Can AI learn to navigate the physical world like a human?

General Intuition is a startup that recently received a massive funding injection, valuing the company at six billion dollars. They are building a foundation model, which is a massive, flexible AI system designed to handle many different tasks rather than just one specific job. The company plans to use this funding to teach their AI how to control robotic bodies. They are building this by using a large dataset, which is a collection of information used to teach the AI patterns. In this case, they are using thousands of hours of video game recordings that show exactly what buttons a human player pressed at every moment, essentially teaching the AI the relationship between a command and a physical action.

When you ask an AI like ChatGPT a question, it predicts the next likely word in a sentence. General Intuition is building something called a large action model. Think of this like teaching a student by showing them a recording of a master at work. By feeding the AI millions of hours of gameplay, the system learns to mimic how a person reacts to visual changes on a screen. If a character sees a ledge, the model learns the button sequence required to jump. By scaling this up, they hope to transfer that same intuition to a robot. Instead of coding every single movement a robot should make, the AI watches human behavior and learns how to navigate space. It is moving the AI out of the digital void and into the physical environment.

The reason this is so valuable to investors is that it addresses one of the hardest problems in robotics: generalisation. Most robots are programmed to perform one specific task in a factory, like picking up a box in the exact same spot every time. If you move the box, the robot gets stuck. By building a model that understands the concept of moving through space, these companies hope to create machines that can adapt to messy, unpredictable human environments. Whether this will lead to helpful home assistants or more efficient industrial machines is still an open question, but the race is now on to see if we can teach AI to understand the physical world as well as it understands language.