Face It, Users Don’t Care: Affinity and Trustworthiness of Imperfect Digital Humans

Digital humans are growing in application and popularity, both as avatars for people and as standalone artificial intelligence-controlled agents. While the technology to make a digital human look more realistic is improving, we know little about how realistic they need to be. Humans are exceptionally good at identifying imperfect digital reproductions of human faces, so it has been reasoned that the slightest imperfections in the visual design of digital humans may translate into reduced acceptance and effectiveness. The broadly held wisdom is that digital humans should be photorealistic and indistinguishable from real people. To examine this common belief we collected data on individuals’ affinity and trustworthiness in photorealistic digital humans when engaged in a product bidding situation, along with a human presenter with varying degrees of video imperfections. The results reveal that participants noticed some of the video imperfections, but this did not adversely affect their willingness to pay, affinity, or trust. We found that once digital humans become close to realistic, users simply do not care about visual imperfections