To teach an artificial intelligence to tell a duck from a cat, you need to feed it millions of labelled images. Millions. Your two-month-old has seen maybe a few hundred objects in their entire life. No tutors. No learning apps. And yet their brain can already tell ducks from cats, birds from trees, dogs from shopping trolleys. A 2026 fMRI study of 130 infants showed that at eight weeks old, your baby's brain is sorting the visual world into categories using patterns that match the most advanced AI on the planet.
What researchers call visual categorisation is the ability to group objects by type, even when the objects look nothing like each other. A Labrador and a Chihuahua are both dogs. A bare oak in winter and a leafy mango are both trees. It is the most basic move your brain makes every time it looks at anything. And in your two-month-old, that move is already online. It is happening in a region called the ventral visual cortex, the part of the brain that answers the question what is that?
Picture invisible folders inside that little head. One folder is filling up with birds. Feathered, beaked, round, spiky, big, tiny. Another is filling up with trees. Another with shopping trolleys. Each folder collects examples that share some deep feature, even when no two examples look exactly alike. And your baby is not thinking about any of this. Their brain is doing it automatically, essentially even before they can hold up their own head.
Based on O'Doherty et al. (2026): 130 infants show distinct neural patterns for different object categories at just 8 weeks old.
A team from Trinity College Dublin, Queen's University Belfast, and Stanford University recruited 130 two-month-old infants and placed them inside an fMRI scanner, the same imaging technology that maps adult brains. This was a monumental undertaking. Babies are wriggly. So instead of the clinical white room you might imagine, the researchers built a soft beanbag nest, gave each baby sound-cancelling headphones, and showed them bright, colourful images for 15 to 20 minutes while the scanner measured their neural activity.
They tested twelve different categories: cat, bird, rubber duck, shopping trolley, tree, and others. And what they found was beautiful. When a baby looked at a rubber duck and then another rubber duck (a different toy, different angle, different lighting), the pattern of brain activity in their visual cortex matched itself more closely than it matched the pattern for a cat. The brain was treating ducks as a single category. It was grouping them. And those neural patterns looked strikingly similar to the ones that appear inside deep neural networks, the AI systems that teach themselves to classify objects by spotting shared features. At eight weeks old, your baby's brain was already doing something that mirrors how the most advanced AI learns to see.
The researchers scanned 65 of the same babies again at nine months old. The category patterns in the visual cortex had become even stronger, more stable, more finely tuned. Like watching a rough pencil sketch harden into precise ink lines. The filing system your baby built at eight weeks was still there, and it had been organising itself ever since.
And this is not just one group of babies. The same fundamental pattern shows up in infants studied across the US, Ireland, the UK, and beyond. It is a universal human starting kit. Every baby, everywhere, arrives with this system already running.
So the next time your two-month-old stares at something with that wide-eyed look, know that behind those beautiful eyes, the filing cabinets are open and busy. A whole visual language is being built, long before they have any way to tell you about it. They are making sense of the world already. You just get to watch it happen.
Join the conversation
Thoughts, questions, or research you've come across? Drop them below.