Error: unable to get links from server. Please make sure that your site supports either file_get_contents() or the cURL library.
Muse AI was trained in video games Bleeding Edge
Microsoft
Microsoft’s artificial intelligence models can replicate realistic video game footage the company says will help designers create games, but experts are not convinced that the tool will help most game developers. yeah.
Neural networks that can generate consistent, accurate footage from video games are nothing new. A recent Google-created AI produced a fully playable version of a classic computer game destiny No access to the underlying game engine. original destiny, However, it was released in 1993. More modern games are more complex with sophisticated physics and computationally intensive graphics, and have proven difficult for AIS to faithfully replicate.
now, Katja Hofmann Microsoft Research and her colleagues developed an AI model called Muse. This allows you to recreate the complete sequence of multiplayer online battle games Bleeding edge. These sequences follow the physics that underlie the game, and appear to keep players and in-game objects consistent over time. This means that the model has a deeper understanding of the game, says Hofmann.
The Muse is trained with seven years of human gameplay data, including both controllers and video footage. Bleeding EdgeNinja Studios is a Microsoft-owned developer. It works similarly to large language models such as ChatGpt. If given input, it imposes predicting the next gameplay in the form of video game frames and their associated controller actions. “To this day, for me, it’s a very moving thing to me, purely from training models to predict what will come next. I learn a sophisticated and deep understanding of this complex 3D environment,” Hoffman said. I say it.
To understand how people use AI tools like Muse, the team researched game developers and learned which features would be useful. As a result, researchers added the ability to repeatedly adjust changes made on the spot, such as changes to player characters or new objects entering the scene. This could help you come up with new ideas and try out what-if scenarios for developers, says Hofmann.
However, the muse is still limited to generating sequences within the original boundaries Bleeding Edge Games – Can’t come up with new concepts or designs. And I say it’s unclear whether this is a model-specific limitation or something that can be overcome with more training data from other games. Mike Cook King’s College London. “This is a long way from the idea that AI systems can design their own games.”
The ability to generate consistent gameplay sequences is impressive, but developers may prefer greater control, says Cook. “If you create a tool that is actually testing the game code itself, you don’t have to worry about persistence or consistency because you’re running the actual game. So these are introduced by generative AI itself. It’s solving the problem.”
It is promised that the model is designed with developers in mind, he says Georgios Yannakakis The Digital Games Institute at the University of Malta may not be feasible for most developers who don’t have that much training data. “Does that come down to the question of it being worth it?” says Yannakakis. “Microsoft has been collecting data for seven years and training these models to demonstrate what they can actually do. But real game studios can afford it. [to do] this? “
Even Microsoft itself is vague about whether AI-designed games could be on the horizon. When asked if there was a possibility that developers in the Xbox gaming division would use the tool, the company declined to comment.
Hofmann and her team hope that future versions of Muse can generalize beyond training data, but they can create new scenarios and levels for the games they are trained to work in a variety of games. I hope that I can do it. Challenge is because modern games are very complicated.
“One way games distinguish themselves is by changing the system and introducing new concept-level ideas. So machine learning systems go outside of their training data and go beyond what they see. It’s extremely difficult to innovate and invent,” he says.
topic:
Source: www.newscientist.com