Measuring Interpersonal Trust towards Virtual Humans with a Virtual Maze Paradigm

We used the combination of the investment game and the virtual maze paradigm to measure users' trust towards virtual humans. A) In the investment game, participants need to decide the number of tokens they want to invest in the trustee. B) In the virtual maze, participants must select one of the two doors to escape when entering a new room. The virtual human appears in the middle of the room. C) A participant interacts with the virtual environment in the virtual maze in an immersive VR experience.

Abstract-Virtual humans, including virtual agents and avatars, play an increasingly important role as VR technology advances. For example, virtual humans are used as digital bodies of users in social VR or as interfaces for AI assistants in online financing. Interpersonal trust is an essential prerequisite in real-life interactions, as well as in the virtual world. However, to date, there are no established interpersonal trust measurement tools specifically for virtual humans in virtual reality. This study fills the gap, by contributing a novel validated behavioural tool to measure interpersonal trust towards a specific virtual social interaction partner in social VR. This validated paradigm is inspired by a previously proposed virtual maze task that measures trust towards virtual characters. In the current study, a variant of this paradigm was implemented. The task of the users (the trustors) is to navigate through a maze in virtual reality, where they can interact with a virtual human (the trustee). They can choose to 1) ask for advice and 2) follow the advice from the virtual human if they want to. These measures served as behavioural measures of trust. We conducted a validation study with 70 participants in a between-subject design. The two conditions did not differ in the content of the advice but in the appearance, tone of voice and engagement of the trustees (alleged as avatars controlled by other participants). Results indicate that the experimental manipulation was successful, as participants rated the virtual human as more trustworthy in the trustworthy condition than in the untrustworthy condition. Importantly, this manipulation affected the trust behaviour of our participants, who, in the trustworthy condition, asked for advice more often and followed advice more often, indicating that the paradigm is sensitive to assessing interpersonal trust towards virtual humans. Thus, our paradigm can be used to measure differences in interpersonal trust towards virtual humans and may serve as a valuable research tool to study trust in virtual reality.