Artificial Consciousness/Neural Correlates/Neural Groups/Unified Neural Group/Columnar Model

From Wikiversity
Jump to navigation Jump to search

A Unified Neural Group/Columnar Model[edit | edit source]

We can clearly see, that if we take the Neural Group Models, and the Columnar Models together, then we have a chance at arriving at a much better cerebral cortex model than Marr could have, and the reason of course is that reasearch has been done that links the different functions together in ways that Marr couldn't have predicted in his time.

The Cerebral Cortex "Out of context" model[edit | edit source]

Now I want you to understand that we are just talking about the native functions of the Neural Groups that make up the cerebral cortex, not how they react to outside circuits, nor even why they might exist in this configuration. Thus this is taking the Cerebral Cortex "out of context" in order to look at how it might function.

Layer 1[edit | edit source]

Inhibitive Highly Interconnected "Mossy Dendrite" Layer[edit | edit source]

In Marr's diagram of the cerebral cortex signals are input into the cerebral cortex at layer 1 where they are interpreted by a layer of mossy dendrite neurons. These operate on the selectionist theory to compete for output to parallel fibers in the Layer2/3 zone, which act as inputs to the Pyramidal Neurons in these two layers. This is somewhat analogous to a layer of combinational logic, feeding an input bus. We assume because of the dynamic nature of opportunistic connections that we will not find anything but local coding on these parallel fibers. Thus coding across the brain is likely to involve redundancy and degeneracy (Dr. Edelman's term) where the same stimulus is coded in a number of different manners.

Because the parallel fibers are local in extent, the Pyramidal Neurons do not have to deal with the breadth of the possible signals, and so degeneracy in the code is not a significant factor. However, neither is location of a stimulus, since rich interconnections at layer 1 link separate local bus like arrangements. This means that the location of a specific codon (Marr's name for the self-classifying content addressable memory units) is indeterminate at this level.

Layer 2/3[edit | edit source]

Content Addressable Memory[edit | edit source]

Marr's description of the Codon as a type of Content addressable memory, is controversial to this day, however one aspect of content addressability is favored by the mechanism of a selective layer feeding local busses, and that is the idea that given that type of input, content addressability simply requires that the neuron detect a particular pattern on the bus, in order to fire the neuron. Thus the content (in the form of a pattern sensitivity) of the neuron selects which neurons fire in the second and third layers.

Marr's description however talks about self-classification as a necessary function of the codon, however this function, while it is not resulting in locational specificity is also done by the combinational complexity of the input layer by selection of which stimuli are placed on which local busses. To some extent however the fan-out of the input neurons limits the breadth of any connection to the cerebral cortex, thus limiting the direct influence any one signal has on the brain. As a result there is a locality bias to signals where signals tend to gather around the particular connection they were transferred through.

Layer 4[edit | edit source]

Bias inducing Interface[edit | edit source]

Although Marr, thought that layer 4 was primarily suited to reducing redundancy in the signals, we have since had reason to believe that his characterization of the Mathematical Functions of this layer were a little optimistic, while this layer might seem suited to reduction of redundancy another interpretation of this layer, is that it might present an interface by which other parts of the brain can influence the processing of the first three layers. For instance Basket neurons which he though would function as division functions, might also work as Somal Shunt functions essentially shutting off the output of a Pyramidal Neuron without affecting its input. There might be good reason to do such a function, if we connected our cortex to the rest of the brain.

Other Neurons which seem to have the effect of exciting or inhibiting their tame pyramidal Neurons, might bias the outputs of the content addressable memory in order to suit functions outside the mere address-ability of the content. An example of where this might be valuable is in the influence imposed on the brain by emotions. Being able to excite or inhibit specific signals might be valuable.

Layer 5[edit | edit source]

Mini-Column Activation Interface[edit | edit source]

If Dr. LaBerge is even slightly correct in his interpretation Layer 5 gives areas outside the actual cerebral cortex the capability to semi-activate areas of the cortex, making them more sensitive to the content of their inputs. I won't get into the linkage to the Bottom-up Attention system at this point in time, but the ability to activate the brain at the mini-column level, is certainly suggestive.

Layer 6[edit | edit source]

Column Activation Interface[edit | edit source]

While we may not be sure that it operates on the Right Hemisphere as strongly as it does on the Left, the ability to cause a center surround effect is important if you are going to have a selective memory that discriminates between similar elements. The linkage here to the concept of Bicameral Function is not insignificant or there would not be as much difference in thought patterns between people with different dominance patterns in their brains.

It is interesting to note that the Apical Dendrites of the sixth layer terminate in the Fourth layer and thus may be influenced by external inputs. {{category: Neural Networks]]