Wolf In Sheep's Clothing

Menu
Campfire
End round stats

Inspiration

Most social deduction games focus on AI trying to trick humans. We wanted to reverse that idea and create a game where the human is the impostor. Our goal was to explore how people perceive AI behavior and challenge players to think like a machine. The project originally began as a voice-based experience with speech recognition, but we eventually pivoted to text responses to create a more reliable and strategic gameplay loop while preserving the core concept.

What it does

Wolf In Sheep's Clothing is a social deduction game where the player is placed among a group of AI agents. Each round, everyone answers the same question. The player's challenge is to respond in a way that appears AI-generated and avoids drawing attention. After all responses are submitted, everyone voices their opinions and vote on who they believe is human. After every voting phase, players can view detailed AI reasoning, voting statistics, and agent decisions.

How we built it

The game places the player in a group of AI agents. Every round follows two phases:

Question Phase : All participants answer a prompt.

Elimination Phase : The AI agents analyze the responses and vote on who they believe is the human.

To survive, players must answer questions in a way that appears logical, structured, and AI-like. Any response that feels too emotional, personal, or human can raise suspicion. The game world is brought to life through our custom-created pixel art assets and a 2.5D platform-style environment, giving the project a unique visual identity while keeping the focus on player interaction and strategy.

Challenges we ran into

One of our biggest challenges was balancing the AI agents. We wanted them to feel intelligent and capable of detecting human behavior, but not so strict that players would be eliminated immediately. Another challenge was deciding to move away from our original voice-recognition concept. While voice interaction was exciting, implementing it reliably within the project scope proved difficult. We pivoted to text-based responses, which allowed us to focus on stronger gameplay mechanics, smoother interactions, and a better overall experience. We also spent significant time refining prompts and agent behaviors to ensure voting felt meaningful instead of random.

Accomplishments that we're proud of

Creating a unique twist on the social deduction genre where the human is the impostor.
Building AI agents that provide transparent reasoning behind their decisions.
Developing a system that allows players to learn from AI behavior and improve over time.
Designing all of our own pixel-art assets and visual identity.
Successfully combining strategy, AI interaction, and replayability into a single experience.

What's next for Wolf In Sheep's Clothing

Our original vision included voice recognition, and we'd love to revisit that idea in future versions to create a more immersive experience. We also plan to:

Add more unique AI personalities and behaviors.
Introduce different game modes and difficulty levels.
Expand the variety and complexity of questions.
Improve long-term progression and player statistics.
Add multiplayer and community-driven scenarios where humans can compete to see who can survive the longest among AI agents.

Can a human learn to think like an AI well enough to fool one?