Symbolic Representation in the Neocortex: What is it about the 6-layers and their connections that makes the neocortex able to do symbolic representation

Symbolic Representation in the Neocortex: What is it about the 6-layers and their connections that makes the neocortex able to do symbolic representation

by Kassie von Stein UA’25 and Jim Stellar

Introduction:

The neocortex holds an intricate and vital role in the human brain. Composing half the entire volume of the brain and comprising the largest section, the neocortex is believed to be responsible for the neural compositions underlying conscious thought, sensory perception, information processing, imaginative creation, and symbolic recreation of the world around us. This symbolic recreation/logic applies to the brain’s own limbic system where it is thought that emotional processing occurs. Exploring that capacity for abstract re-representation of felt knowledge is the purpose of this blog. Other blogs will discuss the function of that re-representation in informing decisions, such as how the felt-knowledge of an internship can impact the college student’s academic studies and the choice of major/career field.

Symbolic representation

The premise of symbolic representation is critical to high level functioning of simpler processes, that is the re-representation of them at the symbolic/neocortical level. For example, although the superior colliculus, a low-level brainstem nucleus, can drive eye movements to lock the eyes onto a moving target. The high-level frontal cortex eye fields are also involved in moving the eyes in humans and in other primates like monkeys where cortical to subcortical connections can be assessed. It’s also important to note that many animals, e.g. lizards, without much of a cortex do not seem to be able to take into consideration an object’s trajectory, but simply maintain that lock-on-target eye-tracking. Another example is in very young children where an object like a ball rolls behind a screen. Their eyes, like the lizard’s, will stay fixed on where the ball disappeared, only to jump to the other side of the screen when it reappears. However, in slightly older children, after object permanence develops at a certain age in infancy, they will jump their eyes to the other side of the screen when the ball disappears and wait for it to reappear, showing surprise if that timing is altered. They have developed a concept of trajectory. Human minds have a much more sophisticated process for following objects that are within this object trajectory concept. While there is still investigation of the exact cortical areas involved and some surprises may still surface, we can say that the neocortex re-represents the function of the colliculus at a higher, more symbolic level.

It is, however, speculative to say that symbolic thought(s) can originate from specific, localized brain connections. There is much ambiguity surrounding the exact function of the neocortex, archicortex, and paleocortex, however, the neocortex serves as a homebase for the projection of representations that we create, as well as the complex information processing directed to many other brain sectors on the receiving end. In fact, hierarchical processing is a key element of the neocortex way of organizing and processing information. Each layer of the neocortex contributes to the processing and abstraction of features and stimuli, varying from simple to complex. This hierarchical organization allows for the representation of abstract concepts and symbols, built upon more rudimentary sensory or motor information.

In addition, the combination of local and distributed processing likely allows for complex symbolic representations by building upon simpler, localized representations. Local processing within each layer can involve the extraction of specific features, while distributed processing involves the communication and integration of information across layers and regions. The fact that the neocortex is highly plastic, meaning that it can readily change its circuitry to adapt and learn, enables the formation and personal modification of symbolic representations through experiences and learning. New symbolic representations can be established and existing ones will be continuously updated as new information is acquired.

This symbolic process has been present with us for centuries. The first human-made symbolic representations (as shown below) were made 200,000 years ago by the modern humans in Ice Age Europe, indicating their superior and advanced cognitive abilities- especially when compared to the nature of the Neanderthals.

Additionally, symbolic representation can date back even further to symbolic communication that was necessary for the great apes to adapt and evolve together. “Ethologists studying chimpanzees in the wild have described extensive social communication based on gestures, the manipulation of objects, and facial expressions. In addition, studies of monkeys have shown that some species normally use a variety of vocalizations in socially meaningful ways, and that these vocalizations may activate regions in the frontal and temporal lobes that are homologous to Broca’s and Wernicke’s areas in humans,” according to an article by Stephan et al in 1981.

Great apes and humans are the only species that possess what are known as von Economo neurons (discussed later), which are large, widely connecting neurons that are thought to be specialized in complex social cognition. Von Economo neurons are concentrated in the Anterior Cingulate Cortex, as well as the Frontal Insular Cortex, which are regions of the neocortex involved in various cognitive functions, including decision-making, emotion regulation, empathy, and the integration of sensory and emotional information (Allman et al., 2012). Due to their large size and unique morphological structure, some researchers have suggested that von Economo neurons may play a role in the rapid transmission of information over long distances within the brain, allowing for quick and efficient communication between different brain regions, and have even been implicated in how consciousness might emerge in books by Damasio. Had it not been for the exquisite evolution of the neocortex and the rapid growth of the primate brain, this processing would never have advanced thus far.

The neocortex, a critical step in evolution

The complex system of neurocircuitry that is the neocortex contains columns made of six layers, which we will later assess as an abstract or symbolic representation of the outside world. We point out the tremendous expansion of the neocortex as evolution progressed in the diagram below.

Note that the Dashed line indicates proportional scaling. The slope of the function is greater than 1.0 in this double logarithmic plot, indicating that in primates, neocortex volume becomes disproportionately greater as brain volume increases.

Symbolic representation is very useful to the subject, as it underlies the capacity for language and symbolically represents our everyday perceptions. On a more basic level, and possibly as a driver of that evolution particularly in primates (and humans), the size of the neocortex relative to the rest of the brain seems to underlie the size of a social group that can function well together. This notion is derived from a British anthropologist who developed a concept that bears his last name – the Dunbar number. With a larger social group, one would need, in theory, more computational power to know which individuals were altruistically contributing to the group and which ones were free-riding on the work of others. In humans, that group size is much larger compared to other intelligent primates, like a Baboon in the graph above where the group size might be a third of that human number. Again, this number could be very important in early human evolution where a larger (but well-functioning) group could be a significant advantage to individual survival.

It is important to note that some researchers argue against Dunbar’s Number due to technological advancements/influences and methodological limitations. These critics further argue that with the current age we live in technologically, specifically with social media, the number of social connections individuals can maintain may have greatly surpassed the cognitive limits Dunbar once proposed. Additionally, some critics found challenges in quantitatively and universally measuring social relationships with a single fixed number, and believe this oversimplifies the complexity of human beings and lacks nuance in the social dynamics they possess.

So, how does the neocortex function cognitively in other domains?  For example, in normal human interaction today, after a decision is made to switch a task or to hold a particular thought, a conscious thought is born. It will immediately be categorized by a particular area of the neocortex, the frontal cortex, as similar or dissimilar to the memory of something else. This categorization process involves the application of symbolic logic, a system of reasoning using symbols to represent concepts and relationships between them. Since one of the frontal cortex’s evolutionary roles is survival and concern for dangers, it is only natural that it would try to categorize things for perceived risk assessment of real or imagined threats and that is would employing symbolic logic to analyze and interpret patterns in the environment for adaptive decision-making. Then, once the danger is categorized as similar or rather different from some thing or another, it usually is tucked away until another similar event occurs. The idea of cortical risk is discussed in another recent blog post.

With continual and repeated conjuring of a particular thought, whether triggered consciously or unconsciously by memory, or by environmental stimulation, it will most likely lead to actions influenced by this thought, and immediately, and ultimately, to more thoughts, provided they aren’t morphed into something unpleasant or irrelevant. As these thoughts become more habitual and motor-memory based, they may begin to be transferred to more unconscious places of the brain where they become habits and then lifestyles – becoming more deeply ingrained in what we subjectively see ourselves to be, which if we live long enough, will likely prove itself to be true.

The 6-layered cortical column as a unit of the neocortex

The neocortex is made up of columns and somehow using this structure it is able to symbolically map and model the world as it is perceived. Moreover, in conjunction with other columns, it can intelligently process external sensory information. That processing begins within its structure. But first, what is a cortical column?

As shown below, each cortical column has a similar structure, consisting of 6 layers. In the primary visual cortex, layer 4 is thick as it receives massive input from the dorsal geniculate nucleus of the thalamus (dLGN) relaying what comes from the retina of the eyes. The information is then relayed for processing through the cortical column layers as follows: “dLGN? Layer 4? Layers 2/3?layer 5?”.

This thickness of layer 4, mentioned above, marks the primary visual area, and it ends abruptly as this part of the cortex ends and another functional part begins as shown in the diagram below. Using this kind of marking in all the layers, Brodman made a map of 52 cortical regions which seems to relate to functions across the whole neocortex. The primary visual cortex is Brodman area 17. In a way we are just learning from tracking connections through modern technology such as functional magnetic resonance imaging (fMRI), diffusor tensor imaging (dTI), and other brain scanners.

Relating back to the columns, the way visual information is extracted from the input from the eyes is quite interesting to examine, and earned David Hubel and Torsten Weisel the Nobel Prize in 1981. They noted that the retina and thalamus neurons really are spot detectors, i.e. those cells fire when light falls on the spot that is their receptive field, and that the cortical column neurons receive a convergence of the thalamus cells from a series of spots that make up a vertical line. So the cortical neurons can be thought of as line detectors. Move a bit to the side in the cortex to the next column and the “line detector” becomes slightly tilted. It has a different pattern of thalamic neurons converging on them to make that tilted line. Moving more to the side and the line detection receptive field tilts more. The thalamic input that activated the first column, described before, does not activate these other columns. They require a tilted line input, and these lines are the elementary features. The line detecting cortical cells then combine to become corner detecting cells in the next brain area (Brodman area 18), and at the next level of complexity we have, in theory, simple object (e.g. a chair) detecting cells where the inputs to those cells are made up of those same elementary features. In theory, we could keep this convergence going and even get to the processing of a face. The repeated columnar structure of the cortex is used in this model to get to more and more abstraction in processing.

Today we think of face detection as something of a cortical circuit property, as opposed to repeated convergence onto a single cell. We even think that the neocortex sophisticatedly models the external world so that what you are seeing in your conscious awareness is really that model of the world continuously updated by your visual experience, what neuroscientist György Buzsáki calls “The Brain from the Inside Out” in his 2019 book.

Over time, this columnar neocortical learning assembly line will sharpen the human brain to better predict the nature and behavior of objects from learned and stored sensory, location, and other data. The plethora of layering systems, varying cell types, neural connections and pathways, and messenger systems of the neocortex attest to an amazing aptitude in symbolically representing the world we perceive around us. The question that remains, is how?

Are repeating cortical columns the key to symbolic extraction in the neocortex?

We do not have a clear-cut answer on how the neocortex, with its repeating columnar structure, builds a model of the world. We are, however, betting that the answer goes beyond single columns. For example, consider a paper by Bennet (2020) with the title, “An attempt at a unified theory of the neocortical microcircuit in the sensory cortex,” and the quote they take from Mountcastle “…the neocortex is made up of repeating subunits of macro columns, each of which is largely identical in circuitry.”  Could these repeating columns, particularly the macrocolumns, be the key structural difference from the lower brain that allows for the process of abstraction in the neocortex? Consider what Bennet depicts as a macrocolumn in the left side of the diagram below. According to Bennet, the human brain has over a million of these.

This repeating minicolumn structure gathered up into macrocolumns in the neocortex operation seems on the surface to be like the repeating network nodes that are gathered in the right side of the diagram above from computer simulations of neurons in artificial neural networks. These levels underlie the premise of Artificial Intelligence (AI) and particularly Large Language Models (LLM) like in ChatGPT. The diagram on the right has two levels between input and output. We understand that current ChatGPT artificial neural networks have 96 layers. While not directly comparable to the repeating columnar structure of the neocortex, the structure, operation (trained not programmed), and other features of this computer computation seem comparable to the neocortex.

While it is beyond the scope of this blog to do a deeper comparison between the brain and AI systems, we do note that the brain has 86 billion neurons (plus glial cells), none of which are simulated. All of our neurons are real and probably have subtle but important operational properties that the computer simulation is only just beginning to imitate. To repeat the idea from the diagrammatic comparison above, the macro columns are a repeated structure like the artificial neurons organized as levels in LLMs. There is something about this repeated interconnected structure that allows the neocortex to generate the symbolic logic that the amygdala as a brain nucleus cannot do. Thus we have a trajectory of objects, as discussed above that goes beyond eye-tracing functions, or we have risk calculations coming from fear reactions in negotiating external opportunities, as discussed in a recent blog. Undoubtedly, AI and the LLMs that underlie ChatGPT as well as similar models in other domains (e.g. DALL-E) will get better. Where we want to go in the future is to discuss how these column and macrocolumn groupings of neurons bring information into their symbolic computations so that something like fear gets processed in the amygdala as Pavlovian conditioning can be elevated to a more symbolic level of risk.

Before we do that, we want to first consider some of the neural mechanisms that the symbolically logical, abstract reasoning neocortex circuitry may use to bring in the limbic system processing.

Special neurons in the neocortex

The von Economo neurons, first identified by Constantin von Economo, represent an exclusive category of widely connected, large, spindled, and bipolar neurons, localized in distinct regions of the cerebral cortex such as the anterior cingulate cortex and the fronto-insular cortex. Damasio referred to these as convergence-divergence neurons and thought they played a role in our conscious experience. These neurons have only been identified in higher cognitive functioning mammalian species such as elephants, dolphins, great apes, and humans, even further alluding to their intelligent orchestration of cognition.

The large size of von Economo neurons could hold the ideal morphology to carry out the rapid and efficient long-range communication of stimuli input or output in the brain, and specifically in the neocortex. Even moreso, von Economo neurons are found in the fifth layer of the neocortex in the anterior cingulate and fronto-insula cortex, and may be responsible for input and output connections, alluding to these neurons having a role in integrating information sent from different brain areas. Previously mentioned was the idea that the neocortex gets information from several different regions and is able to congeal it all into one cumulative information processing; von Economo neurons are certainly a key part of this cumulative information processing and their characteristics and localization allow them to do so. Some researchers speculate that the localization of von Economo neurons in regions that specialize in abstract thinking, emotion processing, and self-awareness. Focusing on emotion processing, this ties into the premise of empathy which can correlate with the group size point made earlier.

The pyramidal neurons of the neocortex are vital to the circuitry in the neocortex as they are either predictive, active, or inactive in state. The only thing that differs is the input it receives and the location of the signals it sends. This very tightly anatomically constrained operation of neurons may be another difference. In the analogue biological computations of the neurons, each of the many hundreds of dendrites can have millions of contacts called dendritic spines with other neurons. The type (neurotransmitter action on the receiving cell, excitation vs inhibition, short vs long acting, etc.) and the location (far from or near to the axonal trigger point) all further the biological complexity. These actual points of contact are always forming and projecting away. They are dynamic and determine the weights that the computer models represent, and they do it in a very energy-efficient manner so the computer fits in the skull and the energy demands match what the body can carry.

Back to symbolic processing

While the overall function of the layered columns in the neocortex is still discussed (e.g. Do and Hasselmo 2021), we do know that each column is able to consistently learn and integrate its obtained object learning, recognition, and permanence over time. The never-ceasing integration of learned sensory and location data are then able to rapidly infer and predict the nature and behavior of objects, i.e., recognizing that a ball rolling past a wall does not disappear, but instead is rolling behind the wall. Each individual column maps and models the world as it is perceived, and in conjunction with one another, can make a rapid and ever-increasingly intelligent processor of cortical and sensory information.

Understanding the major connectors of the brain, i.e, the structure, function, and connections of white matter tracts, for instance, is vital in grasping the brain’s affinity and how it encapsulates and tackles each and every cognitive process. White matter can be described as a vast, intertwining system of neural connections and nerve fibers that encompass all areas of the neocortex, and beyond and allows for the communication between them (see Filley 2005). More specifically, white matter tracts that connect cortical areas are responsible for connecting different regions of the cerebral cortex- where higher cognitive functions are executed, and allow for the integration of information to coordinate these more complex cognitive functions such as the dynamically occurring default mode network. It links the cortex to the thalamus, which brings in much sensory input, and it connects in a classical loop with the caudate nucleus and other subcortical motor structures to allow the neocortex to drive our behavior. It connects the frontal cortex with the amygdala in a way that seems to determine subtle elements of our behavior like how relaxed or anxious we are.

These networks are critical. When you feel an object in your pocket like a key, you immediately have a visual image of that key in your mind. But how does this cross-modal transfer work? We do not know that answer, but we do know that it depends on the white matter connections between the tactical and visual areas. We also know what kind of sound that key will make if we drop it. These are hard things for other animals to do. While you see your cat as an integrated experience of sight, sound, touch, etc., your cat probably only sees you because you bring all of those experiences to the cat by your presence. You have many more connections (white matter) between brain areas than a cat. Each person has a unique and highly complex functioning network of all of those neural interactions between cortical columns, with subcortical structures, and so on down to the brainstem and spinal cord.

Many of these subcortical structures do enormous computations. Consider simple walking where you place your feet to keep your balance and send you in a direction you consciously choose, but likely without awareness of how you keep upright without slipping, stumbling, and falling. Now you can take control of your feet and dance out of the room instead of walking out of the room. But if you do not take that control (and even if you do), the lower brain and spinal mechanism will keep you walking gracefully without your preoccupied thoughts.

The next step

The question we will address in the next major blog in this series is how does the cortex re-represent limbic information, facilitating the extraction of higher functions such as the value of something to go along with its price – from an old quote by Oscar Wilde about the cynic as, “a man who knows the price of everything, and the value of nothing.” Further, how does cognitive-emotional integration, about which we write a lot in this blog series, actually occur? Whether it’s envisioning the trajectory of a rolling ball or evaluating risk in the insula cortex based on fear experiences stored in the amygdala, our journey into the intricacies of neural orchestration continues.