Can reading human brains help teach robots to do chores?

A data company called Encord is now testing a novel way to bridge this gap by recording human brain activity while people perform physical tasks. A trainer wears a sensory headset that tracks their brain waves—specifically looking for signals related to focus, mistakes, and intent—while they use robot arms to stack blocks or move objects. The goal is to see if these brain signals provide the robot with better clues about how a human performs a task, hopefully leading to more capable and reliable robots that don't need to be told every single movement to make.

When you train a computer, you are essentially showing it a series of examples and asking it to find the pattern. For a robot, the data needs to be incredibly high quality. If you want a robot to plug in a cable, recording a low-resolution video is not enough. You need to know exactly how the human arm moves, when they apply pressure, and where they make micro-adjustments. Current methods involve having a human guide the robot remotely. Encord is trying to make this process smarter by attaching sensors to human trainers that track electrical signals in their muscles or brain waves. By tagging the video footage with these biological signals, the AI gets a map of the performance. For example, if a human hesitates or makes a mistake, the sensors flag that moment. This teaches the AI not just what the action looks like, but the logic behind the effort being put into the movement.

The fundamental challenge in physical AI is that we cannot scrape the internet for physical experience. A chatbot could learn from 20 years of digitized writing for free. A robot that folds laundry must have that specific action performed by a human and captured by sensors, which is slow and expensive. This creates an economic bottleneck for companies building human-like robots. By trying to extract more value from every minute a human spends training a robot—using brain waves to make the training more efficient—these companies are essentially trying to make the most of a very limited, very expensive resource: real human experience. Whether this becomes the standard way we build physical AI or simply a temporary experiment in a crowded field remains to be seen.