Learning videos with speaking avatars: Impact on learner performance and cognitive load

In the scientific literature, avatars are representations that can take various forms and fulfill multiple functions and roles for learners. The recent emergence of ultra-realistic avatars generated by artificial intelligence presents new and pressing questions for educators and instructional designers. What is the true impact of these virtual pedagogical agents on learning performance? Do they genuinely enhance knowledge acquisition, or do they represent a significant risk of inducing extraneous cognitive overload?This paper presents a study investigating the impact of a realistic AI-generated avatar in an educational video on learners' performance and cognitive load, compared to an identical condition with no avatar. The study was guided by the hypothesis that the avatar's presence would be detrimental to learning by increasing the cognitive load imposed on the learner.An experiment was conducted with 27 participants who were randomly assigned to one of the two conditions. The results showed no statistically significant differences in either learning performance or perceived cognitive load between the groups. While these findings align with the "persona zero effect," the analysis revealed a critical methodological insight: 81.5% of participants (22 out of 27) completed the experiment on a mobile phone. This massive skew towards mobile use likely diminished the avatar's visual presence and impact, offering a concrete explanation for the observed lack of effect. This paper thus highlights the viewing device as a critical confounding variable in research on pedagogical agents.