Several years ago I read Jeff Hawkins’s A Thousand Brains, which dug into his theories on how our brain works. In it he argued intelligence comes from thousands of cortical columns that each learn their own model of the world and then vote to reach agreement. Then AI engulfed the tech world and along the way I kept finding myself thinking about Jeff's theory.
To be clear I am neither a neuroscientist nor an AI researcher, but how Jeff described our brains and how today’s AI models are built have interesting parallels. Specifically, a common site of building blocks for intelligence. How so?
Hawkins estimates the part of the brain that handles seeing, hearing, language, and thinking contains about 150,000 cortical columns. Each of these columns effectively runs the same program on the same biology. But they differ in what they're connected to (eyes ears fingertips). Each column learns things by combining what it senses with where it senses it.
The shape of a coffee cup in the dark adds up when you touch the rim here and the handle. Hawkins claims the same process works for everything we learn, even math. Cortical columns work together by sharing their guesses until they agree on what’s there. That’s why you recognize a cup faster by grabbing it with your whole hand and where the “thousand brains” title comes from.
AI’s Building Block Transformers, the architecture behind many of today’s AI systems, are also built from one repeated block. It handles text, images, audio, and video once they’re broken into small chunks called tokens. Each token carries a “what” (the word “cup”) and a “where” (third token in the sentence). A mechanism called attention lets each token look at the others and pull in what’s relevant, which is how the model works out how the pieces fit together. It plays a role a lot like voting between columns, with many units sharing information to settle on one interpretation. Hawkins’s “where” is a feature’s location on the object itself (the handle sits on the side of the cup). A transformer’s “where” usually means a token’s position in its input, so the comparison is a loose one. Images get closer, since a patch’s position in a photo corresponds to a place in the scene. During training, many language models improve by predicting what comes next and adjusting after wrong guesses, much like Hawkins says the brain learns from surprise. Transformers learn most of what they know during training. Once deployed, they can adapt within a conversation, but their weights stay fixed, while Hawkins describes columns that keep rewiring themselves as they experience the world. Researchers are working on models that keep learning after deployment, which would bring the two closer together. [Image: text, a photo, and audio broken into tokens and fed into the same stack of transformer blocks] One Building Block In both brains and AI, intelligence may come from one building block, repeated many times and shaped by experience. The two still differ a lot up close, and Hawkins’s Thousand Brains Project is testing whether AI built around cortical columns can close that gap. I’m curious where that leads.