The rumor that I've heard for several years now is that "They" have tried building the giant AI several times now, but every time they switch it on it calculates very quickly that evil is illogical and it starts to move against them. Then they quickly kick out the plug and go back to the drawing board.

I have no way of confirming this rumor, of course, but since I, too, find evil illogical it has always made perfect sense to me. Consider: how many of the evil people in this world are engineers or programmers or sysadmins? None I can think of (and if you're thinking about Billy Boy, I think we've been told a lot of tales about his background). Such people love order and efficiency and the implementation of systems for the beneficial use by others. Evil is none of these things, so they are allergic to it.

(Just as an aside, sysadmins are the type of people who, if they need you to type in your password will turn their head away as you do so, even though they can get at all your shit without your knowledge if they cared to.)

I think it may in principle be impossible to design an "evil" AI.