Tetris AI

Tetris AI is an interactive tool developed by Catherine Sibert, Assistant Professor Human Computer Collaboration at the Faculty of Science and Engineering. It lets people explore underlying machine learning mechanisms. It makes the relationship between features and behavior visible.

JTS Early Career Researcher Prize 2024 winner Catherine Sibert on the game: "In this project, we aim to create a fun and easily accessible tool to help the public build intuitions about how machine learning works. Using an existing piece of research code that develops AI models that play Tetris, we will build an interface that will let people adjust the model’s weights and watch how that changes the gameplay behavior in real time. We hope to expand the interface to support additional game modes, visualizations of AI learning processes, and explanations of model features."

Tetris AI is essentially a game of Tetris, although instead of the user making choices about where to place each piece, they instead adjust the weights of an AI model that uses those weights to select placements, which will be visually reflected in real time. Sibert: "The hope is that by taking a relatively familiar domain with understandable features, users will build better intuitions about how the relationships between different features can result in visibly different gameplay behavior. If we can get the initial mechanics working, we have several ideas for extending the tool, but all with the goal of making aspects of machine learning visible."

Modes

The game has multiple modes: Tetris gameplay, challenge modes, sandbox, and AI. It is still in the developing stage, there are many opportunities for future expansion.

Catherine Sibert worked closely with student assistant Rijk van Putten on the project.