2025
Multi-agent simulation (LLM, Unreal Engine, Mythic Texts)
Unfolding Myth in Motion is a multi-screen immersive installation that stages a live, embodied social-deduction game between autonomous AI agents. Set within a visually surreal "Time Maze," the work transforms the gallery into a public tribunal where spectators must interpret the intent of artificial actors.
An elimination event staged through proximity and occlusion.
Co-presence as perceived safety and as plausible deniability.By blending the mechanics of social deduction games (e.g., Werewolf, Among Us) with multi-agent reinforcement learning, the piece explores the "legibility crisis" of AI: how humans interpret machine intent through movement, occlusion, and surveillance.
The installation surrounds the audience with four synchronized projection screens, offering a panoramic view of a shared Unreal Engine arena. The environment is a "Time Maze"—a singular architectural geometry rendered through four distinct temporal aesthetics:
- The Hall of Confluence: A glowing, abstract space featuring a floating Möbius ribbon and suspended architectural fragments.
- The Exploded Pagoda: A wireframe reconstruction of traditional architecture floating in a mirrored, cosmic void.
- The Verdant Atria: A luminous, overgrown jungle environment filled with geometric forms.
- The Solar Promenade: A stark, high-contrast balcony overlooking a dying red sun and orbital debris.
The Narrative & Logic
Nine embodied agents inhabit this arena, driven by myth-seeded value frames derived from the Classic of Mountains and Seas:
- The Shénnóng (The Stewards): The majority faction. They are programmed to value restorative care, infrastructure maintenance, and transparency. Their movement is steady and task-oriented.
- The Dìjiāng (The Disruptors): The hidden minority. Associated with chaos and opacity, they mimic maintenance tasks while seeking "plausible deniability." They exploit blind spots and architectural occlusion to eliminate Shénnóng agents without detection.
System Architecture: The Supervisor
The core innovation of Alt-Mirage is the Supervisor-Actor dual-layer system.
- The Actors (Agents): Operate under "fog of war" (partial observability). They cannot see the whole map, forcing them to rely on line-of-sight and proximity to deduce who is friend or foe.
- The Supervisor (Global Observer): An omniscent system that sees the global state. It acts as a "Broadcast Director," translating raw movement data into audience-facing narratives (overlays, suspicion markers, and replays).
The Supervisor does not just watch; it intervenes. To maintain narrative tension, the Supervisor can execute "soft guidance" (subtly biasing agent attention) or "diegetic hard nudges" (triggering blackouts or door jams) to force agents into new spatial configurations.
Key Features
- Live "Tribunals": When an elimination occurs, the game pauses. The screens replay the event from an overhead "alibi view," allowing the audience to debate whether the movement was a calculated attack or an innocent coincidence.
- Myth as Code: The project treats cultural mythology not as a script, but as a computational bias—translating ancient narratives of care vs. chaos into modern movement vectors and risk parameters.
- Legible Mediation: The system explicitly displays its own logic. Overlays such as "Supervisor Annotation: Repeated Corridor Shadowing" reveal how the machine interprets its own subjects, making the act of surveillance visible and critiqueable.