Can AI-powered avatars replace human trainers? An empirical test of synthetic humanlike spokesperson applications

Purpose This study aims to investigate the effectiveness of widely adopted but under-studied synthetic humanlike spokespersons (SHS) compared to organic human spokespersons in workplace training videos. The primary aim is to evaluate whether employees will rate training videos more negatively when they perceive their trainer to be synthetic such as those videos made with the AI-powered tools Synthesia or HeyGen. Results suggest that while ratings are more negative when the viewer perceives syntheticness, the change is only mild across most measures deployed here. When measures were calculated by using actual modality and not perceived modality, no significant change in rating was found. This study deployed three measures: actual knowledge transfer, perceived effectiveness and brand impression. It is guided by a convergence of AI Literacy Theory, the Technology Acceptance Model and the Theory of Reasoned Action. Design/methodology/approach Over 250 professionals assessed the effectiveness of training videos in a 2 × 2 trial design. Participants were randomly assigned to view one of four training videos featuring either a synthetic or organic spokesperson for a fictional business. After watching the video, participants answered Likert-scale survey questions and multiple-choice quiz-style questions to provide comparative measurements of actual knowledge transfer effectiveness, perceived effectiveness, brand impression and the effectiveness of the synthetic deception. The study used both ANOVA and multiple regression analyses to control for potential confounding variables and ensure the robustness of the findings. Findings The results indicate no significant differences between SHS and organic human spokespersons in terms of actual effectiveness (quiz scores), perceived effectiveness or brand impression, based on actual modality comparisons. However, when respondents perceived syntheticness, the avatar triggered the negative uncanny valley phenomenon and slightly decreased overall scores in perceived effectiveness and brand impression. Notably, more than half of respondents did not detect that the SHS was synthetic. Demographic variables such as gender, age or household income had no significant impact on the results. Practical implications Organizations can justifiably consider incorporating SHS into their training programs, leveraging these synthetic agents to deliver cost-effective and scalable learning solutions. The findings suggest that SHS can be used effectively for goals such as actual knowledge transfer without compromising training quality, and that other perceptual goals may be within reach as well. This may offer a viable alternative to traditional, organic human spokespersons, opening up new opportunities for enhancing training efficiency and accessibility across various industries. Originality/value Synthetic avatars, as outlined here, are a demonstrably effective new option in the array of available learning technologies. They are being adopted en masse but without significant study of their effectiveness. This research provides foundational quantitative assessments designed to address that gap in the literature and practice. The data presented here is highly valuable for practitioners and scholars interested in cutting-edge learning tools. Given the rapid advancement of technology, this study provides an important benchmark for evaluating future improvements in these tools. Furthermore, it offers actionable recommendations for integrating AI-powered avatars into professional development, enhancing understanding of the roles advanced technologies play in educational and training programs.