[BlueTemplar]

I'm not sure that it would be necessarily the doctor that would actually gets condemned in this case. Could as well / instead be the programmer(s) that made the AI.

But this is why the current crop of AIs (which Watson itself might NOT be part of) is problematic : they're too much black boxes.

So they directly clash with the laws that assume on one hand perfect transparency of the tools used, and on the other perfect responsibility of the people using them.

How long before the use itself of a neural network is deemed to have been illegal because it broke one of the laws mandating the explanation of the algorithm that has been used to make a decision to the person that this decision targeted ?

Eventually, way down the line, this might involve giving some kind of civil status to computer programs so they can actually be made responsible.