The effect of emotional expression in a digital human and gender on psychological and physiological outcomes in healthy adults

Participants were block-randomised by gender to interact with one of six versions of a digital human that varied in terms of their face type (no face/ neutral face/ emotional face) and voice type (neutral voice/ emotional voice): 1. No face, neutral voice 2. No face, emotional voice 3. Neutral face, neutral voice 4. Neutral face, emotional voice 5. Emotional face, neutral voice 6. Emotional face, emotional voice A digital human is a type of embodied conversational agent with a humanlike embodiment and animation (based on a real human), that includes artificial intelligence for emotional intelligence (e.g., classifiers of emotional expression in a users face). The digital human in this study was a mixed race, young adult female based on a real person that used a finite state conversation engine (i.e., pre-programmed language) and responded through speech using pre-recorded voice clips from the human model. The interaction involved completing the Relationship Closeness Induction Task (RCIT; Sedikides et al., 1999) with the digital human. The RCIT is a structured conversation task that involves reciprocal self-disclosure in response to 28 questions which gradually increase in intimacy (e.g., from "what is your name?" to "describe the last time you felt lonely"). The participant took turns at asking and answering personal questions from the RCIT with a digital human. The RCIT has been shown to reliably induce a moderate sense of closeness between human strangers in experimental psychology research, and it has been associated with improvements in wound healing (Robinson et al., 2013). The intervention took place on a laptop computer in a private clinic room with a researcher available (PhD student) in another room to seek help from if needed. Participants completed one 15-minute digital human interaction as part of a 60-minute appointment at the University of Auckland Clinical Research Centre. Audiovisual data were recorded while participants interacted with the digital human which was transcribed and analysed. This data indicated that all participants completed the interaction. The digital human was designed and built specifically for this study and is not readily available to the public.