Just yesterday, Microsoft Xbox rolled out Muse, their new generative AI model crafted specifically for “gameplay ideation.” This was announced alongside an open-access article on Nature.com, an accompanying blog post, and a video on YouTube. If you’re scratching your head over what “gameplay ideation” involves, don’t worry; Microsoft uses this term to describe the generation of “game visuals, controller actions, or both.” However, despite its technical definition, its real-world applications are somewhat limited and certainly don’t replace traditional game development processes.
Nevertheless, the data underlying Muse is rather intriguing. The system was trained extensively using H100 GPUs, requiring roughly 1 million updates to take just a single second of actual gameplay and stretch it into nine additional seconds of responsive, engine-accurate simulated gameplay. The dataset mainly consisted of footage from existing multiplayer sessions.
Interestingly, instead of running this on a typical PC setup, Microsoft utilized a network of 100 Nvidia H100 GPUs. This method demands a lot more money and power, yet it only succeeded in delivering a resolution of 300×180 pixels for the nine seconds of generated gameplay.
Probably the most noteworthy demonstration involved Muse duplicating existing props and enemies in a game environment and mimicking their functions. Given the significant cost and energy required, one could argue if this is really a smarter approach compared to using conventional development tools to achieve similar tasks.
Muse did manage to keep tabs on object permanence and accurately duplicate aspects of the game’s behavior. But when you place its current uses beside well-established game development techniques, they seem unnecessarily excessive.
Although it’s possible that Muse’s future enhancements may lead to more exciting capabilities, right now, it joins a long queue of projects attempting to simulate gameplay entirely through AI. While certain elements like engine accuracy and object permanence are still noted, employing Muse as a cornerstone for developing, testing, or even playing games feels inefficient. After extensively reviewing the material, it’s difficult to see why this would be a preferable option.
If you’re interested in staying updated with the latest from Tom’s Hardware, including top-notch news and detailed reviews, do consider subscribing to our newsletter.
