AI
World models are coming to everyday life—here's how AI will learn to see the real world
As companies from Meta to Google race to perfect digital simulations of physical reality, these AI systems are quietly reshaping robotics, autonomous vehicles, and the job market.
WIRED — AI
The same environmental scans you helped create while hunting Pokémon in 2016 are now feeding a technological shift that will reshape how robots and driverless cars operate. Unlike language models that predict the next word in a sequence, world models process images and generate entire simulated environments complete with physics—turning video feeds into digital training grounds.
What world models do#
Think of a world model as a system that takes a visual snapshot of reality and projects what comes next. According to Oliver Cameron, CEO of Odyssey, an AI lab focused on the technology, these systems can then create simulations that "look and feel real" while remaining purely digital.
For robotics and autonomous vehicles, this capability has enormous practical value. Rather than learning through trial and error in the physical world—which is dangerous and expensive—machines can run millions of simulated trials in a single digital replica of their intended environment. A company could photograph a kitchen, reconstruct it in 3D, and train a robot to navigate and manipulate objects within that space without ever risking a misstep.
The push from industry leaders#
The race to dominate this space has attracted some of AI's biggest names and deepest pockets. Yann LeCun, Meta's former chief AI scientist, secured over $1 billion to fund Advanced Machines Intelligence. Meanwhile, Stanford computer scientist Fei-Fei Li launched World Labs in 2024—which AMD then acquired for more than $8 billion just last year.
Even Niantic, the company behind Pokémon Go, has pivoted its focus toward world models. "We're very much about reconstructing the real world," says Brian McClendon, chief technology officer of Niantic Spatial, noting the company intends to leverage its accumulated environmental data to build accurate digital replicas.
Real-world applications arriving soon#
By 2027, world models are expected to catalyze major breakthroughs in two areas:
- Home robotics: Humanoid robots training in simulated domestic spaces before deployment
- Autonomous vehicles: Driverless cars learning to navigate hazardous conditions without road testing
Wayve, a company building world models specifically for vehicles, uses dashcam footage to recreate dangerous traffic scenarios and edge cases—situations far too risky to test on actual roads.

