An Evaluation of Visual Embodiment for Voice Assistants on Smart Displays

Abstract

Smart displays augment the concept of a smart home speaker with a touchscreen. Although the visual modality is added in this device variant, the virtual agent is still only represented through auditory output and remains invisible in most current products. We present an empirical study on the interaction of users with a smart display on which the agent is embodied with a humanoid representation. Three different conditions are compared in a between-group experiment: no agent embodiment, a digitally rendered character, and a photorealistic representation performed by a human actress. Our quantitative data do not indicate that agent visualization on a smart display affects the user experience significantly. On the other hand, our qualitative findings revealed differentiated perspectives by the users. We discuss potentials and challenges of embodying agents on smart displays, reflect on their continuous on-screen presence, present user considerations on their appearance, and how the visualization influenced the politeness of the users.