Drug Harm Assessment by ChatGPT vs. Medical Students — Agreement, Divergence, and Cultural Context
COMPARISONA comparative analysis examined agreement between final-year medical students and ChatGPT in assessing the relative harm of various substances. Results showed strong agreement on the most harmful substances — those with well-established, high-confidence evidence bases — but significant divergence in assessments of moderately and minimally harmful drugs. The authors interpret these differences as evidence that drug harm perceptions are shaped not only by pharmacology and toxicity data but also by social and cultural context, local patterns of use, clinical exposure, and the specifics of medical training. ChatGPT's assessments may reflect the composition and cultural weighting of its training data, which may not align with regional clinical experience or contested evidence landscapes. The study highlights a specific limitation of AI in evidence-sparse or contested domains: when the underlying evidence base is limited, difficult to trace, or subject to ongoing scientific debate, AI outputs may be less reliable and lower confidence. Medical students also demonstrated lower confidence in their own assessments of minimally harmful drugs, suggesting that uncertainty in this domain is inherent rather than a unique AI failure. The findings have implications for AI use in pharmacology education and clinical decision support, particularly in addiction medicine and harm reduction contexts where local and cultural factors are clinically significant. ---
Subscribe to read full wiki analysis
Full wiki content — including source provenance, contradiction tracking, related pages, and analyst notes — is available to Subscriber members.
View Plans & SubscribeNew users get a 7-day trial — no credit card required.