Evolution is not just a theory invented by Wallace and proved by Darwin, it is an everyday reality for neural systems. Human brains evolved from simple sensorimotor loops in a smooth, uninterrupted series of functionally optimal steps. Each nervous system design was not just good enough to work (ie govern behaviour and learning) but was good enough to produce surplus resources, sufficient not only for survival and reproduction, but also to overcome conspecific competition. This idea represents the COFFEE principle, 'combinatorially optimised familial fecundity explains evolution'. The mention of combinatorics refers to the ecological 'dice-throw' of dominant and recessive alleles caused by sexual reproduction.

There is an tongue-twisting aphorism in genetics, one which, though  no longer thought correct, is still true enough to be quoted here:- 'Phylogeny capitulates ontogeny'. It means that the sequence of intermediate body forms observed as the foetus goes from fertilised egg to newborn infant  summarises (capitulates) the evolution of (all) species. It as if the history of life itself is replayed at fast-forward speed by each foetus in the womb.

For example, each human spends a short time in their mother's womb as a fish-like form, since we evolved from more primitive forms via a fish-like stage. The sequence of extant forms corresponds to a sequence of cellular machines with optimal functionality. Living creatures lurch from one successful 'design' to the next, in a rather discontinuous manner. The non-expressed genetic codes, a.k.a. introns used to be called 'junkDNA' because, since they were never expressed, they seemed to serve no clear purpose. It is becoming increasingly evident that far from being 'junk', these codes represent a 'race memory' of past (by definition, failed) {familial fecundity x foraging fitness} experiments*. Like the cards taken from the deck that are still in the player's hands, introns are BOTH a form of coding redundancy, helping to correct errors (mutations) and eliminate 'noise', AND a font of possible future variations, helping to jump past genetic bottlenecks (eg population crashes) and other spurious minima, a risk whenever competition levels drift outside of normal ranges, such as island populations (low competition from few predators) or habitat shrinkage (high competition for few resources).

*imagine, if you will, you are a time-travelling archaeologist from the future,  moon-beamed to Palo Alto, the Valley of the Kids. Like Howard Carter, you discover that the computer hardware is buried with grave goods. These digital documents are called 'software', and they seem to rule over the underlying hardware, but the question remains open as to the functional role of each part of the software code. After much head scratching, you have a 'eureka' moment. You have found that only a tiny portion of the code actually 'does anything' (a direct quote from your report). You isolate the part that produces output from the laserjet printer attached to the computer. You label all the rest of the software code as 'junkDNA'- junk Digital Numerical Alphatext. 

Figure 4.1

All living and many non-living systems maintain their organisational integrity by means of feedback loops. When a simple feedback circuit forms the basis of a self-regulating system, that circuit is called a homeostat. The upper diagram (figure 4.1 to the left) depicts such a homeostat circuit. When the input P exceeds the threshold T, the output F acts in such a way so as to exactly cancel out the supraliminal differential, such that..
 F= ||P-T||. 
When the threshold T exceeds the input P, the output F also acts in such a way so as to exactly cancel out the differential, such that...
  ||P-T|| >>0  as   F >>0
In each case, the 'polarity' of F is the opposite of that of the differential. F is called the Drive State Differential (DSD).

Homeostats occur in their hundreds, if not thousands, within the metabolic machinery of living creatures, but at unconscious levels, organised in vast, distributed forms, as inverted hierarchies ('lowerarchies'). At each level of the lowerarchy, local T values and local P values form local feedback loops. 

Regulation also occurs on a more centralised basis, which is often manifest as both subconscious reflexes and conscious commands, both possible responses to that limited core set of metabolic signals which affect the survival of the entire system as a whole. For these purposes, the basic homeostat is modified by adding a top-down 'command' signal called a command offset, or global bias input, designated as L^. 

Because L values are global, they usually command many local T-P or P-T differentials. Such circuits are called servostats. The lower diagram in the figure to the left depicts such a servostat circuit. Even when local T-P or P-T differentials aren't sufficiently large to cause corrective actions, central top-down command signals can act as offset values, thereby biasing local 'decisions' with global 'directions'. By precisely this manner, organisms are able to maintain their unique structural shape, while moving across their environments in a centrally coordinated manner.

Automobile cruise-control systems all use some variation of the Servostat to achieve the required function of speed constancy with the actual speed chosen by top-down driver command selection. The same mechanism operates in a domestic refrigerator. The home owner adjusts the 'fridge's temperature indirectly, by biasing its setpoint by application of a command offset (the 'thermostat' knob).

^Motor command models which rely on the flawed idea of efference copy have been disproved by both Feldman and Dyer independently.

Figure 4.2

From Hughlings-Jacksons principle of selective recombination, we know that the higher level cybernetic functions (up to and including computations and consciousness) are formed from simpler, lower level cybernetics. The way this works is that some lower-level functions are enhanced, or amplified, while others are inhibited or attenuated. Jackson remarked at the similarity between his idea and Wallace/Darwin's evolution.

A consequence of this principle is the 'all or nothing' conclusion that if the simplest creature functions as a cybermaton^ (eg a homeostatic machine) then so too does the most complex one, ie us humans. This is not to say that our brains are not computers, rather, if there is computation (ie formulaic execution using finite state automata), it occurs above the level of cybernetic loops. This is a commonsense assertion, since it is the cybernetics which creates the parametric stability required for biological variables to participate in finite state computations.

Figure 4.2(a) depicts the simplest sensorimotor system, the elbow joint and biceps muscle. There is a matching pair of sensor (green arrow) and motor (purple arrow) neurons, and a 'top-down' command (yellow arrow) neuron. The effect of the command neuron is to 'bias' the 'natural', 'resting' or 'equilibrium' state of the synaptic junction, the one between the sensor neuron's axonal button and the motoneuron's somatic dendrites, increasing or decreasing its triggering threshold, and therefore changing the neuron's output (firing rate). The net effect is to reduce/increase the resting length of the muscle, which causes the elbow joint to flex, changing its joint angle. In terms of a mechanical (mass-spring) analogy, increasing the top-down command bias, so the motoneuron is more likely to fire, has the same systemic effect as increasing the preload on the spring (equivalent to reducing the spring's initial, or 'installed'  length).

It is the motoneuron's threshold which is reduced by the top-down (command) input, thereby reducing the sensory input at which the motoneuron starts signalling the muscle to contract. A cybernetic loop is formed by the spinal ganglion subsystem, spindle receptors (= input channel) and motoneurons (= output channel). In abstract terms, this loop is a Servostat, functionally identical to the refrigerator thermostat. 

Figure 4.2(c) attempts to depict the result of multiplying this simple sensorimotor ('servo' or 'heterodyne') system many thousands of times, as seen in both
(i) the pyramidal circuits in the cerebrum, and
(ii) the purkinje circuits in the cerebellum. Note that the motor control principle is globally feedforward but locally feedback. The displacement is set and force levels required to achieve that position (eg elbow angle) are  taken care of (computed locally) by a feedback loop, just as in any true servomechanism*.

*see the author's website W[6], which contains essentially the same analysis and conclusion as that of Anatol Feldman. 

^not the same thing as a cyberman, thankfully. They (the cybermen) scared the hell out of me when I was a child, much scarier than the Daleks. Doctor Who and The Daleks, later just shortened to 'Doctor Who' is a long-running  British children's sci-fi serial noted for its speculative treatment of time and space travel, and wildly speculative forms of alien life, both good and evil.

Figure 4.3

The polyheterodyne, which consists of many parallel (concurrently active) servostat circuits, is found in the central nervous system in the cerebrum and cerebellum (see figure 4.2). 

The polyheterodyne in the cerebrum, which includes the CA1 pyramidal cells, is responsible for constructing the spatial (position and velocity) trajectory corresponding to the primary declarative goal in the subject's memory. 

The polyheterodyne in the cerebellum, which includes the purkinje cells, and parallel fibres, is responsible for executing this trajectory as a function of time, utilizing procedural skills associated with this goal. The cerebellum and basal ganglia (CBG) manages movement by a method called 'saltatory'* motion control. Each purkinje cell is meta-inhibitory, because it inhibits the firing of a neuron which itself inhibits the polyheterodyne (multiple degrees of freedom) posture-sequence transition system, or PSTS. This mechanism operates in an identical manner** to the keyframe/inbetweening animation algorithm used in animated movies (eg Pixar's Toy Story), and CGI post-production visual effects on most other types of movies. 

Each posture, which involves many d.o.f., represents a static snapshot of a moving target. The relationship between the postures  (keyframes) and the actual saltatory motion is identical to the function linking the position of the base of the spring to the sprung mass in Figure 4.2(b) above. When the climbing fibers detect that the target is 'in the crosshairs', they cause the next purkinje cell in the basal ganglia job queue to fire, thus releasing the 'brakes' on all the moving parts represented by the parallel fibres which intersect the dendritic 'fan' of each purkinje cell. 

* the term means 'jumping' ie stepwise, and is normally used to describe the signalling mode of white matter neurons. 

** Nature and Hollywood use the same system because it is both simple and optimal, demonstrating that the use of ORPH3US (Occam's Razor & Pierce's Hook Hybrid Heuristic) is able to reveal axioms common to both biological and artificial worlds.

From GOLEM to QOLEM

The GOLEM models the mind as a pair of independent information channels, but this is far from the true situation. In fact, the data hierarchies which constitute each channel are richly interconnected. We can observe this in the pyramidal cells in the cerebrum, and the Purkinje cells in the cerebellum. These cells reside in the output channels of the cerebrum and cerebellum respectively. Each pyramidal/Purkinje cell accepts dendritic inputs from many tens of thousands of input side  axons, forming what amounts to a crossbar matrix with millions of synapses, each one representing the interaction between input and output data values. 

In the late 1970's Schneider & Schiffrin did an experiment which led to the creation of the GOLEM model of mind.  They clearly demonstrated that perceptual data held in short term memory can be divided according to whether the mind can process them in parallel or serially*. The assumption that this data division is present in both input and output hierarchies is a safe one to make, especially since it supports other facts we know about hierarchies. This suggests a method of thematically analysing the interaction between input and output networks by representing the output (effector, motor) channel as  a column matrix, and the input (affective, sensor) channel as a row matrix. The issue remains as to which order to perform the matrix combination, since matrix operations are not commutative. The issue is not a new one- it was  encountered by 19th Century pioneering psychologist William James who, coining the term 'common coding', proposed that either both channels be coded in terms of percepts, or that motor data be chosen as the  common representational basis. The first option is the one James chose, and is in fact the only choice which is computationally possible. In order that the matrix multiplication be non-degenerative, the output channel (the column matrix) is placed before the input matrix (the row matrix), resulting in an outer product form. The outer product of two matrices each of arity = 2 creates a vector space with arity = 4. This is the qualiate space depicted in the bottommost diagram in figure 4.3.

*Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic human information processing: I. Detection, search, and attention. Psychological Review, 84(1)