Directing Space: Rehearsing Architecture as Performer with Explainable AI

AI can be a creative collaborator in live performance: The paper demonstrates that AI systems, particularly large language models, can actively participate in artistic creation by sensing spatial behavior and speech, then generating real-time environmental responses—suggesting AI's role extends beyond tool-use to genuine creative partnership in performance contexts.

Explainability enables artistic control: By using "reasoning traces" (showing how the AI makes decisions), the system allows artists to understand and shape how the AI interprets their work, supporting iterative experimentation and ensuring human artists maintain creative agency over the final performance outcome.

Rehearsal and testing are essential for responsible AI deployment: The paper emphasizes using virtual blueprints and rehearsal settings to test AI-mediated environments before live performance, highlighting that responsible AI implementation requires iterative refinement and dialogue between creators and the system rather than direct deployment.

Architecture itself becomes an interpretive performer: By treating interactive spaces as "performative agents" that respond to human behavior, the work reframes how we think about AI in physical environments—not as passive tools but as active participants that require dramaturgical (narrative and artistic) principles to guide their behavior responsibly.

Transparency supports human-AI creative dialogue: The system's design prioritizes making AI reasoning visible to artists, suggesting that explainability isn't just a technical requirement but a creative necessity that enables meaningful collaboration and shared artistic vision.