Nerves are elongated somatic cells. The elongated part is called the axon. They are the bioequivalents of electrical wires, with their axons threading through all parts of the body and mind, like a vehicle's wiring harness. Unlike electrical wires, which are passive circuit elements characterised only by resistance, neurons are active circuit elements which use signal amplification and (sometimes) a kind of 'insulation' to maintain signal intensity and speed of transmission over long distances. Neurons are not self-contained specialists, they need assistant cells called glia (eg stellate, basket and Schwann cells, depending on location) to create and maintain viable computational networks.
It is a truism to say that the neural circuits in the brain are cybernetic, since every somatic susbsystem in every living creature is cybernetic. This means that every single one of an organism's somatic and cognitive variables has a ‘native’ (ie genetically determined) 'setpoint' associated with it. This genetic parameter represents the idealised value of a biovariable; it is the value that it 'should' be maintained at, its 'resting' value, if you will. Complex organisms like animals and humans have many millions of these metabolic and biomechanical values, each 'knowing' its ideal magnitude. Neuronal setpoints are equivalent to the resting potential of their membrane.
The conventional view of cybernetics which is limited to setpoints is essentially a static one. To more fully understand the dynamic aspects of cybernetic systems like the nervous system, setpoint values are not enough- offset values are also required. Consider a generalised cybercircuit (a.k.a. servostat, cyberstat or unistat) in which a setpoint is complemented by an offset. The figure below depicts the neurostatic circuit principles which govern the response of all biosystems under external load. Note that the term 'heterodyne' and 'neurostat' are synonymous.
For those who aren't used to visualising electrical currents and voltages, there is a simple mechanical equivalent (or 'analogue') to this situation, which consists of a jointed lever (representing the forearm) whose angular range is limited by a spring (representing the biceps muscle). Crucially, there is a dead-zone, or 'slack', deliberately interposed between the lever and the spring, allowing the lever to move a little bit before it starts to stretch the spring.
If a cybercircuit only has a setpoint, there is (i) no way to control it (ii) no way to link it to other cybercircuits in the organism's body. This is the situation that individual single cell microbes and amoebae find themselves in. To solve these issues for multiple cellular organisms, the basic setpoint cybercircuit evolved into one which also possessed an offset (or 'bias') capability. Offset values could be temporarily added to setpoint values, so that these normally static systems could respond to dynamic environmental challenges, while still maintaining structural integrity. We can observe the interplay between these biovariables in the way the human forearm is controlled by the flexor/biceps and extensor/triceps muscle in combination.
Consider the electrophysiological activity of the neural cell membrane. For the motoneuron (or indeed, any other type of neuron) to fire a signal down its axon (its output 'channel', or 'cable'), the sum of all of its incoming signals, which can be both excitatory (+) and inhibitory (-), must exceed the neuron's membrane threshold, which is a relatively large negative voltage. It sometimes helps to imagine the membrane potential as being stuck in a deep hole in the ground, with the neuron's dendritic inputs being ladders of varying lengths, which must all be placed end-to-end to climb out of the hole. There are only three ways for this to happen- (a) the load is increased, until the stretch sensor alone causes the motoneuron to exceed its membrane threshold (b) the descending spinal command neuron alone causes the motoneuron to exceed its membrane threshold and fire. (c) some combination of (a) and (b) above. Clearly, the 'top-down' addition of the command neuron to a motoneuron's inputs represents a simple and elegant method of governing a limb's motion under load. Moreover, as the organism learns new (global) behaviours, more links can be added to implement constituent limb motions, which are the local adaptations needed to construct those new behaviours .
Unfortunately, the neurostat theory of governance and adaptation presented above, whose truth in the context of human neurophysiology has been demonstrated beyond reasonable doubt*, is relatively unknown. There are two reasons for this. (i) Cybernetics, after being largely ignored by the conservative scientific establishment for almost a century, is only just now getting traction in mainstream ‘system science‘ circles. (ii) The term 'offset', and its relationship to saccadic reflexes, has not yet been published in a reputable journal.
*Feldman, A. G. (2015) Referent Control of action and Perception: Challenging conventional theories in behavioural neuroscience. (Book published by Springer-Verlag). While Professor Feldman makes a convincing case for the complete revision of neural science orthodoxy, he has yet to extend his research to include cybernetic topics like offsets. The importance of neurostat theory to the governance of complex systems (= cybernetics) of all types is this author’s discovery alone. Fortunately, the basic idea has been described independently by Anatol Feldman, and published as the 'equilibrium point' (a.k.a. ‘Referent Control’) theory. Feldman uses EP/RC theory to comprehensively debunk the concept of 'efference copy', the current (but obviously erroneous) explanation of musculoskeletal systems governance.
Problems with the synaptic efficacy model of neural plasticity
Engineering models of 'feedforward' neural computation are called ANN's (Artificial Neural Networks). ANN's almost always assume that the neurodes they contain can have their behaviour adjusted by means of their synaptic conductances, just as we control fluid flows by opening and closing valves, taps, stopcocks and faucets. This seems plausible mainly because it follows established paradigms used successfully in electronic and electrical engineering. The success of ANN's in the field of engineering has influenced medical science, such that the currently accepted neuronal theory attributes neuronal efficacy to variations in synaptic conductance.
The major problems with this explanation are as follows:-
(i) Latency. Existing synaptic learning models, in which teaching signals derived from gradient descent require many thousands of iterations to modify synaptic conductances, are much too slow by an order of magnitude.
(ii) Granularity. Artificial Neural Network (ANN) teaching signals are inescapably global, derived from the entire output/input data map. This stands in direct contradiction to the local learning mode ('one fact at a time') of the Natural Neural Network (NNN) possessed by humans and higher animals. Human minds, like languages, learn incrementally by acquiring, then assimilating knowledge one new symbol at a time.