Microsoft Xbox Unveils Unrealistic AI Model for “Gameplay Ideation”

Just yesterday, Microsoft Xbox introduced a new tool called Muse, claiming it as a breakthrough “generative AI model for gameplay ideation.” Alongside this, they released an open-access article on Nature.com and shared a blog post with a video on YouTube. If the term “gameplay ideation” sounds a bit confusing, Microsoft’s explanation might help: it’s about crafting “game visuals, controller actions, or both.” However, don’t get too excited—its practical applications are quite restricted and definitely don’t bypass the traditional game development processes.

Nevertheless, some aspects of the data are fascinating. The AI training was executed on a massive scale, using H100 GPUs. It took around a million training updates just to stretch a single second of real gameplay into an added nine seconds of engaging, engine-dedicated simulated play. The bulk of this training data came from existing multiplayer sessions, by the way.

Rather than running the game on one PC, Microsoft took things up a notch by training the AI model using a cluster of 100 Nvidia H100 GPUs. This approach is significantly more costly and power-draining, and yet it only results in an output resolution of 300×180 pixels for about nine seconds of additional gameplay.

The coolest trick that Muse managed to pull off was copying existing props and adversaries in the game environment while preserving their functions. But, it begs the question: with all this hardware expense and energy consumption, wouldn’t using standard development tools to create enemies or props be more efficient?

Muse did show some neat abilities, like maintaining object permanence and mimicking the original game’s behavior pretty well. Yet, compared to the already proven methods in game development, its practical applications appear a bit extravagant.

While the team’s future enhancements to Muse might achieve more groundbreaking results, it finds itself among numerous initiatives attempting to simulate entire games using AI. Though the preservation of engine accuracy and object permanence is commendable, this method remains far from ideal for developing, testing, or playing games. Even after thoroughly analyzing the material, it’s tough to fathom why anyone would opt for this method.

For enthusiasts craving more insights, Tom’s Hardware offers top-notch news and detailed reviews delivered right to your inbox.