|Turned up to eleven: Fair and Balanced|
Thursday, May 09, 2002
Consider a "neural nets" type of model of the ideal neuron. Consider this neuron as a vertex in a connected graph. Connected just means, in this instance, that there are no isolated neurons that absolutely have no connections to other neurons. An important, and useful property of neuron connections is they are directional, meaning that a signal can be sent across the synapse from neuron A to neuron B, but not back across the same synapse from B to A. This is important. A back-propagated signal can be sent, but it is separate from the initial signal. I will, for the remainder of the discussion, consider 3 types of neurons, based on their physiology. One is theinput neuron, which takes in a passive signal (usually through an attached chemo-, photo-, or mechano- sensory cell), and passes it on via an action potential (for a review of some of the terms being used here, click here). Any neuron that is directly connected to a sensory input is an input neuron. Conversely, any neuron that is directly connected to an output mechanism (I believe all of these will be motor neurons, which stimulate muscle activity, but I could be mistaken), are going to be considered output neurons. Notice that we are implicitly considering the entire nervous system as "brain", and I think that this is appropriate, in order to get a true picture of function. We may later find that we can discard our consideration of the extra-cerebral nervous system, but I will keep it in my model for now.
Finally, consider what is essentially the rump component of the nervous system, the "intermediary" neurons. I will at this point suggest that there is essentially no cognition occurring directly at the site of input or output, and hereby propose that we consider only (!) the intermediate, transitive neurons as the "seat of consciousness". We can immediately see that a "state function is possible, in theory, to describe this system. There is a (very large) matrix that contains every neuron, and a corresponding matrix of connections. Every neuron is either on or off (we may wish to include a third state, the refractory period, in which the neuron, even when stimulated, cannot fire). Every connection has a weight associated with it, that can straightforwardly be associated with the signal transmitted. Thus, as in the artificial neural network model, we can see that the input into neuron i is equal to I(neuron i)= Sum[I(ai), I(bi)...I(ni)], that is the sum of inputs from each neuron that connects it. In order to make the model complete, we can simply include in every summation every neuron in the model. This makes the computation excessive, but most of the values in every sum will be zero. Each individual term I(ai) is W(ai)*O(neuron a), where O(neuron a) is either 1 or 0, depending on whether the neuron is firing or not. This basic graph is no different from the standard ANN model, on which it is freely admitted it is based.
I will leave you with this teaser-this system seems no different from the standard model, but no one that I know of has claimed that ANN's are conscious. Yet we clearly are. I personally think that this very basic model fairly describes neural function, although it is admittedly stripped down. So, why are we conscious? I don't claim to have an answer, but I may be able to propose something that helps us think about it a bit differently.