Artificial intelligence (AI) systems have recently achieved remarkable breakthroughs in mathematics, solving numerous long-standing open problems with unprecedented speed and efficiency. In early September, OpenAI reportedly deployed 10,000 AI agents to tackle the Navier-Stokes problem, resolving it in just 88 hours. The company has since announced that its internal model has addressed more than 100 enduring mathematical challenges across various fields, a development it is preparing to share with the broader research community.
Yet this rapid progress has sparked a debate among mathematicians about the potential consequences of relying heavily on AI for problem-solving. A group of 27 Fields medal winners—the highest honor in the discipline—has issued a joint declaration emphasizing that the objective of resolving famous problems is not merely to reach solutions, but to foster conceptual understanding and insight. These problems serve as guiding landmarks that help mathematicians develop deeper comprehension of mathematical structures rather than simply targets for mechanized resolution.
Among the signatories is Terence Tao, widely regarded as one of the leading mathematicians of his generation. Tao has expressed concerns that AI-driven problem-solving risks "strip-mining" mathematics by extracting solutions without nurturing the broader mathematical ecosystem. He likens significant open problems to nonrenewable resources that support training and growth within the field. If automated systems rapidly solve these, the opportunity for human researchers, especially early-career mathematicians, to learn and develop intuition may diminish. Tao has even proposed safeguarding certain problems from AI intervention, treating them analogously to protected natural reserves.
This viewpoint has met skepticism from some AI proponents and outside observers, who argue that solving mathematical problems more efficiently should be welcomed as genuine advancement. Critics have questioned whether insistence on human understanding represents a form of professional self-preservation by mathematicians facing competition from intelligent machines.
The debate reflects a deeper divide within mathematical culture between “pure” and “applied” mathematics. Pure mathematicians prioritize the intrinsic beauty and coherence of mathematical concepts, seeking insight and understanding beyond practical outcomes. Applied mathematicians, on the other hand, focus on harnessing mathematical tools to address real-world challenges, often drawing inspiration from physical phenomena and technological needs.
Historically, advances in mathematical understanding have driven transformative progress in science and technology. For example, breakthroughs in electromagnetism in the 19th century laid the foundation for modern wireless communication. Similarly, the development of compressed sensing in the early 2000s—rooted in mathematical theory about signal sparsity—dramatically improved magnetic resonance imaging (MRI) by reducing scan times and minimizing the need for sedation in pediatric patients. This advance emerged from decades of cumulative human insight bridging pure mathematics and applied engineering.
The authors caution that while AI could arguably bypass such complex chains of reasoning to produce faster or more efficient methods directly, this might come at the expense of lost opportunities for new understanding, unforeseen applications, and broader technological innovation. The concern is that without human comprehension, advances might be more limited in scope or brittle in practice.
As AI continues to reshape the landscape of mathematical research, the community faces a critical question: how to integrate powerful automated tools while preserving the human creativity and conceptual depth that have historically fueled scientific and technological revolutions. The resolution of this tension may determine not only the future of mathematics but the wider advancement of knowledge across disciplines.