Google DeepMind is assembling a new AI research team focused on creating “world models” that can simulate physical environments. These models will be used in both games and robotics.
Taking a major step forward in artificial intelligence, Google DeepMind is launching a new research initiative to develop world models. This team will specialize in AI systems that simulate physical environments, led by Tim Brooks, an expert in video production and simulation technology. Brooks previously co-led OpenAI’s Sora project before joining DeepMind in October.
"World models" represent a developing concept in AI with the potential for wide-ranging applications. From creating real-time interactive media for video games and films to providing realistic training scenarios for robots, these models push the boundaries of AI capabilities. Recently, we saw Nvidia’s Cosmos announcement in this area.
.jpg)
AI-based world models enable systems to understand and predict their surroundings. Like humans use mental models to comprehend the world, AI builds internal representations of objects, physical laws, and event dynamics. This empowers it to forecast future events and plan actions to achieve goals.
A football analogy illustrates this a player calculates speed, distance, and external factors like wind to make a precise pass within seconds. Humans do this instinctively, but it’s a complex computational process in AI.
Google DeepMind’s goal is to become the first to realize artificial general intelligence (AGI). By focusing on world models, the team aims to advance large-scale AI systems. DeepMind is hiring engineers and scientists to join this groundbreaking effort.
Job postings emphasize that "training on video and multimodal data is a critical step toward AGI." World models have the potential to revolutionize visual reasoning, planning, and real-time interactive entertainment . Major competitors in this race include OpenAI and Nvidia.
