How the Neocortex works & How that applies to engineering AI

Discussion of the nature of Ultimate Reality and the path to Enlightenment.
Locked
User avatar
Ryan Rudolph
Posts: 2490
Joined: Sun Jan 29, 2006 10:32 am
Location: British Columbia, Canada

How the Neocortex works & How that applies to engineering AI

Post by Ryan Rudolph »

Take three common human senses.

Sight
Hearing
Touch.

Now, Imagine three separate abstract pyramids for each sense, and each pyramid transmits sensory information up a hierarchy to the top of the pyramid. And each pyramid represents groups of specialized neurons within the neocortex. Every level of the pyramid is specialized to handle the sensory information traveling up the hierarchy. Each level of the pyramid can be thought of as a node or a tier.

Now Imagine the mind visualizing an image of a bird.

First, The bottom layer of the pyramid represents specialized cells designed for taking in visual information from the environment as electrical impulses or patterns. However, the cells at the bottom of the pyramid do not understand what information they are collecting, each cell simply takes in a small amount of visual information such as part of the bird’s wing, and then transmits the electrical information up to the next level of the hierarchy.

The next level of the hierarchy processes the electrical patterns received, which correspond to properties of the image sent from the lower tier. The parent tier sends information back down to the child tier to double-check the information. This could happen thousands of times in one single second to double check the causal patterns absorbed from the environment for consistency. Only consistent electrical data results in image creation. So communication is two directional up and down the hierarchy. Every advancement up to the next tier results in a lower number of cells, more complexity, and a slightly advanced picture of the image. Basically, a parent tier sees more than its child tier directly below. This process continues up the hierarchy and each tier or level creates a more advanced image of the bird, until the top tier renders a clear image of the desired input. So each tier renders a slightly different quality image, and each rendered image is referred to as an invariant representation. So each parent tier creates a much more advanced version of the invariant representation than its child. However, the sophistication of the parent nodes are dependent on the information received below by the child nodes.

Moreover, the pyramid for hearing is connected to the pyramid for vision, so one could visualize a comparison of information happening between the different senses to determine if any sensory electrical patterns are being emitted from the same object. It might be useful to imagine communication between hierarchical pyramids as horizontal, while imagining communication up a hierarchy, and within a hierarchy as vertical.

To give an example of sound merging with vision, as one single invariant representation, Suppose the bird was singing, so the pyramid for hearing works exactly the same way as vision, meaning electrical patterns are transmitted up its hierarchy, and they correspond to sound waves absorbed from the environment. And the sound of the bird singing would be compared to the visual image of the bird, and near the top of each pyramid’s hierarchy, both inputs would be compared, and there would be a correlation formed that the image of the bird and the sound of the bird singing are coming from the same object.

This is the basic idea of HTM theory, developed by engineer Jeff Hawkins. The practical value of creating an abstract sort of theory based on how the Neocortex works is that it can be used to create intelligent machines. Moreover, sensory systems put in particular environments will become specialized over time to be able to absorb, render and interpret the most frequent recurring sensory causes happening within the environment. The initial machines will probably be used as weather machines and security machines. You might have hundreds of sensory units distributed over a wide area, and connected to a master unit that interprets the desired results.

However, such a theory could be the rudimentary technical map that will eventually give birth to advanced forms of A.I capable of perception, cognition, and social interaction with humans.
Locked