From Performance to Installation: How Interactive Reinforcement Learning Reframes the Roles of Performers and Audiences
Danny Perreault; Victor Drouin-Trempe; Vincent Cusson; David Drouin; Sofian Audry;

- Format: oral
- Session: papers-3
- Presence: remote
- Duration: 10
- Type: medium
Abstract:
This paper explores how interactive reinforcement learning (IRL) reconfigures the roles of performers and audiences in audiovisual performance and immersive installation. We adapt the Co-Explorer (a software tool originally developed for musical co-creation) to audiovisual immersive contexts and examine its creative potential using a reflexive research-creation approach. Our study reveals how IRL splits the role of the performer into three distinct positions: (1) the designer, who defines the parametric space; (2) the guide, who reinforces the agent’s behavior; and (3) the machine performer, whose actions are shaped by interactive training. As IRL introduces agency into the creative process, it transforms traditional notions of authorship and control, enabling unexpected emergent outcomes. By showcasing an interactive installation/performance, we further explore how audiences contribute to collective creation through reinforcement-based interaction. Our findings underscore the challenges of balancing the temporality of IRL with the demands of public-facing works and of adapting RL-based systems to different exhibition contexts. Our work contributes to the discourse on co-creative systems, emphasizing the evolving roles of artists, artificial agents, and audiences in hybrid creative ecosystems.