Integration

Twas Brillig....

In 'Unified Theories of Cognition', AI pioneer Alan Newell defines a 'Unified Theory of Cognition' (UTC) as a single set of mechanisms that operate together to produce the full range of human cognition - problem solving, decision making, routine action, memory, learning, skill, perception, motor activity, language, motivation, emotion, imagining, dreaming, and daydreaming, just to mention a few. 

By any measure, TQT (the acronym for TDE QOLEM Theory) is exactly the kind of cognitive theory Newell envisaged. The discovery of TQT may seem precocious, but it is in fact a result of following up a series of 'aha' (ie insightful) moments with due diligence and the best available advice.  Some of the sources of good advice are long gone. For example, John Hughlings Jackson was able to state a key architecture principle* for the brain in 1824.  Although this idea fell out of favor over a century ago, and almost no one remembers it, it has been an essential part of the eventual solution.

TDE theory is particularly notable for its use of 'vanilla' computer science concepts to model some quite esoteric phenomena - see figures 18 and 19. Perhaps the most contentious of these is that the TQT models thought as inner speech (see figure 36 below), which is compiled just like any other piece of source code, then executed to produce the desired behavior**. 

*FYI, this principle states that the function of the forebrain (cerebrum) is to compute a spatial trajectory, as compared to the function of the hindbrain (cerebellum), which is to execute body motions over time.

**Could AI be that simple? According to the carefully reasoned arguments presented in this website, this question can be confidently answered in the affirmative.

In figure 35 below, the 'discovery pathway' from TDE to TQT is depicted as a graph. Briefly, the first and main discovery was the TDE, found in 2011 after careful examination of the gross (ie macroscopic) neuroanatomy of the human central nervous system (CNS).  Strictly speaking, the term TDE* only applies to each cerebral hemisphere.  The next step was a speculative one - could the four-lobed TDE pattern be meaningfully applied to the entire CNS? The answer was resolved in the affirmative, by  its close agreement with Tulving's knowledge model.  This stage of the modelling was called TDE-R, where 'R' stands for 'recursive'.  That is, the TDE-R is the global version of the TDE. 
TDE-R is obtained by constructing the recursion 1xTDE3 (4xTDE2 ( 16xTDE1 ) ) ) .......Eq.1
Of course, the number multipliers (the '1', '4' and '16' in Eq.1) are just placeholders for the four functionally differentiated lobe types, as depicted in figure 3(a) and figure 4(a) in section 'TDE-R'.  

*the acronym stands for 'Tricyclic Differential Engine'. Because the adjective 'tricyclic' has more than one meaning in this context, use of the full, expanded term is considered too confusing.  Hence the expanded term is of historical interest only. The most recent definition of TDE is a recursive one- TDE Differential Engine.

In figure 36 below, the discovery processes covered in figure 35 are integrated into a 'cog', a complete, functional ABI design in which the component parts of the TDE/QOLEM theory (TQT) are summarised in graphical form.

A series of models of cerebellar function have been proposed 
Over the last half-century,  a series of models of cerebellar function have been proposed, each one slightly more accurate and sophisticated than the last. Their authors are-

Cerebrum and cerebellum act like software and firmware of a digital computer 
The correct model is the one found by Hughlings Jackson, and rediscovered independently by Dyer. In this model, the cerebrum computes spatial trajectories, while the cerebellum executes them. If the right abstractions are used, this pair of interfunctional mechanisms (IFM's*) is morphologically similar to the digital computer. The cerebrum and cerebellum act in concert in a manner virtually identical to the software and hardware (firmware, actually) of a digital computer, where the software is responsible for the creation of the IFM's organisational part (its structure, the bit that doesn't change shape over time), while the firmware is responsible for the IFM's operational part (its process, the bit that does change shape over time). The software::firmware relation is itself recursive, so within the software, there is a further organisational (structural) and operational (procedural) separation of functions. 

The basic neuronal parameters are well known. Consequently,  there are only two possible types of candidate mechanisms for encoding state-

Varying synaptic conductance requires backprop-like training of a population of closely associated neurons. It is meaningless to use backprop to adjust the synaptic conductance of an individual neuron. In any case, backprop, or at least, backprop-like mechanisms have not been observed in reality. Hence, by a process of elimination** the only mechanism left is to vary membrane bias. This can be varied by adjustments to external circuit parameters.  For several reasons, this is a much more viable choice. It is also the mechanism used to exert top-down (voluntary) control over somatic musculature.

Marvin Minsky noted that one of the evolutionary pressures placed upon neural circuit design is that of increasing size by means of repeating units. Perhaps there is something 'canonical' about this simple somatic muscle control mechanism. The term 'canonical' means ubiquitous in a massively modular sense***. That is, the TDE is the long-sought-after repeating functional unit.

*The 'classic' analogy for the dual parts of the IFM is the clockwork mechanism. The organisational component is the chassis or frame of the clock, which are the pieces of brass which do not move, while the operational component is the bits that do move- the hands, the spring, and gears of the catchment. 

**In this research project, this principle is called 'Holmes' Shroud'. When impossible options have been ruled out (ie when the proven dead have been covered with shrouds), whatever option remains, HOWEVER IMPROBABLE, must be true.

***Carruthers, P.  The case for massively modular models of mind

In figure 36 below, the discovery processes covered in figure 35 are integrated into a 'cog', a complete, functional ABI design in which the component parts of the TDE/QOLEM theory (TQT) are summarised in graphical form. 

In figure 37 below, one of the three QOLEM (qualium) diagrams is expanded to show how the pyramidal crossbar (PXB) circuits are constructed, using Feldman's Equilibrium Point* theory as the primary governance paradigm. The key idea behind EP governance is essentially 'control by servo'.  A feedforward 'target location' signal is added to the feedback 'target identity' signal. In the wider context of the cerebrum, it is the frontal lobe of each hemisphere which provides the top-down 'target location' information (coloured red in figures 37 and 39), while it is the Temporal lobe of each hemisphere  which provides the bottom-up 'target identity' information (coloured blue in figures 37 and 39 ).  This is effectively the same model as the currently favored 'hierarchy-prediction' model of neurocortical function. 

Most texts use the term 'prediction' which corresponds with their active inference** (a.k.a.  'passive action') paradigm, but the TQT model prefers 'declarative programming' to preserve the teleology and explanatory power of the simpler, more direct cause-effect paradigm. This means that the brain sends the motor system the penultimate-to-terminal position of the movement. As explained in a previous page of this website, this methodology is the same as the declarative programming paradigm in computer science, where the desired effect is specified by means of telling the sub-system 'where to go' (declarative specification) rather than 'how to get there' (procedural specification). This is more cybernetically compatible than a procedural approach, because each terminal position acts like an attractor basin, allowing late specification of goal approach direction (a type of late binding). The cybernetic drive differential in this case is essentially a scalar, not a vector interpolation, and so is more compatible with the 'one-sided' terminal position specification. Also, semaphores or Dijkstra 'flags/guards'*** are not needed in declarative programming, since declarative instructions have virtually zero latency. 

*The concept was discovered independently by Charles Dyer, discoverer of the TDE pattern and primary author of this website. Dyer's explanation is initially harder to follow but eventually proves more universal in its application, representing what amounts to a complete theory of subjective, declarative autonomous agent programming. 

** Adams, R.A., Shipp, S. & Friston, K.J. (2012) Predictions not Commands: Active Inference in the Motor System. Springerlink- Brain Structure and Function: 218; 611-643

***in procedurally programmed multitasking systems. there needs to be some means of preventing 'dining philosophers' type resource contention between concurrent processes in the same name space. 

Figure 39 below, which explores the functional roles of the three TDE levels, is a logical extension of figure 37, but depicts the various functions in more detail.  The critical decussation between F-lobes and T-lobes occurs at cortical level IV as shown by green stars in figure 39(c)*. 

* Adams, R.A., Shipp, S. & Friston, K.J. (2012) Predictions not Commands: Active Inference in the Motor System. Springerlink- Brain Structure and Function: 218; 611-643

While their introductory analysis is correct, their further explanation is in terms of an intermediate level of 'neuronal assemblies', which exist in between the functional levels of the single neuron and whole brain areas. Without the 'Rosetta Stone' of fractal cybernetics, any such single-level explanation is doomed to flounder on the rocky shore of inadequate axiomatisation. 

To their credit, they correctly identify the concept of neurosemantic loops with specific activity thresholds maintained by feedback mechanisms, but without invoking the specialist (and arcane) terminology of cybernetics proper, any such discussion will necessarily be ill-focused and therefore lacking a usefully specific conclusion. It is the precise nature of the feedback mechanisms, as discovered independently by Dyer** and also by Feldman*** which provide the answer to these questions.

*Pulvermueller, F., Garagnani, M. , Wennekers, T. (2014) Thinking in circuits: toward neurobiological explanation in cognitive neuroscience. Biol Cybern 108:573-593

**Dyer, M.C. (2012) Tricyclic Differential Engine-Recursive (TDE-R)- A bioplausible Turing Machine. B.Sc Honours Thesis, Flinders University of South Australia.

***Feldman, A.G. (2016) Active Sensing without Efference Copy : Referent Control of Perception. J. Neurophysiol 116: 960-976.

There are several research groups interested in a more detailed study of the cybernetics of neuronal assemblies and networks. However, these groups (eg Williams et al*) typically seem to miss the point, examining trees in great detail while failing to account for the forest, ie losing themselves in comparisons between various learning schemes (ie Hebbian non-Hebbian etc), while overlooking, or dealing only superficially with, the various ways that the neural cybernetics is actually used to do cognitive computations. 

*Williams, A.H., O'Leary, T. & Marder, E. (2013) Homeostatic Regulation of Neuronal Excitability. Scholarpedia 8(1): 1656

**The end result is, however, well worth the extra effort involved.

***This decussation or 'crossing' is referred to as a 'crossbar switch' in the earlier TDE (pre-TQT) documents.