Comment by 827a

I think it comes down to how much data we're comfortable feeding an AI. If the AI has cameras and/or microphones in the room and the patient is directly talking to the AI: I strongly suspect AIs will always achieve better outcomes than humans. However, this kind of configuration will be viewed very negatively in a medical context for the foreseeable future; outside of limited contexts like "let me take a picture of that mole"; and hobbling the AI to only a text input (or dictated text by the doctor) muddies the waters on who is performing better. There's a lot of intuition in the diagnosis of something like "the location of the pain aligns with appendicitis, but they just aren't in enough pain" that cannot come through in just the textual representation of what is happening; you need to hear the person's voice and see how they're holding their body. AI can do that, but will we let it do that?