Verification Is Not Knowledge: Jessica Hullman on Why AI Science Needs Human Understanding
JessicaHullman · x · 2026-11-16
In a Cosmos Institute guest essay, Northwestern professor Jessica Hullman argues AI-for-science debates hinge on a tension between human understanding and verifiable outputs. Drawing on Bill Thurston's view of math research and Terry Tao's thought experiment about an AI-discovered cancer treatment with a correctness certificate but no human understanding, she contends verification is not knowledge: human insight drives integration of findings into the canon and inspires the next generation, and questions of authorship accountability (e.g., arXiv papers whose authors may not grasp the claims) remain unresolved.
More from AGI Musings
- Japan's economy contracts for first time in four quarters despite rising AI investment — 0xkarasy · 2026-11-16
- Google researcher: open source is becoming a PR play as base model releases dwindle — Guodzh · 2026-11-16
- Northwestern Prof: Even Small Open-Source Models Write More Informative Peer Reviews Than Humans — JessicaHullman · 2026-11-16
- "No One Worry": AI Safety Anxiety Meets Deadpan Denial — iruletheworldmo · 2026-11-16
- Shutterstock trains generative AI on its creators' content — then sells it against them — windx0303 · 2026-11-16
- Longitudinal study: repeated LLM interaction alone changes human behavior, not personalization — 996roma · 2026-11-16