You are here

Socially Intelligent Agents in Serious Gaming Environments

IUALL (Interaction for Universal Access)

In this WP we investigate the use of interactive storytelling techniques in serious games used for training purposes. Our focus is on providing artificial agents (“non-player characters” in the game) with social intelligence to support training. That is, the agents need to be able to send social signals and react to social signals sent by other characters in the game. These other characters can be either other non-player characters, or (avatars controlled by) human trainee(s).

Social signals refer to aspects of social relationships between people that interact: agreeableness, dominancy, distance, acceptance, etc. Mostly people are not aware of how social signals affect their behavior or judgment of a situation; they are recognized on a subconscious level. Social signals are not deliberately sent out as a result of some cognitive process. In this WP we focus on social signals that are related to people’s interpersonal relationships or attitude, for example interest or aggression.

The main objective of the WP is to model the interpretation and generation of social signals by artificial agents to support training. An iterative approach will be followed to develop increasingly sophisticated models of social behavior interpretation/generation for use by socially intelligent agents. This involves investigating the following questions about social signals presented by avatars (controlled by artificial agents) in a specific scene in a serious game used in a tutoring scenario.

  1. How can believable social behavior be generated by avatars in a serious game?
  2. How can social behavior by artificial agents be used to support training in serious games?
  3. Do the social signals of the agents contribute to successful identification of the situation by the trainee?
  4. How should feedback to the human trainee’s social behavior be presented?

Another important challenge is to devise techniques to keep the game optimally adaptive to the user’s actions while still incorporating specific training challenges. This will require the agents to have some meta-level understanding of the training goals, enabling them to reason about the best actions to take in order to fulfill the game purpose, while still remaining believable in their interactions with the trainee(s) and each other. The results of the WP should be applicable in different types of training games. These could be social behavior training games, where trainees learn to react appropriately to social signals by others, but they could also be games with different training goals in which social signals nevertheless play an important role (for example to establish interpersonal relationships between players).

For our work in this WP we will build on our experience in building artificial agents for interactive storytelling. There is a close link between interactive digital storytelling and serious gaming. Interactive storytelling is aimed at providing human users with the dramatic experience of being a character in a story that unfolds based on the actions they take in a simulated story world. Serious gaming has the same elements of human players acting in a simulated world, in which they can try out the consequences of their actions. The main difference between the two is that serious gaming aims at providing the player with an educational rather than a dramatic experience.

Emergent narrative is an interactive storytelling approach that lends itself well to serious gaming applications. In this approach, autonomous intelligent agents create a story in a simulated environment, in interaction with a human user or users. The story is not given in advance but emerges from the interactions between the human user(s) and the agents. This means it will develop differently depending on the players’ interaction with other characters in the story (played by other humans or by artificial agents). This enables human users to directly experience the effects of their own actions, including their social behaviors. Note that in this approach the human user has full agency, which goes far beyond making choices at selected points in the story to determine its outcome (as in simple “branching narrative” approaches). The emergent narrative approach is used in the Virtual Storyteller system developed at the University of Twente, which will form the starting point for the research in this WP. So far, the Virtual Storyteller has been oriented toward entertainment, but here we will investigate its use for serious gaming and in particular, for serious training games.

WP Leader: