A framework for emotional influences on problem solving by synthetic agents

We develop a framework for emotional modelling that is based on a plan selection architecture. The emotional module determines emotional state based on goal and plan outcomes. The first novel aspect of the framework is that the current emotional state in turn influences the expected utility computations of untried plans, affecting performance on the core problem solving task. A second novel aspect is the inclusion of plan-based behaviours that are triggered by the simulated emotional state and cause that state to dissipate. These behaviours compete for execution with the core problem solving plans. On abstract planning problems, agents modelled with these two emotion framework elements exhibited clearly different problem solving choices than agents modelled without them. Using narrative traces of concrete problem solving behaviour, a study with human subjects found that agents modelled with the emotional module are perceived differently than those that are not, on some personality traits.