AI's really come into their own in "expert" and related decision support systems where they act as prompters, reference engines, and data organizers for human experts. Back in the 80s some (such as CADUCEUS
) were developed for medical use where the system was designed to assist
doctors in coming up with a diagnosis. The tests using such systems showed improved diagnostic accuracy and fewer misdiagnoses when compared to purely doctor-generated diagnoses. Unfortunately, the medical profession of the time wasn't too happy about the idea that a machine might "second guess" them, and generally declined to become involved with, or support, what might have led to a major improvement in medical care.
There was an interesting discussion about this in David Gelernter's book Mirror Worlds
. Out of print, and slightly dated in places (it was written before the Internet took over our world), but still a book worth tracking down and reading.