preparing for AI that may never be invented...
I think it's a mistake to let the zealotry of the singularity weirdos cast doubt on everything that they're predicting; yes, they've got some pseudo-religious delusions about what AI will do for us, but I don't think they're far off when they say that it's coming, fairly soon, and there's nothing anyone can do about it.
There are some very clear numerical reasons that AI has failed thus far (we simply don't have the computing power right now - people hoped that we were clever enough to find algorithmic workarounds, but apparently we weren't), and whatever happens in the actual field of AI research, we're all but guaranteed to succeed at brute force AI some time over the next 50 years or so. The only real question is whether someone is able to figure out a hack that makes brute force unnecessary and bumps the timetable forward, but as I mentioned above, so far nobody has, so we may just have to wait for full brain simulation to be feasible.
Apart from that, I'd agree with you: with a few notable exceptions, most Singularity inspired "work" has been very low ROI, amounting to little more than speculation about the future and how great it will be rather than actually doing something to ensure a good outcome. There also seems to be a common delusion (again, with a few important exceptions) that AI will automatically be friendly without us needing to spend significant efforts towards that goal, which is very misguided.
...especially since a lot of the best funded and most focused AI research is taking place in China right now. In the rest of the world the past failures of AI have made people assume that it's a fool's errand, so there's not a lot of money going towards it, which could be a real disaster - regardless of how friendly AI can be made in principle, it's pretty much assured that if the Chinese government gets there first, we're going to see some pretty nasty stuff (though an argument could be made that once any government gets there we're all screwed).