XAI Micro-Feedback, Cognitive Load, and Learning in Secondary Physics: A Quasi-Experimental Study

Secondary school physics presents significant conceptual challenges, particularly in high-intrinsic-load topics such as electromagnetism. Traditional end-of-task feedback is often inadequate for addressing misconceptions during active problem solving. Explainable Artificial Intelligence (XAI), which provides transparent, step-by-step reasoning, offers a promising approach for delivering formative micro-feedback during problem solving. This quasi-experimental mixed-methods study (N = 145 Grade 11 physics students; five intact classes) examined the impact of GPT-4-generated XAI micro-feedback on conceptual understanding, transfer performance, and delayed retention, while also investigating whether intrinsic, extraneous, and germane cognitive load mediate the relationship between XAI feedback and learning outcomes. Students in the XAI condition (n = 87) received four-component structured micro-feedback; the control group (n = 58) received traditional end-of-task teacher feedback. Multilevel regression indicated that XAI micro-feedback significantly predicted posttest performance (β = 26.1, p < .001; Cohen's d = 2.21). Transfer task scores (d = 1.32) and delayed retention scores (d = 1.33) were also substantially higher in the XAI condition. XAI feedback reduced extraneous cognitive load (d = −1.99) and increased germane load (d = 1.22), with no change in intrinsic load. Multilevel mediation analysis confirmed that extraneous load (indirect effect = −3.1) and germane load (indirect effect = +8.2) were significant mediators. Qualitative thematic analysis of student written reflections corroborated three themes: conceptual clarification through transparent reasoning, metacognitive activation, and reduced confusion. The findings provide school-based empirical support for the XAI-ED framework and demonstrate that structured, teacher-supervised GPT-4 micro-feedback can produce substantial and durable learning improvements in secondary STEM contexts.