World Models Will Soon Be Inescapable. Here’s What You Need to Know

Niantic, known for Pokémon Go, has transitioned into the world of AI's cutting-edge topic: world models. "We're very much about reconstructing the real world," states Brian McClendon, CTO of Niantic Spatial. These world models utilize visual data to generate digital environments that mimic physical laws. This technology is expected to become more prevalent, influencing areas like home robotics and autonomous vehicles by 2027.

World models can predict future visual states of the world given a current visual state. This capability allows them to function as realistic simulations, serving as training grounds for robots. Rather than physically moving into environments, robots can "millions of simulations" to learn about their surroundings. This principle applies not only to humanoid robots but also to autonomous vehicles, which rely on split-second decisions informed by physics, causality, and the behavior of other objects and people.

Developers are also exploring the potential of world models in video games, given their relevance to physics and 3D rendering. Major players such as Tencent are investing in this technology. High-profile tech companies like Google and Meta are continually improving their world models, attracting significant investment. Yann LeCun, Meta's former AI chief, raised over a billion dollars for his startup focusing on advanced machine intelligence. Stanford professor Fei-Fei Li founded World Labs, which AMD acquired for over $8 billion.

While you may not directly interact with these world models, your life's data is increasingly contributing to their development. To create accurate simulations, AI labs require real-world data. One individual captured hours of video footage of their daily activities using an iPhone, selling this data to AI companies. As more businesses aim to perfect world models, the need for this kind of data collection will likely grow, potentially leading to more gig work focused on this field.