An evenly matched opponent AI in Turn-based Strategy games

Recent researches in turn-based strategy (TBS) games point to the development of artificial intelligence to beat players. Guaranteeing that opponents will be beaten, however, is not the focus of commercial Turn-based Strategy games. For commercial games, if human players do not win, most of them quit the game. This can result in horrific future sales. Therefore, keeping player engage in the game is much more important. This paper presents an artificial player that learns to adjust its skills to match a player it is playing against, without displaying unreasonable moves or performing sudden changes in its difficulty level. A Final Fantasy Tactics-like game is used in our experiment. We introduce evaluation functions for calculating the score from each unit´s action. By evaluating a human player´s score, our artificial player can estimate his skill and play at the same level or adjust the level´s difficulty based on the player´s skill throughout the game.