In a matter of days, artificial intelligence has upended the field of mathematics. OpenAI disproved a conjecture by the legendary Hungarian mathematician Paul Erdős that had stood since 1946, and Google DeepMind soon after announced solutions to nine problems, including two that remained unsolved for half a century. These breakthroughs have pushed a discipline long considered safe from automation into an urgent debate about the role of human intellect in discovery.
"It's a considerable mathematical achievement," said Jeremy Avigad, a professor of philosophy and mathematics at Carnegie Mellon University, referring to OpenAI's result. "In contrast to previous results, this problem is well-known, and its solution could be published in the world's best journals." The speed of these advances has rattled the community, prompting a mix of fear, indifference, and the drafting of the Leiden Declaration on Artificial Intelligence and Mathematics, which recommends disclosing AI use in scientific articles and confirming human authorship.
A tool, not a replacement
Beneath alarming headlines like "Is Mathematics Obsolete?" the conversation among researchers is more measured. Petra Schwer, a professor at Heidelberg University, said mathematicians have faced similar existential threats before. "At the end of the day, AI is a tool. Mathematicians have already been threatened by the calculator, the computer, and computer algebra systems," she said. "There will continue to be a place for mathematicians."
The real risk lies not in AI's capabilities but in how people use it without proper training. Javier Gómez Serrano, a Brown University professor who previously collaborated with Google on the Navier-Stokes equations, said fluency with these tools creates an edge. "But critical thought is necessary. Someone who has the tool without the training would produce trash they won't even be able to detect, and that is probably what will happen in great quantities." Demis Hassabis, founder of DeepMind, reinforced this limitation: "Today's systems are extremely far from what would be a true invention or someone like Ramanujan, no matter how many Erdős conjectures they solve."
The business of solving famous problems
The focus on famous, long-standing problems is no accident. Startups and major labs target high-profile conjectures because each headline-grabbing solution attracts attention and potential investment. "Mathematical research can be competitive, but these communities have solid ethical norms," Avigad said. "It is unacceptable to use someone's work and ideas without giving them credit. In academia, skipping over those norms can endanger one's reputation. In business, it's harder to follow those norms."
Seewoo Lee, a researcher at University of California, Berkeley, said the public image of a mathematician as someone who simply solves hard problems fuels misunderstanding. "AI helps mathematics progress more quickly. For mathematicians, that's not always a comfortable advance," he said. A mathematician constructs theory, finds appropriate definitions and theorems, and helps explain the world-tasks that go far beyond cracking a single conjecture.
Uneven adoption and questionable quality
Despite the splashy results, AI has not yet triggered a dramatic acceleration in mathematical discovery. Thomas Bloom, a mathematician at the University of Manchester who maintains a catalogue of Erdős problems, said the recent solutions do not yet constitute a huge leap. Sam Livingstone, a mathematician at University College London, noted a more troubling trend: "I have heard that the most prestigious journals are seeing an increase of around 20% to 30% in submissions in comparison to two years ago, and that they suspect it could be AI-related, but also that the majority of those additional submissions are not considered good studies."
Adoption among mathematicians remains patchy. Livingstone said he knows several prominent researchers who have never touched AI tools. "Mathematicians, generally speaking, are a conservative group. That could change if it becomes clear that AI is accelerating mathematical discovery, but for the moment that's the state of things."
Why this matters for science and research professionals
If AI continues to gain ground in mathematical reasoning, researchers across disciplines will face the same question now confronting mathematicians: what work remains for humans? Avigad drew a distinction between mathematics and games like chess, where AI dominates. "Mathematics is not just a game; it is an important part of how we give meaning to the world, we reason and discuss with one another. I don't think that will change." The adaptation will also reshape education. Gómez Serrano said students must develop competencies in AI for Science & Research and learn to connect apparently distant fields within mathematics-skills that will define who thrives in the AI era and who produces undetected errors. For working scientists, the takeaway is practical: AI is a powerful assistant, but the advantage belongs to those who can tell when it is lying.
Your membership also unlocks:
