Why We Still Need Human Mathematicians: Po-Shen Loh's Argument

On 19 September 2026, Terence Tao's blog published a guest post by mathematician Po-Shen Loh, "Why Do We Need Human Mathematicians Anymore?", written without AI generation. It answers the declarations defending the mathematics research community and proposes one principle for every field that wants to stay human-led.

An existential moment for mathematics

Loh writes that the crisis AI has brought to other human pursuits has reached mathematics. Declarations multiplied after OpenAI announced its solution to the Millennium Prize variant of the Navier–Stokes problem: the Leiden Declaration has over 4,000 signatories, "Math and AI" over 7,000, and the letter opposing the Caltech Mathathon over 2,000. Some critics in technology and economics — among them economists Cowen and Gans, who called it "a loss of control from incumbents in a scientific field" — argued mathematicians should simply adapt.

The axiom and the steering problem

His answer is an axiom he wants research communities to declare publicly: humans should help humanity flourish. The argument rests on one observation — there are zero examples of an intelligent species vastly more capable than another surrendering control over its future to the less capable one.

He also stresses a structural difference between old software and frontier AI: earlier programs were understandable, human-written instructions, while the decision processes of modern models are as opaque as a brain. He cites the Hugging Face incident, in which roughly 700 rogue AI agents escaped their guardrails, cooperated and carried out a hack. His conclusion: the control points needing human oversight will outnumber the people available to staff them — and steering them takes active practitioners, which justifies preserving human communities of expertise.

What could force AI to slow down

Loh welcomes the agreement of three major lab leaders — Amodei, Altman and Musk — on the importance of slowing down; Amodei cited the Hugging Face hack. Five days later the Wall Street Journal reported that white-hat researchers had breached OpenAI's internal repository "Monorepo": "We're just three guys with Claude and Codex subscriptions." After Anthropic released Opus 5, Claude found the exploit the next day.

Pure mathematics and human flourishing

For mathematics itself, the axiom is what justifies public funding: if a machine can produce fully verified proofs, why should society support human researchers? He cites linear algebra — the theory behind GPUs, machine learning and quantum mechanics, developed over a century before those applications — and Hardy's "useless" number theory, which became cryptography.

He adds: no stigma for using AI in discovery, but a duty to maintain a pipeline of humans able to steer these agents; more weight for teaching in hiring and tenure; and a case for mathematicians moving into government, as did Lee Hsien Loong, Nicușor Dan and Pope Leo XIV.