Empathic Virtual Agents

Within the EU Framework V project VICTEC we develop a running model of empathy that is based on both empirical results and a state-of-the-art general psychological theory and can be implemented in virtual agents. Empathy focuses on the ability of one person (the so called “observer”) to perceive, to identify and to deal with the emotions of a social partner (whether virtual or human agent; “target”). Most contemporary empathy researchers (e.g. [1] [4]) agree that two aspects concerning the result of an empathic process have to be distinguished: the cognitive and the affective aspect. Thus, one speaks of cognitive empathy when the outcome of an empathic process is that the observer knows what the target feels. Affective empathy is spoken of when the observer feels something due to the perception of the target. The aim of the work was to create a running model that allows reproducing those empathic processes with the help of general psychological architecture, the PSI model [2] [3]. One of the key aspects of the PSI model is a theory on emotion which defines emotions as parameters of cognitive processes like resolution or flexibility. To evaluate the behaviour of the running model two basic approaches were pursued. The first approach focuses on the system’s behaviour which should be stable and plausible. Therefore, an analysis of the behaviour of the key variables (cognitive and affective empathy, need for affiliation and certainty, observer’s care and observer’s external behaviour) under certain conditions (systemic variation of input variables like similarity of observer and target) was conducted. This analysis revealed stable behaviour. Plausibility of the system was rated by psychology professionals. After explaining the model to the experts, it was their task to state all implausible variables and relations that they were able to recognize. No major implausible constituents of the system were detected by the expert evaluators. The main part of the evaluation focuses on the verification of the system by comparing the system’s behaviour to empirical results. Encouraging results were received from this type of evaluation.