Cybernetic roots
Twas brillig...
Unified Cognition
Don't be tempted to skim this section. Without a firm grasp of the following material (which I have called neocybernetics because it doesn't yet have a name) TDE/GOLEM theory won't make sense. The core concept is truly amazing- biomachines generate dynamic behaviours by means of animating sequences of statically determinate postures ('pose' is the robotic term that captures position and orientation within a single tuple / dyad). Biomachines and systems never need to do complex reverse kinematics/kinetics maths, only us silly old humans do that stupid stuff!
Proper scientific explanation of brain function cannot exist without also explaining mind, consciousness, free will, and self. In many ways, it is wrong-headed that we have so many different words to describe the various unified aspects of the same thing- cognition. It is equally wrong-headed to think that this is a problem 'beyond solution'. If scientists can discover the invisible submicroscopic structure of the atom (protons, quarks, quantum effects etc), then they can certainly expect to solve the complex mechanism of cognition in a reasonable time frame. TDE/GOLEM theory claims to be just such a solution. The solution it presents differs from other research efforts of a similar nature (eg SOAR, CLARION) due to its dependence on the forgotten science of cybernetics.
Mathematics is optimised for off-line modelling, while cybernetics is best at on-line governance
Cybernetics and mathematics are distinct, though related, bodies of knowledge. While mathematics is best at off-line (predictive, feedforward) modelling, cybernetics is best at on-line (reactive, feedback) governance. As a very broad rule of thumb, mathematics is generally associated with computers, while cybernetics is generally associated with robotics. Traditionally, math treatment of dynamic situations and phenomena involves circular functions, complex numbers and differential and integral calculus. Cybernetics offers a non-parametric* alternative to mathematics in the design of control systems, and the modelling of dynamic situations. Most importantly, cybernetics can offer a way to control systems that CANNOT be modelled mathematically, even in principle. The prime example is the governance of somatic muscle systems, which cannot be achieved with pre-programmed functions, even in principle**. Feldman cites 'mechanical reductionism' as the instinctively adopted, but provably flawed, mindset that is responsible for our collective (ie science's) failure to capture the true mechanism of cognition.
Inadequate Investigation
Scientific study of consciousness and its correlates is held back by investigators who explore fruitless lines of inquiry. They fail to make progress for several reasons, the main ones being inadequate background research, and insufficient understanding of computational, cybernetic and psychophysical interdisciplines.
As an example of inadequate investigation, a simple (ie transparent and mechanistic) solution to the so-called mind-body problem has been available since the 1930's (Uexkull), was rediscovered in the 1970's (Powers), and then again in 2011 (Dyer). Yet there are papers accepted by learned journals after 2000 which use as a key part of their substantive argumentation the statement that the identity (ie mind-body) problem is as yet unsolved.
An example of the second sort (inadequate grasp of the interdisciplinary nature of the AI endeavor) is the issue of modularity of mind. For most of the 20th Century, the distributed concept of the 'engram' enjoyed great popularity. Yet overwhelming evidence for the modularity of mind has existed for over a century, such as description of cognitive functional specialisation between cerebrum and cerebellum was first described in 'modern' language by Hughlings-Jackson in 1824.
* The reader's indulgence is sought in the non-standard use of this terminology. 'Parametric' has come to refer to normally distributed statistics, and non-parametric refers to data which does not follow Gaussian ideals, eg ordinal (ranked) data. In this context, parametric refers to the original meaning of the word, ie scaling factors in algebraic models- eg the constants a, b, c in the polynomial terms a.x ^(n) + b.x^(n-1) + ...(use in 'hard' sciences vs the 'soft' sciences).
**In Anatol Feldman's analysis of somatic muscle control, he demonstrates clearly how a feed-forward formulation (his terminology is 'pre-programmed') fails to model the feedback information flows needed. --Anatol Feldman, (2015) Referent Control of Action and Perception. Springerlink.
Basic Homeostat
The most basic form of cybernetic mechanism is the homeostat [1], exemplified by the thermostat, a control circuit which keeps the temperature of a system (eg the room of a house, the cooling water of an engine...) at, or close to, a desired level called the 'setpoint'. The idea of a basic homeostat is a generalisation of the thermostat, in which almost any scientific or engineering quantity (eg length, position, velocity...) is maintained at or near to a required or desired setpoint. There are almost as many ways to depict homeostats as there are variables to be regulated, however, the graphic convention shown in figure 1 is the one that will be used in this analysis. Note that the difference (setpoint T-current value P) is used to compute the system goal, or 'desired' state of affairs, in the simplest case only. In all the examples used in TDE theory, the system goal equals (T - P) + L, where L is the command offset.
Servomechanism
Figure 1(a) depicts a basic homeostat, which constantly (or periodically) calculates the scalar difference ( T - P ) between the desired, stored level T and the actual, instantaneously measured level P. Figure 1(a) does not specify whether T is static (constant) or dynamic (varied)*. Although the normal assumption is that T is static, if this condition is relaxed, and T is allowed to be deliberately varied, the basic 'vanilla' homeostat is transformed into a servomechanism or 'servo', surely one of the most useful mechanisms ever invented**- see figure 1(a)[ii] above.
The vanilla homeostat is conventionally a unidimensional (asymmetric) regulator***. For example a thermostat switches a heater on when the difference between the setpoint T and currently measured temperature P exceeds some small threshold. In the simplest model, there is no way of cooling the system. Instead, the system cools down through 'natural' processes, until the difference D=T-P, called the 'delta' for convenience, exceeds the threshold, and the heater is switched on again.
*Note carefully that the basic homeostat and the servo have identical graphic depictions, therefore the choice in any particular situation depends on interpreting its engineering context. Also note that the term 'servo' is also used to (wrongly) denote a simple 'slave' cylinder eg in automotive hydraulics, and also in electronics where it is sometimes (also wrongly) used to denote a simple open-loop linear actuator without positional feedback.
**The contemporary trend is toward increasing use of declarative coding (software) and declarative specification (systems), . Declarative coding is derived directly from a systems-theoretical abstraction of the servo concept.
***The use of modern air conditioners (so-called 'climate controllers') confuses this issue, since these devices are heat pumps capable of cooling and heating, ie changing system temperature in a bidirectional (symmetric) manner. To further muddy the waters, the spring-mass system discussed is also a symmetric regulator, because corrective (reactive) feedback loops exist when matter, typically metal, is strained (elongated) as well as when it is compressed (shortened) . This assumes that the spring and mass are firmly connected, so they can transmit forces in a bidirectional manner.
Offset + Setpoint
Figure 1(b) depicts an augmented homeostat, such that the basic servo system has an added offset/bias sub-system added. For several reasons, current cybernetic texts do not include this sub-system variant. Understanding cybernetic functionality at this level of complexity requires a deeper level of analysis, as in figure 2. Firstly, imagine a homeostat which regulates linear position of a mass m by means of a spring of resilience k (.: compliance = 1/k).
Command + Control
Now imagine that we need to move the mass in a direction parallel to the spring along a distance several times the length of the spring. How and where do we apply this 'command' signal? Won't the mass try to return to its 'natural' position, due to the reaction of the feedback forces produced by changes to the 'resting' length of the spring? We now find it necessary to distinguish clearly between the deliberate command actions (which are ALL by definition classified as 'desired', just like the setpoint), and the unpredictable external disturbances, which are compensated by feedback-modulated reactions. In figure 2(a), the desired command signals are conceptualized as acting at the grounded base of the spring, L. Like the setpoint value T, commands (declarative goals labelled L) are 'desired', which is why they have the same (+/-) polarity in the governance equation -
F = L + ( T - P ) ...........................Equation 1
Pyramidal circuits
Figure 2(b) depicts the heterodyne for one variable only. The dotted lines between the setpoint and saccade units indicate how the basic heterodyne can be modified to govern multiple concurrent variables. The dotted lines are called 'parallel fibres' (PF's) because their regulatory function is analogous to the PF's found in the vertebrate cerebellum. Similarly, the offset/bias units (saccade generators) act in a functionally identical manner to the purkinje cells* in the cerebellum. Note that the vertebrate cerebrum (forebrain) ALSO contains cells similar to cerebellar purkinje cells, in that they both seem to have a characteristically shaped soma and dendritic arborisation, described by Cajal and others as 'pyramidal'. Not unexpectedly, form dictates function. Both cerebral and cerebellar pyramidal assemblies govern the motion of many thousands of simultaneous somatic position variables. The cerebral pyramidal circuits govern spatial movement, while the cerebellar pyramidal (parallel fibre-purkinje cell) circuits govern temporal motion**.
Moore and Mealy machines
The simultaneous coordination of thousands of somatic, situational or semantic variables doesn't just happen by accident, it must be computed. When computer designers first faced this most practical of problems, going back to the theory of Turing machines and Von Neumann architectures didn't prove very helpful at all. These famous models are theoretical abstractions invented way back in the dawn of (computing) time - the 1940's - by digital age pioneers, the Mancunian*** Alan Turing and the 'Martian'**** Jon von Neumann. The function of these canonical mechanisms is to make clear the simplest possible model of computation, its 'boiled down' version. But they are not good models of real computers. So scaled-up versions had to be invented to provide the early computer manufacturers with more powerful, industrial-strength microarchitectural paradigms. These more powerful models were ROM (read-only memory) designs called the Moore machine, a F(inite) S(tate) M(achine) which describes synchronous (coordinated) clocked logical circuits, and the Mealy machine, a FSM which describes asynchronous (uncoordinated) logic circuit.
By way of comparison, Figure 2(c) presents the conventional method of depicting hybrid feedforward-feedback controllers. This block diagram methodology is only useful because so many engineers have been taught to use it over the years. It is of limited use in this current context because it is causally opaque, thereby perpetuating an outmoded and logically contradictory view of teleology. However, this form will be used to depict something called an axiomaton (to be subsequently defined). The reader must keep the analytical limitations of this graphic format in mind.
* The purkinje cells are meta-inhibitory, meaning they release (inhibit the inhibitory action of) the inhibitory neurons which control motor tasking (the additions and subtractions of tasks to the 'job queue' ) in the basal ganglia motion cells.
** Amazingly, this task separation between cerebrum and cerebellum was first discovered in the 1820's by pioneer neuroscientist John Hughlings-Jackson. It seems that science has (metaphorically speaking) a type of institutional dementia- it is forever making discoveries, forgetting them and then rediscovering them much later, like that silly old geezer who just can't find his reading glasses - because they are on his head! (I plead guilty)
*** from Manchester
**** from Hungary- together with von Karman, Polya, and Kemeny, Edward Teller and von Neumann are amongst many post war refugees fleeing communism/fascism in Hungary. The label 'martians' is attributed to Leo Szilard, Ed Teller is particularly proud of his initials (E.T. = the extraterrestrial).
