Player Modeling for Adaptive Games

w en we say that a game has "good AI," we typically mean that the characters in the game exhibit consistent and realistic behavior, reacting appropriately to the actions of the player and other characters. For certain genres of games-for example, first-person shooters and real-time strategy game~-~~good AI" also refers to the ability of the game to challenge the player on a tactical and strategic level. While these are certainly worthy and important goals to strive for in your game, they tend to overshadow a third, seldom-mentioned component of good game AI: the capacity to adapt over time to the quirks and habits of a particular player. An adaptive AI can drastically increase the replayability of your game, and make the game experience much more intense and personalized for the players. Nonetheless , most games today are limited to what might be called "manually adaptive AI"; in other words, the game provides difficulty sliders and configuration parameters that allow the player to directly control how game characters act at some very coarse level. While this isn't a great solution, it certainly seems much less daunting than introducing costly (and often finicky) machine-learning algorithms into your game. Although many learning algorithms are computationally expensive enough to make game programmers run screaming, there are some approaches that are relatively lightweight, simple, and flexible enough to be useful. One in particular is a technique we call player modeling, borrowed from the similar notion of "student modeling in intelligent tutoring system research. The basic idea is simple: the game maintains a profile of each player that captures the skills, weaknesses, preferences, and other characteristics of that player. This model is updated by the game as it interacts with the player. In turn, the game AI can query the player model to determine how best to adapt its behavior to that particular player, such as by asking which of several possible tactics will be most challenging to the player. Using player modeling, a game's A1 can adapt both during the course of a single play as well as over multiple sessions, resulting in a computer opponent that changes and evolves with time to suit the player.