C :: Brainsofsand

Brains of Sand C Consciousness & Conclusion

Consciousness is a limited resource

When there are few constraints on a problem, there are many possible solutions. However, when a phenomenon is so richly and robustly over-constrained as 'common' consciousness, there can only be but one possible solution, a unique mechanism.

The solution to the problem of extinction faced by each species over the long term can be recursively broken down into the aggregation of sub-problems of each individual's reproductive fitness in the short term. As a result, each animal seems to be a seamless product of a functionally unified process of deliberate design, rather than a result of multiple partly independent evolutionary processes. When we ask what the purpose of consciousness is, we are really asking how does it contribute to each individual's survival? What was its original function? How did that initial mechanism change over evolutionary time scales, as more organisms evolved, as well as organisms evolved more- more complexity of organisation, more flexible operativity and more mobility to exploit more situations.

Imagine an organism with a nervous system which has a concrete number of parts, say ten branches*. Let's allocate ten 'consciousness units' (CU's) to each of the ten sensorimotor subsystems. The reasons we may wish to do this range from the merely convenient (10 x 10 = 100%) to the deeply and critically profound -we wish the total neuropharmacological amount of 'wakeness' to be as limited as its real-life analogue, which is the number of daylight hours, or length of time (in suitable units) from waking up/dawn to falling asleep/dusk.

* sometimes choosing such so-called 'magic' numbers aids in visualisation during the early stages of modelling. When is a number not a number? When it is a 'numbol', a number used as a symbol, acting as a mental placeholder.

Figure C.1

This part of the discussion refers to figure C.1 above, which depicts the body of an unspecified organism O, and figure C.2 below, which depicts the transfer of inhibitory neurons from the organism's central serotonin (5-Hydroxy-tryptamine or 5-HT) reservoir to individual unbalanced sensoryimotor sub-systems .

Under ideal conditions, central neural ganglion N sends signal T to unspecified peripheral effector P, which sends back return signal R. Now consider what happens when the situation changes, such that R is reduced to R'. R normally acts to suppress error signal E, so reduction from R to R' allows E to increase to E'. While operating state is normal ('business as usual'), R and E are in cybernetic balance. When the external situation changes, the organism should change its operating state parameters to restore the previous balance between inside and outside- there are two possible choices, either do it immediately, when O is on-line, or later when O is off-line? To choose to wait until off-line, O delays the process of restoring balance by temporarily suppressing the warning signal E' with inhibitory neural link C until it is next off-line. This occurs, ideally, according to the universal adaptation mechanism (UAM), which follows the familiar minimax rule, minimising the maximum error (or cost), which in this case refers to the sub-signal E'. The reason for the delay in parametric readjustment becomes relevant when multiple values of T (all symbolized by T*) all contribute to a change in R . As in other similar cases (eg engineering best practice, mathematical optimisation), the finding of the correction set involves a selective clamping (and subsequent elimination) of all but one of the potential causes T*. This mechanism resolves two great mysteries for the price of one mechanism- it is both how we learn and also is the reason why we sleep. During the day, accumulated position^ errors (R'-R) are identified and labelled with tags C(i) which also act as temporary suppressors, pushing the warning signals from sub-system faults below the level of awareness, until the next off-line period. When we sleep, our brains convert temporary (sensor-side) tags C into equivalent permanent (motor-side) links C', in a process known as long-term memory consolidation. There is only a limited supply of C tags, reflecting the limited window of active opportunity over each 24hr cycle. Normally, at waking, the C tags (typically using the inhibitory neurotransmitter endorphin or EP) are all used to inhibit the serotonin reservoir. Serotonin is a sleep-producing substance, and its soporific effect on the brain is known to be the source of sleep pressure increase during waking hours. This occurs when C tag inhibition is progressively removed, by the progressive transfer of C tags from the serotonin 'pool' to each of the body's unbalanced sensorimotor systems.

Figure C.2

We will suspend our skepticism around the issue of the conscious perception of pain, by temporarily sidestepping the issue, then returning to it later.

Each organism has a series of traumatosensory fibres, specialised loops whose default condition is electrochemical integrity, similar to the closed loop continuity of an electric light circuit. When the loop in arm #2 (say) is severed, we would want the hypothetical circuit design to perform optimally, whatever that means. A commonsense interpretation would arrange for a neuromotor response which renders as much 'assistance' (delaying exact definitions, because we will repeat this cycle for each design at each level until we are happy) as there are unallocated system resources. If we design optimally, two things will ring true (i) while injured, the organism will heal, due to the enforced reduction in activity caused by the pain/equivalent circuit. Thus survival to fight for another day is assured (ii) the circuit logic (again, defined in a flexible, sensible, qualitative way, nothing resembling quantitative precision will be allowed at this early, tentative stage of conceptualisation ) will be based on a pattern which is both simple enough to climb fitness gradients, while also being a sufficiently complex 'canon' or 'prototype' to encode the wide range of future-functional and anticipated-behavioural possibilities within the menu of dominant and recessive alleles.

First, and foremost, our design should allow the uninjured animal to enjoy maximum sensorimotor performance. Each peripheral loop, after starting at the ganglion[2] which resembles the 'central station' in a typical railway network, loops through the major fleshy parts of the 'arm', then returns to exert an inhibitory influence on a centrally located, globally effective excitatory neurovector. If injury occurs, it is easy to see what happens next. The excitatory neurovector is no longer suppressed by the long inhibitory loop, and produces amplified output, which is experienced as intense pain when it occurs in identically formed and routed pathways in human nervous reticulae (Latin for net). The pain/equivalent is referred, which means that even though it is experienced as a conscious entity centrally, the sensorimotor system is trained to automatically associate it to the peripheral location of the original cause. Whatever the true nature of its pseudoconscious experiential analogue, the effect on the creatures behaviour is the same- the animal rests the damaged body part. It cannot ignore the intense signal, and in the act of attending to it, causes inhibitory neurovector 'tags' (eg neurons with endorphin or nitric oxide neurotransmitters) to be attached on a temporary, peridiurnal (for a period of about a day)basis, thus downregulating the sum total nociceptor input (noise) measured at the head ganglion, and improving signal/noise ratio, which translates immediately to improved ability to focus on the next task enqued by the goalsetter's agenda. A complete understanding of referred sensations will have to wait until more of the theory has been covered.

What if more than one arm were injured? At some point, the pain or equivalent permanent injury signal (PIS) will be so large as to block all attempts to focus attention on the next task to be done. The response will be identical to the single-injury case, but multiplied by the number of sites with traumatically severed connections- an automatic (largely involuntary) focussing of limited attentional (inhibitory) resources to the out-of-balance excitation, creating a centrally located, globally focussed anti-PIS suppression signal. Note that in all cases, the number of suppression tags is strictly limited to a sum total equal to the available number of waking hours, or other suitable time unit, which is also the same total sum of attentional tags, counted in consistent unity's, naturally.

These two types of neurovector 'tags' are each managed in a way that reflects their underlying purpose.As described in the above discussion, sensory tags are (automatically) deployed(a) each time there is a ganglionic (central)imbalance caused by peripheral injury (at the zeroth branchpoint/layer of the nervous system's organisational tree/ operational hierarchy. We have adequately covered this case.(b) each time there is a central ganglionic response caused by an equivalent situation to that of physical injury, except it is a virtual analogue of trauma, and occurs in sensorimotor systems at higher orders of response abstraction. We will now cover these higher order cases.

To adequately explain higher order effects, we must now consider the flip side of the sensory coin, namely motor-system actuation. The reason for the diurnal method of temporary allocation of suppression tags is so that the pool of temporary neurovectors (eg an endorphin or nitric oxide interneuron) is returned to its full complement by the time the organism wakes up.

If the sensor mechanism is refreshed each morning, what's become of all the daytime effort involved in the temporary silencing of problematic sensor input? Surely there is some way of amortising the learning in the form of stateful predicates of the procedural knowledge subtype.

Of course there is, evolution is nothing if not thrifty with its subject's hard-won, hand-woven experiences. All animals have evolved a process of nocturnal replacement of temporary (sensor-side) neurovectors with their permanent (motor-side) equivalent neurovectors. Unlike the temporary 'neurotags', which are of necessity loaned from, and paid back to a strictly limited inhibitory neurotransmitter (ARAS serotonin) reservoir, there is no theoretical limit to the number of permanent (eg dopamine) neurotags that can be used in the nocturnal replacement process a.k.a 'memory consolidation'. This conforms to commonsense reasoning, since the process of correcting for physical and virtual injury analogues will continue unabated for each and every day that the organism is alive and awake. It would therefore make no sense to place a numerical limit, since there is no matching real-world limitation, as there is for the sensor-side tags.

There is one key fact that makes possible this rather elegant mechanism. Every sensor-side input stimulus hasan equivalent motor-side response, similar to the stereotypical form of procedural programming scripts which link IF phrases to THEN causes. This reflects the cybernetic nature of all biosystems (see Jakob Uexkull's cirkreis) in which metabolites, energy supplies and control signals circulate continuously in tightly regulated, but not magnitude limited, closed and semi-closed loops. It is not quite true to say that sensor-side stimuli can be replaced exactly by their motor-side responses. Even if the signal magnitudes are exactly the same, the sensory inputs occur as feedback loops, while the motor outputs occur as feedforward precomputations, representing preparations, or setup adjustments.The Behaviourists realised that biological I/O (input/output) was organised around reflexes, matching pairs of sensory stimuli and motor responses, arranged to occur sequentially, therefore forming procedural loops in which the IF condition must first test true (ie its stimulus level must be superliminal, ie above the sensory threshold value, equivalent to its cybernetic set-point) before the THEN function can be successfully executed as a neuromuscular response.If an organism only has a simple set of pain/equivalent circuits (PEC's), its response to challenges (threats and opportunities) is limited to signalling that an injury has already occurred. An obvious improvement, which became evolution's next step forward, was the development of neural circuits which sent signals before injury, not after. This could be achieved with the original PEC by having two sets of sensors with low and high firing thresholds. The low threshold sensor detects when the cause of the injury (eg a predators teeth) is touching the skin, but has not yet breached it.Let's imagine what happens when something touches, but does not cut, its skin. The organism's PEC circuit is disrupted, and the central warning system distracts the organism from its normal business of goal-seeking and task completing. Hey! Suddenly it senses a threat of injury in part of its periphery, even when that injury was avoided. This simple arrangement, in which PEC's are used to detect BOTH near misses AND direct hits, where risk of injury as well as actual injury is worth disrupting 'core business', represents consciousness in its simplest possible form.

As organisms evolved improved somatic motor systems, eg longer arms and legs, they also evolved distal sensory organs. These in turn gave improved spatial awareness and hence better control over movements. These evolutionary advancements led to improved PEC's which could predict physical contact long before it ever occurred. Let's play devil's advocate, and imagine an alternative evolutionary pathway in which advances in long range perception were not exploited by PEC circuits. Clearly such a disconnected organism would be at a significant survival disadvantage to one whose proximity and injury detection circuits were interconnected.

It is important to realise that the order of the causes is the reverse of that of the effects, as described above. There wasn't any such thing as the option of connecting the distal object detection circuits to the proximal injury detection network. Instead, the distal object system evolved FROM the increasing sophistication of the injury warning system. It is NOT the case that pain is a special, enhanced type of consciousness. In fact, exactly the reverse is true. Consciousness is a special, attenuated, diluted type of pain. We think of consciousness as this subjective experience which is primarily spatial, as if we are floating in some kind of super-sensory fluid. Distant events produce tiny ripples which rush towards us across its info-fluid surface. These ripples may exert key influences on the moment-to-moment decisions which constitute organism's self-governance. Whether we notice certain tiny signals and ignore other much larger ones depends on how well these signals help us to predict the future and assess the past.

Effective Detection has been characterised as the selection of those parts which best represent the whole, eg feature detection to confirm identity). If that detection concerns time, it is called prediction eg selecting little events which best predict big ones, from the frequent (called 'episodes') to the rare (called 'occasions').

Effective assessment is the other side of detection, since they represent opposite sides of the importance coin. As before, discussing causes means covering topics in the reverse order to that used for effects. If organisms (or industries or bureaucracies) never made complex assessments of past activities, there would be almost no need for complex detection of present activity, or prediction of future ones. It is only by assessment that complex behaviours can be confidently broken down into causes and effects, core contributors and collateral recipients, drivers and passengers.

A complete discussion of the evolution of consciousness is incomplete unless we also consider emotions. The latest idea in pain science is that pain should be thought of as a kind of emotion, over which we seem to have some, though limited, voluntary control, rather than as an involuntary aspect of consciousness to be endured. However, choosing this way of looking at things means we must go right back to the start of this discussion, and put in those aspects which were hiding in plain sight, if you will.

Before pain (or pain equivalent sensations/ PES's), there must have been emotions, or pre-emotions of some kind, defined simply as an internal, global, system-wide measure of past effectiveness. The logic behind this claim is the same as the above argument linking assessment of past experiences with prediction of new ones. There is no point in making predictions without basing them on the results of assessments. Expecting useful prediction without first making assessments is wishful thinking, a course of action even worse than random guesswork.Consider the imaginary worm-like creature upon which we originally inflicted injury. How was it injured? From experience, we know that most injury is self-inflicted, the result of human error- inattention, of poor eduction, insufficient or inappropriate training, or careless preparation. Predators know that they can't catch healthy, alert, well-fed prey- they don't catch their prey so much as that it is the prey that let themselves get caught. Predation, though obviously brutal, is not intentionally cruel. While it is not supposed to be fair, its commitment to the core principle that the best should survive is absolute. Evolution, like the scales-carrying lady of justice, is fair precisely because it is blind. When the healthy are killed, it contradicts Darwin's and Wallace's claim that evolution is an isomeritocratic animalgorithm; a compass needle floating on a bubble of time, reading true north, pointing vertically up, way past good enough and even better, to simply the best. When healthy prey die, Bestworld's computer prints out an error message that the system has failed.

Consciousness and emotionality

Consciousness and emotionality are regarded as simple capabilities which nevertheless remain beyond the ken of current and foreseeable AI systems, while more complex aspects of human cognition such as knowledge and memory are much better understood*.

TDE theory offers a resolution of this apparent conundrum. It suggests that consciousness and emotionality are what animals use in lieu of explicit manipulation of knowledge and semantics. It also suggests that, without the prior evolution of vertebrate consciousness, followed by primate emotions, the evolution of language (underpinned by memory structures specialised for symbolic knowledge) might not ever have occurred.

This section deals with the differences and similarities between consciousness and emotionality. The main shared similarity is that for a given level of intensity, these mental states integrate information sources of roughly equivalent importance, whether found at low, medium or high levels. The main difference is that consciousness is an immersive (local) representation of form (real/unreal), while emotionality is an intentional (focal) evaluation of preference or utility (good/bad). One is conscious OF something in the shared self-world environs in a not dissimilar manner to one's consciousness of oneself. The low-level, physical sense of being located within a life support vehicle, the body is somehow melded with a medium level sense of purpose, teleology and volition/agency. Functioning as a thematically unifying semantic envelope is the final part of the triad, the knowledge of one's own experience and of the facts that build up our sense of a persistent identity, acting with agency within a concrete but dynamic universe.

Figure C.3

This comparison is depicted in figure C.3 above. Consciousness reflects a qualitative (subjective) assessment of situational form/familiarity, while emotionality is a qualitative (subjective) assessment of situational function/utility. Both mental states are necessarily subjective and qualitative because animals do not possess either the capability of thinking objectively, dispassionately, or the Cartesian skill to construct accurate and realistic mental maps of their surrounds. Indeed, they have little need of such complex machinations, since the world that they shared with early humans was a world ideally governable by qualitative mental processes of consciousness and emotions. Animals care nothing about the human ideals of accuracy, repeatability, reasoning and keeping written records of both the stories of the living and the glories of the dead.

The following model is a sensible one. Simple sensory inputs whose magnitude exceeds a threshold reduce to pain signals. There remains the question of how to handle simple sensory signals which are important but do not exceed the threshold. The answer is to escalate these to compound sensory inputs, via an appropriate but as yet unspecified learning mechanism. An example of this is the biceps flexor neuromusculoskeletal mechanism. The hand touches the candle flame and the arm is quickly withdrawn- 'ouch!'. The side-effect of this is the new ability to change the range of bodily extent, to introduce proximal bodily space as it were. Compound sensory mechanisms create proximal space which is immediately predictive of physical contact, and prevents injury, as in the case of the candle flame. .

The next step in the evolutionary process involves the same process again, the increase in predictive discrimination. There will be some compound sensory inputs which are sub-threshold, but still very significant, ie predictive of future proximal events. A typical example is eyesight, and the eyeball mechanism. This is really nothing more elaborate than a virtual biceps flexor system, in which the reach of the virtual arm is not anatomically fixed but variable. This complex sensor system nevertheless fulfils the same relative (recursive) incremental functionality - it extends the creatures predictive envelope even further out from the body surface, the skin. Instead of being triggered by hot wax on skin, as with the compound sensor, it is triggered by saccadic targeting, or fovea scanning. When the fovea crosses the target, or rather, when the targeted pattern is focussed upon the foveal area of the retina by saccadic eyeball motion, the next-in-line P-cell's climbing fibre receives a 'jolt' from the inferior olive ganglion, causing it to fire, and inhibiting the 'braking' neuron in the dopaminergic area of the basal ganglia. This releases the brakes momentarily, causing the (cybernetically primed) voluntary motion system to increment forward by one animation 'key frame'. In a cybernetically primed system, effectors are excited into constant readiness, and are waiting for the step-function-like release of meta-inhibitory 'braking' or clamping. The sequence of key frames exists in cerebral cortex because of the voluntary command system, which evolved at the previous secondary compound sensorimotor stage. The sacaddic sweep system operates at its heart in precisely the same manner as the biceps' flexor mechanism's descending volitional gate signal. However, the complex or tertiary stage of sensorimotor evolution involves an additional level of inhibition. The virtual flexor response is itself virtualised - it exists only as a potential for motion. At the tertiary, or complex stage, the motion itself is saltatory, jerk-like, and proceeds in fits and starts. To our minds it seems smooth because its true saltatory, meta-inhibitory dynamics are hidden below the consciousness threshold.

This evolutionary sequence, as described above, forms the basis of the embodied, embedded situated sense of consciousness possessed by all vertebrates. Virtualized sensors, developed by evolutionary pressures, create a predictive region of space surrounding the body. This represents an early warning system, if you will, in which highly predictive secondary (compound) and tertiary (complex) levels augment the primary signaling capacity of simple (pain-level) sensorimotor reflexes. The basic crude consciousness of injury-based pain messages has become augmented by evolution into a feedback-level automatic flexor reflex whose predictive span now avoids most direct injury. This mechanism evolved yet again into one whose pseudo-pain signal is even more predictive and sophisticated, and is now able to act pro-actively, in a feedforward manner. Volitional input evolved at this stage because of the extra range of choices provided by increased degree of predictive look-ahead. These developments resulted in the well-known introspective features of consciousness.

Humans evolved knowledge and language. Procedural knowledge is necessary for syntax, to be able to memorise and retrieve multiple phones (acoustic sound 'bites') for construction of phonemes. Declarative knowledge is necessary for semantics, which is the linguistic term for hierarchical states in persistent memory which carry meaning.

Humans are able to seamlessly and flexibly convert the knowledge-form (level 3) of consciousness into the visuospatial (level 2) form, via the features that memory states share with visual states (optical representations). The evolutionary advantages of storing complex images as simpler statements of fact are rather obvious. The symbol string 'A red flower' contains an amount of information in the order of bytes, whereas a visual image of the same flower contains an amount of information in the order of kilobytes to megabytes- a factor of between 10^3 and 10^6 (1,000 to 1,000,000)- certainly large enough to influence evolutionary processes.

The precise mechanism used by the human brain to create meaning is not known, but TDE theory makes an educated guess, choosing a simple function, the visual set overlap (as per Venn diagram) where each noun labels (substitutes for) a set of exemplar hierarchical visual representations (HVR's). In hominids the mental speed up occurs by substituting echoic memory symbol sequences (chunks) for visuospatial memory mappings. This substitution would result in much more compact forms of proto-logical expressions and computations. The hominid with the ability to use smaller memory symbols for a given reality representamen will be able to fit more symbols in a given region of cortex, and therefore be able to construct more complex expression functions, resulting in an enhanced ability to predict complex future states of self-in-world. Better, more accurate predictions means less missed opportunities and more successfully avoided threats at any given level of interactive environmental complexity (read: improved ability to manage social situations, rely on other cospecifics for cooperative advantages).

*thanks to Endel Tulving's work. There is enough evidence supporting his model (which uses knowledge categories to functionally subdivide human memory) to indicate that the underlying theory is probably correct