Google’s New AI Team Aims to Simulate the Physical World
In a fascinating move towards the future of artificial intelligence, Google is assembling a fresh team focused on developing AI models capable of simulating our physical world. Tim Brooks, who recently transitioned from OpenAI’s video generator project, Sora, to Google DeepMind, is set to lead this promising endeavor. Brooks unveiled this exciting update in a post on X, confirming the team’s integration into Google DeepMind.
“DeepMind has ambitious plans to make massive generative models that simulate the world,” Brooks expressed. “I’m hiring for a new team with this mission.”
— Tim Brooks
The job listings he shared provide insight into the collaborative nature of this project. The new team will work alongside Google’s existing Gemini, Veo, and Genie teams. Their mission? To tackle critical challenges and scale these models to unprecedented levels of computational power.
- Gemini: Google’s flagship AI series designed for image analysis and text generation.
- Veo: Google’s proprietary video generation model.
- Genie: An AI model that can simulate games and 3D environments in real time.
The latest Genie model, showcased in December, is already capable of generating diverse 3D worlds. The potential applications are vast, ranging from visual reasoning and simulation to planning for embodied agents and offering real-time interactive entertainment.
“We believe scaling [AI training] on video and multimodal data is on the critical path to artificial general intelligence,” a job description explained.
Artificial General Intelligence (AGI) refers to AI that can perform any task a human can. Brooks’ team aims to develop “real-time interactive generation” tools, integrating them with multimodal models like Gemini. This area is buzzing with interest from startups and tech giants alike, including Fei-Fei Lee’s World Labs and Israeli startup Decart.
However, this technological progress isn’t without its challenges. A recent Wired investigation highlighted how game studios like Activision Blizzard are leveraging AI to boost productivity amidst workforce reductions. Additionally, a 2024 study by the Animation Guild anticipated significant disruption in U.S.-based film, TV, and animation jobs due to AI advancements by 2026.
Some startups in the world modeling arena, like Odyssey, have committed to working alongside creative professionals rather than replacing them. It remains to be seen if Google will adopt a similar approach. Meanwhile, copyright concerns arise as reports suggest some world models use unlicensed video game playthroughs for training purposes. While Google claims it has permissions under YouTube’s terms of service for training its models on platform videos, specifics remain undisclosed.
