The Dawn of AI in Mathematics: A Mathematician's Uneasy Vision

When a seasoned mind like Steven Strogatz, the brilliant 67-year-old Cornell University mathematician, admits to feeling "terrified" by artificial intelligence's recent strides in his field, well, you can bet something truly significant is unfolding. It’s not just a passing worry; it’s a profound uncertainty about the very essence of mathematics and humanity's place within it. He's openly musing that groundbreaking mathematical discoveries might soon be inseparable from AI's involvement, potentially nudging human researchers into a role more akin to interpreters of machine-generated insights—results they might not even fully grasp on their own. For Strogatz, 2026 looms large, a potential watershed moment that could redefine mathematics as we know it, depending entirely on how these seismic shifts are assessed and understood.

Consider some of the specific AI feats that are stirring such a powerful reaction. Take OpenAI, for instance, and their bold claim regarding the Navier-Stokes equations. This isn't just any old problem; it's a monumental, 90-year-old puzzle, complete with a tantalizing $1 million prize for its solution. On September 8, 2026, OpenAI publicly announced an analytical proof and a Lean formalization related to a finite-time singularity within the notoriously complex forced, three-dimensional, incompressible Navier–Stokes equations. Now, it's crucial to note this claim is still awaiting independent verification. It hasn't yet, let's be clear, been accepted as the definitive solution to that coveted Millennium Prize problem. Yet, the very fact that such a claim can be made, building upon a strategy developed by Spanish mathematicians Diego Córdoba and Luis Martínez-Zoroa, is undeniably impressive, and a sign of things to come.

Then there's Anthropic, a formidable player in the AI arena, showcasing its own prowess with Claude. Just a few days earlier, on September 4, 2026, Anthropic shared that their AI system had successfully formalized an existing proof of Fermat's Last Theorem using Lean. For those unfamiliar, Lean is a sophisticated programming language specifically designed to rigorously check the logical steps of a mathematical proof. It's important to clarify here: Claude didn't discover a new proof of Fermat's Last Theorem, which, of course, has already been famously proven. Instead, it meticulously converted an established argument into a machine-verifiable format. What’s truly astounding is the scale of this achievement: Anthropic stated that Claude worked largely autonomously for a staggering 11 days, with only occasional, high-level human guidance, to prove some 29,500 intermediate theorems and churn out a colossal 13 million lines of Lean code. Imagine the sheer computational power and logical precision required!

These developments, as Strogatz keenly observes, point to a significant redefinition of the mathematician's role. He envisions humans increasingly engaged in what he calls "proof digestion." This isn't about conjuring new proofs from scratch; rather, it’s about deciphering and explaining these intricate machine-generated arguments, interpreting their significance, and figuring out precisely why they matter to the broader mathematical landscape. It’s a shift from primary creation to critical interpretation. However, Strogatz isn't entirely pessimistic. He also sees a silver lining, a compelling potential benefit: AI could genuinely democratize mathematics, opening its doors to a much wider audience, including those who might not possess years of highly specialized training. It’s a fascinating, if somewhat daunting, prospect—a future where human intuition and AI's boundless processing power might, in fact, forge an entirely new path forward for the oldest of sciences.