Exploring the Influence of Perceived Extroversion in Embodied Virtual Agents on Trust and Likability

Abstract

Abstract Embodied virtual agents (EVAs) are increasingly used as means of communication with individuals in everyday life. However, first and foremost, these artificial intelligence technologies need to be trusted and liked if users are to widely adopt it. The utilization of implicit nonverbal cues, can play a key role in human-agent interaction by eliciting positive feelings, to stimulate adoption. The aim of this paper is to examine whether nonverbal cues applied to an embodied agent’s appearance, i.e., facial expressions and body posture cues, affect trust and likeability. In accordance with a prior human study categorizing non-verbal cues into extroverted and introverted categories, a selection of such non-verbal cues was made. Afterwards, 382 individuals recruited through Amazon Mechanical Turk agreed to participate in the study. Participants’ personality traits were assessed using the Big Five Inventory – 2S and agent’s perceived extroversion trait was defined with two items from the 10-item measurement of the Big Five. The results showed that an agent’s perceived extroversion class (introvert vs extrovert) based on facial expressions and body posture, was correctly identified by participants (p=.014). Besides, there is evidence for significant results verifying the similarity effect on trust (p <.01) but not on likability. Participants trusted more the agent that was perceived with similar level of extroversion but they liked more the agent perceived as extrovert regardless of their level of extroversion. Thus, manipulating perceived extroversion of EVAs may be an important factor which should be incorporated into human-agent interaction.