AI Reads Brain Waves to Guess What You Just Looked At

Scientists have taken a small but real step toward decoding what the human eye is seeing. In a new paper, researcher Harshit Goyal recorded the brain activity of a single volunteer using EEG — a cap of electrodes that reads electrical signals on the scalp, like a fitness tracker for the brain. While wearing it, the person looked at photos from a set of 200 images. The AI was then asked: which one did they just see?

It got the right answer about 13 percent of the time on its first guess, and picked the correct image within its top 10 guesses 58 percent of the time. Random guessing would land the right image only half a percent of the time, so the system is genuinely picking up meaning — rough ideas like "an animal" or "a food" rather than a photographic memory.

That gap between "rough idea" and "actual picture" is the whole story. The AI never rebuilt the photo in the person's mind. When the researcher tried generating images directly from brain signals, the results were basically static. Every image also had to be shown many times so the noisy brain readings could be averaged out, and when the model was tested on nine other people without retraining, performance fell off a cliff. Brains, it turns out, are as unique as fingerprints.

So what does this mean for you? Not that anyone can pull images from your head — not remotely. But it does show that cheap, non-invasive headgear can extract meaningful information about what you're paying attention to. That is the groundwork for attention-tracking ads, medical tools that check whether a patient is consciously processing what they see, and eventually better prosthetic or communication devices for people who cannot speak. Today it's one person, 200 photos, and a lot of repetition. Tomorrow is where the interesting questions start.