AI-to-Audience

AI-to-Audience was a year-long artist-research project investigating how performing artists can co-author AI character systems rather than serve as source material for them. Funded by the Canada Council for the Arts Digital Greenhouse grant, the project brought together three core artists, five performing artists working in opera, theatre, comedy, waacking and Kathak, a software engineer and a concept artist to build LEO: a real-time, audience-aware digital character exhibited publicly in Toronto in January and February 2025.

This report documents the full process. It covers the reciprocal workshop method, in which performers learned motion capture and AI fundamentals while engineers learned performance craft; the technical architecture, including a nine-state mood matrix, a temporal knowledge graph for conversational memory, and integration across Unreal Engine 5.3, speech synthesis and facial animation systems; the installation design, using holographic projection to place LEO in the same apparent space as the visitor; and a public exhibition where visitors spoke with LEO in scheduled sessions.

Findings address perceived responsiveness in pre-authored versus generated animation, the effect of latency on conversational believability, safety implemented through character design rather than technical filtering, and the data and consent architecture adopted for a system that detects but deliberately does not recognize individuals. The report also sets out an unresolved question the project raises: how performing artists should be compensated for ongoing use of captured expression and movement.