Recent advances in generative models have expanded the possibilities for AI-based music composition. However, tensions persist between the essentially imitative and automated nature of these systems and the kind of artistic value that emerges from the deliberate, deeply personal compositional practices of human creators. Enabling effective user control over such systems remains a central research challenge across a range of artistic disciplines.
This paper presentation provides an overview of Ricercar, a symbolic AI-based composition system developed to support experimental, contemporary compositional practice. We present insights into the system’s capabilities and interface, and share observational findings from its use by emerging composers training at four established music universities. These participants, working within contemporary music contexts, developed a range of artistic concepts and interaction strategies to shape the system’s output in ways that reflect their individual aesthetic goals.
Our findings suggest that meaningful collaboration in composition requires a balance between artistic direction and system autonomy, where composers experiment with strategies to influence the AI without fully controlling it, allowing space for surprise and creative divergence. This underscores the importance of designing generative systems that are not only controllable, but also open-ended and adaptable to diverse artistic intentions.