Polygon - Mimic

Mimic — A social deception game built with AI, played against it.

Comment

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned# 🧠 About the Project — Mimic: Built with AI, Played Against It

🎮 The Concept

Mimic is an AI-collaborative social deception game where human players and AI agents compete to outwit each other — one word at a time.
Each round begins with a 4×4 grid of words within a topic (e.g., Food, Movies, Science).
Only the non-Mimic players know the “secret” word. Everyone submits a single clue to hint at it, while the Mimic fakes their way through the round.
After clues are revealed, all players — human and AI alike — debate and vote on who they think the Mimic is.

The twist: our AI player is not just a narrator or referee; it’s an active participant that makes contextual decisions using association scoring and probabilistic logic.

🧩 The AI Logic — A Probabilistic Approach

The Mimic’s decision process was modeled as a simple conditional probability problem:

[ P(\text{word} \mid \text{topic}) = \frac{f(\text{association(word, topic)})}{\sum_i f(\text{association}(w_i, \text{topic}))} ]

where

  • ( f(\text{association}) ) measures semantic proximity between a clue and the known topic grid.
  • The Mimic (AI) selects the word ( w^* ) that maximizes this conditional likelihood without directly matching the hidden word.

Similarly, during voting:

[ P(\text{suspect}_j) = \frac{1}{Z} \cdot \text{Sim}(clue_j, \text{consensus})^{-1} ]

This ensures the AI “suspects” players whose words deviate most from consensus similarity, allowing dynamic, believable behavior.

🧠 The Build Story — Vibecoding the Game

Mimic wasn’t just coded — it was co-created.
Our team used a development philosophy we call Vibecoding — a process where human creators and AI systems collaborate conversationally to design, document, and implement in real time.

Stages of the Process

  1. Ideation through Dialogue:
    The concept, name, tone, and mechanics emerged through iterative sessions with AI — similar to a human writer’s room, but with generative input.

  2. Live Documentation:
    Every design choice and technical adjustment was versioned in Markdown (project_overview.md, gameplay_loop.md, pitch_rfp_proposal.md, trailer_creation_stages.md).

  3. Agentic Integration:
    The same AI models that co-authored design documents later became in-game agents, closing the creative loop.

  4. Rapid Prototyping:
    We built a functional prototype in Next.js with Firebase, leveraging JSON wordpacks (topics.json) for semantic clue selection.

⚙️ Technical Architecture

Layer Technology Role
Frontend Next.js (TypeScript + React) Game UI and phase logic
Backend Firebase / Firestore Real-time multiplayer state
AI Module Local Python micro-agent + JSON datasets Word association + voting
Data /src/data/topics.json Curated category boards
Collaboration AI-assisted Markdown docs Continuous code + design dialogue

🧩 The Outcome

  • A working proof-of-concept demonstrating AI as a co-player in real-time social games.
  • A fully AI-generated documentation and gameplay loop.
  • A reusable “agentic architecture” that can extend to any multiplayer logic loop.
  • A visible demonstration of the future of AI-human co-development.

🪄 The Lesson

The greatest discovery wasn’t just that AI could play — but that it could build alongside us.
Every design file, game state, and line of dialogue became a shared creative space between humans and machines.

📐 Summary Equation

If we model traditional development as:

[ G = f(C, D, I) ]

where ( C ) = code, ( D ) = design, ( I ) = iteration,

then our Vibecoding process extends it to:

[ G' = f(C, D, I, A) ]

where ( A ) = active AI collaboration.
This expansion isn’t just additive — it transforms how creation itself scales.

🚀 Tagline

Mimic — A social deception game built with AI, played against it.

What's next for Polygon - Mimic

Built With

Share this project:

Updates