Thursday, 7 March 2024
------------------------

In the sixth chapter and appendix D, we learn a bit more of the underlying implementation of concepts in the brain. In order for the brain to have this powerful concept capability in practice, it must be implemented in an efficient manner to reduce energy and structure requirements. In short, it uses the principal of degeneracy.

In my simplified understanding, in the brain, at the bottom, we have sensory and motor regions. Here data is expressed as tangible single-type senses. Opposite, at the top of the brain, data is expressed at its most abstract form of concepts. The most abstract expression of a concept is its goal or function that it serves to accomplish by transforming into a larger expression of tangible senses.

A concept is not a real thing that exists, a thing that you can find somewhere in the brain. Imagine if every little variation of all concepts had a physical place in the brain, each having similar copies of the same data. You'd run out of space quickly, not get much abstraction or variability in concepts. Instead, the brain expresses data from tangible single-sense data to abstract intangible multi-sense data in small bits throughout the brain, each bit of data being different. A concept first comes to life by many, many predictions constructing it bit by bit, reusing the same bits of data for multiple instances of different concepts, choosing which bits of data to flow through by probability in the given context the concept is instantiated. This means a small piece of highly abstract data can turn into a large expression of many different tangible senses that represents the concept in a specific context. Very efficient as you can now express a lot more by repurposing small bits of data, but also instantiate large meaningful expressions (lots of different tangible senses) from small pieces of abstract data, that doesn't contain the end result expression itself.

The abstract bits of concepts come to be by the process 'bottom-up', as in tangible single-type senses expressed in sensory regions in their purest single-type form, become more abstractly expressed in multi-type (expressing more than one sense) bits of data, linked together in near endless ways by probabilities dependent on context. This explains how the brain learns to perceive or make sense of the world through concepts. It starts with the raw tangible senses (very noisy and nonsensical information), and by statistical learning finds patterns in the world, narrowing down how it pays attention, what it finds meaningful, and what it ignores, to ultimately make better predictions of reality to survive. How it makes up these bits of data in any given context is fine-tuned along the way when it predicts incorrectly and makes adjustments to the probability a given data bit is chosen (or neuron is fired) in a context.

That explained how the brain forms concepts to improve prediction. How does it use them to make predictions? That process is called 'top-down'. You can say that an instance of a concept is a representation of many, many probabilistic predictions. In any given moment, the brain makes thousands of predictions to instantiate (or construct) the correct representation of concepts at that moment. At the top, it starts the process of constructing a concept by overall goal or function it expresses. Let's make an example with the word 'Apple', and I write words like yummy (I think red apple), sticker (I think of those you never end up using but keep for some reason), ticker (I think tim apple better make green bars). (These examples of what you imagine when you read them are pure (or mostly) simulations, as they're still expressed very abstract in your imagination, and no actual sensory input to match to check for prediction errors I think. Unless reading my examples serves as a form of input or context on this higher, abstract level of concepts we're simulating in our mind?). Many of these concepts, are much simpler expressions that you don't think of much, they just come together to form a cohesive experience. Our thoughts are usually concepts expressed in a highly abstract level. Imagine having to consciously figure out how to walk step by step. For us, it just seemingly magically works. Well not in the start. Same as how words you hear and read means something to us, and how you say and write words with ease. Imagine all the steps involved from low-level concepts / patterns (like an angle seen in two lines coming together) up to more high-level concepts. And these concepts don't just contain one sense like vision, but can contain all in various ways, to make up things like a duck. How it looks like a duck, how it quacks like a duck, so on.

Another way to look at it is that we compress lots of data into compact data, then we decompress it. Only how it gets compressed and decompressed is varied by probabilities based on context.

Predictions run parallel, starting with the lowest level of sensory and motor regions that simulate the tangible sense and send actions to the body. This happens at once, like we experience it, so these different parallel running predictions synchronise in various hubs in the brain (if a hub doesn't work properly it means bad time for you).

In short, concepts don't exist as things in our brain. An instance of a concept is made up of many multipurpose components/neurons that are probabilistic put/fired together dependent on the context the concept is instantiated. The good things of this is: 1. More efficiently structure/store information. 2. More efficiently adjust how the information comes together by fine-tuning the probabilities of which component/neuron to put/fire together in a given context. 3. More efficiently process and act on incoming information from bottom-down to top-down (lots of categories to quicker dissect information, ignoring useless noise, though world is perceived more narrowy or refined as result).

The more refined your concepts are, the more efficiently your brain can predict, saving your brain precious energy. If you perceived the world with less refinement and more nonsensical noise, you'd get overwhelmed quicker (we start on that level from womb/infant and lesser noise as we get older unless you have a bit of a wonky brain that make you less good at making concepts). That's why a new setting you're unfamiliar with can be more overwhelming, but as your brain improves its concepts / prediction, it has to spend less energy. Everything just becomes normal after a while. It is a familiar feeling right?