AI Chatbots Show Promise in Medicine, But Critical Errors Raise Concerns About Over-Reliance
SANTIAGO, Chile – Artificial intelligence chatbots are demonstrating an ability to pass medical licensing exams, but significant errors in key areas like internal medicine are fueling debate about the risks of prematurely integrating AI into critical healthcare roles. While exceeding average approval scores on Brazil’s National Single Exam of Medicine Knowlege (EUNACOM) - a key requirement for practicing in the public health system – the chatbots’ performance reveals that medicine demands more than just facts recall.
The findings, stemming from a recent study evaluating three ChatGPT versions, highlight the limitations of AI in contextualizing complex medical scenarios. One version excelled in psychiatry,while two showed aptitude in surgery,but all stumbled in internal medicine. As one of the study’s authors noted, “that medicine is much more than handling information, it is necessary to contextualize and consider multiple aspects.” This underscores the potential for serious consequences if AI-driven systems are relied upon without human oversight,a concern echoed by research published in The New England Journal of Medicine which found AI can generate erroneous or confusing medical advice.
Despite these risks, experts emphasize that AI is already being incorporated into medicine to optimize consultations, improve exam analysis, facilitate diagnoses, and enable remote patient follow-up. The challenge lies in finding the right balance between leveraging AI’s capabilities and preserving the critical thinking and nuanced judgment of human physicians.
C. González
El Mercurio (Chile) – GDA
