Microsoft has unveiled Muse, an AI that can generate gameplay and game environments. It is the first World and Human Action Model (WHAM) that will help developers not only create new games, but also optimize existing ones for modern platforms.
Muse was developed by Microsoft Research. This new AI model will be able to create game environments based on visual elements or player controller actions. The model understands the 3D world and game physics and can respond directly to player interactions with the game.
The Muse model was trained on gameplay data from Ninja Theory's Bleeding Edge (2020). This is a 4v4 online team-based game. Match recordings were only collected if players agreed to the End User License Agreement (EULA).
“It’s been amazing to see how Microsoft Research is using the Bleeding Edge environment and data to explore new methods in the rapidly evolving field of artificial intelligence. From the hackathon that started it all, where we first integrated AI into Bleeding Edge, to creating AI agents that could behave like human players, to a world and human action model that can invent entirely new sequences of Bleeding Edge gameplay under human guidance, it’s an eye-opener to the potential this type of technology has,” said Gavin Costello, CTO at Ninja Theory.
Studying data from Bleeding Edge, the Muse model has access to seven years of gameplay, using billions of images of player actions. Although still in its early stages of development, it can now generate gameplay visuals at a resolution of 300×180 pixels, which is better than previous results (128×128) but still far from the 1080p standard.
Microsoft has shown off a number of examples of gameplay generated using Muse. The company says the new AI model could potentially be used for early-stage game development, but it's important to note that Muse is not intended to be used to create entire games or replace developers.
Microsoft is exploring the use of Muse to enhance classic games and adapt them for modern systems. The company is also looking at using AI to prototype games and add intelligence to existing projects. In addition, Copilot Labs will feature interactive games so users can experience new technologies for themselves.