Abstract
Current generative AI tools for novice musicians are predominantly prompt-based, enabling users to create music without learning about the music-making process. To advance beyond this paradigm, this paper presents the design of a co-creative AI tool and learning environment designed to support music learning by scaffolding the creative process. We draw on a theoretical framework of creative thinking as a contextual, domain-specific process and review prior work on generative AI in music composition, focusing on its potential impact on learning and creativity. To investigate the tool, we conducted workshops with 25 secondary school students creating music with computer code. We present an ethnographic account of one learner’s creative process, highlighting how his composition evolved through interactions with peers, facilitators, and the tool. We describe how learners evaluated AI-generated music by its fit with their existing work, often in terms of “vibe” rather than technical features, and how this engagement prompted reflection on authorship, ownership, and the distributed nature of creativity. Finally, we outline design principles grounded in our conception of creativity and our empirical work. We argue that tools supporting learners’ creative activity should scaffold specific creative practices rather than generate entire products, and that this approach can support learning more effectively than one-size-fits-all systems that rely primarily on natural language.