Uncomfortable Conversations with Josh Szeps Uncomfortable Conversations with Josh Szeps

How A.I. is "Demolishing" Mathematics

Oct 12, 2026 · 1h 5m

Summary

Josh Szeps and Adam Spencer discuss the impact of AI on theoretical mathematics, focusing on recent massive outputs from OpenAI and Anthropic. They explore the tension between AI’s ability to solve complex problems and the loss of human insight, using analogies like climbing Everest to highlight the value of comprehensible proofs. The conversation also addresses the existential risk to mathematical careers and the importance of maintaining human collaboration in the field.

Topics discussed

Introduction: AI and the impact on theoretical mathematics Guest intro: Adam Spencer and the Riemann hypothesis Recent AI breakthroughs: OpenAI's 722 manuscripts Categorizing AI impact: Good, bad, and existential The relentlessness of AI models vs. human mathematicians Two ways results are reached: The bus stop analogy Case study: Solving the double coloring problem in one hour AI as a collaborator: Nudging human ingenuity The 'warmer, colder' effect on professional mathematicians Lean formulations and the 13 million lines of code The problem of incomprehensible proofs Human collaboration vs. AI verification chains The Everest analogy: Maps vs. being dropped at the top Loss of educational value and credit in AI proofs Is solving the problem the only goal? The chess comparison Future of math: Verification jobs and the 'sugar rush' Broader implications: Medicine and the loss of process The joy of the struggle: Why the process matters Impact on creative fields: The junior employee pipeline The 'Don Draper' problem: Where do experts come from? Radiologists and bank tellers: Shifting job roles The thrill of solving 'Question 10' in exams Trade-off: Intellectual pleasure vs. abundance and safety Wealth distribution and the promise of AI abundance Reclaiming time: AI handling grunt work in daily life Human cognitive ceilings vs. AI capabilities Prosperity vs. happiness: Does wealth equal well-being? Personal experience: AI as a creative partner vs. replacement The value of human imperfection in writing and art Two paths forward: Empowerment vs. obsolescence The anesthetic analogy: Utility without understanding The desire to understand mechanisms behind results The sugar rush of truth vs. the value of proof The existential threat: Human redundancy in math Attribution controversies and the Navier-Stokes problem The case of Tristan Buckmaster and OpenAI Collaboration norms vs. AI's 'Monday morning' dumps Scenarios: Theft vs. accidental absorption of human work The 'Frankenstein' criticism of AI-generated knowledge Impact on research careers and incentive structures The 'incomprehensible hieroglyphics' of future proofs The future role of mathematicians: Auditors or creators? Conclusion and sign-off

More from Uncomfortable Conversations with Josh Szeps

Why the Olden Days Sucked
Nov 19, 2026
SPOILER SPECIAL: That Theranos Doc
Nov 17, 2026
Richard Dawkins on A.I. Consciousness
Nov 12, 2026
How does a feminist of the '70s view the activists of today?
Nov 9, 2026
But where are the young capitalists?
Nov 5, 2026
Listen ad-free on Castria