This, to me, is a greater risk than "evil AI." If humanity does cease to exist, it will be a slow, painless, barely observable extinction (perhaps even paraded as ascension) as we integrate with and are then replaced by the technology we've created.
I don't think this is entirely realistic, just because there's an anthropocentric bias in me that believes that humans are wonderfully useful and robust self-replicating machines, and that economically we're (1) fairly energy efficient and (2) valuable to "throw at problems" we don't want to mass produce specific silicon-based solutions to.
But I did want to thank you for pointing this out. I do think it's wonderfully naive we often ascribe human characteristics and fallacies to hypothetical general AI that we experiment with or build, as though distinctly human consciousness is somehow the epitome of achievement. It's not even in our best interest, economically or ethically, to strive for that.
Argument about conscious Google spam is quite unfair and unwarranted. In fact, if you actually consider what Metzinger has to say about necessary conditions for consciousness (for example in "Being No One" or "The Ego Tunnel"), then obviously Google spam is nothing like that.
Either way it seems to me that it may be better to err on the side of not making conscious AI in the first place rather than unintentionally making one with capacity to suffer. Especially that it may be non-obvious that it has capacity to experience pain. For example, we have already made this mistake by using curare alone as an anesthetic - presuming that a lack of automatic pain response, and ability to communicate feeling of pain means that one doesn't feel pain. That was a considerable misjudgment.
If that argument is taken seriously then it should already be starting to apply to things like the Google spam filter, but once you take it out of the space of hypothetical possibilities like that and apply it to real code it becomes obviously absurd. There is less than no reason to believe that the only possible artificial intelligence will be simply humans that live in machines and have all their needs and desires, and even such AIs that are nascent in the world today already have grotesquely different architectures than that, with little reason to believe making our spam filters or car navigation systems "more human" in the existential sense would even remotely improve them. Why would my car's navigation system even need to be able to represent "a sadness that I'm just a car navigational system instead of a Real Boy", let alone actually manifest it?
By far the greater risk in that world is the economy/universe optimizing out everything that makes humans human until the entire universe is simply a highly efficient philosophical zombie, rather than the risk that every last subsystem of everything in the world is secretly going to be a human-class homunculus trapped in some boring task.