I'm a game designer and developer with 11 years of gamedev experience and decent understanding of ML (couple of online courses, nothing serious). I don't see how using ML for AI would be benificial in any way for any games that I've worked on.
ML makes AI stronger and makes it easer for it to beat players, but as a game designer, it's never my goal. My goal is to make it lose in competitive, interesting, but most importantly, understandable way: I want players to gradually learn patterns behind AI behaviour, understand them and subvert them.
May be it's because my experience is more about simpler and more action-focused games, and in 4x straregies things are different, of course - it's just my two cents.