Po-Shen Loh argues AI moves faster than humans can steer

Po-Shen Loh posted a guest essay on Terence Tao's blog on September 19. His argument is that advancing AI needs more human mathematicians, not fewer. As he puts it, "Driving a car faster than you can run is fine. But not faster than you can steer." His claim is that AI capability creates control points faster than the supply of qualified experts can cover them. Labs end up throttled by hiring reality, not by the technology itself.

Math departments are the ones who'd have to act on this. Loh counts more than 4,000 signatories on the Leiden Declaration alone, part of the backlash he's answering. His ask is concrete: weight teaching and human-facing work in hiring and tenure decisions. He also wants the stigma dropped against mathematicians who use AI as a research assistant.

The opposition goes beyond that one declaration. More than 7,000 people have signed onto Math and AI, and more than 2,000 have opposed the Caltech Mathathon. Both followed the backlash after the Navier-Stokes controversy.

One detail cuts against reading this as pure AI advocacy. Tao notes the essay's prose was written entirely in vim, with no AI generation. The page layout, though, came from Claude Code.