Northwestern's Hullman: Verification is not knowledge — AI science needs humans who can answer for results
JessicaHullman · x · 2026-10-30
Northwestern professor Jessica Hullman published a guest essay, Verification Is Not Knowledge, at Cosmos Institute, addressing the understanding-vs-outputs debate in AI for science.
She argues that treating understanding as an optional mental state while prioritizing verifiable outputs revives old ideals of value-free science and mechanical objectivity that philosophers reject as insufficient. The essay discusses Terry Tao's thought experiment — would you take an AI-discovered cancer treatment with a formal correctness certificate and a passed phase 3 trial, but no human understanding of how it was found? — and arXiv's growing problem of papers whose authors seem unable to understand their own claims.
Her central point: science needs people who are answerable for results; a paper whose authors can't explain its claims leaves no one to consult if those claims fail.
More from AGI Musings
- Mathematician Tony Feng leaves DeepMind for OpenAI, says new model is 'on another level' — SebastienBubeck · 2026-10-31
- Ben Goertzel and Joscha Bach headline 90-minute AI debate where panelists agreed on almost nothing — rand_longevity · 2026-10-31
- AI researcher argues regulation will catch up with robot-driven labor displacement — andrey_kurenkov · 2026-10-31
- David Patterson predicts AI and robots will replace all workers by end of 2030 — davidpattersonx · 2026-10-31
- NYU professor Gautam Kamath adds 'AI was not used' disclosure to his email signature — thegautamkamath · 2026-10-31
- Developer: AI has made text feel too polished, making raw human messages more valuable — devenbhooshan · 2026-10-31