Google has unveiled Genie 3, a cutting-edge “world model” designed to train AI robots in simulated spaces like warehouses, a development the tech giant claims is a major step toward human-level artificial general intelligence (AGI).
The new technology, announced by Google DeepMind, is built to create realistic, interactive environments where AI systems can practise performing tasks without needing physical deployment. According to Google, such models could accelerate the training of robots and autonomous vehicles while also offering humans entirely new ways to experience virtual simulations, from navigating industrial spaces to skiing down alpine slopes.
A Step Towards Artificial General Intelligence
Unlike traditional AI, which is often limited to single tasks such as translation or chess, AGI aims to match or exceed human capability across a wide range of activities. Google says its Genie 3 system could help achieve that by giving AI agents the ability to understand, interact and adapt to complex virtual worlds.
“World models are foundational for the development of autonomous AI agents capable of making decisions and taking action in dynamic environments,” DeepMind said in a statement. “We expect this technology to play a critical role as we push toward AGI, and agents play a greater role in the world.”
While Genie 3 remains in development and is not yet available for public use, Google says the model can generate simulations instantly from text prompts. In demonstrations, it created environments with fully interactive physics—such as warehouses complete with staff and machinery—and even recreational settings like ski resorts, where scenarios could be adjusted in real-time, including adding herds of deer on slopes with a simple text input.
Competing in the AI Race
The announcement comes as competition intensifies in the AI sector. Just days before, OpenAI CEO Sam Altman teased the company’s forthcoming GPT-5 model, signalling the high-stakes race to build next-generation AI capabilities.
For Google, Genie 3 builds on a broader effort to integrate AI into immersive environments. The company previously introduced Sima, a virtual agent capable of performing tasks within video games, though like Genie 3, it has not been released to the public.
Experts see these world models as pivotal to the future of robotics and AI. Professor Subramanian Ramamoorthy, Chair of Robot Learning and Autonomy at the University of Edinburgh, said,
“To achieve flexible decision-making, robots need to anticipate the consequences of different actions to choose the best one to execute in the physical world. World models are extremely important in this process.”
Andrew Rogoyski, Director at the Institute for People-Centred AI at the University of Surrey, added,
“If you give a disembodied AI the ability to be embodied—albeit virtually—it can explore and learn, effectively growing its capabilities in ways not possible with static training data.”
From Data to Action
Traditional large language models, which underpin popular chatbots like ChatGPT, excel at processing information and generating text. However, they lack the ability to take physical action. World models like Genie 3 offer a bridge by enabling AI systems to simulate interactions in virtual environments, where they can test decision-making strategies before applying them in the real world.
Google has not given a timeline for Genie 3’s public release, citing ongoing limitations. Yet, the company believes its approach could transform industries ranging from logistics and manufacturing to entertainment and training simulations.