Hmm, ooli, this article seems really light. Did you link to the right thing or am I missing something? Either way, it reminds me of two things. Recently, NPR had an article about using AI in medicine and the challenges in ensuring such programs are safe, accurate, and transparent. I also listened to a really good interview the other week about medicine and AI, where researchers discovered that if given loose enough parameters, AI programs like to cheat to figure out results. The example they gave was a program that was supposed to interpret images to create a diagnosis would, if allowed access to additional patient data, use that to determine its results. I thought it was interesting, because on the one hand, the program wasn't working as intended, nut on the other hand, doctors often try to rely on as much data as possible to try and make diagnoses, so it would make sense that a self learning AI program would do the same if given the opportunity.
Everything I've read about Watson for the past two years indicates that Watson has been an almost total failure. It doesn't mean that AI is doomed. It probably means that health Care is hard and that we have a long way to go toward applying AI to many real world situations.