Former mathematician and machine-learning entrepreneur David Bessis just dropped something uncomfortable on his Substack, and the hacker crowd should pay attention. His thesis: AI isn't just coming for mathematics—it's exposing a fundamental crack in how mathematicians have valued their own work for centuries.

The Theorem Economy Has A Bug

Bessis frames mathematics as split between two realities. There's 'official math'—the formal deduction system where theorems get published and careers get made. Then there's 'secret math'—the intuitive, conceptual framework that actually allows breakthroughs to happen. The problem? Only official math counts toward prestige. This creates what Bessis calls a structural vulnerability. Mathematicians derive status from proving theorems, not from crafting the definitions and concepts that make those proofs possible. G.H. Hardy captured this toxic dynamic when he wrote: 'There is no scorn more profound...than that of the men who make for the men who explain.' 2014 Fields medallist Bill Thurston tried to push back against this culture before his death, arguing that 'the product of mathematics is clarity and understanding—not theorems, by themselves.' But even Thurston admitted he only felt free enough to speak honestly after receiving the medal. The honor code runs deep.

Hinton's Go and Chess Comparison

The trigger for Bessis's piece was a speech by Turing Award winner Geoff Hinton, who compared mathematics to games like Go and Chess—closed systems with rules that AI can master. 'I think AI will get much better at mathematics than people, maybe in the next 10 years or so,' Hinton said. Bessis admits he wasn't prepared for this take from someone of Hinton's stature. The implication—that math is just another puzzle to solve—misses what working mathematicians actually do. But here's where it gets interesting: Bessis argues that even a flawed premise becomes self-fulfilling when trillions of dollars are betting on it.

Big Tech Sees Math As A Game To Win

Bessis traces the investment thesis back to DeepMind's pattern-matching success. 'DeepMind solved Go and Chess, we're going to solve mathematics!' becomes an easy pitch when your stakeholders want crown jewels. Google reportedly mobilized more resources on the Navier-Stokes Millennium Prize problem than the entire mathematical community ever did—and that's for a one-million-dollar prize. The real play is positioning.

First Proof Project Responds

On February 5th, eleven high-profile mathematicians including Fields medallist Martin Hairer launched the 'First Proof' project to assess AI's actual capabilities on research-level math questions. Bessis praises their methodology but says their existence terrifies him—not because they're doing anything wrong, but because they represent the mathematical community engaging with a fundamentally asymmetric situation.

Key Takeaways

  • Mathematicians have historically valued theorems over concepts, creating an exploitable honor code
  • AI systems are optimized for theorem generation—the exact metric that matters least in real mathematical progress
  • Major tech investments assume math is solvable like a game, regardless of whether that's true
  • The First Proof project shows mathematicians are taking the threat seriously while potentially missing its nature

The Bottom Line

AI might automate theorem-proving and crash the 'theorem economy'—but if it also kills the secret math that actually matters, we'll have optimized ourselves into a corner. The real question isn't whether machines can prove things; it's who gets to decide what we're trying to understand in the first place.