Microsoft Unveils Another Generative AI Model but Leaves Developers Questioning its Practical Applications
By Microsoft has announced that it has created a novel generative AI model capable of producing “intricate gameplay sequences.”
The Xbox manufacturer’s Research Game Intelligence team unveiled the World and Human Motion Model (WHAM), informally called Muse, earlier today, referring to it as a “generative AI model for video games that can create game visuals, controller inputs, or a combination of both.”
Okay. At this juncture, we need to address the ‘generative AI is not actually making video games’ disclaimer.
Microsoft stated that Muse, which was trained using data provided by Hellblade creator Ninja Theory, can currently generate visuals at a resolution of 300 x 180 pixels. It also required 1 million training iterations before producing an output that includes realistic model behaviors and mechanics.
Microsoft elaborated that recent iterations of Muse were trained on over 1 billion images and controller actions, mirroring more than seven years of continuous human gameplay. The company affirmed that the data was sourced “ethically” from Bleeding Edge, the four-versus-four multiplayer game created by Ninja Theory.
The company had previously utilized player data from this title to conduct navigation experiments mimicking human behavior, but in 2022, they began to explore what might occur if this information was directed into a machine-learning model in a foie gras fashion. This led to the conception of Muse.
“As the team began working, one of the primary challenges was scaling up the model training. We initially employed a V100 cluster, which enabled us to determine how to expand training to utilize up to 100 GPUs; this ultimately facilitated large-scale training on H100s,” Microsoft explained, detailing how the technology was developed.
“Key design choices made early on concentrated on how to best utilize insights from the large language model (LLM) community, including decisions about effectively representing controller actions and images.”
Microsoft celebrates generative AI ‘milestone’ but Xbox studio head asserts that technology won’t be used for content creation
Microsoft referred to the development and reveal of Muse as a “milestone” that potentially illustrates how generative AI models can “meaningfully assist human creators.”
For example, in one demonstration, Muse was activated with a modified visual that introduced a completely new character. It subsequently generated a gameplay sequence that exhibited how that character could be “integrated” into the experience.
In a discussion about Muse in a video blog, Xbox leader Phil Spencer remarked that developing the model involved merging “the artistry of science” to empower creators to “accomplish things they’ve never attempted before.” It remains to be seen what those ambiguous “things” might entail.
Spencer also speculated on how AI models might be utilized to enhance game preservation, envisioning a scenario that might alarm some preservationists.
“You can envision a future where, using gameplay data and video, a model could absorb knowledge from old games and effectively adapt them for any platform that supports these models,” he stated. “These models and their capacity to comprehend how a game functions—without needing the original engine to operate on the original hardware—I believe opens up vast opportunities.”
Ninja Theory studio leader Dom Matthews articulated his particular interest in understanding how AI can assist teams in “advancing their creative aspirations.”
“This technology, for me and for our team at Ninja Theory, isn’t about using AI to create content, but rather about establishing workflows and methodologies that enable our team of 100 creative professionals to accomplish more, progress further, iterate faster, generate ideas more quickly, and bring their concepts to life in a tangible manner,” he added.
“Even though we haven’t applied this technology in our game development—and we do not plan to employ this technology for content creation—the intriguing aspect for us is figuring out how we can leverage such technology to streamline the game development process, allowing our talented team to concentrate on what makes games remarkable, which is the essence of human creativity.”
Ultimately, there might be significantly less human creativity showcased at Microsoft today, following the studio laying off approximately 2,550 game developers in 2024.
Those interested can discover more about Muse (and view example outputs) on the Microsoft Research website.