Sufficiently intelligent machines, like sufficiently intelligent humans, can and should be trained to behave as required and can and should be held accountable when they don't.