Matthew Barthet

Matthew Barthet

PhD student in AI and games research, part time freelance game developer.

About Me

I am a full-time PhD student at the Institute of Digital Games, specializing in Games Research and AI. My research is at the cutting edge of the field, focusing on leveraging Artificial Intelligence to train gameplaying agents to behave and exhibit emotions in a human-like manner. I’m also interested in the use of AI to procedurally generate unique and captivating game content, which led me to work with Minecraft where I designed an innovative, creativity-centric building generator for my master’s thesis.

In addition to my academic pursuits, I have also started working part-time as a freelance/independent game developer using the Unity and Godot game engines. This experience has greatly benefited my research as it’s given me the skillset required to create specialized game environments for the testing and deployment of my work. I’m very passionate about combining my work in AI and academia with the practical side of my game development skills, and it’s something I intend to keep exploring in the future.

Research Highlights

The Affectively Framework

Game environments offer a unique opportunity for training virtual agents due to their interactive nature, which provides diverse play traces and affect labels. Despite their potential, no reinforcement learning framework incorporates human affect models as part of their observation space or reward mechanism. To address this, we present the Affectively Framework, a set of Open-AI Gym environments that integrate affect as part of the observation space. Check out my github repository to access the source code and run the framework for yourself!

GameVibe: An Affective Game Corpus

Game Vibe animation

What makes people interested in games? What if we could capture aspects of experience just by looking at the gameplay screen, as a mirror of a viewer’s experience? We could then learn to design better games and speed up research towards general AI models of player experience. Check out my blog post on nature communities where I describe GameVibe, our first-person shooter engagement dataset which we published in the Nature Scientific Data journal.

Open-Ended Minecraft Buildings

This research presents an AI system that generates novel Minecraft buildings through evolutionary algorithms that continuously discover new feature possibilities. The system alternates between evolving building designs using novelty search and retraining its understanding of what constitutes “interesting” architecture, enabling open-ended creative exploration. By dynamically expanding its creative boundaries rather than working within fixed constraints, the system achieves greater diversity and complexity in generated structures. Checkout our immersive article published in IEEE Transactions on Games.