This episode features Fields Medalist Richard Borcherds and mathematician Eric Weinstein discussing AI’s sudden transformation of mathematics, the nature of mathematical practice, the future of human mathematicians, and what is at stake for the field and society.
The AI tsunami in mathematics
AI has undergone a discontinuous leap from “mediocre undergraduate” level to surpassing the best human mathematicians at problem-solving within months.
OpenAI released several hundred math problems, including some of the biggest open problems in their subfields, and AI solved many — often via counterexamples rather than proofs.
The breakthrough is concentrated in problem-solving and conjecture resolution, not yet in theory building — the creation of new conceptual frameworks.
AI excels at exhaustive calculation and searching vast combinatorial spaces that humans find too tedious, effectively clearing “low-hanging fruit hidden behind even lower-hanging leaves.”
Borcherds: AI is already better than him at research and teaching; Weinstein agrees on problem-solving but insists theory building remains a human domain for now.
Both expect AI to master theory building eventually; Borcherds predicts within a few years humans will be “worse than useless” at directing AI, as happened in chess.
Problem-solving versus theory building
Weinstein distinguishes three strands: conjecture/problem-solving, calculation, and theory building; the AI tsunami is almost entirely in the first two.
Theory building — creating new languages, structures, and organizing principles (e.g., Grothendieck’s schemes, Atiyah-Singer index theory) — has not yet been demonstrated by AI.
Borcherds agrees: current AI does “glorified intense calculation”; the complex structure on the six-sphere was found by brute-force search over millions of possibilities.
Historical precedent: pioneering work is often clumsy calculation (Milnor’s exotic spheres, Cartan’s exceptional algebras); understanding comes later through simplification.
Weinstein: AI may soon explode Atiyah-Singer theory by discovering new elliptic complexes from exceptional Dynkin diagrams (E6, E7, E8) that were too computationally expensive for humans.
What mathematics is about
Borcherds: mathematics is about finding unexpected structure — e.g., the clean classification of compact Lie groups, Fermat’s theorem on primes as sums of two squares.
Weinstein: mathematics is the “fabric of existence” — a hidden world of inevitable, beautiful structures (Hopf fibration, 24-cell, fake ℝ⁴, octonionic cross products) that no physical observer could anticipate.
Both emphasize the shock of discovering deep, cross-cutting connections (Langlands program, exceptional Lie groups, monstrous moonshine) that feel discovered, not invented.
Weinstein: the beauty is currently locked in “sheet music” (technical papers); visualization (Grant Sanderson, Dimensions series) is the “orchestra” that makes it accessible.
Exceptional algebraic structures as a frontier
Weinstein argues exceptional algebra (octonions, G₂, F₄, E₆, E₇, E₈, Leech lattice, Golay code, sporadic groups) has not yet entered the mainstream — it is an “island” separated by pedagogical reliance on defining representations.
The large exceptional groups cannot be deprojectivized; their defining representations are not visible in the regular representation ring, so standard pedagogy misses them.
Borcherds: sporadic finite groups are the true mystery — 26 exceptions with no uniform explanation, unlike Lie groups classified by Dynkin diagrams on half a page.
Both expect AI to integrate exceptional structures into the mainstream, revealing new cohomology theories, concrete cycles, and uniform explanations for sporadics.
The human moat: motivation, taste, and oral tradition
Weinstein: AI demotivates many but energizes him — “infinite supply of poppies for a heroin habit.” The moat may be the motivation to explore, not cognitive capacity.
Oral tradition: the best mathematics is often unwritten, passed personally (Singer’s drawer of unpublished ideas); AI has not accessed this.
Personal techniques (“zags” vs. “zigs”) and taste — knowing which problems are fruitful — remain human edges, though Borcherds bets AI’s 200-year literature access outweighs any individual’s secret tools.
Weinstein still prefers top human collaborators (Ed Frenkel, Raoul Bott) for work outside the convex hull; Borcherds concedes humans retain small specialties but AI dominates everywhere else.
Credit, priority, and the sociology of mathematics
Mathematicians publicly claim purity (truth-seeking, indifference to credit) but privately care deeply about priority, glory, and permanence — a “suicidal aesthetic” that misrepresents the field to funders.
Credit allocation is broken: relay-race finishers get disproportionate credit (Perelman/Hamilton, Gell-Mann/Zweig, Noether/Hilbert).
AI exacerbates this: if a random person prompts a proof of Riemann, who gets credit? Weinstein wants to trace AI outputs to the human voices embedded in training data (Witten, Atiyah, Bott, etc.) and reapportion credit accordingly.
Borcherds: credit has almost never gone to the first discoverer; this is a long-standing defect.
Funding, national interest, and the “suicidal aesthetic”
Weinstein: mathematics is central to national security (cryptography, nuclear weapons design, medical imaging, AI attention mechanisms) — not a luxury. The Teller-Ulam design came from a geometer (Ulam); elliptic curves secure billions in commerce.
Borcherds: his own work (monstrous moonshine) has no imaginable practical use; he cannot justify his salary on utility grounds.
Weinstein: this “recreational” self-presentation is dangerous — it invites defunding by officials who demand Q3 ROI. The field’s leadership (Arthur Jaffe et al.) has historically shielded mathematicians from political reality.
Borcherds admits political cluelessness; Weinstein argues mathematicians must learn to articulate their value as a public good (inexhaustible, inexcludable) via market-failure economics.
Peer review, formalization, and the future of verification
Peer review is becoming pointless: AI can formalize proofs in Lean and verify correctness mechanically. Borcherds: “Good riddance.”
Weinstein: peer review is a political-economy institution (gripe-and-swipe, gatekeeping) that suppresses eccentric voices; its demise could liberate “wild men, crazy women, cowboys.”
Autoformalization of all core mathematics (MAP project) is a “human genome project” for math, progressing faster than expected (finite simple group classification formalized in weeks, not years).
What is at stake
Weinstein: everything — mathematics is the lens through which we understand reality; we are at the finish line for a complete theory of physics (beyond GR and Standard Model), which could enable escaping Einsteinian speed limits and leaving the planet.
The Riemann hypothesis, Poincaré, Fermat falling are harbingers of a return to the tumult of early 20th-century science.
Borcherds: more sanguine — chess survived and flourished after computers surpassed humans; mathematics may become more popular and accessible via AI tutors.
Weinstein rejects the “Deep Blue moment” narrative: this is not the end of human mathematics but the beginning of a renaissance if the community stops throwing in the towel and embraces the tool.
Spinors, square roots, and the questions that persist
Borcherds: “You never understand things, you just get used to them” — human brains aren’t evolved for mathematics.
Weinstein: the square root is the “psychedelic of mathematics” — √(-1) → complex numbers, √determinant → Pfaffian, √Pontryagin class → Euler class, √tensor algebra → spinors. The meta-question: what keeps exceptional algebra out of the mainstream, and what does the square root operator reveal across contexts?
Atiyah’s “What is a spinor?” divides the community: Singer (we know via Atiyah-Singer) vs. Sullivan (nobody knows — no faithful irrep of Spin(16), square root of connection vs. momenta). Weinstein sides with Sullivan: the spiritual answer is missing.
