Technology that reconstructs the physical world is moving from experimental labs into mainstream applications. Companies are now using world models to create digital replicas of environments, allowing machines to practice tasks without physical risk. This shift marks a departure from text-based artificial intelligence toward systems that understand visual inputs and physical laws.
Niantic and the Shift to Spatial Computing
Niantic, the company behind the augmented reality game Pokemon Go, has pivoted its focus toward spatial computing. The firm sold its gaming division last year to concentrate on building digital twins of real-world locations. Brian McClendon, chief technology officer at Niantic Spatial, stated that the company aims to reconstruct the real world using data gathered from millions of users.
Unlike large language models that predict text, world models process visual data to predict how a scene will change. Oliver Cameron, CEO of the AI lab Odyssey, explained that these systems take an image of a current state and generate the next visual state. This capability allows developers to create simulations that look and feel real, providing a safe space for training complex machines.
McClendon noted that robots can now practice in these digital environments. By taking photos of a specific location, engineers can generate a 3D model and run millions of simulations. This process trains robots to navigate obstacles and interact with objects without needing physical presence in the target area.
Autonomous Vehicles and Safety Simulations
The automotive industry is also adopting these technologies to improve safety. Jamie Shotton, chief scientist at Wayve, emphasized that world models must understand cause and effect. The software needs to predict how other objects and people will react to the actions of a vehicle.
Wayve uses dashcam footage to train its generative world models. The startup recreates hazardous road conditions that are too dangerous to test in real life. This approach allows autonomous cars to learn how to handle rare but critical scenarios, such as sudden obstacles or extreme weather, without putting human drivers at risk.
Investors are pouring money into this sector. Yann LeCun, former chief AI scientist at Meta, raised over a billion dollars for his startup Advanced Machines Intelligence. Additionally, AMD acquired World Labs, a company founded by Stanford computer scientist Fei-Fei Li, for more than $8 billion. These investments highlight the growing confidence in world models as a foundational technology.
Impact on Labor and Gaming
The demand for training data has created new gig work opportunities. AI labs need real people to record daily activities to build accurate simulations. Individuals have been paid to record videos of chores like folding laundry or washing dishes. This data helps machines learn the nuances of human movement and object interaction.
Video game developers are also integrating these tools. Major companies like Tencent are investing in world models to enhance physics and 3D rendering. As these systems become more sophisticated, they will likely influence how games are designed and how virtual environments are populated.
Experts predict that 2027 will be a significant year for the expansion of home robotics and driverless cars. The underlying technology is already shaping the labor market and the development of autonomous systems. As these digital replicas become more detailed, the line between physical and simulated practice will continue to blur.
Source: Wired
