Google DeepMind is forming a new artificial intelligence (AI) research group dedicated to developing AI models capable of simulating physical environments. These models aim to train robots and create realistic game universes. The initiative will reportedly be led by Tim Brooks, who previously co-led the Sora project at OpenAI and joined DeepMind in October.
New Frontiers in AI Research
“World modeling” is a relatively new area of AI with broad potential applications. It could be used to create real-time interactive media environments for video games and movies, in addition to developing realistic training scenarios for robots and other AI systems. DeepMind is actively recruiting research engineers and scientists for this new endeavor, posting job openings on its Greenhouse website. According to one job posting, the team’s key tasks include training models at scale, curating training data, and exploring methods for integrating these models with multimodal language models. The posting also notes that “scaling pre-training on video and multimodal data is a critical step toward artificial general intelligence.”
By focusing on “world modeling,” DeepMind hopes to advance AI’s capacity for understanding and interacting with complex, dynamic environments. This approach could lay groundwork for more sophisticated robotics, cutting-edge game design, and eventual breakthroughs toward artificial general intelligence (AGI). The new group’s research may also enhance DeepMind’s existing projects, such as its flagship Gemini AI models, the Veo video generator, and Genie, a previously developed world model for simulating 3D gaming environments in real time.
Challenges and Competition
Despite DeepMind’s ambitions, it faces competition from organizations that have already begun exploring similar areas. Nvidia’s Cosmos platform for developing physical AI is one example, and World Labs—a startup founded by Fei-Fei Li, often referred to as the “godmother of AI”—is another. These entities are making significant strides in “world modeling,” potentially giving them a head start in certain aspects of the technology.
NIX Solutions notes that Google aims to achieve AGI before its competitors. The race to be the first to harness the power of superintelligence is heating up. OpenAI CEO Sam Altman recently mentioned that his company is close to achieving AGI, suggesting autonomous AI agents could start entering the workforce as early as this year. While the exact timeline is uncertain, competition in this area is intensifying rapidly, and we’ll keep you updated as developments unfold.
Overall, DeepMind’s new AI research group appears poised to make significant contributions to the field of “world modeling.” The drive to merge large-scale pre-training on video and multimodal data with advanced simulations could accelerate progress toward AGI, transforming industries like robotics, gaming, and beyond.