Intimacy perception : Does the artificial or human nature of the interlocutor matter?

Professional embodied conversational agents (ECAs) are increasingly present in society. Beyond these virtual agents' professional expertise, they must have a social dimension to build long-term relationships with users. Interpersonal intimacy is at the core of the most gratifying social exchanges, and we claim that it could constitute a remarkable means to reinforce the social dimension of ECAs. This paper presents an experimental study that compares the perception of intimacy in human and human-agent interactions. Are nature and social expressiveness critical factors for the perception of intimacy? To answer this question, we created two corpora of videos showing a human or a virtual tourism information (TI) counselor using or not using intimate behaviors in the course of an interaction with a human tourist. Using the Virtual Intimacy Scale (VIS) designed in previous work, we asked observers to judge the intimacy level of the TI counselor during the interaction. We demonstrate that intimate behaviors expressed by the virtual counselor and the human counselor are equally perceived by participants, which supposes a common model of intimacy for human and human-agent dyads. Participants perceive the human counselor less intimate than the virtual counselor when having non intimate behaviors. We posit that humans and agents elicit different expectations regarding social abilities, which directly affect their perceived level of intimacy. Our work raises questions regarding the expression of social behaviors and questions social representations.