Abstract
The ability to acknowledge social context is of importance to believable agents that wish to express human-like characteristics for applications ranging from digital entertainment to education. The traditional approach for incorporating social context into agents is for system builders to script the social relations with each individual agent. A major limitation of this scripting approach is its lack of flexibility and accessibility – such agents can hardly adapt the social context to possible changes in social relations, whereas scenario designers have difficulties to author and identify social relations in a macroscopic perspective. A central challenge for believable agents aware of dynamic social context is to establish social models that are readily understood by the agents as such by the authors, and able to undertake changes in runtime without the risk of falling apart. In this thesis, I integrate the form of ontology with the idea of social constraints to propose a social model that allows designers to author the social context in the global view, and allow agents to retrieve individual social relations. To that end, the thesis makes three central contributions. First, I design a social schema that identifies individual relations and institutional rules in terms of social constraints to permit changes in the social model. Second, I describe how these constraints can be incorporated into agent architecture in a way that guides the agent to generate social behaviors without strictly preventing agents from violating social constraints when necessary. Finally, I present a conflict detection algorithm that reasons about different social constraints to support a further process of decision making. I provide illustrations of deliberation processes to justify how this model can situate agents in the social context.