Google launches team to build AI models simulating the physical world

• Published On: 07 January 2025 at 21:04

Google is forming a new team dedicated to creating artificial intelligence models that can simulate the physical world.

The team will be led by Tim Brooks, a former co-lead on OpenAI's video generation model, and Sora, who joined Google DeepMind's AI research lab in October. Brooks announced the initiative in a post on X on Monday, emphasising the project's ambitious goals.

"DeepMind has ambitious plans to create massive generative models that simulate the world," Brooks told me. "I'm hiring for a new team with this mission."

The new team will expand on the work of Google's existing Gemini, Veo, and Genie teams, aiming to address key challenges and scale models to unprecedented computational levels. Gemini, Google's flagship AI model, handles image analysis and text generation, whereas Veo specialises in video generation.

Genie, a key component of the project, represents Google's approach to "world models," AI capable of simulating real-time 3D environments and interactive games. The most recent version of Genie, which was previewed in December, can create a wide range of playable 3D worlds.

Brooks explained that scaling AI training with video and multimodal data is critical for achieving artificial general intelligence, which refers to AI that can perform any task that a human can.

The goal is to use world models to power domains like visual reasoning, real-time simulation, autonomous agent planning, and interactive entertainment.

The new team will concentrate on real-time interactive generation tools, integrating their models with existing multimodal models such as Gemini to enhance their capabilities.

World models are gaining popularity among startups and large tech companies, including AI researcher Fei-Fei Lee's World Labs, Decart, and Odyssey. These models have the potential to revolutionise fields such as interactive media, video games, movies, and robotics training simulations.

Some startups, such as Odyssey, have committed to working alongside creative professionals rather than replacing them. It remains to be seen whether Google will adopt a similar strategy.

Furthermore, there are unresolved copyright issues with world model training, especially if it involves video game content or other unlicensed material. Google, which owns YouTube, says it has permission to use YouTube videos for model training but hasn't revealed which videos are being used.