There have been a couple of trials involving AIs and medicine. They seem to do well on exams, and when evaluating multiple choice input. But when it comes to practical clinical medicine, they fail miserably. A trial where AIs were presented by a typical clinical situation, where it has to interview patients doesn't work very well. One of the reasons is that patients typically doesn't know what's important, so the doctor has to ask for information. The best scored around 25%.
They work well in specific, limited areas, such as evaluating x-rays, where they can point to areas requiring a closer look. Another example is evaluating ECG's. But in a general doctor-patient situation, doctors still (like mr Segal?) thrash them.
pibbuR whose performance, given his long time away practical medicine, probably compares to those systems.
PS. Knowledge and experience is important. After being a GP for 3 years, I was a sensor in the final oral examination of med. students in Tromsø. One of the cases presented was a woman with suspected extrauterine pregnancy (the fetus developing outside the womb - no way the fetus could survive and very dangerous to the mother). While I found that the diagnosis was almost written on the patient, the student asked a lot of (irrelevant) questions without any direction or purpose. No doubt I did the same on my exams, but those years of experience certainly made me a better physician. As long as it lasted. DS.
wccftech.com
