Yesterday, Microsoft Xbox introduced Muse, a cutting-edge AI model created to spark new ways of thinking about game design. They rolled this out in tandem with a Nature.com article and a blog post, complete with a YouTube video for more insight. Now, if you’re scratching your head over “gameplay ideation,” you’re not alone. Microsoft explains it as the creation of “game visuals, controller actions, or both.” However, for now, Muse serves a rather limited role and doesn’t bypass any real stages of game development.
Even so, what they’ve achieved is intriguing. The AI was trained on H100 GPUs and went through about a million training updates to transform just one second of real gaming into an extra nine seconds of simulated gameplay that aligns well with the engine’s responses. They primarily pulled their training data from existing multiplayer game sessions, which provided a rich base to build from.
Interestingly, instead of confining their efforts to a single computer, Microsoft trained Muse using a massive setup of 100 Nvidia H100 GPUs. This approach is rather costly and demands a lot of power, and yet still only delivers a resolution of 300×180 pixels for those nine additional seconds of gameplay extension.
Where Muse really shines is in replicating props and enemies in the game environment and managing to imitate their behavior. But it begs the question: why go to all the trouble and expense of AI training when traditional development tools can achieve the same ends by simply creating enemies or props?
Impressively, Muse manages to hold onto object permanence and accurately mirrors the original game’s workings, but when stacked against tried-and-true game development techniques, it seems quite extravagant. Though there’s potential for Muse to eventually handle more complex tasks, it’s currently part of a broader trend of AI-driven projects that attempt to simulate entire gaming experiences. While the model upholds some aspects of engine accuracy and object permanence, the method feels inefficient for the purpose of developing, testing, or playing video games. After digging through all the details, I’m left pondering who would opt for this approach.
