STEM-5E Socio-Scientific Argumentation with Generative AI-Driven Scaffolding: Exploring the Interplay between Epistemic Beliefs and Learning Outcomes

This study addresses the limitations of traditional STEM education by integrating socio-scientific issues within the STEM-5E instructional framework, using generative AI-driven scaffolding for self-regulated learning (GAISSR). We investigated how different types of GAISSR affect students' argumentation performance over time and how students’ epistemic beliefs about generative AI (GenAI) relate to their improvement in argumentation skills. Ninety-four eighth-grade students participated, randomly assigned to metacognitive GAISSR, cognitive GAISSR, or a control group. The curriculum involved three consecutive tasks on energy policy issues related to climate change. A linear mixed model was used to analyze changes in students' argumentation performance over time, while epistemic beliefs and self-regulation were assessed via adapted questionnaires. Findings revealed a significant interaction between scaffolding type and task sequence; cognitive scaffolding initially improved performance but did not sustain gains, whereas metacognitive scaffolding supported continued improvement by the final task. Additionally, students who demonstrated improvement reported significantly stronger beliefs in the certainty, source, and structural coherence of GenAI-generated content compared to peers who did not improve. These results suggest metacognitive scaffolding promotes deeper, enduring argumentation skills, and students’ epistemic beliefs about GenAI significantly influence their learning outcomes.