of producing thousands of important theorems would radically increase the quantity of mathematical knowledge. But quantity and meaning are not the same thing. If mathematicians spend their lives reading an endless stream of machine-generated results, the profession may become extraordinarily productive while becoming less capable of deciding what deserves attention.
The alternative is more hopeful. If machines remove tedious work, expose unexpected connections and allow researchers to explore structures that would otherwise remain inaccessible, mathematics could enter one of the most creative periods in its history.
The outcome may depend less on the intelligence of the machines than on the choices made by the people using them. Mathematicians will have to decide what they want to preserve: the struggle, the apprenticeship, the authorship, the social culture, the freedom to work without commercial pressure, or simply the accumulation of correct results.
Sweeting has described mathematics as a vast and beautiful human project connecting thousands of years of thought. That description points toward the question underneath all the others: does that human project continue if the act of thinking itself gradually ceases to be human? And if accelerating the production of knowledge comes at the cost of the experience that made the knowledge meaningful, is that acceleration really progress?
For now, the mathematical community is caught between those futures. Some researchers see an unprecedented intellectual tool; others see the beginning of the end of a uniquely human practice. Many are somewhere between the two, using AI because it is useful while worrying about what its usefulness will eventually cost.
The argument is unlikely to be resolved soon, because the technology itself is changing faster than the profession can establish rules for it. Mathematics may therefore become one of the first places where humanity is forced to confront a difficult possibility: a machine can know something without experiencing the process of knowing it.
The question is no longer whether artificial intelligence can enter the world of mathematics. It already has. The real question is whether humans will remain the ones deciding what mathematics means.
In the end, what is happening inside mathematics is not just a technological shift; it is an anthropological one. Machines are entering a territory that for thousands of years belonged to human intuition, and every new AI‑generated result becomes a test of our ability to understand what “knowing” really means. Perhaps the future of mathematics will not be defined by how fast new theorems appear, but by how deeply we can interpret them. In a world where AI accelerates everything, understanding may become the last truly human act.
AI Mathematical Intuition: When Machines Seem to Find Unexpected Paths in Mathematics A visual and conceptual exploration of how AI is beginning to move through mathematical landscapes once accessible only to human intuition.
Model Hardware Standard Anthropic: A Powerful Step Toward Real‑World AI Agents A look at the evolution of AI agents and the infrastructures behind them — a key piece in understanding how AI is reshaping scientific and mathematical work
