I have heard that real medicine is really all about diagnosis. Identifying the problem. Perhaps AI is not smart, not intelligent. Perhaps AI will never be able to diagnose as well as a human.
Fine.
How about surgery? How about dosing prescriptions? I think it is inarguable that machines will do these things better than humans. So, humans will retain a role in medicine (diagnosis, etc.) but machines will take some medical jobs away from humans.
My basic argument is never metaphysical (”AI will be smarter than humans” or however one wants to phrase it) my basic argument is always economic — “There are a lot of humans who will not have skills or knowledge that remains useful within the national economy in the 21st century. There are machines that will perform tasks which humans used to perform. Will we have to adjust to a reality where a big percentage of humans no longer contribute to the economy.”
Perhaps medicine is a good field for humans. But how about banking? Insurance? Architecture? Logistics? There are some jobs that will no longer be suitable for humans, and it has nothing to do with “machine sentience” or anything like that.
I have read studies claiming AI makes the correct diagnosis 100% of the time, and when you add a human MD co-pilot the accuracy drops to 23%.
This represents diagnosis as a series of boxes to check or uncheck, according to an algorithm.
There is an old joke, “the operation was a success, but the patient died”.
Algorithmic medicine is fraught with problems. The post-1998 Quality movement started with the proposition that “you can’t manage what you can’t measure”.
So, “measuring” diagnosis has mostly come down to “were all the steps followed, and do the steps, if all followed, point to a diagnosis?”
We did a study in 2018 where we evaluated the first 100 charts with a discharge diagnosis of “pneumonia”, and we confirmed the diagnosis in 38. Now, in that study setting, AI would have confirmed 100/100, because all the steps were followed.
This is starting to cause real problems.