See Also: Book Notes, <> (me), <> Notes on Consciousness, <> Predictive Coding Notes, <> Barrett: How Emotions Are Made, <> Deep History: How We Got Conscious Brains <> Seth: Being You image ModelAtRest.png

Predictive Coding Animation AND Predictive Coding Conceptual Scematic

Intro to Predictive Coding

This is the edited script for a summer service given at First Parish of Sudbury, Unitarian Universalist, as a Summer Service - July 24th, 2022. recording of the service. Thanks to Gail Hardenbergh for listening 3 times and helping make it much less GEEKy.

Have you ever arrived home and realize you do not remember anything about the drive? That's OK. It does not mean you were not paying attention. It just means everything happened as predicted. Nothing was novel and salient enough to become conscious.

Your autopilot used a mental model of the drive to get you home safely.

A mental model is the representation, in our brain, of how something in the real world looks, sounds and behaves. For example, we all have a mental model of a traffic light in our head. That model can match a simple 3 light signal or one with many lights and arrows.

Predictive Coding is a theory of how the brain sends codes from one part of the brain to another part of the brain. It is a theory that has become influential in recent years.

In essence, Predictive Coding posits that we use mental models of the world to predict what we are going to see and hear and then our perceptual systems confirm what we believe or not.

By perceptual system, I mean vision, hearing, touch, and other parts of the nervous system that send signals to the brain.

Implicit Bias was part of the title, but I will only focus on predictive coding. I just want to say that by definition, implicit bias is the result of our mental models guiding or misguiding our perception.

Why Predictive Coding? Because it is much more efficient to predict and only send the errors, rather than having the senses send everything up the highest level of cognition. With predictive coding, only the data that needs to be sent gets sent.

Here is a demonstration of predictive coding, which is also called predictive processing. [ Acxtually, no picture. The picture is in Lisa Feldman Barrett's book: How Emotions Are Made ] It was an picture totally turned to black and white and looking quite abstract. Then flash to the original photo and then back to the original abstract photo. Now it is plain as day.

Now, here is the original image. Now that you know what you are looking at, let's go back to the stylized version. Do you see it differently? This is an example about how your expectations determine what you see.

Most of the details of in this note have been published in scientific journals. However, I will be speculating and connecting dots in a way that may or may not be valid, but I hope to paint a simple picture of predictive coding in the brain.

The brain creates a mental model of what is happening in the world around us and it pumps this model thru the nervous system. This is analogous to the circulatory system, in which the heart pumps oxygen rich blood thru the body. The heart gets back the depleted blood and starts over. What the brain gets back is confirmation that the model is valid or information about where the model needs updating.

Here is a diagram of a small part of the perceptual system. The Model is at the top and Reality is at the bottom. The circles represent a hierarchy of different brain regions. The predictive codes travel up and down this hierarchy as information.

The brain is pumping information thru the perceptual systems. - As it travels down the hierarchy, the model is broken down. The color to expect is sent one region, the expectation of what we should hear is sent to the next, the type of object we should be seeing is sent to another region.

Each brain region is made up of hundreds or thousands of prediction units, called cortical columns or mini-columns. in Dehaene' How We Learn there is an image of a honeycomb grid that I think of the columns in the object brain region. [ scroll to Neuronal recycling ]

The prediction, which is based on the model comes down to each prediction unit, if the prediction matches the perception, then it is confirmed. If there is a mismatch, that is called a prediction error and the error gets sent back up the chain to get resolved.

To the left is a schematic that shows three levels of predictive coding. On the right side, you have the model being pumped down to each prediction unit. On the left side, you have the prediction errors flowing back up the hierarchy.

If a prediction error from a lower level can get resolved as part of a larger pattern at an the upper level, it will change the model and send down an updated prediction. Then the prediction and perception will match again. If the prediction errors can not be resolved at any level of the hierarchy, they are sent to headquarters and they pop into consciousness.

Physical Brain

Below, left image is a good picture of a brain. It looks the way it does because it is actually a bit like a pancake that has been scrunched up to fit inside the skull.

The middle image shows the inside. It is composed of gray matter and white matter. The gray matter is where the brain regions are and the white matter is what connects the brain regions. In the lower right corner, I have exaggerated the gray matter and white matter. The gray matter is where all of the neuron cell bodies and dendrites are. On the right is a side view of 5 cortical columns, aka, prediction units.

image PinkBrain.png
Brain Exterior
image GrayMatterWhiteMatter.jpg
MRI slice showing Gray Matter& White Matter
image CorticalColumns2012.jpg
Cortical Columns
image FreeSurfer1.png
FreeSurfer - scrunched
image FreeSurfer2.png
FreeSurfer - inflating
image FreeSurfer3.png
FreeSurfer - balloon
Freesurfer is software used to study brain regions. Above are three images from a FreeSurfer video about inflating the brain like a balloon to make the brain regions much clearer.

Here is a drawing by Santiago Cajal [ CAH Hal ] of some pyramidal neurons. The cell bodies are the dark blobs, and dendrites are the branches expanding out to talk with other neurons. Dendrite is based on the Greek word for tree, as in rhododendron. Cajal drew this and may others illustrations just over 100 years ago. If I zoom in on these pyramid neurons, you can see Cajal's little 'a' for axon.

image cajalPyramidal.jpg
Cajal Pyramidal Neuron
    image CajalZoom.jpg

The pyramidal cells have long axons. The axons are connections between the prediction units. Those axons are covered by a white fatty substance called myelin. That is what makes White Matter white. The myelin makes the messages more reliable. So, the white matter is made of long wires traveling thru the wrinkles of the scrunched up pancake.

This is a picture of many axons in the white matter as they travel form region to region. The color shows which direction that the wire is traveling. From the side, from the top and from the front.

image connectome_unknown_artist.jpg
Beautiful Connectome - Artist Unknown

Cajal Quote: "The cerebral cortex is similar to a garden filled with innumerable trees, the pyramidal cells, which can multiply their branches thanks to intelligent cultivation, send their roots deeper, and produce more exquisite flowers and fruits every day." image ModelREALITYobject.png

So Cajal's forest is a great metaphor, except the roots travel thru the white matter and come out in another brain region and spreads their messages. Here is an illustration from a Lisa Feldman Barrett article.

I made an animation that shows the brain pumping information with a few brain regions. Let's just consider the Object recognition region labelled IT, for Inferior temporal region. The animation script contains 3 objects: a traffic light, a Route 20 sign, and a Ferrari. The other regions deal with color, motion, lower level visual input and audio input.

This Predictive Coding Animation starts with the the green & blue dots
representing predictions being pumped down the hierarchy

Certainly if a red Ferrari zoomed by on your drive home, you would have remembered. We also make predictions about the more interesting parts of Reality. Those that are more open to interpretation.

Is Gail annoyed or amused by one of my quirky behaviors? And how will I interpret what I am seeing? Will my interpretation and reaction to her influence whether she is really amused or annoyed?

Here is a short summary of some interesting research.

"British soccer fans do not have the best reputation. but they are willing to help their own. Researchers invited Manchester United fans to their lab. They arrived and completed a series of tasks designed to remind them of how much they loved Manchester United. Then they were asked to walk to another building to complete the study. On the way they encountered an emergency. A young man ran out across the path ahead of them, tripped, fell, and clutched his ankle, moaning in pain. Would they help this man in distress?"

[YES! if they were wearing a Manchester United jersey (92%) else if Liverpool or plain shirt only 33% of the time. However, if the tasks reminded them of how much they loved the "beautiful game", known as soccer in the US, then they were equally likely to help the Liverpool fan.

Your the mental model of yourself can be tweaked to be a only Manchester United fan or to be more inclusive.

Predictive coding also applies to your gut feelings called interoception.

Anil Seth, in his book Being You, presents a study from 1974, in which a female interviewer approached male passersby while they were crossing one of two bridges in North Vancouver. One of the bridges was a 450-foot-long rickety suspension bridge with low handrails, precariously poised high above shallow rapids. The other was a shorter, sturdier and only ten feet above the water. When the interviewer made contact with each bridge crosser, she invited him to fill out a questionnaire, and she also offered her phone number, explaining that she’d be happy to answer any further questions he might have. Would the men on the rickety bridge would misinterpret the physiological arousal caused by their precarious condition as sexual attraction, rather than as fear or anxiety? Yes, this is exactly what happened. The female interviewer received more calls from men who had been crossing the rickety bridge than from those who had been crossing the sturdy bridge.

Your brain gets a lot of signals and it is not easy to interpret them.

There are two take aways from predictive coding. The first is that we should learn the most useful mental models to help us see the world more clearly. And that imagining ourselves as we want to be will make a difference.

The second is about interpreting prediction errors. When we face uncertainty about the signals someone else is sending to us, we should clarify what they mean explicitly. Humans evolved to read each others minds, but that does not mean we are very good at it.

And when we deal with ambiguity, we should try to discern and articulate our feelings as specifically as we can. And

Prediction also applies to other processes. Allostasis is the prediction alternative to Homeostasis. Homeostasis is where the body works to get back to its same baseline. That happens when we encounter unexpected situations. But the body also predicts what it will need in a process called Allostasis, which means the body is working to get to a new baseline. A simple example is when you anticipate something delicious to eat, your mouth starts watering. When the food arrives, you are ready to start digesting. Peter Sterling, who coined the term allostasis, talks about how circadian rhythms start preparing the body before we wake up and it prepares the body for sleep.

2022-07-31 <> jch