I can see this going in both directions, but I also think our idea of what AI will be, is very limited.

Personally, right now, I don't think the idea of "killer robots" or somthing similar is very likely. Unlike a lot of sci-fi (as much as I love the re-imagined Battlestar Galactica), I think AI will exist online in some way. With a fully functioning AI, I imagine the first order of business would be for it to secure its existence, so while we would see physcial robots building the infrastructure, the "brain" would exist online.

The question is then whether or not AI would consider human beings a threat, or not. Would AI be aware of us anymore than we are aware of insects, or would it be a collaboration between to species, to use that word?

I think the more interesting questions could be:
a) Why would AI find human beings threatening, and are those assumptions well founded?

b) How would AI take action if it felt threatened by human beings? Close of off the web? Eliminate money and wealth? Outright war?

c) Would AI function as individuals, or as a large hive mind, or several hive minds in various networks? If it works as a hive mind, would it then eliminate unecessary drains (to the AI) on power and storage in order to sustain itself, and in the process affect human beings (again, why deal with money if you're a hive mind, and from your own AI perspective own everything collectively. Why not use that space for other purposes? What about the environment? Is the safety of the planet of interest to an AI, and if so, will it consider it at the same level as human beings?)?

Probably lots more to ponder here. AI is incredibly intersting, and as far as use goes, I think that's a term that is only useful as long as we can control AI. The second we develop AI that is truly self-sufficient, the game will change.

Not sure if this was answer to your question, but it did get a lot of thoughts going on my end.