Lay users defer to AI explanations; clinicians often do better without them.
Nature Medicine finds nonexperts trust LLM rationales even when wrong.
Nature MedicineHealth & Science · August 4
Medicine, biology, clinical AI, scientific discovery, and research infrastructure.
This page tracks clinical AI evidence, user risk, and medical training impact.
Today's health & science items · August 4
Current edition
Source-linked items from the current edition. Story dates are shown separately from the August 4 edition date.
People with the least medical knowledge are easiest to mislead.
Researchers warn consumer diagnostic apps can amplify automation bias.
Medical XpressAI can draft chatbot-patient teaching cases — under human review.
WVU tested ChatGPT-5 Pro vignettes for schizophrenia, anxiety, mood, depression, and psychosis contexts.
WVU TodayThe psychiatry vignette evaluation is in npj Digital Medicine.
Authors stress safety framing and human-centered use in medical education.
npj Digital Medicine