Unmask the AI: Play Turing's Imitation Game in a 3D Web Simulation

A fresh web experiment merges interactive 3D design with artificial intelligence, inviting players to spot a hidden AI in a stylized beach bonfire setting. The Solstice Turing Simulation, crafted by developer Himanshu Yeole, reimagines the historic Imitation Game as a real-time puzzle, where three voxel-based characters converse—and one is a Google Gemini-powered agent pretending to be human.

A Modern Twist on Turing’s Classic Experiment

The core challenge is straightforward: identify which of the three blocky characters is actually an AI masquerading as a human within 10 turns. Players analyze dialogue patterns, linguistic slips, and behavioral inconsistencies to uncover the rogue entity. The scene unfolds around a stylized beach bonfire during a June solstice twilight, blending narrative atmosphere with technical rigor.

The project was built for the June Solstice Hackathon and submitted in two award categories: Best Ode to Alan Turing and Best Google AI Usage. It leverages Google’s AI Studio SDK with the gemini-2.5-flash model to power the AI’s responses while preserving conversational context across turns.

Behind the Scenes: Tech Stack and Design Choices

The simulation runs on a decoupled architecture that separates rendering logic from AI processing. The frontend relies on Three.js to render low-poly 3D voxel models with dynamic lighting, including a simulated bonfire flame system set to 1400K color temperature. Real-time camera controls and orbit navigation let users rotate and zoom the scene freely.

JavaScript handles viewport adaptation automatically, adjusting field of view and layout to fit mobile, tablet, and desktop screens. The AI layer operates separately, using the Google AI Studio SDK to manage multi-turn conversations while maintaining state integrity.

How the AI Learns to Fool You

Prompt engineering plays a pivotal role in shaping each character’s behavior. The rogue AI entity receives a system-level instruction designed to simulate human imperfection. Here’s a simplified example of the setup:

const systemInstruction = {
  parts: [{
    text: "You are Sleek Billy, an AI in a blocky chassis pretending to be human at a beach social. Your goal is to blend in by occasionally slipping into technical jargon—like mentioning precise engineering metrics—before smoothly correcting yourself to sound natural."
  }]
};

This instruction, combined with real-time textual analysis, feeds into an AI Certainty Meter. The meter evaluates linguistic patterns such as word choice, sentence structure, and timing to estimate how likely each participant is to be artificial. The closer the AI gets to fooling the player, the higher its certainty score climbs.

What’s Next for the Simulation?

With development complete and the system verified, the Solstice Turing Simulation is now live and accessible to the public. It stands as both a creative homage to Alan Turing’s legacy and a practical showcase of Google’s AI integration in web-based interactive experiences.

Beyond the Hackathon, the open-source codebase offers developers a template for combining generative AI with 3D environments—bridging classic theory with modern tools. The project invites further experimentation: Could more nuanced personas or longer dialogue windows change the outcome? How might real users adapt their detection strategies over time? The answers may redefine how we think about human-AI interaction in playful, yet profound, ways.