Difficulty scaling of game AI

Abstract

“Difficulty scaling” is the automatic adaptation of a game, to adapt the challenge a game poses to a human player. In general, a game of which the challenge level matches the skill of the human player (i.e., an “even game”) is experienced as more entertaining than a game that is either too easy or too hard. In practice, when difficulty scaling is implemented in a game, it only adapts a few parameters. Even state-of-the-art games do not apply it to game AI, i.e., to the behaviour of computer-controlled opponents in a game. In prior work, we designed a novel online-learning technique called “dynamic scripting”, that is able to automatically optimise game AI during game-play. In the present paper, we research to what extent dynamic scripting can be used to adapt game AI in order to elicit an even game. We investigate three difficulty-scaling enhancements to the dynamic scripting technique, namely (1) high-fitness penalising, (2) weight clipping, and (3) top culling. Experimental results indicate that top culling is particularly successful in creating an even game. We conclude that dynamic scripting, using top culling, can enhance the entertainment value of games by scaling the difficulty level of the game AI to the playing skill of the human player.