3 :: conscious-computation

Animal GOLEM

Like the basic TDE, the basic GOLEM model is structurally and functionally equivalent to one cerebral hemisphere of the vertebrate brain [1]. It is therefore an emulation of animal thought, one which maps one single GOLEM to each of the left and right cerebral hemispheres respectively. It is also an emulation of the sensorimotor (GOLEM level 1) and spatiotemporal (GOLEM level 2) aspects of human thought, minus the cognolinguistic (GOLEM level 3) components. This is because animals and humans share the lower two levels of the GOLEM model, and (perhaps with higher primates, some birds, some cetaceans, some cephalopods) some functions of the third level.

A core claim of GOLEM theory is that the hierarchical data structures common to both channels transform objective quantities (feeding into them in a bottom-up manner) into subjective ones, and vice versa. Due to common coding (a global effect) as well as the 'counterflow' crossover between input and output information processes (local effects), the cumulative effect of these hierarchies is that the primary form of memory storage in the neocortex is as semantic categories, which are percept classes not entirely dissimilar to data types in programming languages. Their existence is also a direct consequence of the GOLEM model.

The input channel (equivalent to the Parietal and Temporal lobes in a vertebrate brain) bridges the gap between the existential level (the 'perceived') and the experiential level (the 'perceiver'), while the output channel (equivalent to the Limbic and Frontal lobes in a vertebrate brain) bridges the gap between the experiential level (the 'controller') and the existential level (the 'controlled'). The iconoclastic NYU Philosopher Ned Block puts it in the following way- "a reflexive conscious state is one that is phenomenally presented in a higher order representation of it". This state of affairs is depicted in the figure below. Although the brain shown is human, this discussion refers to animal brains as well. Figure 2.1 below depicts three equivalent lateralised views [2] of a vertebrate cerebrum, (a) anatomic, (b) cybernetic and (c) computational.

(i) The left side diagram depicts the basic data path that information follows when going from dorsal spinal input to ventral spinal output. When simplified in this rather extreme way, the data path resembles an inverted, twisted 'U'.
(ii) The middle diagram in the figure depicts the abstract neocybernetic view of the brain, in which the 'what' stream (sensor-side setpoints) and the 'where' stream [3] (motor-side offsets) are clearly visible, and labelled as T and L lobes, respectively.
(iii) The right-side diagram depicts the four quadrants of GOLEM, and their anatomic equivalents (L, T, F, and P 'lobes'). The equivalent computational (programming) model has the acronym GOES, standing for Goals-Objects-Events-Scripts. The goal-oriented part of the GOLEM computational model uses a declarative, not imperative (procedural) paradigm.

(a) (b) (c)

Figure 3.1 - GOLEM Theory 'Rosetta Stone'

With little loss of generality, we can divide the input (sensor-side) channel of each hemisphere into two mutually exclusive sets of representations-

(i) consciously experienced entities

(ii) those unavailable to conscious perception that we label as unconscious.

In this latter category, we include some but not all sleep states, plus deep anaesthesia such as that induced during major surgery and other medical procedures which require the subject to be in an off-line state.

Similarly, we can divide the output (motor-side) channel of each hemisphere into two distinct function groups-

(iii) voluntary actions, those goal-pursuits under command of the agency of the self

(iv) involuntary actions, those which are self-triggered, or initiated automatically as a part of a learned sequence, typically stored in implicit memory. Involuntary actions are usually subservient to voluntary ones. This is the same kind of parts-whole relationship that groups of stimuli possess w.r.t. the higher-level (eg object) percept that their combination signifies.

(a) (b) (c)

Figure 3.2

Lets start with the GOLEM quadrant labelled P for its proposed similarity to our Parietal lobes. This part of the GOLEM detects sensory events, which are theoretically indistinguishable from those processes the Behaviourists called 'stimuli'. Now lets move to the GOLEM quadrant labelled T for its proposed similarity to our Temporal lobes. The T quadrant lies 'above' the P quadrant, meaning that the information output from the P quadrant flows into the T quadrant. Specifically, the brain learns which patterns of sensation (arrays of stimuli) correspond to familiar objects. Typically, patterns of stimuli from the P-quadrant match learned patterns stored in the data memory in the T quadrant. Just as stimuli at low levels are consciously thresholded based on their magnitude [4] so patterns of stimuli are allowed [5] to enter consciousness based on their higher-order resemblance to the shape and dynamic behaviour of known objects.

The input channel (P-lobe and T-lobe) does not act in isolation, of course, but constantly receives information feedback from the output channel, consisting of the L-lobe and the F-lobe. We have not said how the T-lobe can recognise the patterns of stimuli presented to it by the P-lobe. After all, these are not static in the general case, but consist of many different facets of the same solid object, rotated in various ways according to the given situation and individual behaviour. It is the F-lobe that comes to the rescue! It contains 'scripts', which are a kind of formulae [11] involving functional concatenations, describing the behaviour of sensory events (spatial motion of stimuli) under various types of dynamic regime. The dynamic functions of the F-lobe, combined with the T-lobe's data store of static shapes, constitute the missing link to our consciousness recognition of solid moving objects.

the brain is a duplex counterflow device

These interactions arise because the brain is a duplex counterflow device - see figure 3.2(b). The GOLEM models this intersection between the input streams and output streams as a crossbar switch. Separate, though connected crossbars exist at all three levels of the model. for example, the pyramidal cells of the cerebrum form the efferent fibres of the level 2 (spatiotemporal) crossbar, while the parallel fibres and purkinje cells form the afferent and efferent fibres (respectively) of the cerebellar (level 1 -sensorimotor) crossbar.

As to the neural circuits needed to implement these functions, all that is needed is simply the many-in-many-out, feedforward structure that all neurons have - taking many inputs, thresholding them according to their collective significance, and then sending copies of the output signal to all the target neurons, the ones whose dendritic bushes lie on the axon's 'party line'. Neurons like these naturally join to form hierarchies of significance (HOS). These HOS's can't help but form the higher order representations (HOR) and higher order HORs (HOROR) that form the basis of our conscious awareness and related kinds of spatialized knowledge.

infants learn about their body by motor 'babbling'

As stated before, the ability of the T-lobe to statically recognise solid objects is dynamically enhanced by the F-lobe's ability to learn how fixed patterns transform under three dimensional rotations. This ability, once acquired by animal (incl. human) infants is immediately put to good use, by learning the most important patterns of all, those that correspond to the sensory changes that accompany each individual's own voluntary motion of its limbs and its entire body [6].

More important than the changes are the invariants. It has been said that all cognition can be boiled down to the computation and memorisation of invariants. The infant quickly learns that the body structures that it goes to sleep with at night are the same ones that are there when it wakes up next morning. The different parts of the world (including others like you) may vary from day to day, but there is one part of the world that is always there, yet cannot be seen in toto (without a mirror) - one's own body. Without output feeding back into input, (see diagram) ie without neural correlations that link perceptual change to the voluntary motor commands that create them, these insights would not be possible.

animal GOLEM has GOES programming model

(i) The left diagram of the figure 3.2(a) above depicts the input and output channel of one cerebral hemisphere. Each channel is shown with its own data hierarchy. The basic GOLEM has an associated goal-oriented (agent-based) programming model called GOES. This acronym is formed from its four constituent operations- GOALS-OBJECTS-EVENTS-SCRIPTS (not shown in figure 3.2).
(ii) The purpose of figures 3.2(b) is to remind the reader of the neuroanatomical decussation [7] that enables us to combine the C,U of the input with the V,I of the output channel, making the QOLEM or qualiate GOLEM.
(iii) the four compound subjective states of the QOLEM are shown in the right-hand diagram.

QOLEM = qualiate GOLEM

The veracity of the QOLEM is supported by retroduction [10]. Each of its four quadrants corresponds to four common operating regimes of the brain -see figure 3.2(c). They are:
(i) active attention
(ii) passive awareness
(iii) automation of action sequences- voluntarily triggered, but involuntarily continued
(iv) autonomous operation, eg of viscera and metabolism.
Each subject's phenomenal (egocentric) mental space within the QOLEM has cartesian dimensions of consciousness and volition. Practically, the QOLEM offers a reasoned, scientific explanation for Libet's Paradox, far more acceptable than the impossible explanations Libet's work initially generated. Note Libet's 'sensible' 1990's rewrite[8] of his original 'nonsensible' reversal-of-subjective-time interpretation [9].

1. It is well-known that the (animal) level 2 functions of each cerebral hemisphere governs the opposite handed half-space. This is true for both animals and humans, and fully explains the structure of the optic decussation.

2. this type of diagram is referred to metaphorically as a 'Rosetta Stone'.

3. the awareness that vision (spatiotemporal level 2) consists of 'what' and 'where' streams arises from clinical studies of 'blindsight' patients.

4. relative to cybernetic setpoints

5. These distinctions govern what Ned Block calls the 'phenomenal' (subjective, introspective) aspect of consciousness. Sure, some important low level stimuli (eg pain) does enter subjective consciousness, but mostly we are only conscious of 'the big stuff', such as our own body, emotional reactions and motivations, those of the people we interact with, and big fast moving objects like cars which may collide with us.

6. This process is called 'motor babbling'.

7. this decussation is functionally equivalent to a duplex (counterflow) crossbar switch, not unlike those found in telephone exchanges from the pre-digital era.

8. Libet, B. (1993) Unconscious cerebral initiative and the role of conscious will in voluntary action. Neurophysiology of Consciousness. Contemporary Neuroscientists.

9. Libet, B. (1999). Do we have free will? Journal of Consciousness Studies,6,47-57.
Also see Libet, B., Gleason, C. A., Wright, E. W., & Pearl, D. K. (1983). Time of conscious intention to act in relation to onset of cerebral activities (readiness-potential): The unconscious initiation of a freely voluntary act. Brain, 106,623-642.

10. originally called abduction by its discoverer, 19th Century logician C.S. Pierce.

errata: in figure 3.2(c), “3rd party” should be “3rd person”

11. The products used in F-lobe scripts are Cartesian, ie outer products which act set theoretically, and are space-preserving, not Euclidean, or inner products, which turn vectors (spaces) into numbers (scalars).