Introduction to Non-Genetic Darwinism/Neural Selection
Neural Selection[edit | edit source]
Aims[edit | edit source]
- To introduce Selectionism
- To explain and put into perspective Dr. Edelmans work in Neural Selection
- To Introduce the idea of a Self-Organizing Map
Lesson[edit | edit source]
Selectionism: How neurons compete for output[edit | edit source]
In order to best understand this topic, I want to familiarize you with the state of Neural Networks in the 1970s. At that time David Marr was theorizing about the nature of the memory circuits in the brain, under the assumption that the complex connection structures involved in Natural Neural Networks could be converted into mathematical calculations, which would explain the nature of the neural function. Unexpectedly it was proven that there was too much individualization in the connections between even brains of the same species. There wasn't even equivalence between the mathematical functions as they were translated. Up until this time it had been expected that DNA would drive the placement of connections, and that therefore there would be some commonality. As Dr. Edelman points out in Neural Darwinism: The theory of Neural Group Selection, DNA does not have the combinational complexity to individually determine the placement of the billions of synapses in the brain. What was needed were some very simple laws, that determined the placement of synapses, and some way to explain how order was found at higher levels of organization. Dr. Edelman who had recently become a Nobel Laureate for his work on the immune system, recognized a pattern that was equivalent to what he had found in the immune system, a self-organizing system that took seemingly random connections and turned them into usable information. He called this pattern, Neural Darwinism, because he suspected that the mechanism, like the one he had found in the immune system was a form of Darwinism that did not depend on genetics.
Inhibitive networks[edit | edit source]
The first problem was to find a mechanism that worked as a form of selection. While I am not sure who came up with this idea first, Neural Network Designers that gave up on Neural Calculation but not on Neural Networks, came up with the idea of an Inhibitive Network that acted to select from among multiple options through inhibition of other solutions. The idea being that the answer that survived the selective process was the output of the neural network. An example of a Natural Neural Network that has similar properties is Layer one of the Cerebral Cortex, which Marr, labelled as Mossy Fiber Neurons, Inhibitive Neurons with a high degree of connection between them. This technique was called Selectionism, but it has it's detractors, especially since most neurons in the brain are not inhibitive in nature.
Group Formation in neurons[edit | edit source]
If Inhibitive Networks, resulted in neurons competing for something, what where they competing for? Scientists noted that Neurons in the Cerebral Cortex formed into clusters called groups and that within each group, a pattern of firing was noted where one neuron in the center was amplified, and the surrounding neurons around it were suppressed. From time to time, possibly because of fatigue, the neuron that was amplified would change, but the group as a whole would respond consistently. So the net effect of the competition was to be the neuron that was amplified instead of one of the neurons that were suppressed.
As well, Marr's work had suggested that different types of neurons worked together to form circuits that did specific jobs, for instance in A Theory of Cerebral Cortex Marr suggested that the mossy fiber neurons, some parallel connective neurons and the Pyramidal Neurons in Layer 2 and 3 of the cortex, modified by neurons he thought were mathematical functions such as addition and division, that populated layer 4 would be a structure he called a Codon, and which he claimed acted as if it were a self-classifying content addressable memory. I call these groups of heterogeneous Neurons, Heterogeneous Groups. Unlike Marr's theory, I suspect that these heterogeneous groups are to some extent still connected by opportunistic connections.
The combination of self-classifying content addressable memory and Neural Group formation is a useful concept and finds its way into my memory theory.
The Theory of Neural Group Selection[edit | edit source]
Ok, now we get into the hard part of understanding Dr. Edelman's theory, Dr. Edelman went on to suggest that through the medium of feedback and feedforward signals he called reentry, that joined Neural Groups, through a rich set of interconnections, a second layer of selection happened that caused the Neural Groups to self-organize into maps. He called this theory the Theory of Neural Group Selection (TNGS). Unfortunately he went on to extend this theory in unfortunate ways that have since been questioned, however the basic TNGS theory before he extended it is an important piece of the puzzle especially for understanding the Cerebral Cortex.
Self-Organizing Maps[edit | edit source]
The next clue to self-organization in maps, lies in the work of Teuvo Kohonen, a Finnish professor who published a book called Self-Organizing Maps. Essentially he showed how an Artificial Neural Network could be designed that acted like scientists thought parts of the cerebral cortex acted, and mapped high dimensional data into basically 2 dimensional maps without losing important topological connections. Unfortunately while we could predict that in the brain, the placement of the natural maps would be influenced by the placement of the major connections within the brain, the location of a specific map, remained highly individualized.
An interesting side effect of the use of these Self-Organizing Maps (SOMs) was the use of a device with an SOM in it, to map the location of particular signals in the brain, based on an array of contacts Embedded within the brain. This technique allowed scientists to connect robotic arms to monkey brains, and have the monkey move the robotic arm in order to give itself a treat. By connecting the monkey brain through the device, and having the device analyze the action of the specific brain, a mapping of that brains specific reactions to specific stimuli could be made despite individualization of the map locations. All that was needed was to embed the array in the right general area, and the device would learn the individual neural structure of the brain in that area.
Assignment[edit | edit source]
- Consider that seemingly completely random connections at the neural levels can Learn information by Natural Selection, to form gradually more complete Maps of the Environment, and that at the level of gyrus and sulci, (The folds of the brain) we can map consistently across brains in the same species the same general functions, does this not suggest a self-organizing system? Should we expect incremental increases in the amount of order at each layer of self-organization?
- Consider that the mechanisms by which the brain adds order, are different at different layers of self-organization, does this suggest a common thread to self-organization, or does it suggest that there are many different types of self-organization? or Both?
- Consider, that we can read a monkeys motor signals to its arm, and drive a robotic arm with them, does this mean we can read the monkeys mind? Or are there other barriers to reading the mind, than merely reading the impulses of the Neural Groups in the Motor Cortex? Considering that the SOM learns to read the monkeys movements by comparing the signals to the movement of the monkeys arm, how would we for instance read the formation of words, in the Vocal areas?
- Why do you think that we do not have computer implants that can read our minds? Are there ethical considerations? Technical Considerations? Health considerations? Recently a professor at a major university injected himself with a RFID tag, why was this controversial? Some Nationalities are producing RFID credit cards you inject into the body, why might this seem premature?