Sources
Terence Tao argues AI aimed at open problems without expert supervision destroys the detours that produce the real discoveries
In a three-part Mathstodon thread posted 2026-09-10, Tao frames pure mathematics as one of the few disciplines where childlike, curiosity-driven play is put to serious use, and names premature goal-optimization as the thing that kills it. His argument about AI is specific: human mathematicians used both childlike and adult skills to solve designated open problems even when they only publicly admitted the latter, but AI tools directed without expert supervision can be aimed straight at the nominal goal with no incentive to take the slow path. The flag gets captured at the cost of lessons learned, insights gained, collaborations formed and new targets located.
↳ Follow the thread