The real danger isn't that AI will solve all mathematical problems, but that it will make the human act of mathematical discovery seem pointless. David Bessis, a mathematician-turned-machine-learning entrepreneur, argues that AI poses an existential threat to mathematics not by replacing theorems, but by destroying the cultural and intellectual ecosystem that has sustained the discipline for centuries .

What's the Difference Between "Official Math" and "Secret Math"?

Bessis draws a crucial distinction that most people never consider. Official mathematics is what appears in textbooks and peer-reviewed papers: formal deduction systems where you start from axioms and mechanically derive theorems. It's clean, logical, and objective. But secret mathematics, the human part, is where the real work happens. It encompasses intuition, conceptual breakthroughs, the messy process of deciding what's worth proving, and the creative act of inventing the right definitions to make hard problems easy .

The problem is that mathematics as a discipline has spent centuries devaluing secret math in favor of official math. Mathematicians are trained to hide their intuition, downplay their creative process, and present only the polished final theorems. As G.H. Hardy infamously wrote, "There is no scorn more profound, or on the whole more justifiable, than that of the men who make for the men who explain." This cultural bias has created a system where exposition, teaching, and conceptual clarity are treated as second-rate work .

How Does AI Threaten the "Theorem Economy"?

Bessis illustrates the problem through his own career. His best mathematical result, a theorem that crystallized one morning in Lausanne, Switzerland, never made it to publication. He announced it informally on the last slide of a conference presentation, hoping a young mathematician would eventually formalize it. But by claiming the result publicly, he killed any incentive for someone else to develop it. His second-best result, Theorem 0.5 on Garside categories, was so straightforward to prove that he never bothered submitting the paper. The real innovation wasn't the theorem itself, but the definitions and conceptual framework that made it obvious .

Here's where AI becomes dangerous: if machines can generate theorems at scale, the entire incentive structure of mathematics collapses. Why spend years developing a proof if an AI can produce thousands of valid theorems overnight? The theorem, which has been treated as the currency of mathematical prestige and career advancement, becomes worthless. And if theorems become worthless, the discipline loses its primary motivation system .

But the deeper threat is more subtle. Mathematics has always been about clarity and understanding, not theorems themselves. As Fields medalist Bill Thurston noted, "The product of mathematics is clarity and understanding. Not theorems, by themselves." Yet the profession has built its entire reward structure around theorem-proving. If AI can automate theorem generation, mathematicians will face a reckoning: either admit that theorems were never the point, or watch their discipline become irrelevant .

Ways Mathematics Could Adapt to the AI Era

• Revalue Conceptual Work: Shift prestige and career advancement toward mathematicians who develop new definitions, frameworks, and conceptual breakthroughs rather than those who merely prove theorems, since AI will eventually handle routine proofs.
• Embrace Exposition and Teaching: Elevate the status of mathematical writing, exposition, and pedagogy, recognizing that clarity and communication are irreplaceable human skills that AI cannot fully replicate.
• Focus on Problem Selection: Concentrate on the deeply human work of deciding which problems are worth solving and why, rather than on the mechanical execution of proofs.
• Preserve the Intuition Club: Create space for the informal, intuitive, and exploratory aspects of mathematics that currently remain hidden, making secret math part of the official curriculum and culture.

Bessis expresses a complex emotional response to this moment. He feels vindicated, having anticipated this shift. He feels excited about the genuine revolution in what's possible. But he also feels puzzled by the speed of change, worried about the existential threat to mathematics, and nostalgic for a lifestyle and value system he walked away from but which may soon disappear entirely .

The irony is profound: AI could destroy mathematics by making it too easy, not by making it impossible. If machines can generate valid theorems faster than humans can read them, the discipline loses its organizing principle. Mathematics would survive as a formal system, but the human culture that gave it meaning would evaporate. The question facing mathematicians now is whether they're willing to admit what they've always known but never said: that the real mathematics happens in the thinking, not in the publishing .