Google Launches New AI Team to Simulate the Physical World

Google has announced the formation of a new team focused on creating artificial intelligence models that can simulate the physical world. The team will be led by Tim Brooks, who joined Google DeepMind in October after co-leading OpenAI’s Sora video generation model.

In a post on social media platform X, Brooks shared the ambitious goals of the project, stating, “DeepMind has ambitious plans to create massive generative models that simulate the world.” He emphasized that the team would be tasked with tackling significant challenges and scaling these models to unprecedented computational levels.

This new initiative will build on existing efforts from Google’s Gemini, Veo, and Genie teams. Gemini is Google’s flagship AI model, handling tasks such as image analysis and text generation, while Veo focuses on video generation. Genie, which is integral to the new initiative, is a system for creating “world models,” or AI capable of simulating interactive 3D environments and games. The most recent version of Genie, previewed in December, can generate various playable 3D worlds.

Brooks explained that video and multimodal data are essential for training AI systems to reach artificial general intelligence (AGI), which would be capable of performing any task that humans can. The team’s work will focus on creating AI tools for real-time simulations, autonomous agent planning, visual reasoning, and interactive entertainment.

The new team will aim to integrate these world models with existing systems like Gemini, advancing capabilities in interactive generation. World models are receiving growing attention from both startups and large tech companies. These models are seen as key to transforming areas like video games, interactive media, movies, and robotics training simulations.

While startups such as Odyssey focus on collaborating with creative professionals, it remains unclear whether Google will adopt a similar approach. Moreover, there are ongoing concerns regarding the use of unlicensed content in training world models, particularly in video games. Google, which owns YouTube, asserts it has permission to use YouTube videos for training purposes, though it has not disclosed the specific videos being used.