Now that AI can effortlessly imitate, how will we deviate?

Deviation Game is a game-based experiment exploring the human potential to create beyond the knowledge of AI models. In this game, players must draw prompts like ‘cat’ or ‘democracy’ in a way that fellow humans can recognize—but an image-recognition AI model can’t. If at least one human guesses correctly and the AI does not, humans win; if the AI guesses correctly, humans lose. After each game, participants decide whether their drawing can be used for AI training or be deleted, influencing how the model evolves. 

Inspired by the Imitation Game (1950) proposed by Alan Turing—later expanded as the Turing test—Deviation Game inverses the premise. Instead of using AI to imitate humans, this project challenges humans to deviate from established patterns within AI models, positioning AI as a collaborative force that enhances, rather than replaces, human creativity. 

The game seeks to critique and expose the biases embedded in various mainstream AI models, such as Google’s Gemini and OpenAI’s GPT-4.5. Through experiencing the game, the work allows players to discover and analyse each model’s limits, inherent biases, and unique characteristics. 

Originally developed by a small team of artists based in Tokyo and London, Deviation Game has been exhibited globally at venues including FACT Liverpool, Now Play This, Taipei Digital Art Festival, as well as the initial version at Ars Electronica. Through active engagement from broad and diverse audiences, the game continues to inspire meaningful conversations about creativity, bias, and humanity’s evolving relationship with artificial intelligence. 

Video Game | 2023 – 2025