Artificial Consciousness/Neural Correlates/Neural Models/Tentative Neural Model
A Tentative Neural Model[edit | edit source]
A Neural model aimed at today's models of the brain, must incorporate what is known about the Molecular Traces of Memory, must deal with Habituation, Facilitation, and Long Term Potentiation, properly model the aggregation and processing within the neuron, as well as pay at least some attention to the second order aspects to neural output., It must deal well with the complexity of dendrites, and axons, and it must somehow deal with the fact that there are over 150 different types of neurons in the nervous system.
This is a complex model, if it is to be so comprehensive. Complex enough that we must understand it well before we try to build a neural network simulation out of it.
The Post-Synaptic Sensitive patch[edit | edit source]
At the heart of the Neural Model is the synapse. From our tentative model of the synapse we know that what we need is a Taggable Synapse that will sequester either ephemeral tags, or survivable tags depending on the supply at the time it is active. The synapse must be cloneable, and must be able to be pruned if its weight becomes 0.
To make the synapse work we need the equivalent of a membrane replacement mechanism that will remove ion channels as they become partially denatured, and add ion channels if there is an ephemeral tag on the synapse. It should remove ephemeral tags, and sustain survivable tags, only as long as there is a supply of survivable tag proteins.
Our synapse model must be able to implement Excitation, Inhibition, Shunting, and Moderating effects.
The Pre-Synaptic Bud[edit | edit source]
The other half of the synapse, the output half, should Habituate unless it's output ages long enough between firings. Evidence suggests that this Habituation should operate at at least two levels, the level of short term Habituation where the cell habituates if a small number of repetitions happens quickly, and long term Habituation where the cell habituates if a larger number of repetitions happens over a longer time.
To counter this, the S synapse should cause the facilitation of the cell, whenever it is fired. The NMDA synapse on the other hand, should cause the facilitation of the cell, as long as its secondary Neuro-Transmitter keeps it active, thus forming long-term facilitation. As well the NMDA synapse should modify the second order aspects of the output in order to potentiate the neuron when it is active. Both the S synapse and the NMDA synapse should trigger learning mechanisms when activated.
DNA Phenotyping[edit | edit source]
Since not every cell has either the S synapse or the NMDA synapse, we need some way to turn off and on these characteristics depending on which cell our simulation is simulating. As well we need some way of adjusting the cell for the phenotype variations that exist in normal Neurons. Examples might be the depth of the hidden network layers associated with a layer of cells, or how many synapses there are normally on a single branch of the dendrite.
Learning Mechanisms[edit | edit source]
When triggered by the proper ion channel activations, we must activate the equivalent of the chemical cascade reactions, that trigger in turn, the different types of learning mechanisms. One aspect that we need to deal with, in the simulation will be how many threads to launch within a single neuron. Since the chemical cascade reactions time the launch of the mechanisms, there will be a tendency to use too many threads per cell, thus causing synchronization problems at the network level.