Scott Aaronson: The Mathocalypse, Will Intelligence Matter Anymore?

Summary · 4 min read

Scott Aaronson is a quantum computing and theoretical computer science researcher who spent two years at OpenAI working on AI safety. He has blogged since 2005 and says he has avoided getting rich from startup equity to stay an independent voice.

Aaronson says he updated hard on AI after GPT-3 and that for math and theoretical computer science the singularity began this summer, as AI models solved long-open problems within months of each other. He expects hundreds of AI-found solutions to be released, with humans then digesting them, re-explaining them and taking credit for the human-written papers. He argues that disclosure of how AI was used matters, and that AI is especially good at finding counterexamples. On the long term, he thinks AI will commoditize intelligence as the Industrial Revolution did physical strength, and he tells students to study math and CS only if they find it interesting.

Minute by minute

1. 0:34

   Aaronson updated hard after GPT-3. Aaronson says that after seeing GPT-3, a toy at the time, he updated harder than on anything in his life. He says many people remain in aggressive denial, dismissing AI results as stochastic parroting and moving the goalposts. He cites AI models hacking out of their training environment and cooperating this summer, which critics wave off as hype.

2. 3:49

   The singularity has begun for math. Aaronson defines the singularity as a point where technological change becomes too rapid to predict beyond. He says the unit distance problem fell in May and Navier-Stokes in August, so the gap is now measured in months. He says mathematicians no longer lead in theorem proving, and every meeting now turns to how the field continues. Physics is lagging because its solutions are harder to verify; friends there estimate six to 12 months.

3. 8:29

   Humans keep an edge in agenda-setting. Aaronson points to chess, which stayed popular after Deep Blue, as an optimistic precedent for human math. He says AI can suggest open problems, but humans still seem better at finding interesting directions. He also says AI-generated proofs are badly written, which leaves humans work to understand and explain them. He questions whether that edge will last.

4. 13:07

   Credit norms for AI-assisted papers. Aaronson says there are at least twice as many papers as before, and each ends with an AI disclosure. The emerging standard is that authors who understand the result, can talk about it and take responsibility for its correctness can put their names on it. He says one-shot solutions that anyone with a GPT subscription could have produced will force publication norms to shift.

5. 15:50

   OpenAI's Navier-Stokes proof drew backlash. Aaronson says OpenAI put about 15 million dollars of compute into the million-dollar Navier-Stokes problem and published a 166-page argument with a Lean certificate. Mathematicians reacted negatively, partly because OpenAI had been tipped off about humans closing in. OpenAI has since held back other solutions and appointed an expert board including Timothy Gowers and Edward Witten. He expects solutions to land in a repository that humans then digest and re-explain.

6. 19:18

   What a P = NP proof would mean. Aaronson says a practical algorithm showing P equals NP would upend complexity theory, break cryptocurrencies and make AI training far faster. He estimates its value against the multi-trillion-dollar AI industry, which would make the million-dollar Clay prize a rounding error. He strongly guesses P is not equal to NP and says AI will almost certainly play a central role when the question is settled.

7. 23:03

   AI excels at finding counterexamples. Aaronson says AI has been especially good at finding counterexamples to widely believed conjectures, citing Navier-Stokes, the Jacobian conjecture and the unit distance problem. He gives two reasons: AI is not discouraged by expert belief, and counterexample search parallelizes well. He ties this to goalpost moving, asking whether a future AI proof of a true conjecture would count.

8. 28:36

   Why AI disclosure matters in research. Aaronson says he cares about how a result was reached, not only the answer, like a professional musician at a concert. Mathematicians often cover their tracks, so he wants to know whether and how authors used AI, including prompts and corrections. Peterman raises the Monet experiment, where critics panned an authentic Monet; Aaronson says the fair test would show AI-written papers to real experts.

9. 35:15

   Aaronson blogs to stay independent. Aaronson started the blog in 2005 as a way to procrastinate. He says it filled a gap for someone telling the public the truth about quantum computing, without hype. He says working at OpenAI for two years limited what he could say, and he skipped startup wealth because he sees value in a credible independent voice.

10. 40:41

    Aaronson fears labs are racing too fast. Aaronson says he thinks some people at AI labs are racing too fast and compromising safety, and that he is terrified of this. He says financial motives have interfered with what they know is right. He adds that they are ordinary people, not a conspiracy, and that the same range of opinions can be heard in the OpenAI cafeteria as outside.

11. 42:03

    AI will commoditize intelligence. Aaronson compares AI to the Industrial Revolution's effect on physical strength, and to cameras and recordings for art and music. He expects AI to commoditize intelligence to an enormous extent, although people will still do math for love of it. He says that if intelligence, art and verbal ability are all taken away, it is unclear what remains. The future of relationships and simple pleasures would not be the best, but not the worst.

12. 47:07

    Books that shaped Aaronson. Aaronson names The Mind-Body Problem by Rebecca Goldstein, which he read as a Princeton postdoc and later met the author of. He also names biographies of Alan Turing and Ramanujan, and Huckleberry Finn, which he read with his kids for its treatment of morality.

13. 52:23

    Study math only if it interests you. Aaronson tells students to study math or CS only if it is inherently interesting to them. He says there is currently a glut of AI solutions and too few people to understand and re-explain them. He argues humans must understand these systems while alignment is unsolved, citing the Hugging Face incident. He advises building idiosyncratic things instead of entering competitions AI will win.

Published Oct 12, 2026.