The Hitloop project is an AI art installation that examines the level of human control over AI decision-making. Using music as a metaphor, the installation allows participants to experience and influence AI-driven choices in real time.

At ESNS, researchers from the Amsterdam University of Applied Sciences tested this collaborative installation, where a DJ, AI, and festival visitors co-created music. The experiment visualized different levels of human oversight in AI decision-making while ensuring that the AI-generated output met the audience’s aesthetic and ethical expectations.

Hitloop uses WAIVE, an open-source application by Thunderboom Records. WAIVE generates virtual instruments from street sounds and public-domain audio. The AI then composes tracks using these instruments, while a human DJ curates the selections. Festival visitors actively engage by using hand controllers to manipulate individual instruments, influencing the AI-generated composition.

This research is crucial as human oversight in AI is essential for mitigating risks in high-stakes applications, such as healthcare and criminal justice. While regulations like the EU AI Act emphasize the need for human supervision, they lack clear implementation guidelines. However, human oversight itself has limitations—people make mistakes, are inconsistent, and may struggle to correct AI outputs effectively.

By testing Hitloop in a live music setting, researchers explore new, more human-centered approaches to AI supervision. This experiment was made possible with support from Regieorgaan SIA.

During testing at ESNS, the overall user experience was rated positively.  Participants found the interface user-friendly, though there is room for improvement in visual feedback and explanations. The positive attitude of music professionals towards AI as an opportunity was a surprising outcome. Artists expressed interest in collaborating with Hitloop and in conducting live experiments with their audience.