Agentic AI
New AI Agent Systems MACAT and MACataRT Enable Real-Time Collaboration Between Musicians and Machine
Researchers have developed two AI-driven musical agent systems that work alongside human performers to enhance improvisation and creative music-making through personalized, ethical training approaches.
A new research initiative explores how AI agents can serve as creative partners in live music performance and improvisation. Two complementary systems—MACAT and MACataRT—represent distinct approaches to building intelligent musical systems that respect both artistic vision and human agency.
How the Systems Differ#
MACAT is built for autonomous agent-driven performance. It uses real-time sound synthesis paired with self-listening capabilities, allowing the AI to continuously monitor and refine its own output as it performs. This architecture enables the system to respond dynamically to musical context without requiring constant human direction.
MACataRT, by contrast, prioritizes collaborative flexibility. Rather than pursuing autonomous performance, it excels at creating improvisation environments through audio mosaicing—a technique that samples and reassembles fragments of existing recordings—and sequence-based learning that adapts to musician inputs over time.
Emphasizing Ethical AI for Music#
Both systems share a core philosophy: they train on small, personalized datasets drawn from individual artists' work rather than massive generic music corpora. This approach serves multiple purposes: it yields more musically coherent outputs, maintains transparency about where the AI's training material comes from, and ensures artists retain control over how their creative signatures influence the system's behavior.
Expanding Musical Creativity#
The research frames these agents as tools for expanding what musicians can express in real time. Rather than replacing human artistry, they operate in what the researchers call "co-creative spaces," where the boundary between performer and instrument becomes fluid. Improvisers can explore new sonic territories, experiment with unfamiliar harmonic or rhythmic territories, and discover unexpected creative directions—all while maintaining artistic agency throughout the process.
This human-in-the-loop design reflects growing interest in generative AI that amplifies rather than automates human creativity.
