Study Finds Safety Gaps in AI Medical Advice
Click to Enlarge
A new study, First, Do NOHARM, evaluated AI-generated recommendations across 1,100 clinical consultation scenarios and found potentially severe errors in 2.9% to 24.6% of cases, depending on the AI system. More than 80% of severe errors involved omissions of important clinical recommendations.
The study also found that AI assistance improved physician performance, highlighting both AI’s potential benefits and risks in clinical decision-making. The findings underscore the importance of clinical oversight and evaluating AI tools for patient safety.
First, Do NOHARM: A Medical Safety Benchmark and Randomized Study of Physician and AI Teaming on Clinical Consultations (Wu et al., arXiv, 7/13).
