Why every attempt to close mathematics for good has expanded it instead — and why an AI that treats it as closed cannot learn the one thing that matters about it.
"We must know. We will know."
— David Hilbert, epitaph, Göttingen
24. July 2026
Geoffrey Hinton recently argued that mathematics is a closed system — a playing field with fixed rules, the kind an AI can train on the way it trained on Go. The claim is striking, not because it's wrong, but because it's wrong in a way mathematics itself can prove. Three times in its history, mathematics has tried to close itself. Three times, that very attempt has blown it back open.
Three Closures, Three Ruptures
The Pythagoreans wanted to build the world out of ratios of whole numbers. Everything was supposed to be rational — in both senses of the word. Then the diagonal of the unit square landed at their feet: √2. No ratio, anywhere. Legend has it they drowned the man who found it. The number stayed. The number space expanded to include the irrational — forced open by the very attempt to keep it rational.
The algebraists wanted every equation to have an answer. A modest goal. But x + 5 = 3 demands numbers below zero. x² = −1 demands numbers that "couldn't exist." They called them negative, imaginary, absurd — defensive vocabulary, every one. Today every AC power line runs on them. The demand for solvability, conceived as the system's completion, produced number after number the system had never made room for.
Hilbert, finally, wanted to close the lid for good: ground all of mathematics in a complete, consistent system of axioms. Every true statement provable, every doubt mechanically decidable. In 1931, Gödel showed that any consistent formal system powerful enough for arithmetic contains true statements it cannot prove — including a statement of its own consistency. This isn't opinion or cultural pessimism. It's a theorem. The most ambitious closure attempt in the history of mathematics proved that mathematics cannot be closed.
The Mechanism
Three times, the same structure:
The intent is rational. Wanting completeness isn't malicious — it wants clarity, reliability, solid ground. The Pythagoreans, the algebraists, Hilbert: not dogmatists, but precision workers.
The result is inversion. The very claim to closure exposes exactly where the system doesn't close. √2 exists as a scandal only because everything was supposed to be rational. Gödel's theorems exist because Hilbert demanded completeness — they are the precise answer to a precise question that would never have been asked without the will to close.
All are guilty. None are at fault.
This is Paradoxical Interaction in its cleanest form: no villain, no failure, no conspiracy. Just rational actors whose rational conduct systemically produces the opposite of what was intended — not as accident, but as structure.
The Münchhausen Move
You could put it another way: mathematics pulls itself out of the swamp by its own hair. That's supposed to be impossible — no system can ground itself from within, which is exactly Gödel's point. And yet it keeps happening.
The trick is in the detail: the hand doing the pulling doesn't belong to the system that's sinking. It belongs to a mathematics that doesn't exist yet at the moment of sinking. √2 sinks rational arithmetic — it's rescued by the real numbers, which first had to be invented. x² = −1 sinks real algebra — rescued by the complex plane, previously dismissed as "impossible." The crisis forces the invention of the ground on which it retroactively turns out to be solvable.
Münchhausen only lies as long as he insists on staying the same man. Mathematics climbs out of the swamp because it's willing to climb out as someone else.
What This Means for AI
If mathematics were a closed game, Hinton would be right: you could play it out like Go. Self-posed problems, checked proofs, reward for winning. Much of that will work — within the rules.
But mathematics is not rule-following. Historically, verifiably, it is the event where the rules stop being enough. The moment √2 shows up. The moment someone takes the imaginary unit seriously. The moment someone poses the completeness question so precisely that the answer moves the foundation itself.
An AI that learns mathematics as a closed system learns the game. Whether it learns the moment when the game blows itself open is an open question — quite literally: a question that cannot be settled from within the system it concerns.
That's exactly what Gödel would have predicted.
One clarification, in fairness: Gödel's incompleteness theorems apply to formal systems powerful enough for arithmetic. They don't say "everything is open" or "nothing is provable" — quite the opposite: they are themselves a proof, carried out with the utmost rigor. The openness of mathematics isn't arbitrariness. It's a precise, proven structural feature. That makes it stronger, not weaker.
Related Posts
A Paradoxical Interaction in Three Acts
How Rejecting Glory Becomes the Most Glorious Act – And Why That Cannot Be Escaped
How systems theorists reproduce the enclosure milieus Luhmann warned against
On piinteract.org
- [The Framework] — The underlying structure this case fits into: rational actors, a collectively irrational result, no one at fault.
- [Examples: Science & Academia] — More cases where science works against its own claim to completeness.
- [Examples: Technology & AI] — The Hinton reference in the context of further PI cases from AI development.
See also (external links):
Geoffrey Hinton in conversation with Yascha Mounk — Full interview transcript, primary source for Hinton's claim that mathematics is a "closed system."
Gödel's Incompleteness Theorems – Stanford Encyclopedia of Philosophy — The formal, peer-reviewed account of both incompleteness theorems underlying the Hilbert section.
Pythagoras – MacTutor History of Mathematics (University of St Andrews) — Institutional source for the historical account of the Pythagoreans' discovery of the irrational.
Real Numbers – MacTutor History of Mathematics — Further source on the still-disputed question of who first recognized √2 as incommensurable.
Paradoxical Interactions (PI): When rational actors consistently produce collectively irrational outcomes — not through failure, but through structure.
All are guilty. None are at fault.
Peter Senner Thinking beyond the Tellerrand
contact@piinteract.org
https://piinteract.org
Co-created with Claude (Anthropic) — two incomplete systems making each other's gaps visible.