Google Assembles a New Division to Develop AI Capable of Simulating Physical Environments

Google has launched a new team within its DeepMind division dedicated to advancing AI models capable of simulating physical environments. This initiative, led by Tim Brooks, a former co-lead on OpenAI’s video generation project Sora, signifies a major step forward in the pursuit of artificial general intelligence (AGI).

Tim Brooks, who joined Google DeepMind in October 2024, shared the news on X (formerly Twitter), emphasizing the ambitious goals of his new division. According to Brooks, the team will focus on building massive generative models designed to simulate the physical world. He also announced hiring efforts for this mission, indicating that the company is actively expanding its research capacity in this field.

The new modeling team will collaborate closely with other prominent Google AI projects, including Gemini, Veo, and Genie. Each of these projects contributes unique capabilities to the overall goal of world modeling. Gemini is Google’s flagship AI model, capable of tasks like image analysis and text generation. Veo, on the other hand, specializes in video generation, while Genie focuses on simulating interactive 3D environments and video game worlds in real time.

Genie, which was recently updated to its second version, has demonstrated the ability to generate highly diverse and playable 3D worlds. This advancement aligns with the team’s focus on developing models that can simulate complex environments, a crucial component in the path toward AGI. The job listings associated with Brooks’ new team also emphasize a focus on scaling models to leverage the highest levels of computational power available, a requirement for developing such expansive and detailed simulations.

The Path Toward Artificial General Intelligence (AGI)

Brooks’ team is driven by the broader goal of achieving AGI—artificial intelligence capable of performing any intellectual task a human can. According to the job postings, the team will focus on scaling video and multimodal data training, which involves integrating different types of media, such as video, text, and 3D models. This approach is believed to be critical for developing systems capable of general reasoning, planning for embodied agents, and enabling real-time interactive experiences.

The research and development process will also explore real-time interactive generation tools. These tools could potentially transform industries such as game development, animation, and robotics by enabling the creation of realistic virtual worlds and interactive simulations. The implications extend beyond entertainment, as these models could also be used in scientific simulations and robotics training environments.

Mixed Reactions from the Creative Industry

While the technological advancements in AI world modeling are groundbreaking, they have sparked mixed reactions within creative industries. Some see these models as powerful tools for enhancing creativity and productivity, while others view them as a threat to job security.

A 2024 study commissioned by the Animation Guild warned that over 100,000 jobs in the U.S. film, television, and animation sectors could face disruption by AI technologies by 2026. Companies like Activision Blizzard have reportedly turned to AI for increased productivity, sometimes at the cost of significant layoffs.

In contrast, some startups in the world modeling space, such as Odyssey, have committed to working alongside creative professionals rather than replacing them. It remains to be seen whether Google DeepMind will adopt a similar approach as it advances its world modeling initiatives.

Copyright Concerns and Ethical Challenges

Another unresolved issue in the development of AI world models involves the sourcing of training data. Some models appear to be trained on video game playthroughs and other copyrighted content, raising concerns about intellectual property rights. While Google, as the owner of YouTube, claims it has permission to use videos on its platform for AI training under its terms of service, the company has yet to disclose the specific datasets used.

The debate surrounding the ethical use of copyrighted content in AI model training continues to unfold, with industry experts calling for greater transparency and clearer legal frameworks.

As the race to develop advanced world models accelerates, Google’s new DeepMind team stands at the forefront of this technological frontier. Whether it will balance innovation with ethical considerations remains a pressing question for the industry moving forward.