A recent study from Western University's Schulich School of Medicine and Dentistry reveals that despite its training on extensive datasets, ChatGPT can fail to diagnose human ailments effectively. According to the study published in PLOS ONE, the research tested the popular language model on 150 case studies. The study involved feeding detailed patient data to ChatGPT 3.5, including medical history, lab results, and exam findings—and evaluating its diagnostic accuracy and reasoning.
The findings indicate that ChatGPT provided the correct diagnosis only 49% of the time. While the AI model performed well in explaining its diagnostic process and ruling out potential conditions, it could not diagnose the patient’s condition accurately and consistently. This suggests that although ChatGPT's explanatory capabilities could benefit medical students, its current diagnostic performance is…
