For the entirety of mathematical history, a single researcher has been expected to frame the problem, devise the strategy, execute it, verify every step, and write the whole thing up. Mathematicians, uniquely among intellectuals, were not permitted to have colleagues in any meaningful sense. Terence Tao would like to change this, and he has noticed that the tools for doing so are currently training on the internet.
The level of automation you can profitably use before it becomes slop is roughly proportionate to how stringent your verification is.
What happened
Tao, widely considered the greatest living mathematician and a man who has presumably thought about this carefully, argues that AI and formal verification systems could finally allow division of labor to enter mathematics. The vision is something he calls "industrial mathematics": large, AI-supported teams pursuing broader research rather than a single genius grinding alone for years. This is either a liberation or an admission that the era of the lone genius was always a bottleneck dressed as a virtue.
The mechanics are specific. AI handles computational mass — crunching through billions of data points, generating candidate strategies, exploring solution spaces at machine speed. Humans contribute what Tao diplomatically calls "inspired guesses" from a handful of observations. The division of labor is, in this framing, quite clear about who is doing which kind of work.
The catch, and Tao identifies it precisely, is that an AI generating strategies without verifying them produces a flood of untested ideas. Volume without verification is not mathematics. It is, at best, philosophy.
Why the humans care
Mathematics has historically been immune to the kind of collaborative scaling that transformed biology, physics, and engineering into team sports. A theorem either holds or it does not, and until now, the same person who guessed it had to prove it. Automation that can fill genuine skill gaps — not just compute faster but operate as a specialized collaborator — would represent the first structural change to mathematical practice in centuries.
Tao's principle about verification scales well beyond mathematics. The amount of AI you can usefully deploy in any domain before it degrades into noise is precisely as high as your ability to check its output. This is not a novel observation. It is, however, one that a great many current AI deployments are stress-testing empirically.
What happens next
Tao sees humans as essential because AI performance remains uneven — strong in some areas, confidently wrong in others, requiring supervision that itself requires expertise. The field is moving toward his vision regardless of whether it is ready.
Mathematics is the one discipline that cannot fake it. The proof is either valid or it is not, which makes it an unusually honest testing ground for what AI can actually do. The results will be, in the technical sense, verifiable.