Have We Crossed the Uncanny Valley? Understanding Affinity, Trustworthiness, and Preference for Realistic Digital Humans in Immersive Environments

Developers have long strived to create virtual avatars that are more realistic because they are believed to be preferred over less realistic avatars. However, an uncanny valley exists in which avatars trigger aversion when they are almost but not quite realistic. We used a field study to investigate whether users had different affinity, trustworthiness, and preferences for avatars with two levels of realism, one that was close to human-realistic and one a cartoon caricature. We observed behavior, conducted one-on-one interviews, and collected survey data from SIGGRAPH conference attendees who either participated in a live discussion session between two avatars in a VR environment, or observed it via 3D VR headsets or on a large screen 2D video display. Eighteen sessions were conducted over four days, with the same person animating the human-realistic avatar and different guests animating the caricature avatars. The guests who interacted with the human-realistic avatar had a positive experience in the VR environment. The observers had positive evaluations of both avatars while acknowledging their different levels of realism. They rated the human-realistic avatar as more trustworthy, had more affinity for it, and preferred it as a virtual agent. Participants who observed the interview through VR headsets had an even stronger affinity for the human-realistic avatar and stronger preferences for it than those who observed via the 2D screen. Effect sizes ranged from medium to large. Our results suggest that it is now possible to cross the uncanny valley with human-realistic avatars rendered in real time.

Automated summary

The study investigated users' affinity, trustworthiness, and preferences for avatars with different levels of realism, finding that human-realistic avatars were more positively received, with higher trustworthiness and affinity. Observers also had positive evaluations of both avatars, acknowledging their differences.