Are aggressive agents as scary as aggressive humans?

Abstract

Most research into intelligent virtual agents focuses on agents with a positive stance towards the user. Nevertheless, the development of virtual agents that show aggressive behavior may also be interesting for a range of application domains, varying from aggression de-escalation training to anti-bullying education. However, ensuring that such aggressive agents achieve the desired effect is not easy, as they need to be believable in a number of aspects. In particular, they need to bring their human conversation partners into a serious state of anxiety. To investigate to what extent this can be achieved using state-of-the-art virtual agent technology, an experiment was performed in which the impact of an aggressive virtual agent was compared with that of an aggressive human. By randomly distributing a group of 28 participants over two conditions (virtual and human) and measuring their physiological and subjective emotional state before and after an aggressive outburst of their conversation partner, the difference between virtual and human aggression was studied. The results point out that both types of aggression induced a substantial stress response, but that the impact of the human aggression was higher than that of the virtual aggression.