I definitely understand the concerns about the overlap between CrossValidated, Data Science, and this site. What we need to do, to help the site get more traction, is to define that boundary in a useful way. At a high level, it wouldn't make sense to reject a site about statistics because a perfectly good mathematics site already existed. Statistics has different goals, conventions, notation, and concerns--even though it's almost all mathematics.
I'd argue that the failures of the previous sites were more a question of timing than content. Serious interest in AI is on the horizon again, very recently, precisely because of advances in ML. That doesn't mean, however, that AI proper is the same thing as ML, or needs to be focused on implementation issues. There's a large amount of theory that isn't necessarily data science, either.
We went through some of the same growing pains on Signal Processing. The approach we took there (and I'm not saying it's the right approach for AI), was to concentrate mostly on theory, and avoid implementation details. It's something that didn't exist, and it gave us a way to attract experts who weren't programmers.
Explicitly making the history of AI on-topic, however technical, might be a good starting point to help clarify what a site dedicated to AI can add to the SE network. I'm not saying that it's necessarily off-topic now, but given that MathJax isn't even enabled yet, there's currently a strong bias toward strictly non-technical questions.
I think the AIXI agent is a good example to begin discussing these issues. It's heavily mathematical, based on reinforcement learning, inspired by statistical reasoning (ala. Solomonoff's Universal Prior), and uses non-computable concepts (i.e. Kolomogorov Complexity). So, there's a potential overlap with any number of fields, but really it's proper AGI. It's a much more practical definition of intelligence than, say, the Turing Test--precisely because it's defined mathematically. At the very least, it seems like definitions of intelligence should be on-topic, and we need math for those.
It might warrant a completely separate meta question, but I'll offer one thought on how to help clarify the scope of the site (in addition to including AI history). Let's start with Peter Norvig's definition of AI (from Frank Dernoncort's slides):
We think of AI as understanding the world and deciding how to make
good decisions. Dealing with uncertainty but still being able to
make good decisions is what separates AI from the rest of
computer science.
Any discussion of decision making under uncertainty will almost necessarily involve probability and statistics. However, the challenges involved in automating those decisions effectively, in my opinion, are the domain of Artificial Intelligence, whether general or specialized. That definition also includes all of the potential social issues.
