Content and Consciousness Page 10

  A nesting of hierarchies could probably be generated from a single ‘seed’, a pre-wired set of controls for some one simple pattern of goal-directed behaviour. For example, if some ‘hunger’-afferent had the genetically established effect of producing continuing, widespread, but ‘unspecified’ efferent activity – leading, one might suppose, to behaviour in the form of random scrambling about – this efferent activity could be pruned along the lines of various different ways of getting food by the evolutionary principles we have already proposed. Once these different sub-routines had been established, sub-goals within them could develop from the convergence of new afferent and efferent activity, and so forth, spawning in pyramidal fashion a series of goal-direction controls in overall response to the obstacles in the environment preventing the direct achievement of the pre-wired end-states. In such a view the pressure of the hydraulic models still exists in not entirely metaphorical form: the afferent initiation of the behaviour produces a pressure of efferent activity which seeks to be relieved via various routes, these routes controlling various attempts at achieving the goal.

  I do not want to lean at all heavily on this rather cloudy view of goal-generation, but only use it as an illustration of one possible avenue for the researchers and programme devisers. In all likelihood considerably more hierarchical structure is genetically transmitted; for example, food-seeking tropisms and reflexes (including the sucking reflex of a baby) of considerable complexity are very widespread in nature, and in higher animals these could serve as a much more elaborate starting point for goal-generation than was suggested above.

  The preliminary sketch of a centralist theory of behaviour developed in this chapter is intended only to reveal the general shape such a theory must take if it is to deal with the problems set for it by what are largely a priori conditions. Although there has been some guarded comment about the ‘significance’ of certain neural impulses, and the ‘ambiguity’ of certain others, no strict justification has been yet proposed for what must be the crux of any centralist theory: the ascription of content or meaning to particular central states of the brain.

  4

  THE ASCRIPTION OF CONTENT

  IX FUNCTION AND CONTENT

  So far the argument has been that if there is to be a rapprochement between the extensional physical sciences and ‘the language of the mind’ – whether our ordinary Intentional discourse or the statements of a ‘science of Intention’ – we must find a rationale and justification for ascribing content to certain internal states and events of the behavioural control system. And since Intentional explanations presuppose the appropriateness of the sequences of events they purport to explain (see Chapter 2), part of the burden of such content ascription is providing an account of the generation of structures to direct these generally appropriate sequences. It was to meet this requirement that we proposed hypotheses about evolution, both of species and of neural structures. Put another way, since environmental significance is extrinsic to any physical features of neural events, and since the useful brain must discriminate its events along lines of environmental significance, the brain’s discriminations cannot be a function of any extensional, physical descriptions of stimulation and past locomotion alone. Rather, some capacity must be found in the brain to generate and preserve fortuitously appropriate structures. It was then argued that a close analogue of natural selection of species would be a system that could provide this capacity and could itself be provided for by natural selection of species. The system was developed just enough to provide some answer to the question of whether it could control goal-directed behaviour, but it will provide us with footholds for the next task: determining the conditions under which one could justifiably ascribe content to neural states.

  For a start it is clear that for any system to be called Intentional it must be capable of discriminating and reacting to fairly complex features of its environment (e.g., external physical objects and not just changing conditions – temperature, contact, pressure – on its outer surface), and for any system to do this it must be capable of interpreting its peripheral stimulation. That is, it must be capable of producing within itself states or events that normally co-occur with generalized conditions of objects within the system’s perceptual field. I do not think this is a formal requirement for any Intentional system so much as one designed to satisfy our intuitions; no system that lacked this capacity could engage its environment in ways interesting and sophisticated enough to make it plausible to say that it had beliefs, desires, intentions – even if in the end we could find no logically necessary trait for, say, belief, that the system lacked. This capacity for afferent analysis does not suffice in itself, however, to establish a system as Intentional, for the information produced by such an analysis, for all its abstraction from its source in peripheral stimulation, will still be only non-intelligently held information unless something else is added. The something else is a certain association between the results of afferent analysis and structures on the efferent side of the brain. This can be brought out by example. Suppose that in an organism O there is a particular highly interpreted afferent output A (summing, we can suppose, signals from visual, tactile and olfactory sources) that fired normally if and only if food was present in O’s perceptual field. The firing of A might have any of a vast number of effects on O’s behaviour. If it happened, for example, to have the effect of terminating a series of ‘seeking’ sub-routines and initiating a series of other, ‘eating’ sub-routines, we would have evidence for saying that O had achieved its goal of finding food, had recognized that the goal was achieved, had discriminated the presence of food as the presence of food. If, on the other hand, A did not have this effect, if O did not commence eating or in other ways behave appropriately to the presence of food under the circumstances, then regardless of any evidence we might have about the specificity of the stimulus conditions determining the firing of A, there would be no reason to say that the animal had discriminated the presence of food as the presence of food.

  This point has often been missed, probably because of a misplaced analogy between our introspective experiences while problem-solving and the state of affairs that exists in the brain. The point is missed when some S-R behaviourists pose their most intractable problem: how does the novel stimulus (meaning what it does) get to or select the appropriate response? This is a hopeless question, for it presupposes an impossible state of affairs in the brain, in which the brain somehow recognizes or discriminates a stimulus as, say, one of pain or of a white triangle or of the dead-end of a maze alley, but does not ‘know’ yet what the appropriate motion is in the face of such a stimulus. This view of the situation is apparently harmonious with our experience in that we often come upon things in our environment about which we wonder ‘what shall I do about that?’, so it seems plausible enough that discrimination of stimuli and doing something about them are perfectly separable. But this extrapolation from human experience is not justifiable on the level of explanation involved when one is talking about brains rather than people. In the brain, discrimination of afferents according to their significance just is the production of efferent effects in differential response to afferents, and hence it does not make sense to suppose that prior to the production of an efferent event or structure the brain has discriminated its afferents as anything at all.

  No afferent can be said to have the significance ‘A’ until it is ‘taken’ to have the significance ‘A’ by the efferent side of the brain, which means, unmetaphorically, until the efferent side of the brain has produced a response (or laid down response controls) the unimpeded function of which would be appropriate to having been stimulated by an A. This is not the epistemological point that as behaviourists we cannot tell whether the organism’s brain has discriminated its stimulus as having the significance ‘A’ until the organism manifests this in its behaviour, but the logical or conceptual point that it makes no sense to suppose that the discrimination of stimuli by their significance can occur so
lely on the afferent side of the brain. The epistemological point is a canon of the experimental method in psychology: since the animal cannot tell us whether it can tell a circle from a square, we must set up the situation so that its behaviour tells us. This canon, as it stands, hides an ambiguity. Surely there is another alternative which we are prevented from using only by the limits on our present research techniques. We could in principle record the afferent activity in the animal when its eyes were presented with circles and squares and, on the basis of vast knowledge of the principles of afferent function, determine that the animal’s afferent analysis system had unique and different outputs for circles and squares. Would this show that the animal discriminated circles from squares? In one sense, it would. This is the sense of discrimination of interest in research into pattern recognition devices, where all that is at issue is whether or not the system is capable of producing outputs – whatever they may be – that co-occur with the critical patterns of the inputs. In principle we could know that in this sense an animal could discriminate circles from squares without ever examining its overt behaviour. This is not yet discrimination by significance, however. We would not give as the conclusion of this experiment that the animal could discriminate circles as circles and squares as squares. Furthermore, for all animals lower than human beings there is no behavioural experiment we could perform that would have this as its conclusion, since circles and squares, even under laboratory conditions, could have no bearing as circles and squares on the life and activities of the animal. They could have bearing as left-turn indicators or as warnings of an electric shock, but not as circles and squares. This can be seen by contrasting circles and squares with food pellets. One could set up an experiment in which a food pellet served as a left-turn indicator for a rat in a maze, and once the rat had learned this we could say its behaviour showed that it discriminated the food pellet as a left-turn indicator. To discriminate the food pellet as food, on the other hand, is to try to eat it. There is something appropriate a rat can do with a food pellet such that it makes a difference whether it is a food pellet or a marble, but there is nothing a rat could do with a circle such that it makes a difference whether it is a circle or a square or a triangle. This limitation is due, of course, to the very limited interests and activities of rats. Were rats interested in making wagon wheels the situation would be different. The significance an item in the environment can have to a creature is limited by the creature’s behavioural repertoire, but this limitation only comes into force at the level of afferent-efferent intermeshing, which is, therefore, the first point at which we can speak of discrimination by significance. Since, then, effects on behavioural controls are conceptually required for there to be discrimination by significance, and since a stimulus, as a physical event, can have no intrinsic significance but only what accrues to it in virtue of the brain’s discrimination, the problem-ridden picture of a stimulus being recognized by an animal, meaning something to the animal, prior to the animal’s determining what to do about the stimulus, is a conceptual mistake.

  An idealized picture of content ascription emerges from this from which we can draw some conclusions before complicating it out of existence. The content, if any, of a neural state, event or structure depends on two factors: its normal source in stimulation, and whatever appropriate further efferent effects it has; and to determine these factors one must make an assessment that goes beyond an extensional description of stimulation and response locomotion. The point of the first factor in content ascription, dependence on stimulus conditions, is this: unless an event is somehow related to external conditions and their effect on the sense organs, there will be no grounds for giving it any particular reference to objects in the world. At low enough levels of afferent activity the question of reference is answered easily enough: an event refers to (or reports on) those stimulus conditions that cause it to occur. Thus the investigators working with fibres in the optic nerves of frogs and cats are able to report that particular neurons serve to report convexity, moving edges, or small, dark, moving objects because these neurons fire normally only if there is such a pattern on the retina.1 However mediated the link between receptor organ and higher events becomes, this link cannot be broken entirely, or reference is lost.

  The point about the link with efferent activity and eventually with behaviour is this: what an event or state ‘means to’ an organism also depends on what it does with the event or state. This suggests a tempting but not altogether reliable analogy with the logical distinction of extension and intension: the stimulus conditions dependence ensures that neural ‘expressions’ will have reference or extension, while the efferent effect dependence ensures that they will have sense or intension. Our paradigm here is the case of the simple nervous system of strain A, in which there is an inherited arc linking a certain stimulus with a withdrawal motion, and the stimulus conditions of which are, as a general rule, harmful to the well-being of strain A. The effect of this link and the conditions under which it operates give us reason for calling the afferent side of the arc a signal of pain, or perhaps danger, but there would be no reason for this ascription if the organism responded inappropriately or not at all to the stimulus, whatever the conditions of stimulation. That the stimulus did not mean danger to him would be abundantly clear from his reaction. The criterion for intelligent information processing must involve this behavioural link – however mediated – since propitiousness or adaptiveness of behaviour is at least a necessary condition of intelligence. This immediately establishes a limit on the events and states within the brain to which the investigator can ascribe content. Where events and states appear inappropriately linked one cannot assign content at all, and so it is possible that a great many events and states have no content, regardless of the eventual effect they have on the later development of the brain and behaviour.

  This point is important enough to be worth further development. Let us concoct an artificial case in which the behaviour is wildly inappropriate to the perceptual environment. Fido, who has not been fed all day, is handed a large chunk of beefsteak, but instead of eating it he carefully gathers together a little pile of straw, puts the meat in the middle, and sits down on the meat. Now suppose further that we have voluminous data on Fido’s neural states. Afferent state A is the outcome of the convergence of olfactory, visual and tactile stimulation, and is the normal outcome of afferent analysis when Fido in the past has discriminated food. But this time its efferent continuation leads to the bizarre behaviour. Since Fido has not behaved appropriately, we cannot say that state A has the content (roughly) ‘this is food’ for him, but if not, no other candidate is supported either. Fido’s behaviour would be appropriate to a belief that the beef was an egg and Fido was a hen, and since state A has the efferent effect governing this behaviour it might seem that solely on the basis of our second factor, the generation of behaviour controls, we can ascribe content to state A: ‘this is an egg and you are a hen’. But Fido’s behaviour is also appropriate to other beliefs, e.g., ‘this is beef, but if you pretend it’s an egg you’ll get twice as much beef tomorrow’, or ‘it is worth starving to throw these psychologists into confusion’ or ‘sitting on beef improves the flavour’. Since any behaviour will be appropriate to a variety of different beliefs and desires, the only feature that can be counted on to determine the correct hypothesis will be the afferent source of the structure that governs the behaviour, and the afferent source will favour one of the hypotheses only in the event that the behaviour is appropriate to the conditions of this source. Where there is an inappropriate liaison, the response to the environment ‘makes no sense’, and, since it makes no sense, no Intentional (putatively sense-making) account of the liaison will be justified.

  So, one can only ascribe content to a neural event, state or structure when it is a link in a demonstrably appropriate chain between the afferent and the efferent. The content one ascribes to an event, state or structure is not, then, an extra feature that one discovers in it
, a feature which, along with its other, extensionally characterized features, allows one to make predictions. Rather, the relation between Intentional descriptions of events, states or structures (as signals that carry certain messages or memory traces with certain contents) and extensional descriptions of them is one of further interpretation. If we relegate vitalist and interactionist hypotheses to the limbo of last, desperate resorts, and proceed on the assumption that human and animal behavioural control systems are only very complicated denizens of the physical universe, it follows that the events within them, characterized extensionally in the terms of physics or physiology, should be susceptible to explanation and prediction without any recourse to content, meaning, or Intentionality. There should be possible some scientific story about synapses, electrical potentials and so forth that would explain, describe and predict all that goes on in the nervous system. If we had such a story we would have in one sense an extensional theory of behaviour, for all the motions (extensionally characterized) of the animal caused by the activity of the nervous system would be explicable and predictable in these extensional terms, but one thing such a story would say nothing about was what the animal was doing. This latter story can only be told in Intentional terms, but it is not a story about features of the world in addition to the features of the extensional story; it just describes what happens in a different way. Supposing one could have complete knowledge of the mechanics of a computer without the slightest inkling of the rationale of its construction, one would be in a similar situation: one would see type pressed against paper, relays open and close, and all this would be predictable and explicable in terms of physics, but one would have nothing to say in this account about the logic of the operations, about adding, subtracting and comparing, or even about operations at all.