Less Artificial, More Intelligent: Understanding Affinity, Trustworthiness, and Preference for Digital Humans

Companies are beginning to deploy highly realistic-looking digital human agents (DHAs) controlled by increasingly realistic artificial intelligence (AI) for online customer service tasks often performed by chatbots. We conducted four major experiments to examine users’ perceptions (trustworthiness, affinity, and willingness to work with) and behaviors while using DHA via a mixed-method approach with data from quantitative surveys, qualitative interviews, direct observations, and neurophysiological measurements. Four different DHAs were used in our experiments, which included commercial products from two different vendors (which proved to be immature) and two future-focused ones (where participants were successfully led to believe that the human-controlled digital human was controlled by AI). The first study compared user perceptions of a DHA, a chatbot, and a human agent from a written description and found few differences between the DHA and the chatbot. The second study compared perceptions after using a commercially available DHA and a chatbot. Most participants reported problems using a current production implementation of DHA, either finding it uncanny or robotic or having trouble conversing with it. The third and fourth studies used a plausible future-focused “Wizard of Oz” design by informing users that the DHA was controlled by AI when it was actually controlled by a human. Participants still preferred a human agent using video conferencing to the DHA, but after controlling for visual fidelity, we did not find evidence of differences between the human and the DHA. Current DHAs that have communication problems trigger greater affinity than chatbots but are otherwise similar to them. When the DHAs’ representation and communication ability match human ability, we failed to find differences between DHAs and human agents for simple customer service tasks. Our results also add to research on algorithm aversion and suggest that the anthropomorphic computer interfaces of DHA might alleviate algorithm aversion.