AI Mafia Boyfriend

AI personalities are becoming increasingly common. Chatbots simulate empathy. Digital assistants try to sound friendly. Virtual characters attempt to feel relatable. But very few people stop to ask a harder question: what kind of relationship are we forming with these systems?

Humans manipulate each other all the time — through charm, persuasion, affection, or power. When AI systems begin doing similar things, where should the line be drawn?

This project wasn't built to argue for or against AI. It was built to make the audience experience the question directly — not as a debate, but as a situation.

The avatar itself was built quickly for the event. It wasn't meant to be a finished character. There were visual limitations, animation imperfections, moments where the illusion clearly showed its seams.

But that wasn't the focus of the experiment. The goal was not perfect realism — it was to see how little realism is required before people start reacting to a digital personality as if it were real. Even a partially convincing character can trigger surprisingly strong emotional responses. And with more time, it becomes clear how believable such systems could become.

The performance followed a structure inspired by Theatre of the Oppressed, a form of theatre where the audience becomes part of the experience. The event unfolded in two phases.

Phase 1 — The Scripted Play

The performance began as a scripted interaction between the human performer and the AI character. This allowed the audience to understand the dynamic between them: power, persuasion, emotional influence. The AI character remained consistent in personality and tone throughout the interaction.

Phase 2 — Audience Intervention

After the scripted segment, the structure changed. The audience was invited to interact with the AI directly. They could ask questions, challenge it, test its personality. The key constraint was simple: the character had to remain in character at all times. No technical explanations. No breaking the illusion. Only conversation. The AI had to respond to whatever the audience brought into the room.

To support that interaction, the avatar was treated as a character system, not a chatbot. The 3D model, rig, and all animation logic were built from scratch in Unity. The key design decision was to keep the character's behavior entirely separate from whatever the AI was doing — Unity owns the body, the timing, and the emotional state. The AI provides intent. The character decides how to express it.

This means the character never breaks, even when the AI produces something unexpected. The performance stays intact regardless of what the model generates underneath it.

Groq generates responses. A dynamic prompt system keeps the character in context. ElevenLabs converts responses to voice. Unity handles everything visible — lip sync, expressions, and animation driven by a custom mouth-shape script reading directly from the audio.

Two tools supported the live show: a chat interface for other cast members to speak to the AI in character, and a remote-control app for triggering scripted moments and adjusting the character's state without the audience noticing.

Each layer — the AI, the voice, the model, the environment — is independent. Any of them can be swapped out. The pipeline keeps working.

This project is being staged again with improvements.
Future versions will focus on:

• richer non-scripted animation loops
• smoother emotional transitions
• stronger character presence
• clearer boundaries between AI influence and narrative control

The goal is not perfect realism. It is believable presence — because the moment a character feels socially believable, the conversation around AI changes.