@404... Re: Ever get the idea we're doing this all wrong?
You have two issues...
1) Coming to a conclusion and a diagnosis based on a patient interview...
2) Coming to a conclusion and diagnosis based on diagnostic imaging.
Its not just asking the better question, but also how you interpret the answer.
To give you an example...
My father was a Radiologist who, back in the day, actually had patient contact. ( Radiologists today, for the most part, don't really interact with the patient, except in a couple of procedures. )
He used to tell me that when he looked at a patient, he sometimes had to ask if they were a smoker or a drinker. If they drank, he would ask how much... e.g. how many beers in a sitting or per day.
Then based on what he saw, and how they answered... he would come up with a swag. Why? Because a good doctor knows when a patient is lying or not being completely truthful.
So imagine when you have an AI attempt to do the same thing... not so easy.
But when you get to diagnostic imaging processing, you have more potential for AI to help augment Radiologists. (e.g. CAD systems for mamograms) This is where both the AI and the Radiologist are on par in that it takes training and attention to detail in order to find an artifact or something that would require a biopsy.
And of course the X-rays aren't just binary... you may find calcification / remodeling from earlier breaks and healing. How does the AI interpret that? Again, it takes a lot of data and experience to be able to interpret the data, so the AI systems can improve... IMHO... I'd still trust the human over the machine.