Microsoft wants to use generative AI tools to help make video games

An artificial intelligence model from Microsoft can recreate realistic video playing, as the company says, could help designers make games, but experts are not convinced that the tool will be used for most game developers.

Neural networks that can produce coherent and accurate recordings from video games are not new. A recent Google-made AI generated a fully playable version of the classic computer game Doom Without access to the underlying game engine. The original Downfall, However, was released in 1993; More modern games are far more complex with sophisticated physics and computational intensive graphics, which has proved more difficult for AIS to faithfully recreate.

Now Katja Hofmann at Microsoft Research and her colleagues has developed an AI model called Muse, which can recreate full sequences of the multiplayer -Online -fight Bleeding edge. These sequences seem to obey the game’s underlying physics and keep players and objects in the game over time, which means the model has a deep understanding of the game so far, says Hofmann.

Muse is trained on seven years of data on human gameplay, included both controller and video recordings provided by Bleeding edgeMicrosoft-owned developer, Ninja Studios. It acts as large language models like chatgpt; When it gets an input, in the form of a video game frame and its associated controller actions, it is tasked with predicting the gameplay that may come next. “It’s really quite astonishing, even for me now, who cleans from training models to predict what’s gooir to appear next … It teaches a sophisticated, deep understanding of this complex 3D environment,” says Hofmann.

To understand how people can use an AI tool like MUSE, the team also examined game developers to learn what features they would find you. As a result, the researchers added the ability to iteratively adapt to changes made on the go, such as a player’s character that changes or a new object, a scene. This can be useful for coming up with new ideas and trying what-fanarios for developers, says Hofmann.

But Muse is still limited to generating sequences within the bounds of the original Bleeding edge GAME – It can’t come with new concepts or designs. And it is unclear whether this is the inherent restriction of the model, or something that can be overcome with more training data from other games, says Mike Cook at King’s College London. “This is a long, long way away from the idea that AI A system can design games on your own.”

While the ability to generate consists of gameplay sequences are impressive, developers may affect more control, says Cook. “If you build a tool that is actually testing your game, running the game code itself, you don’t have to work on persistence or consistency because it runs news games. So these solve problems that generative AI itself has introduced.

It is promising that model is designed with developers in mind, says Georgios Yannakakis at the Institute of Digital Games at the University of Malta, but it may not be possible for most developers who do not have as much training data. “It comes down to the question of is it worth doing?” Says Yannakakis. “Microsoft spent seven years collecting data and educating these models to demonstrate that you can actually do it. But would an actual gaming studio can afford [to do] This? “

Even Microsoft itself is ambiguous about how A-designed games could be on the horizon: When asked if developers in his Xbox Gaming Division may have used the tool, the company refused to comment.

While Hofmann and her team are hopeful that future versions of mouse will be able to generalize beyond their training data – come up with new scenarios and levels of games on which they are trained, as well as work for games – this will be signal challenge, Says Cook, Becau’s modern game is so complex.

“One of the ways in which a game separates itself is by changing systems and introducing new conceptual levels. It makes it very difficult for machine learning systems to get outside their training data and innovate and invent beyond what they are, ”he says.