Human-like behaviors are an important factor in achieving entertaining computer players. So far, the target of human-like behavior has been focused on actions in a game with the goal of winning. However, human behaviors might also be performed with other purposes not directly related to the game’s main objective. For example, in FPS games, some human players create illustrations or graffiti with a weapon (i.e., gun). In co-operative online FPS, when chat is not allowed in-game, some players shoot the nearest wall to warn an ally about danger. This kind of action for an indirect purpose is hard to reproduce with a computer player, but it is very important to simulate human behavior and to entertain human players. In this article, we present a survey of the possible actions in a game that are not directly related to the game’s main objective. Study cases of these behaviors are collected and classified by task and intention (i.e., warning, notification, provocation, greeting, expressing empathy, showing off, and self-satisfaction) and we discuss the possibility of reproducing such actions with a computer player. Furthermore, we show in experiments with multi-agent Q-learning that such actions with another purpose can emerge naturally.