Games

Microsoft CEO Satya Nadella Calls Muse AI ‘a Massive Moment of Wow’ As Xbox Looks to Train it on ‘Catalogue of Games’

BBG News

After a successful debut a week ago, Microsoft’s Muse AI has earned praise from CEO Satya Nadella, who referred to it as an “extraordinarily impressive” moment.

As per Nadella, Microsoft intends to educate its Muse AI using a collection of video games, aiming to begin utilizing these games for gameplay generation soon. This advanced AI technology will serve as a tool to produce gameplay videos, thereby accelerating the development process.

During the Dwarkesh Patel podcast, as reported by Eurogamer, Nadella stated that he is looking for a system capable of creating games that are both consistent in their design and offer diversity in representation. Additionally, these games should be resilient to user modifications. In other words, he’s seeking a tool that can strike a balance between maintaining the core game mechanics while allowing for a wide range of variations and adaptability to user-generated content.

He expressed his enthusiasm about introducing a collection of games that would be powered by the models we currently have or trained, followed by actually playing those generated games.

When Phil Spencer initially demonstrated it to me, he was holding an Xbox controller and this device essentially transformed input into output in a way that matched the input, aligning with the game. Nadella continued, “That’s a truly awe-inspiring moment for me, reminiscent of the first time we witnessed ChatGPT generating coherent sentences, or Dall-E creating art, or Sora performing tasks. This is one instance of such groundbreaking moments.

Although Microsoft is excited about its Muse AI tool, some, like gaming AI expert Dr. Michael Cook, have expressed skepticism about its practicality. In a recent blog post, Cook challenged Xbox CEO Phil Spencer’s claims that it could be used as a preservation tool. To put it simply, Cook found Spencer’s comments on the topic to be misguided or uninformed.

They created a tool that enables game developers to modify a game stage employing pre-existing game elements, such as placing a jump pad in a location where none previously existed. Afterward, they provided this freshly designed level to their model and asked it to predict what a video of a player playing from this new position might appear like,” Cook elaborated.

He pointed out that it’s remarkable how it manages to produce these results based on visual data, considering factors such as lighting, camera angles, user interface, and more are complex for an AI model to process. Despite the vast amount of data collected, careful annotation of datasets, and so forth, it was still barely capable of predicting player behavior in the footage.

The researchers who developed this likely anticipate that it will become increasingly effective and streamlined with time, potentially making it cost-effective or manageable for smaller development teams. Yet, an important query arises: How do we secure video recordings of individuals playing our game initially?

Cook noted: “While this model could perfectly mimic the original game software, it doesn’t represent the ultimate solution for game conservation. A generative model showing possible game footage might serve as an interesting side note in the larger preservation effort, but it will never match up to other strategies we employ for addressing this issue.

Read More

2025-02-24 21:42

Previous post Hollywood Mourns: John Lawlor, Star of The Facts of Life, Passes Away at 83
Next post Avowed’s director hopes for an expansion or sequel