학습자와 생성형 AI 간 상호작용의 교육적 효과 메타분석 = A meta-analysis of the educational effects oflearner-generative AI interactions

다국어 초록 (Multilingual Abstract)

This study examined the educational effects of interactions between learners and generative artificial intelligence (GAI) through a meta-analysis. Following the PRISMA 2020 guidelines, 169 effect sizes were extracted from 54 experimental and quasi-experimental studies and analyzed. To address the dependency of multiple effect sizes within studies, a three-level meta-analysis and robust variance estimation (RVE) were applied. The results showed that learner-GAI interactions had a large overall effect (Hedges’ g = 1.053). By interaction type, elaborative feedback-centered interactions and conversational interactions showed relatively large effects. Moderator analyses revealed significant effects of digital literacy, task structure, and intervention duration. In terms of learning outcomes, large effects were found for cognitive and metacognitive outcomes such as self-efficacy, self-regulated learning, and academic achievement, whereas learning motivation showed a moderate effect and learning engagement showed an inconclusive effect. These findings suggest that the educational effects of generative AI depend not simply on technology adoption, but on the quality of interaction design, learner readiness, task-technology fit, and the learning context. This study highlights GAI as an interactive cognitive partner that can support learners’ thinking processes and offers implications for effective GAI-based instructional design.