I actually tried this with someone who is currently attending med school and got him to quiz ChatGPT on diagnosing common and esoteric ailments. According to him, ChatGPT assumed to know the answer when in most cases a good doctor would have asked questions to clarify or be specific to make an accurate diagnosis. In a few of those instances, it misdiagnosed when it could have easily gotten it right just by asking one or two simple follow up questions.
I have had this experience with incurious human doctors as well. We do not need mechanical diagnosticians. We need minds that can think through problems using a mix of logic and intuition built through intensive study and experience.
