How Does it Relate to Recipients’ Intrinsic Cognitive Load and Feedback Adequacy, and Can it be Fostered with AI? Recent and ongoing research in the fields of Educational Psychology and Empirical Educational Research will be presented and discussed in the research colloquium.

Peer feedback can be an effective pedagogical approach to support learning (Double et al., 2020). Researchers assume that the more features a peer feedback message contains, the higher the quality of the feedback message is (e.g., Wu & Schunn, 2020). Yet, this “one-size-fits-all” assumption may be questioned, as not all students benefit from peer feedback to the same extent (Li et al., 2020). We assume this discrepancy to be due to a dependency on the recipients’ skill levels, in essence, a lack of learner-adaptivity within the feedback (Shulgina et al., 2024). Further, recent research has started to explore to what extent AI can be used to support students during peer feedback (e.g., Darvishi et al., 2022; Guo, 2024). Yet, whether AI-tools can help students develop learner-adaptive feedback messages is an open question and likely depends on how it is prompted. This study therefore, explores the questions: To what extent do students adapt their peer feedback messages to the recipients’ skill levels? How is learner-adaptive feedback related to the feedback recipient’s intrinsic load and perceived feedback adequacy? And a third research question: What are the effects of AI-support on the learner-adaptivity of peers’ feedback messages? 297 pre-service teachers participated in a study across four weeks. In Week 1, they were asked to analyse a case vignette of a fictitious classroom situation by aid of two educational theories. In Week 2, they created feedback drafts for two peers and revised their feedback messages in Week 3. In Week 4, they received their peers’ feedback and revised their initial problem analyses. To compare the effects of natural vs. two kinds of AI-supported peer feedback, we employed a 1x3 experimental design with three conditions (“human-only”, “best-of-two-feedback messages”, “feedback-on-feedback”). Dependent varia bles include feedback features; evidence-informed reasoning skill, intrinsic cognitive load, and feedback ade quacy. Concrete results will be presented at the Colloquium.