I read the book Complications by Atul Gawande a while back and it touched on this issue. He mentioned how a computer was more accurate at detecting heart attacks than an experience doctor. He also talked about how if the computer is better than the doctor at reading things like these, it really doesn't make much sense for the doctor to have to evaluate/approve the results. Kind of reminds me of James Simons' thought process on quantitative trading at RenTech.

Sorry about the lack of technical knowledge into the CS stuff, first post here and I haven't really put in the time to learn about CS and AI yet.