In the face of rising generative AI use in science communication, this study investigates how the realism and gender of AI-generated avatars affect perceived trustworthiness (expertise, integrity, benevolence). Using an experimental design with 491 participants, the research examined avatars of varying degrees of realism by avatar realism (very high vs. cartoonish style) and gender presentation (male vs. female), all communicating the same scientific content via video. Contrary to the “uncanny valley” hypothesis – a phenomenon coined in robotics and cognitive science referring to the unsettling feeling experienced when humanoid representations appear almost, but not entirely, human-like – more realistic avatars were seen as more trustworthy, especially in terms of integrity and benevolence. However, gender bias persists: male avatars were rated as more competent than female ones. The findings highlight both the potential and the pitfalls of using synthetic figures in science communication.
The Journal of Science Communication (JCOM) also highlights the study in a press release. In an interview, Jasmin Baake discusses the key findings of the publication. The full interview can be read here: https://www.eurekalert.org/news-releases/1079762
The paper is available open access:
Baake, J., Schmitt, J. & Metag, J. (2025). Balancing Realism and Trust: AI Avatars In Science Communication. JCOM 24(2), A03. https://doi.org/10.22323/2.24020203