In any game, regardless of it's revenue, it is the direct impact of AI that makes it possible to make any revenue at all. Beginning our journey from Quake and Unreal games we've noticed the major shift of frags (aka bits) from simply following you to squatting and building tactics.
However, what is more noticeable and important in gaming AI is the speech and pattern recognition. We say, it is possible to raycast object by it's shapes and recognize certain values of it, both in Unity and Unreal engines, but how about shifting a bit further?
By memorizing a player AI with quantity and quality (color, personal habit, etc.). That needs to set up a whole theory of probability in use. Or does it? A lot of research has gone up to the point of automated differentiation, Markov Modes, automated Calculus, which basically combine number theories and more and less computation.
It is possible to set variables of imaginary number probability in complex number theory using Bayesian type inference.
Which makes basic and primitive calculation of AI used in games, to be predictable and (what is more important),