Content and Consciousness Page 7
Non-intelligent information storage is nothing more than reliable plasticity of whatever lies between input and output, and hence we can see that the capacity for non-intelligent information storage must be the basis for the capacity for intelligent information storage. Typically, definitions of information storage in the literature are definitions of non-intelligent storage, although they sometimes carry undesirable connotations of intelligence. A definition of MacKay’s, for example, is ‘any modification of state due to information received and capable of influencing later activity, for however short a time.’1 One must not read ‘rationally influencing’ or ‘appropriately influencing’ for ‘influencing’; information is non-intelligently stored whenever the effect of an input is to contribute to the determination of a later output, whatever this contribution is.
We should reserve the term ‘intelligent storage’ for storage of information that is for the system itself, and not merely for the system’s users or creators. For information to be for a system, the system must have some use for the information, and hence the system must have needs. The criterion for intelligent storage is then the appropriateness of the resultant behaviour to the system’s needs given the stimulus conditions of the initial input and the environment in which the behaviour occurs. Since appropriateness is not an intrinsic physical or formal characteristic of any thing or event, no examination of the relations between intrinsic characteristics of input and output will give us a clue about intelligence. A system that did not exhibit this capacity for environmentally advantageous response might be in fact a brilliantly conceived device for mathematical calculation (as could be determined by an examination of intrinsic relations between its input and output), but it would be in the end only a tool; it would have no intelligence of its own, and would store no information for itself. The capacity to store and use information intelligently, then, does not emerge automatically at any degree of size or complexity of the information storage and processing mechanisms, but is an additional and separable capacity. The question now before us is what features a system must have if it is to acquire this additional capacity.
VII THE EVOLUTION OF APPROPRIATE STRUCTURES
The useful brain is the one that produces environmentally appropriate behaviour, and if this appropriateness is not utterly fortuitous, the production of the behaviour must be based somehow on the brain’s ability to discriminate its input according to its environmental significance. If the brain cannot react differentially to stimuli in appropriate response to the environmental conditions they herald, it will not serve the organism at all. How is the brain to do this? No physical motions or events have intrinsic significance. The electrical characteristics of an impulse sequence, or the molecular characteristics of a nerve fibre could not independently determine what the impulses mean, or what message the nerve fibre carries, and therefore what a stimulus – however complex – heralds cannot be a function of its internal characteristics alone. Therefore the capacity of the brain to discriminate by significance cannot be simply a capacity for the analysis of the internal structure, electro-chemical or cryptological, of the input sequences. It is easy to lose sight of this when we see how straightforward a task it is for researchers to determine the ‘significance’ of neural ‘signals’ in experimental animals. Whereas we, as whole human observers, can sometimes see what stimulus conditions cause a particular input or afferent neuron to fire, and hence can determine, if we are clever, its ‘significance’ to the brain, the brain is ‘blind’ to the external conditions producing its input and must have some other way of discriminating by significance. The criteria, for example, by which the MIT group2 determines that certain afferent signals from the frog’s retina signify sustained contrast or moving edges or convexity cannot be used by the brain of the frog to discriminate these signals, because the frog’s brain cannot observe the frog’s retina, cannot tell where these signals are coming from.
Since environmental significance, even in the attenuated sense in which retinal impulse streams signify certain retinal conditions, is not an intrinsic physical characteristic, the brain, as a physical organ, cannot sort by significance by employing any physical tests. The only other explanation that would be acceptable to the physical sciences is that the brain’s capacity to discriminate appropriately is based on chance. That is, a particular pathway through the brain might just happen – entirely fortuitously – to link an afferent (input) event or stimulus to an efferent (output) event leading to appropriate behaviour, and if such fortuitous linkages could in some way be generated, recognized and preserved by the brain, the organism could acquire a capacity for generally appropriate behaviour.
Let us mean by a functional structure any bit of matter (e.g., wiring, plumbing, ropes and pulleys) that can be counted on – because of the laws of nature – to operate in a certain way when operated upon in a certain way. Obviously just about anything can be viewed as a functional structure from one point of view or another. A functional structure can break down – not by breaking laws of nature but by obeying them – or operate normally. A nail is a functional structure; so is a gall bladder, and an open telephone line between Washington and Moscow. Given a brain with an initial plasticity or capacity for producing different functional structures as a result of input, the key to utility in the brain must be the further capacity to sort out these functional structures, keeping and using those that are useful to the survival and comfort of the organism, and eliminating or refraining from using the harmful ones. We cannot suppose that harmful structures suffer in themselves from any physical defect – e.g., chemical instability or a tendency to atrophy – nor that useful structures are particularly robust, so if we are to have an analogue of natural selection do the sorting of structures, it cannot operate on a principle of physical fitness; nor can it be, as we have seen, that useful structures are useful in virtue of any distinguishing intrinsic physical characteristic that could be keyed on by a sorter. There are other ways of establishing a sorting principle, however, and an externally grounded sorting mechanism that meets the requirements we have enunciated can be described with the help of a very elementary excursion into a hypothetical evolutionary history.
At a very early point in evolutionary history, organisms appeared with simple nervous systems; contact with their surfaces produced electrical activity similar to that of neurons. The value of this phenomenon depended on the result it happened to trigger. Suppose there were three different strains of a certain primitive organism in which a certain stimulation or contact caused different ‘behaviour’. In strain A the stimulation happened to cause the organism to contract or back off; in strain B the only behaviour caused by the electrical activity in it was a slight shiver or wriggling; in strain C the stimulation caused the organism to move towards or tend to surround or engulf the point of contact causing the stimulation. Now if the stimulation in question happened to be caused more often than not by something injurious or fatal to the organism, strain A would survive, strain B would tend to die off and strain C would be quickly exterminated (other conditions being equal). But if the stimulus happened to be caused more often than not by something beneficial to the organism, such as food, the fates of A and C would be reversed. Then, although all three responses to the stimulation are blind, the response that happens to be appropriate is endorsed through the survival of the species that has this response built in. This observation taken one way is tautological; what is appropriate tends (by definition) to aid survival; what is inappropriate tends (by definition) to kill off the organism. The species that survive are the species that happen to have output or efferent impulses connected to the afferent or input impulses in ways that help them survive.
As the evolutionary process continues, the organisms that survive will be those that happen to react differently to different stimuli – to discriminate. Thus if strain A backs off for both stimuli X and Y, while strain B backs off for X and approaches for Y, and if X happens more often than not to announce injury and Y
happens to be caused more often than not by nourishment in the environment, strain A will die of starvation since it runs from both danger and food, while strain B will survive by discriminating appropriately. The discriminatory behaviour of strain B is only blind, dumb-luck behaviour; that is, it is the fortuitous and unreasoned result of mutation, the appropriateness of which is revealed by the survival of the strain. In this way a variety of simple afferent-efferent connections can be genetically established, and once they are firmly ‘wired in’ the afferent stimuli can be said to acquire a de facto significance of sorts in virtue of the effects they happen to have, as stimuli-to-withdraw-from and stimuli-to-remain-in-contact-with. Moreover, natural selection ensures that the former will be in fact danger signals and the latter, beneficence or security signals – harbingers of good in one respect or another. Of course nothing in the organism will recognize these stimuli as danger or security signals, unless one wants to say that the organism’s good fortune to be so wired as to react appropriately to these stimuli amounts to its recognition of their import, but this would surely be an overly fanciful way of speaking.
So far so good: natural selection can provide for the dullest sort of appropriate reflex responses to stimuli discriminated by their meagre, in fact binary, ‘significance’. Other genetically grounded connections besides those rudimentary arcs would be possible and in fact likely. In all species the pain network is at least to some extent wired in (and we shall see later why this must be so), and the transmission of the controls of rigid ‘instinctual’ behaviour must also be genetic. Any afferent-efferent connection that was regularly appropriate would have survival value, the likelihood of survival depending on how regular the beneficial environmental results of the response motion are. It is presumably possible in principle for evolution to produce an organism with a useful brain that was entirely genetically pre-wired in this way and had no plasticity at all. This would depend first on the genes’ having sufficient information-transmission capacity to transmit complete wiring-diagrams from generation to generation. Such an organism would ‘know it all’ from birth and be unable to learn – not that it would need to. This could only happen where the environment in which the species lived consisted of utterly stereotypic situations and remained extremely uniform throughout the aeons of evolution. Only where the appropriate response to a stimulus remains unchanged from individual to individual and generation to generation would pre-wiring on such a scale have any survival value. And of course no matter how precocious the organisms would appear in their natural habitat, in an alien environment they would be worse than moronic. Such rigid behaviour patterns, or tropisms, are of course common among insects and other lower animals, and, for example, if fires became regular features of the environment of the phototropic moths, they would soon become extinct.3 Thus a preponderance of tropistic behaviour controlled by pre-wired afferent-efferent connections can become an evolutionary trap for a species if the environment changes.
If too much inherited wiring is a bad thing, a certain amount is absolutely essential. Nothing about an afferent impulse by itself could mark it as positive or negative ‘feedback’ and thus start the learning process. Afferent impulses alone could have no useful bearing on the behaviour of an organism, so there is no hope of achieving utility unless some afferent impulses are pre-wired to the appropriate responses. The problem then becomes: how does the pre-established ‘significance’ of some afferent impulses allow the brain of a learning organism to discriminate appropriately the other impulses, which are not genetically endowed with any ‘significance’?
A fairly common picture of the brain that might suggest itself in response to this question must be rejected now. This is the picture of the brain as composed at birth of two sides, afferent and efferent, with a few pre-wired connections between the two sides (reflexes and tropisms), but the rest of the gap free of connections, awaiting the dual gift of efferent coordination and afferent analysis, before the afferents are connected to the ‘right’ efferents. This view may be a hangover from the telephone switchboard motif of a few decades ago, which made it comfortable to envisage two essentially separable switchboard systems: the afferent caller announces his business and the operator plugs him in to the appropriate efferent receiver. This is a hopeless way of looking at the brain, since it still requires ‘the little man in the brain’ who understands, reasons, and in general intelligently uses the brain, thereby robbing the brain of just those intelligent functions we are trying to endow it with.
The insoluble problem of getting the stimulus to find the right response, of making the right connection across the great divide, can be avoided only if it is supposed that the afferent and efferent sides of the brain are richly, if to some extent randomly, interconnected from birth. The classic function of natural selection is to cull repeatedly the few good from an abundance of candidates, and if the process of evolution is to be brought into the brain, there must be an initial abundance from which to cull the survivors. Skinner has the concept of ‘operant’ behaviour, which is not stimulated or ‘elicited’ but just ‘emitted’ by the brain – apparently by the efferent side of the brain acting alone.4 The meaningless babbling of an infant and its apparently random limb movements are examples of operant behaviour, and Skinner holds that somehow operant behaviour can be refined and connected to a stimulus cue. Thus the child learns to speak and walk. Skinner’s problem is how to make the afferent stimulus cues jump the gap to the efferent side and select the appropriate ‘emissions’. If instead of supposing with Skinner that the apparently random operant behaviour is in fact randomly emitted by the efferent side, we suppose that it is stimulated – entirely inappropriately – by the as yet unstructured and unanalysed afferent barrage, the problem is no longer how the afferents get to their appropriate efferents, but how the appropriate interconnections among the many inappropriate ones get weeded out for survival. What is needed is some intra-cerebral function to take over the evolutionary role played by the exigencies of nature in species evolution; i.e., some force to extinguish the inappropriate. A capacity for propagation is also needed to provide continued abundance for intra-cerebral selection. In inherited pre-wiring we have the basis for such capacities, but in order to explain how this might work we must delve deeper into the physiology of the nervous system.
The nervous system is composed of two major types of cells, neurons and glial cells. The glial cells are generally supposed to have the function of providing life support for the neurons, but perhaps they also participate in the functional plasticity involved in information storage. Whether they do in fact have this latter function is immaterial here. The neurons, which number in the neighbourhood of 10,000,000,000 in the human brain, are the transmission and switching elements of the brain, and may contain within themselves the whole capacity for information storage, leaving the glial cells to their more mundane role.
Schematic Diagram of a Neuron
If the neuron has a threshold of + 2, it will fire impulses along its axonal branches only when it receives impulses simultaneously from at least two of A, B, C, but not D – or all four.
Each neuron has an input end, consisting of many terminals to which are led the outputs from other neurons or, in the case of neurons on the periphery of the nervous system, from receptor cells in sense organs. The neuron has a single output line, the axon, which branches after leaving the cell body into many outputs which lead to the input terminals of other neurons. The endbulbs of the axon branches do not quite touch the input knobs on the receiving neurons; the gap between them, or synapse, is crossed only when the impulses arriving at the synapse achieve a certain minimum frequency. The millions of neurons, particularly the afferent neurons, are arranged in regular ranks, so that all the output branches of neurons in one rank connect to inputs of neurons in the next rank up. There are important exceptions to this directionality, such as the ‘descending effects’ that seem to be critically involved in the process of perceptual analysis, but th
ey do not concern us here.
Of paramount importance to the theory to be proposed is the phenomenon of threshold. Some synaptic crossings contribute to the excitation of the neuron and some inhibit its excitation. Each neuron has a ‘statistical’ or threshold mechanism so that it fires its output only when the weight of excitatory crossings at a given time exceeds the weight of inhibitory crossings by a certain value. To simplify for our purposes, if each excitatory crossing is given a weight of +1 and each inhibitory crossing a weight of -1, a neuron with an excitation threshold of 2 would fire its output only when, for a short moment, the sum of all crossings >2. The threshold of a neuron is variable. Frequent firing of a neuron tends to lower its threshold while inactivity raises the threshold.
This much seems quite well established by the neurophysiologists although the importance and roles of these features are widely debated. It is in any case enough for the hypotheses we need, and in some ways more than enough. All we need is a multitude of switching elements arranged with enough directionality to allow us to speak in a general way of higher and lower levels, and a general rule (though it need hold only over a certain range of values, and need not be unexceptioned) that the firing of a switching element increases the likelihood of its firing again. This last condition, for which there might be equally suitable analogues, gives us our principle of ‘species’ propagation.