Hank Yang: AI Excels at Well-Defined Problems, So the Real Skill Is Defining New Ones

hankyang94 · x · 2026-09-10

Hank Yang observes that Astra shines precisely where humans have spent years building knowledge — Millennium Prize math, scene reconstruction, robot control — and that accumulated human effort enables the tipping point where AI surpasses individuals.
On ill-posed, ambiguous problems AI remains weak: he wouldn't hand several of his own group's research questions to Astra, not because the math is harder but because the problems can't yet be clearly formulated.
The dilemma: work on hot problems and AI may soon beat you; work on niche problems and they may not matter.
His proposed research strategy: discover and frame problems that don't yet exist properly, then use AI to solve them — human value shifts from solving to problem-finding.

More from AGI Musings