Those are all good points, Geko. It is assuming that AI will be able to surpass humans, but that is a reasonable assumption. Computers have been surpassing humans at specific tasks for the last few decades, it’s reasonable to assume that a general purpose AI would surpass human thinking for any task.
If your really interested in this stuff, at least read Bill Joy’s famous article Why the Future Doesn’t Need Us
BTW an AI with emotion is actually a more terrifying scenario. I would rather be in the Terminator Movies than in “I Have No Mouth and I Must Scream”.

Having a powerful AI that’s indifferent towards humankind is almost as bad as having a powerful AI that’s malevolent. It might, for instance, destroy all oxygen producing plants in order to reduce the rate of decay of its chips and just not care about how that will affect humans.

Caio, you are right that my example requires that the AI is motivated by self-preservation. It’s are reasonably assumption since most other goals are predicated on continued existence, but I’ll grant that it might not always be the case. My point is still that an apathetic AI can still be dangerous.
RSIxidor, beyond what the programmers intended, you also have to consider any unindented consequences. For example, suppose google makes an AI and give it the tasks of “answer all questions that people ask you”. A sufficiently advanced AI might decide that the best way to achieve that goal is to make sure no one asks any questions. Asimov and Clark wrote a lot of stories about situations like that (and were much better writers than Herbert’s corpse fucking son).
With genetic algorithms, emergent behavior and cascading development, the situation would get far beyond the control of the initial programmers very quickly.

@asdf
“Fire Upon the Deep” by Vernor Vinge
“Singularity Sky” by Charles Stross
“Blood Music” by Greg Bear
All of the “Culture” books by Ian M Banks, “Consider Phlebas” is the first one, but they don’t have to be read in any order.