The Neocortex Builds a Model of the World

How your brain is a prediction machine

Over the past few weeks I've introduced a few concepts related to Jeff Hawkin's A Thousand Brains theory of intelligence. In The Organ of Intelligence I introduced the neocortex and its astonishing organization into trillions of neurons organized into layers across different cortical columns, each about the size of a grain of rice.

In Vernon Mountcastle's Big Idea I introduced Mountcastle's ground-breaking and controversial claim that all of intelligence can be explained by one universal algorithm inherent in every cortical column across all different kinds of intelligence from vision to logic to language.

Since writing those pieces, I've wanted to dig deeper into exactly how these cortical columns work but I've realized that we first need to zoom out and make sure we are thinking about the neocortex in the right way from a high level.

One of the foundational claims of Jeff Hawkin's theory is that the neocortex builds a predictive model of the world.

Everything you know is stored in this model including how objects look, feel, and sound, where things are located in space, and how things change when you interact with them.

Stored in your neocortex are tens of thousands of models of objects, words, and ideas. Intelligence requires learning a model of the world and updating it continuously.

What is a model?

At this point you might be wondering what exactly a model is, and if you are, then you are like me and might be bothered when experts take for granted words with fuzzy definitions like "model"—so here is my attempt to understand and explain it.

A model is a structured way of thinking about an object or concept. You might be familiar with the use of models in architecture where they are used to help us visualize and think critically about a structure as we plan it:

Models can also represent high level, intangible concepts such as a liberal democratic society:

I like to think of models as shortcuts for objects and concepts that we need to hold in our brain and use everyday. For example, in our political discourse everyday we throw around enormous concepts like "liberty" or "individual rights." Each of us has our own working model of what these concepts mean and how important they should be. Models are critical to our use of our intelligence.

What is a prediction?

If we accept that our brain uses models for our knowledge of the entire world, where does the prediction part come into play? Well, if intelligence is about updating our models of the world then we need a mechanism for detecting changes. Our brain does this by constantly making predictions about what we will see, touch, hear, or taste based on our model of the world.

Most of our prediction-making happens unconsciously. Just the other week I left work to head home. On my way out I picked up my backpack. My brain holds a model of my backpack and everything inside. It knows what it feels like, about how heavy it is, and what it is like to wear it.

On this particular day, after a few minutes of walking, I noticed that my backpack felt lighter than it usually feels, so I stopped and inspected my backpack for missing items. In other words, my brain made a prediction about the backpack based on its model and realized that it might need to update. Maybe my backpack actually is this light? This time, however, I found that my water bottle was missing. In this case, my model was right the first time and my brain noticed a difference because it made a prediction.

This story represents, I think, the concept of prediction-making: the neocortex builds a model of the world and makes predictions about what we will sense each second based on that model. Learning is the process of changing our models. My model of the heart in elementary school has been vastly updated since taking anatomy and physiology in high school, for example.

So that is the neocortex in a nutshell: a model-building and prediction-making machine. Jeff Hawkins has since showed that models of complete objects are actually stored inside individual cortical columns, and that predictions occur inside individual neurons, but more on that later. For now, I need to go update my model on what my lunch will be.

🎉 Some Personal News 🎉

This week, I will be starting medical school as a part of the Medical Scientist Training Program in the U.S. This is a huge career milestone for me as I pursue a career as a physician scientist.

What does this mean for Synapse? I'm not quite sure yet, but I imagine my future newsletters might contain more clinical or health-related themes as we continue to explore the nature of the human experience from a neuroscientific perspective.

I'm also thinking about starting an irregular sub-column within Synapse called Inside a Doctor's Training where I will informally share my experience as a medical student within our healthcare system. We all have connections to the healthcare system and I know some people might be intrigued to know what it is like to train to be a doctor. If this is something you'd enjoy reading, let me know by replying to this email or commenting below!

Anyways, this was a longer edition, but I hope you enjoyed it. In the coming weeks I'm hoping to wrap a bow on this discussion of intelligence and move onto interesting subjects as they come up! As always, if there is something specific you'd like me to write about, please let me know.

Thanks for reading.

⚡️P.S. If you're new here and want to read more of the Synapse Newsletter every other Sunday, subscribe below!⚡️