Current AI/LLMs are just very complex probability matching with some frills to make it work. Our brains may be something like that too, as chemical and electrical signals can be reduced to math. It’s all math.
What you described before is pathing, and that can be a simple routine or very complicated and breakable, depending on the needs of the game. The really sophisticated ones would even chart the player’s behavior and react or plan a path based on past actions (even on some C-64 games, which is impressive). There was one karate game where (subjectively and not well tested) if you let it run the opening demo or played it a while, the game’s character got better. And it wasn’t just a higher level thing, you could tell (again, just a feeling) that it started to anticipate your usual moves.
Instant gratification culture. Got to have things faster and faster.