Validating Affective Agent Ecosystems

Non-player characters, or agents, in video games have not advanced at the same pace as computer graphics. Newer games have become more aesthetically pleasing and game worlds have grown in size and detail, but the agents living in them are often perceived as stupid and predictable by players. Enriching these agents with emotional capabilities is one way to make them more believable and engaging. Computational models of emotion can give agents autonomous emotional behaviour based on events in the game world. Nonetheless, agents with such models are rare in modern games as studios opt for simpler hard-coded solutions. We seek to formally answer the question why video games lack emotional agents. We believe lack of control over these agents to be a key factor. With autonomous emotional reasoning it becomes hard to predict their behaviour, as many things influence their emotional state. This is particularly true for large game worlds with many agents and plethora of possible situations. Results from a questionnaire on the subject of emotional agents in games, sent out to game designers in the gaming industry, showed that there is little pressure from players and publishers to include these agents. The recipients said that it is expensive to include them and that there are more cost effective ways to enhance the gaming experience for players. Their responses indicated that it is hard to incorporate emotional agents in the game design and difficult to test their behaviour. We took a first step towards solving this problem by designing and implementing a prototype version of a general purpose testing tool for emotional agents. Our tool was designed to be emotion engine independent and generic. It receives data, logged during a game session, from web services and stores it in a database. The data can be queried and visualized from a web-based front-end by interacting with several data visualization. Users can look at high-level map visualizations and spot something interesting and then zoom in on that part to inspect it further using low-level timelines. We validated the tool by conducting two rounds of user tests with a pilot group consisting of three individuals with industry experience. The results suggest that such a tool can be useful to debug and test the behaviour of emotional agents and help motivate game studios to include them in their games.Media and Knowledge EngineeringComputer ScienceElectrical Engineering, Mathematics and Computer Scienc