- Author
- Jun-Yong Park, School of Mathematics and Statistics, University of Sydney
- Published
- 7 July 2026
- Kind
- Policy essay posted to arXiv under math.HO
- Original
- arxiv.org/abs/2607.06377
- Subject
- Review, taste and expertise
- Themes
- Skill under automation, Who holds the power
What it is
Jun-Yong Park, a University of Sydney mathematician, wrote a policy essay for a US audience. It sets out to show that cuts to mathematics weaken human oversight of AI. It ends with four recommendations.
What it says
- Park reports an OpenAI model disproved an Erdős conjecture, checked by nine mathematicians.
- Park says over $14 million in math grants was clawed back; Congress rejected most cuts.
- Park argues machines may make proofs, but humans must keep the capacity to understand.
- A network was decoded only because researchers recognized Fourier structure, Park says.
- Park wants consequential AI claims exposed in formal, machine-checkable form.
- Park says one Lean formalization attempt hid gaps behind placeholder structures.
What evidence it gives
The essay cites news reports, an OpenAI announcement, companion papers and a 1984 committee report. It uses analogies to water systems and surgery. It contains no new data, experiments or calculations.
Interests and limits
This is advocacy by a mathematician for funding mathematics. Funding figures and the Erdős result come from cited sources the essay does not reproduce. The essay answers the view that fewer mathematicians may suffice only briefly. Its recommendations are untested and aimed at the United States.
Connections
These links are this wiki's own reading, not claims the sources make.
- The threat is comfortable drift toward not understanding what you're doing. Commenters say the program exists to produce scientists, not results. Park makes a similar case for funding mathematicians as machines begin to prove things. Both ask what research really outputs. Is it a proof, or someone who understands it?