A Two-Study Approach to Explore the Effect of User Characteristics on Users’ Perception and Evaluation of a Virtual Assistant’s Appearance

Abstract

This research investigates the effect of different user characteristics on the perception and evaluation of an agent's appearance variables. Therefore, two different experiments have been conducted. In a 3 × 3 × 5 within-subjects design (Study 1; N = 59), three different target groups (students, elderly, and cognitively impaired people) evaluated 30 different agent appearances that varied in species (human, animal, and robot) and realism (high detail, low detail, stylized shades, stylized proportion, and stylized shade with stylized proportion). Study 2 (N = 792) focused on the effect of moderating variables regarding the same appearance variables and aims to supplement findings of Study 1 based on a 3 × 5 between-subjects design. Results showed effects of species and realism on person perception, users' liking, and using intention. In a direct comparison, a higher degree of realism was perceived as more positive, while those effects were not replicated in Study 2. Further on, a majority evaluated nonhumanoid agents more positively. Since no interaction effects of species and realism have been found, the effects of stylization seem to equally influence the perception for all kind of species. Moreover, the importance of the target group's preference was demonstrated, since differences with regard to the appearance evaluation were found.

Age-Related Differences in the Evaluation of a Virtual Health Agent’s Appearance and Embodiment in a Health-Related Interaction: Experimental Lab Study

“…Because of the Wizard-of-Oz setting, the agent responded with higher accuracy than current state-of-the-art agents, which might affect the evaluation and responses toward the agent. Although the quick responses of the wizard might match the expectations participants had of the humanoid agent, when these expectations are not fulfilled in real application, people might lose interest to interact with the agent [14]. Thus, further long-term and field studies are needed to support the current findings and to derive better insights for possible design guidelines.…”

“…As virtual assistance is applied in health-related domains in this research, the results of Straßmann and Krämer [13] might be better applicable. Further research [14] complements the assumptions and demonstrates that the species is more important in the evaluation process of a virtual agent's appearance for seniors and that seniors evaluate machine-like agents less positively than humanoid ones. Based on these features, for seniors, no differences between different degrees of realism are assumed, whereas they are expected to evaluate a humanoid agent more positively.…”

“…Concerning the interaction effects of appearance variables and user age, several target group-specific differences in participants' evaluations were assumed. Because there is scant prior research testing the preferences of senior users with regard to an agent's appearance, according to the evidence of prior research [13,14], it was assumed that seniors prefer a humanoid appearance over a machine-like appearance (H3). Moreover, with reference to the statements in qualitative interviews [13], where seniors stated that they preferred an embodied agent because an interaction with it is more familiar and that they would like to have an interlocutor who can be addressed (eg, looking in the eyes) during the interaction, it was assumed that seniors evaluate an embodied agent more positively than a nonembodied agent (H4).…”

“…However, the agent had the same behavior under all conditions, so this cannot explain the differences between the two target groups. It was shown that seniors rely more on the species of the agent and that realism is not a crucial variable for them [14]. Therefore, the differences in perceived realism might have been positive side effects of the generally more positive perception of Billie.…”

Virtual Agent Representation for Critical Transactions

“…In [15], for instance, species and realism are identified as the key variables of agent appearance. The research indicates different species (human, animal, robots, objects, and mystical creatures) and realism levels (stylization, resolution, and detailedness) which, in later work by the same authors, was demonstrated to have an impact on older people's willingness to interact with an agent [16,17].…”