I've never specialized in AI but from my experience on game teams most of the challenges revolve around making the AI 'fun' to play against, which is not genera

I've never specialized in AI but from my experience on game teams most of the challenges revolve around making the AI 'fun' to play against, which is not generally very well defined. It's also about taking feedback from designers and playtesters and figuring out how to adjust the AI to address that feedback. Machine learning techniques aren't really used much in game AI, partly due to these aspects. It's not necessarily obvious how to adjust some black box ML model in response to feedback from playtesters on balance issues or from designers on making the AI more or less 'agressive' or other qualitative feedback. Usually it's more a collection of rules and heuristics with various parameters that can be adjusted to tune the behavior and allow for different difficulty levels etc.