Yeah given the tree of possible moves with my game that would turn into a virtually infinite number of games very quickly. Some kind of directed A* search is probably needed, and that is gonna be tough. Either that or hand-craft some AI behaviours that make some sense, but probably aren't all that hard to beat. As I said, ideally I'd have some kind of reinforcement ML expertise on my team and the AI could be trained to make good moves through the whole reinforcement neural net magic. :-)