The use of Pedagogical Agents (PAs) in digital learning environments is very common, for example, to strengthen social aspect of the learning experience (according to Social Agency Theory). There are research gaps regarding the combination of choice and the use of PAs to enhance motivational regulation via the basic need of autonomy as suggested by Self-Determination Theory (SDT). The central research question of this study was whether allowing learners to select their own PA for learning purposes can enhance learning performance, perceived autonomy, and motivation, as opposed to the more traditional approach of being assigned an agent. We therefore explored the impact of learner autonomy in digital learning contexts as well as learning outcomes and motivation.

To address our research question a single factorial between-subjects design with three levels (choose PA vs. assigned PA vs. no visual PA) was conducted. 87 Students (76% ♀, Mage = 22.58, SDage = 3.25) from the University of Technology Chemnitz were recruited. Participants watched an 18min instructional video on lucid dreaming. The highly realistic, human-like PAs were designed using MetaHuman and the narration was synthesised using ElevenLabs.

Learning performance was assessed using twelve multiple-choice questions derived from the instructional video. Additionally, perceived autonomy and motivation were measured using validated scales: the Intrinsic Motivation Inventory (IMI), Multimodal Presence Scale (for social presence), and Godspeed Questionnaire (evaluating anthropomorphism, animacy, sympathy, and perceived intelligence).

The results of the statistical analyses did not support the hypotheses. Learners who chose their PA did not show significantly higher learning outcomes, nor did they report higher levels of autonomy or motivation than participants with an assigned PA or without PA. But we did find a significant strong difference in social presence, which we also tested as a manipulation check. Possible explanations for these unexpected results may be that the specific form of choice provided was redundant or lacked sufficient personal relevance to significantly influence the learners' autonomy or motivation.

Additionally, the relatively short exposure duration to the PA and limited interaction with the learning material may have mitigated potential positive effects. Future research should focus on providing more meaningful or individualized choices in the context of sustained learning experiences to enhance learners' perceived autonomy effectively. Also, our study underlines some limitations like a relatively small sample size, possible lack of identification with the agents, and the restricted range of choice. Future research should explore more personalized and impactful forms of choice within digital learning environments. The findings underscore the complexity of autonomy-supportive design in educational technologies and the need for careful consideration of individual learner preferences, context relevance, and task meaningfulness.

Keywords: Multimedia learning, instructional design, pedagogical agents.