Computational Influence for Training and Entertainment

2) a set of abstract drama manager; 3) a model of player response to drama manager actions; and 4) an author-specified evaluation function. The drama manager's task is to select appropriate actions given the player model and the current sequence of plot events. In its original formulation, the DODM drama manager selected actions by performing a variant of expectimax game tree search. After its revival, this approach was shown not to scale to larger narrative domains (Nelson and Mateas 2005). In order to improve the scalability of DODM, I helped develop a reinforcement learning-based approach to replace search (Nelson et al. 2006). This new technique scales significantly better but it does have a few issues of its own. In order to achieve good results, variants of narrative domains where the drama manager was given synthetic causers (actions that are highly manipulative to the player) were used. As a result, the drama manager seemed to effectively cause a narrow set of narrative ex...

• Type: Conference
• Year: 2008
• Where: AAAI
• Authors: David L. Roberts
• Keywords: AAAI 2008, Drama Manager, Drama Manager Actions, Drama Manager Task, Intelligent Agents