Google AI to Simulate the Physical World - New Team Formed
Google Establishes New AI Team Focused on Physical World Simulation
A new team is being assembled within Google to concentrate on the development of artificial intelligence models capable of simulating the physical environment. This initiative signifies a substantial investment in advanced AI research.
Leadership and Integration with DeepMind
Tim Brooks, formerly a co-leader of OpenAI’s video generation technology, Sora, will spearhead this new team. He transitioned to Google’s AI research division, Google DeepMind, in October and announced his new role via a post on X. The team will operate as an integral part of Google DeepMind’s broader efforts.
Brooks articulated that DeepMind is pursuing ambitious goals to create extensive generative models that accurately simulate the world around us. He is currently recruiting individuals to join this mission.
Collaboration with Existing Google AI Projects
Job postings linked by Brooks indicate that the new modeling team will work in conjunction with, and build upon, the advancements made by Google’s Gemini, Veo, and Genie teams. Their focus will be on addressing “critical new problems” and expanding model capabilities through increased computational power.
Gemini serves as Google’s primary series of AI models, designed for tasks such as image analysis and text generation. Veo represents Google’s dedicated video generation model, while Genie is Google’s approach to creating a world model.
Genie: A Real-Time 3D World Simulator
Genie, specifically, is an AI designed to simulate games and three-dimensional environments in real time. A recent preview of Google’s latest Genie model, showcased in December, demonstrated its ability to generate a diverse range of interactive 3D worlds.
The Path to Artificial General Intelligence
One job description highlights the belief that scaling AI training using video and multimodal data is crucial for achieving artificial general intelligence (AGI). AGI refers to AI systems capable of performing any intellectual task that a human being can.
World models are anticipated to have a significant impact across numerous fields, including visual reasoning, simulation, planning for embodied agents, and the creation of real-time interactive entertainment.
Developing Interactive Generation Tools
Brooks’ team will concentrate on developing tools for “real-time interactive generation” based on the models they create. They will also investigate methods for integrating these models with existing multimodal models, such as Gemini.
Competition in the World Modeling Space
Several startups and established technology companies are actively pursuing the development of world models. These include World Labs, founded by influential AI researcher Fei-Fei Lee, the Israeli startup Decart, and Odyssey. The potential applications of world models are vast, ranging from interactive media like video games and movies to realistic simulations for robot training.
Creative Industry Concerns
However, the emergence of this technology has elicited mixed reactions from creative professionals.
A recent Wired investigation revealed that game studios, including Activision Blizzard, are utilizing AI to streamline operations, boost productivity, and address workforce shortages. Furthermore, a 2024 study commissioned by the Animation Guild estimated that over 100,000 U.S. jobs in film, television, and animation could be affected by AI by 2026.
Some startups, like Odyssey, have committed to collaborating with creative professionals rather than replacing them. It remains to be seen whether Google will adopt a similar approach.
Copyright and Data Sourcing Issues
The issue of copyright remains unresolved. Certain world models appear to be trained on video game playthroughs, potentially exposing their developers to legal challenges if the videos were used without proper licensing.
Google, as the owner of YouTube, maintains that it possesses the necessary permissions to train its models on YouTube videos, in accordance with the platform’s terms of service. However, the company has not disclosed the specific videos used for training purposes.
Stay informed with TechCrunch’s AI newsletter! Subscribe here to receive it in your inbox every Wednesday.