As someone who studied computer science... my two cents:
Diagnostic Radiology WILL NOT be the first to go. We can barely do face recognition. EKG analysis is far from perfect (and it's a LINE, not a CT/MRI). Taking a complicated 3d image and analyzing it in a classification algorithm is not trivial. It's extremely difficult
Alternatively... having someone check off a list of symptoms, in conjunction with vitals and a few numerical lab values is much easier to classify. Check out the diagnostic algorithm Watson... it's already being done
. So yeah: I would say ER could be automated to some extent but we'll always need physical exam. Things that are not acute that simply need a decision of: admit/not-admit could be handled by a computer since the decision and inputs are pretty simple. Also, social issues are also probably pretty hard to automize and we all know that they make up a significant portion of medicine.
At the end, I don't think this would be bad for physicians. Doctors would spend less time on "trivial cases" and more time on the 10-20% of things that don't follow the pattern. They would probably also have more time to talk to patients about social issues that often get missed due to time constraints.