A new team at Google, under the leadership of Tim Brooks, will focus on creating artificial intelligence models capable of simulating the real world
Brooks, was a former co-lead on OpenAI’s video creation model, Sora, who joined Google DeepMind’s AI research group in October.
In a post on X on Monday, Brooks made the announcement, highlighting the project’s lofty objectives.
“DeepMind has ambitious plans to create massive generative models that simulate the world,” Brooks said. “I’m hiring for a new team with this mission.”
Read also: Google, Samsung unveil Eclipsa Audio, a free spatial 3D audio solution
Goal of Google’s proposed AI models
The goal of the new team is to address significant challenges and scale models to unprecedented levels of computing power. This effort will build on the foundational work done by Google’s current AI teams, including Gemini, Veo, and Genie.
Google’s primary AI model, Gemini, manages activities like text generation and picture analysis, whereas Veo is more focused on creating videos.
Google’s approach to “world models,” AI that can replicate interactive games and real-time 3D settings, is Genie, a crucial part of the project. Genie’s most recent iteration, which was shown off in December, is capable of creating a large variety of playable 3D landscapes.
According to Brooks, expanding AI training using video and multimodal data is seen to be crucial to reaching artificial general intelligence, or AI that can carry out any work that a person can.
World models are intended to support fields like interactive entertainment, real-time simulation, autonomous agent planning, and visual reasoning.
Read also: Microsoft plans $80 billion AI investment, half to be allocated to U.S.
New team to focus on developing real-time interactive generation tools
By combining their models with already-existing multimodal models like Gemini, the new team will concentrate on developing real-time interactive generation tools.
Both startups and major IT businesses, such as Decart, Odyssey, and AI researcher Fei-Fei Lee’s World Labs, are paying close attention to world models.
Video games, movies, interactive media, and robotics training simulations are among the industries that these models are thought to have the potential to revolutionise.
Some firms, such as Odyssey, have committed to work with creative people instead of taking their place. It’s unclear if Google will adopt a similar strategy.
World model training also raises unsolved copyright issues, especially when video game content or other unlawful materials are included.
Google, which owns YouTube, has not disclosed which specific videos are being used for training purposes but insists that it has obtained the necessary permissions to utilise YouTube videos in the process.