I wouldn't be so skeptical @littlenewthings :)

> when AI really starts to learn and "think for itself", will it start to judge (without compassion - the lack of human element) the human race where the stories of the dreaded sci-fi horrors will come to pass?

In my opinion developers will make sure that AI has implemented human values (or at least three laws of robotics), before they will even start some learning process. There is also a possibility that this process will snowball out of control, but to be honest we currently know really little about the way it would work. This is also the most pessimistic and rather unlikely scenario.