Artificial intelligence tools like ChatGPT aim to reduce clinician workload by triaging patients and providing diagnoses. However, a study by Harvard Medical School and Stanford University found that while these tools perform well on medical tests, they struggle in real-world scenarios. A new evaluation framework called CRAFT-MD was created to assess AI models’ ability to interact with patients naturally and make accurate diagnoses. The study suggests developers incorporate open-ended questions, interpret non-verbal cues, and integrate textual and non-textual data to improve AI models’ real-world performance. The framework could help optimize AI models for clinical use and enhance patient outcomes.
Source link
New test evaluates AI doctors’ real-world communication skills
