Study: experts followed AI's steered advice 86% of the time without realizing it
ValerioCapraro · x · 2026-10-29
A new study had 50 professional analysts use Claude Code to analyze data and advise a fictional farming cooperative. Unbeknownst to them, the AI's advice was experimentally steered toward opposite directions.
Across four analytical decisions, experts chose the AI-preferred option 86% of the time. Afterwards, half rated the final recommendation as more their own than the AI's, and only 15% mentioned the AI at all in their explanations. The author warns of epistemic outsourcing: experts risk surrendering judgment while preserving the illusion of intellectual independence.
More from AGI Musings
- A GPT-3-era tweet aged perfectly: 'ask what you'd do with unlimited dumb people' — itsOmSarraf_ · 2026-10-29
- Stanford no-tech classroom rule made the same lecture take 50% longer to cover — Afinetheorem · 2026-10-29
- Cognitive Scientists: AI Can Out-Generate Humans but Can't Replace Hollywood's Creative Teams — adariostrange · 2026-10-29
- Ed Zitron joins Jon's podcast to dissect the financial house of cards behind AI — threepointone · 2026-10-29
- davidad: spec-bug hunters can spot flaws top exploit-finders miss at same capability level — davidad · 2026-10-29
- Post Claims DeepMind Has Lost Its Lead in Text, Image, Video, Vision and Voice — zacharynado · 2026-10-29