"Expected Applications for Robots and Autonomous Vehicles"
On June 11 (local time), Meta Platforms, the parent company of Facebook, announced that it will launch a new artificial intelligence (AI) "world model" designed to better understand physical rules in real-world environments.
The "world model" is a system that enables AI to learn physical rules and simulate them independently. Meta explained that its proprietary model, called "V-JEPA2," is capable of understanding and predicting the physical world. The company added that this allows AI to learn, plan, and make decisions in ways similar to humans.
For example, the model helps AI understand principles such as a ball rolling off a table and then falling, or realizing that an object that disappears from view is merely temporarily obscured rather than having vanished completely. Meta emphasized that this model could offer significant advantages for the development of technologies such as delivery robots and autonomous vehicles.
Yann LeCun, Meta's chief AI scientist, stated, "Teaching machines to understand the physical world is very different from teaching them language," highlighting that this model is different from existing large language models (LLMs). He further explained, "The world model is like a digital twin of reality for AI to understand the world and predict the outcomes of its actions," adding, "This enables AI to devise action plans to accomplish assigned tasks."
As efforts continue to surpass the technologies underlying current generative AI, interest in world models that understand the physical world is growing. DeepMind, Google's AI organization, is also developing its own world model capable of simulating games and 3D environments in real time.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
