Thinking more about the agent/actor/character distinction, and how this triadic structure of the simulated character makes interpretive readings of games that have them a problem. We don’t attribute agency to characters in traditional narratives (that is, we attribute the intention to act to an actor – his actions are always mimetic, and attributed to that of the character he is portraying. This isn’t the case with a simulated agent, whether that agent is in a game, as a simple bot, a non-player character, a monster or even a strategy-making AI.
My earlier post about the MASSIVE software being used to make the Lord of the Rings film suggests that this problem has gone “upstream” to film. The agent responds to that subset of game-space that are its percepts and generates action accordingly; the actor may be the interface component of the agent, a virtual player reaching for a gun or walking through a door or laying seige to a fortress; the character is the unifying logic of the executive function, to which the player attributes the intelligent pursuit of a goal described in some narrative aspect of the game or determined by the overall context of it (in a war-game, the desire to conquer; in a murder mystery, the desire to avoid detection.)
[On edit:] and it looks like Michael Mateas has been looking at these problems for a while now, under the rubric of “believable agents,” from an AI and IF perspective.