Generative AI in music learning: A mixed-method study of skill development, AI dependence, and the moderating role of...

This study examines how generative AI (Gen-AI) is associated with music learning outcomes by shaping underlying cognitive and self-regulatory mechanisms. Drawing on Self-Regulated Learning (SRL) and Cognitive Load Theory (CLT), the study investigates how students' use of Gen-AI is associated with music-specific domain skills and dependence on AI tools, and how these dual pathways are associated with the quality of musical performance. It further explores the moderating role of AI literacy in shaping these associations. A mixed-method approach was employed. Quantitative data were collected from 1142 music students across six leading university music schools in China using an online questionnaire and analyzed through Covariance-Based Structural Equation Modeling (CB-SEM). In addition, qualitative data were obtained from 15 experts in music education and AI-assisted music practice and analyzed using thematic analysis. The quantitative results show that Gen-AI use is positively associated with domain skills while also being positively associated with dependence on AI tools. Both domain skills and dependence are significantly associated with performance quality. Mediation analysis indicates that Gen-AI is indirectly associated with performance through a dual pathway: a positive association via skill development and a negative association via AI dependence. Moreover, AI literacy significantly moderates the associations between Gen-AI use, domain skills, and dependence, strengthening the positive pathway while weakening the negative one. The qualitative findings support and extend these results by highlighting how Gen-AI facilitates skill development, contributes to cognitive offloading, and how AI literacy functions as a metacognitive regulatory capability in guiding effective AI use. The study contributes to theory by extending the application of SRL and CLT to AI-mediated creative learning environments and by highlighting the dual cognitive mechanisms through which Gen-AI is associated with learning outcomes. Practically, the findings provide guidance for integrating AI in music education while emphasizing the importance of developing AI literacy to support balanced and effective AI use. Because the design is cross-sectional, the reported relationships are interpreted as associations rather than causal effects.