Microsoft Xbox Unveils an Impractical AI Model for ‘Gameplay Inspiration’

Yesterday, Xbox unveiled Muse, a new tool from Microsoft that’s powered by generative AI, aimed at injecting creativity into game development. Along with an article on Nature.com and a blog post, a YouTube video accompanies the launch. Now, if “gameplay ideation” feels like a head-scratcher, you’re not alone. Microsoft explains it as crafting “game visuals, controller actions, or both.” However, don’t get too excited—it’s clear Muse isn’t a magic wand that bypasses the traditional game development pipeline.

Even with its constraints, there’s some fascination in the details. Imagine this: they trained Muse using H100 GPUs at a pretty massive scale. It required approximately one million training updates just to stretch a single second of real gameplay into a further nine seconds of reactive and engine-accurate simulated action. What’s even more intriguing is that they sourced their data largely from existing multiplayer games.

Now, rather than just firing up the game on one ordinary PC, Microsoft had to harness a fleet of 100 Nvidia H100 GPUs. That kind of power and expense, only to produce a nine-second snippet with a resolution of 300×180 pixels, certainly raises eyebrows.

One standout capability demonstrated by the team using Muse was the duplication of props and adversaries within the game’s world while mimicking their original functions. The irony? They could have simply used usual development techniques to create new enemies or items, saving all those resources and costs.

Although Muse does manage to keep up with the game’s original behavior and object permanence, its practicality is questionable against the backdrop of tried-and-true game development methods. Despite the possibility that future iterations of Muse could achieve more impressive functions, for now, it seems like it’s just another project attempting to fully simulate gameplay using AI. And even though it’s nice to see the accuracy and permanence of the gaming environment retained, this doesn’t seem like the most efficient way to develop, test, or enjoy a video game. After thoroughly examining all the material, it’s hard to see why anyone would willingly choose Muse in its current form.