Microsoft Unveils New AI Model to Generate Video Game Gameplay

In collaboration with Xbox game studio Ninja Theory, Microsoft Research has introduced an innovative AI model called Muse. This model can generate gameplay visuals and predict controller inputs, which the company regards as a significant advancement in AI for gaming.

Trained on over a billion images and controller actions from the 2020 game Bleeding Edge, Muse clearly understands 3D game environments and physics, responding dynamically to player interactions. Microsoft has categorized Muse as the first model in a new group called World and Human Action Models (WHAM).

However, the current prototype has its limitations. It can only produce low-resolution gameplay at 300x180 pixels, and during real-time demonstrations, it ran at just 10 frames per second, according to Microsoft researchers.
Phil Spencer, CEO of Microsoft Gaming, proposed that this technology could contribute to game preservation by enabling classic games to run on modern hardware without needing the original engine. The company also envisions it playing a role in game prototyping and development.

In light of industry concerns over AI replacing human developers—especially following major gaming industry layoffs in 2024—Ninja Theory's studio head, Dom Matthews, clarified that Muse is not intended for content creation. "For us, the fascinating aspect is how we can use this technology to streamline the game-making process for our talented team," Matthews said.

Microsoft plans to offer some interactive AI gaming experiences through its Copilot Labs, though no specific release date has been provided. The company also mentioned that individual Xbox studios will be free to decide how they integrate generative AI into their projects.

Source: Times of India