Jeffrey A. Gray
How does the brain generate conscious experience? Twenty years ago most scientists thought this a question for the philosophers; and most philosophers regarded it as a symptom of some kind (though what kind, never became clear) of deep linguistic confusion. In sharp contrast to this picture there is now a large measure of agreement among both scientists and philosophers that, not only is there a real problem about consciousness, but that it is a scientific problem and that the time has come for the scientists to tackle it. Hardly anyone today doubts that consciousness is in some way a product of the brain, a product that is intimately connected with the brain's role in behaviour and the processing of information. Cartesian dualism -- the notion that brain-stuff and mind-stuff are essentially separate, though able to communicate with each other -- has virtually no contemporary followers. The main debate (Marsh 1993) now centres on just how to go about determining the way in which consciousness in fact relates to brain function.
On the one hand, contemporary functionalists appear to hold the view that there are no major philosophical or theoretical issues that need to be resolved. It will be sufficient merely to gather more, and more detailed, data regarding the empirical relationships that link together environmental events, brain function, conscious experience, information (computational) processing and behaviour; the resulting data set will itself then provide a sufficient account of how consciousness fits into the overall scientific picture. Dennett (in Marsh 1993, pp. 7-8), for example, argues that one can already take what is known about the way the brain codes and recodes visual stimuli, put it together with existing data from psychophysics, and use this knowledge to predict, successfully, new visual phenomena, such as illusory experiences. One can even 'inject' such experiences into the brains of experimental animals, as demonstrated, for example, in elegant experiments (Newsome & Salzman 1993) in which monkeys responded (behaviourally) to microstimulation of a circuit encoding a particular direction of motion in the same way that they had been trained to respond to an exteroceptive stimulus having the same directional value. What more than this, Dennett (in Marsh, loc. cit.) asks, do we need? "If your science can tell you under exactly what conditions a subject will hate a certain stimulus or prefer one circumstance to another, it seems to me that to say, 'but that just gives you the observable physical surroundings and circumstances of phenomenology' isn't a fair representation of the facts. As far as I can see, what's left to be inaccessible is pared down to something which one has to take on faith as making a difference even though it doesn't make any detectable difference. At that point, it seems to me, science would say, 'there's nothing left to explain'." Kinsbourne (1993, p. 43) makes the same point still more firmly: "I take as my premise a view that I call neurofunctional -- that awareness is an irreducible property of the activity of functionally entrained neuronal assemblies and therefore is amenable to no further explanation."
Some, however -- probably the majority at a recent symposium (Marsh 1993), including myself -- remain unconvinced that it is going to be so simple. On this view, more than just dedicated gathering of experimental data is going to be needed. What is needed, rather, is a new theory, one that will render the relations between brain events and conscious experiences, to use Nagel's (Marsh 1993, p. 4) felicitous term, 'transparent' (as, say, the theory of heat renders transparent the relation between the gas flame and the boiling of the kettle), rather than a series of brute correlations between environmental inputs, brain events and behavioural outputs, on the one hand, and conscious experiences on the other. This, after all, is the standard set in all other scientific domains, so why not here?
What would such a theory need to explain? I have considered this issue before (Gray 1971, p. 251) and concluded that: "We stand in need of a scientific account of how conscious experiences (in the sense which is best illustrated by the experience of primitive sensations, that is, qualia) (1) evolved and (2) confer survival value on organisms possessing them (the evolutionary questions); and of how they (3) arise out of brain events and (4) alter behaviour (the mechanistic questions)." A successful theory of consciousness would be one from which answers to these questions could be deduced from one unifying set of concepts. Consider the analogy with the dual wave and particle aspects of elementary physical particles (Marsh 1993, pp. 242-243). It is a brute fact that such a duality of aspects exists in our universe, just as it is a brute fact that there is a duality of brain events and conscious experiences. But the theory of quantum mechanics provides a single set of equations from which both the wave-like and the particle-like aspects of the behaviour of fundamental particles can be predicted and understood; whereas no such theory yet exists from which one might derive the duality of brain events and conscious experiences. Moreover, such a theory is at present unimaginable -- though only in the sense that no-one could have imagined relativity or quantum mechanics before they were invented, not because we are dealing with the unknowable or a bad language habit.
With this background in mind, we can now state the main aim of this paper. This is to propose a new and specific hypothesis concerning a limited version of question 3 distinguished above, viz: how do conscious experiences arise out of brain events? (The hypothesis is termed 'new' principally in order to distinguish it from the overall theoretical model to which, as indicated below, it is an addition; it has, of course, many antecedents, as made clear in later sections of the paper.) According to this hypothesis a specific set of brain processes, linked to a specific set of psychological (information- processing) functions, provides the basis for conscious experiences; and I shall attempt to show how the nature of these processes and functions might give rise to a number of specific features of conscious experience. In this sense, the paper takes a theoretical step beyond just the gathering of data on brute correlations between brain events and conscious experiences. But it is a step that still falls far short of Nagel's standard of transparency, as I shall show in the final section of the paper. (In general terms, this enterprise is similar to one undertaken previously by Edelman [1989]. In more particular terms, it resembles most closely the model of consciousness proposed by O'Keefe [1985]. This shares with the one proposed here allocation of central roles to the hippocampus, to the hippocampal theta rhythm, and to match/mismatch decisions. In spite of this family resemblance, however, the two models have been developed independently and differ in many details.)
The new hypothesis is stated fully in section 3, below. In brief, it proposes that the contents of consciousness consist of the outputs of a comparator system (Gray 1982a; 1982b) that has the general function of predicting, on a moment-by-moment basis, the next perceived state of the world, comparing this to the actual next perceived state of the world, and determining whether the predicted and actual states match or do not ('mismatch'). The 'contents of consciousness' in this hypothesis refer to the subjective experiences that make up what Jackendoff (1987, pp. 3-4) calls 'primary awareness', including above all the perceived world with all its various qualities, but also bodily sensations, proprioception, mental images, dreams, internal speech, hallucinations, etc. What Jackendoff (loc. cit.) calls 'reflective awareness' -- including e.g. beliefs and self- awareness -- lies outside the scope of the hypothesis. The hypothesis states, therefore, that the contents of primary awareness consist of subjective qualities whose neural and computational equivalents have been processed by the comparator system, and determined by that system to be familiar or novel. Since the comparator system has itself previously been described within the context of a detailed neuropsychological model of anxiety (Gray 1982a; 1982b), intermediate memory (Gray & Rawlins 1986; Rawlins 1985), the positive psychotic symptoms of schizophrenia (Gray et al. 1991a; 1991b; Schmajuk et al., under revision) and goal directed behaviour (Gray, in press), the new hypothesis necessarily also contains proposals concerning the specific neural mechanisms that generate the contents of consciousness. The general neuropsychological model is outlined in sections 1 and 2 below. No attempt is made to summarise the data or arguments on which it is based; for these the reader is referred to the original sources. In section 4, I consider the level of analysis occupied by the new hypothesis; and in section 5, I attempt to show how it is able to account for a range of special features of consciousness noted in previous discussions of the problem, e.g., O'Keefe (1985), Marcel & Bisiach (1988) and, most recently, Marsh (1993). Finally, in section 6, I consider the limitations in principle of this general kind of hypothesis.
In proposing any hypothesis concerning the contents of primary awareness, one encounters the problem posed by the private nature of conscious experience, in contrast to the public, intersubjectively confirmable domain of scientific data (Meehl 1966). It is not, however, the privacy of conscious experience per se that poses the problem. Scientists need have no more difficulty in principle in agreeing on observations about conscious experiences than they do in agreeing about meter readings; witness the whole of psychophysics. Furthermore, reports of such experiences can be used to provide tests of specific hypotheses concerning their nature; research on the rotation of mental images (Cooper & Shepard 1973) provides a good, if controversial (Pylyshyn 1984), example. From a scientific point of view, the problem may be put as follows (Gray 1987). One's own conscious experience is a datum that stands in need of explanation; the conscious experiences of others, however, can function only as a hypothesis by which to explain their behaviour. The reason why the problem posed by consciousness seems, at least to non-functionalists, so acute is the following. Nothing that we so far know about behaviour, physiology, the evolution of either behaviour or physiology, or the possibilities of constructing automata to carry out complex forms of behaviour, is such that the hypothesis of consciousness would arise, if it did not occur in addition as a datum in our own experience; nor, having arisen, does it provide a useful explanation of the phenomena observed in those domains. The hypothesis proposed here approaches the problem principally (but see Section 4, below) from the point of view of consciousness-as- datum; that is, it seeks to account for some of the phenomenological features of conscious experiences by postulating specific neural and psychological functions that might give rise to them. In this respect, it resembles the approach taken, on a much larger scale, by Jackendoff (1987).
Sections 1 and 2 also consider a further issue: how far does the normal scientific process of building neuropsychological models of specific psychological phenomena -- as we have done for anxiety (Gray 1982a; 1982b) and positive psychotic symptoms (Gray et al. 1991a; 1991b) -- lead ipso facto to progress in solving the general problem of consciousness? If Dennett (in Marsh 1993, pp. 7-10) is correct, the successful construction of such models should lead to sufficient, and sufficiently detailed, understanding of the relationships between brain, behaviour and conscious experience for the general problem of consciousness gradually to wither away.
1. The comparator model: anxiety
As developed so far, the comparator model encompasses three interlinked levels of analysis: behavioural, neural and cognitive (i.e., information-processing). Notice that these levels of analysis have not hitherto included an experiential component (except in a limited sense that is clarified below). Indeed, the aim of the present paper may be seen as extending the model to that, fourth, level.
The model was first developed as a theory of anxiety (Gray, 1982a; 1982b). At the behavioural level, this theory postulated a 'behavioural inhibition system' (BIS), as presented in Figure 1. The critical eliciting stimuli for activity in the BIS are
----------------------
Figure 1 about here
----------------------
conditioned stimuli associated with punishment, conditioned stimuli associated with the omission or termination of reward ("frustrative nonreward"; Amsel 1962; 1992), and novel stimuli. The behaviour elicited by these stimuli (right-hand side of Figure 1) consists in behavioural inhibition (interruption of any ongoing behaviour); an increment in the level of arousal, such that the next behaviour to occur is carried out with extra vigour and/or speed; and an increment in attention, such that more information is taken in, especially concerning novel features of the environment. Any one of the inputs to the BIS elicits all the outputs; furthermore, a range of interventions is capable of blocking all the outputs to any of the inputs, while leaving intact other input-output relationships (including some that involve inputs to or outputs from the BIS but not both). These are among a number of reasons for regarding the BIS as indeed a unified system, rather than a congeries of separate input-output relationships.
Among the interventions which specifically abolish the input-output relationships that define the BIS is the administration of drugs, such as the benzodiazepines, barbituates, and alcohol, which reduce anxiety in human beings (Gray 1977); indeed the study of such drugs was a major impetus to the formation of the concept of the BIS (Gray 1982a; 1982b). It is on this basis that I identified the subjective state that accompanies activity in the BIS as anxiety. This identification gains plausibility from the fact that it leads to a face-valid description of human anxiety: i.e., a state in which one responds to threat (stimuli associated with punishment or nonreward) or uncertainty (novelty) with the reaction, "stop, look and listen, and get ready for action" (right-hand side of Figure 1).
In this sense, then, the model did include an experiential component from its inception. However, let us suppose that the identification of activity in the BIS with the subjective state of anxiety is correct in every detail. Note, even so, that we are still left with no more than a brute correlation: the identification offers no account of how such activity gives rise to the specific subjective features of felt anxiety, nor of why activity in the BIS should give rise to any subjective features at all. Thus the attribution of subjective features to the model in this way does nothing to solve the central problem of consciousness, it merely evades it.
Neurologically, the set of structures which appear to discharge the functions of the BIS are as illustrated in Figure 2, including notably the septohippocampal system (SHS) and its associated 'Papez loop', areas of the temporal and frontal neocortex, and the ascending noradrenergic and serotonergic pathways that innervate these forebrain regions. We shall leave till later a more detailed consideration of the neural activity mediated by these structures. The proposal that the BIS, as defined behaviourally (Figure 1), consists of such activity depends upon a variety of sources of information (Gray 1982a). For the sake of the argument pursued here, let us again suppose that this proposal is correct in every detail: would this bring us any closer to an understanding of the relationship between, on the one hand, the behavioural and neurological aspects of anxiety and, on the other, its experiential features? Clearly not: we merely have another brute correlation. To see that this is so, suppose that, instead of the set of brain structures depicted in Figure 2, another set in fact turns out to mediate the functions of the BIS. Would this lead to any testable predictions about change in the subjective features of anxiety? The ability in principle to derive such predictions would take us beyond brute correlations and towards 'transparency'; but the theory that links the hypotheses depicted in Figures 1 and 2 to each other and to anxiety offers no way of deriving them.
-----------------
Figure 2 about here
-----------------
Now, the main function of the brain is to process information. Thus, faced with a neurological flow diagram, one should ask not only how the structures illustrated therein produce their relevant behavioural outputs, but also what cognitive (i.e. information processing) operations they perform in order to do so (see below, section 4, for a detailed discussion of how such information processing relates to the physiological events -- firings of neurons etc -- by which they are instantiated). The information-processing functions attributed in the model to the interlinked set of structures depicted in Figure 2 are themselves illustrated in Figure 3; detailed justification for the ideas contained in this figure can be found in Gray (1982a; 1982b). The key concept is that of the comparator, i.e. a system which, moment by moment, predicts the next likely event and compares this prediction to the actual event. The experimenal evidence upon which this concept rests is complex and taken from many sources, especially those summarised by Gray (1982a; 1982b) in relation to anxiety. Eichenbaum et al. (1994) have recently used a similar conceptual analysis in the attempt to integrate experimental evidence concerning the memorial functions of the hippocampal system.
-------------------
Figure 3 about here
-------------------
The system depicted in Figures 2 and 3 (i) takes in information describing the current state of the (perceived) world; (ii) adds to this further information concerning the subject's current motor program; (iii) makes use of information stored in memory and describing past regularities that relate stimulus events to other stimulus events (derived from the process of classical conditioning); (iv) similarly makes use of stored information describing past regularities that relate responses to subsequent stimulus events (derived from instrumental conditioning); (v) from these sources of information predicts the next expected state of the world; (vi) compares the predicted to the actual next state of the world; (vii) decides either that there is a match or that there is a mismatch between the predicted and actual states of the world; (viii) if there is a match, proceeds to run through steps (i) to (vii) again; but (ix) brings the current motor program to a halt (i.e., operates the outputs of the BIS; Figure 1) if there is a mismatch, or (x) if the predicted state of the world is one associated with punishment or nonreward; and (xi) in that case takes in further information to resolve the difficulty that has interrupted the current motor program.
Figure 3 depicts, as it were, the software of the comparator proposed by Gray (1982a); as we have seen, the corresponding hardware is as illustrated in Figure 2. At this neural level, the core structure is the SHS, composed of the septal area, entorhinal cortex, dentate gyrus, hippocampus, and subicular area. Here we note only the following points (refer to Figure 2 for the circuitry).
First, the heart of the comparator function is attributed to the subicular area. This is postulated (1) to receive elaborated descriptions of the perceptual world from the entorhinal cortex, itself the recipient of input from all cortical sensory association areas; (2) to receive predictions from, and initiate generation of the next prediction in, the Papez circuit (i.e. the circuit from the subiculum to the mammillary bodies, the anteroventral thalamus, the cingulate cortex and back to the subiculum); and (3) to interface with motor programming systems (not themselves included in Figure 2; see below) so as either to bring them to a halt or to permit them to continue.
Second, the prefrontal cortex is allotted the role of providing the comparator system with information concerning the current motor program (via its projections to the entorhinal and cingulate cortices, the latter forming part of the Papez circuit).
Third, the monoaminergic pathways which ascend from the mesencephalon to innervate the SHS (consisting of noradrenergic fibres originating in the locus coeruleus, and serotonergic fibres originating in the median raphe) are charged with alerting the whole system under conditions of threat and diverting its activities to deal with the threat; in the absence of threat, the information-processing activities of the system can be put to other, nonemotional purposes (Gray, 1984).
Lastly, the system depicted in Figure 3 needs to be quantized in time, to allow appropriate comparisons between specific perceived states of the world and corresponding predictions, followed by initiation of the next prediction and next intake of information describing the world. This function is attributed to the hippocampal theta rhythm, giving rise to an 'instant' within the model of about one-tenth of a second. (More specifically, the theta rhythm in animals such as the rat has a frequency ranging from about 6 to about 12 Hz; so an "instant" must have a duration of approximately 80-160 ms.)
In the application of this model to anxiety (Gray 1982a; 1982b), the main focus of the analysis was on steps (ix)-(xi), outlined above, and the further consequences of these steps. A detailed attempt was made to show how the combination of the three levels of analysis shown here in Figures 1-3 (henceforth collectively termed simply 'the BIS') was able to account for a variety of features of human anxiety and anxiety disorders, including many highly subjective features (e.g., the compulsive ruminations of patients with obsessive-compulsive disorder). A key feature of this application to human anxiety was the distinction between two modes of operation of the BIS. The first, 'checking', mode applies when the comparator repeatedly declares 'match' and recursively runs through steps (i)-(vii) listed above. The second, 'control', mode applies when either a predicted threat or a mismatch is detected, leading to operation of the behavioural outputs of the BIS (Figure 1). The phobic symptoms of human anxiety were attributed to activity of the BIS in control mode; the cognitive symptoms (worry, rumination, etc), in part to the operation of step (xi) in control mode, and in part to certain special features of activity in checking mode that become possible uniquely in human beings because of descending control over the BIS from the prefrontal cortex, conveying influences from language-based cortical systems.
Does the addition of the information-processing level of analysis, summarised in Figure 3, to the other two levels (Figures 1 and 2) bring us any closer to a transparent theory of the linkage between brain-and-behaviour and conscious experience? Certainly, there are many passages in Gray's (1982a) detailed account of features of human anxiety which might give the unwary reader the impression that real progress in this direction has been made (see, e.g., pp. 442-444). This impression is strengthened, in particular, by the use in the description of the software of the model (Figure 3) of terms that have a rather ambiguous status insofar as consciousness is concerned. Consider, for example, "the use of information stored in memory and describing past regularities of experience in order to predict the next state of the world" (steps iii-v), or "the comparison between this predicted state and the actual state of the world" (steps vi, vii). Are these conscious or unconscious processes? If we were to model them in a computer program (as can be done; e.g., Schmajuk et al., under revision), we would surely suppose them to lack conscious components; and it is in this way that I intended to employ these concepts. But the fact that such predictions and comparisons are familiar to us also as elements in our conscious experience allows one to slip all too readily into the assumption that, by the simple postulation of equivalence between certain aspects of neural activity (e.g., the passage of impulses around the Papez loop) and certain features of the processing of information (e.g., the making of a prediction as to the next input into the system) -- itself a controversial, but not a mysterious step in theory building (see Marsh 1993, pp. 273-275) -- one has made a contribution to the substantive problem of consciousness. But it is precisely the fact that we may leave the status -- conscious or unconscious -- of such processes undefined in the construction of this kind of information-processing model that most clearly demonstrates our lack of progress in understanding consciousness. For consciousness must make a behavioural difference, otherwise it could not have been the subject of Darwinian selection and evolution (Gray 1971). If our theories make identical predictions whether the processes they postulate are treated as conscious or not, then this critical difference that is due to consciousness has clearly been left out.
2. The comparator model: schizophrenia
More recently, Gray et al. (1991a; 1991b) have extended the comparator model and applied it to some of the bizarre cognitive features of positively symptomatic acute schizophrenia. Whereas application of the model to anxiety was concerned principally with the consequences of disruption to the monitoring of motor programs (when threat or mismatch is detected), the extended model is more concerned with the details of the monitoring process itself, and the way in which this interacts with motor programs. The running of the latter is attributed to a system separate from the BIS, the 'behavioural approach system' (BAS; Gray, in press). Like the BIS, the BAS has been submitted to three interlinked levels of behavioural, neurological and information-processing analysis.
The input-output relations that define the BAS at the behavioural level are set out in Figure 4. In essence, this
-----------------
Figure 4 about here
-----------------
depicts a simple positive feedback system, activated by stimuli associated with reward or with the termination or omission of punishment ("relieving non-punishment"; Mowrer 1960), and operating so as to increase spatiotemporal proximity to such stimuli. By adding the postulate that conditioned appetitive stimuli of this kind activate the BAS to a degree proportional in their spatiotemporal proximity to the unconditioned appetitive stimulus ('goal') with which they are associated, we have in Figure 4 a system that is in general capable of guiding the organism to the goals it needs to attain (food, water, etc.) for survival (Deutsch 1964; Gray 1975, chapter 5; Gray, in press).
At the neurological level the last decade has seen rapid progress (Groves 1983; Penney & Young 1981; Swerdlow & Koob 1987; Gray et al. 1991a) in the construction of plausible neuropsychological models of the BAS (though, in the relevant literature, this phrase has not itself been used, terms such as 'motor programming system' being preferred). The key components are the basal ganglia (the dorsal and ventral striatum, and dorsal and ventral pallidum); the dopaminergic fibres that ascend from the mesencephalon (substantia nigra and nucleus A 10 in the ventral tegmental area) to innervate the basal ganglia; thalamic nuclei closely linked to the basal ganglia; and similarly neocortical areas (motor, sensorimotor, and prefrontal cortex) closely linked to the basal ganglia. These components are best seen as forming two closely interrelated subsystems, as illustrated in Figure 5 (based on Groves 1983; Penney & Young 1981; and chiefly Swerdlow & Koob 1987). The upper part of this Figure shows the interrelations between non-limbic cortex (i.e. motor, sensorimotor, and association cortices), the caudate- putamen (or dorsalstriatum), the dorsal globus pallidus, nn. ventralis anterior (VA) and ventralis lateralis (VL) of the thalamus, and the ascending dopaminergic pathway from the substantia nigra; for the sake of brevity we shall refer to this set of structures as the 'caudate' motor system. Similarly, the lower part of Figure 5 shows the interrelations between the limbic cortex (i.e. prefrontal and cingulate cortices), n. accumbens (ventral striatum), the ventral globus pallidus, the dorsomedial (DM) thalamic nucleus, and the ascending dopaminergic projection from A 10; for brevity we shall call this set of structures the 'accumbens' motor system. Importantly, n. accumbens also receives projections from two major limbic structures: the subiculum (output station for the SHS) and the amygdala. The overall system, and its links to the BIS, are shown in Figure 6.
-------------------------
Figures 5 and 6 about here
-------------------------
At the information-processing level, the key proposal (Gray et al. 1991a) is that a particular set of neurons firing at a particular time in the basal ganglia: (1) represents a step in a goal-directed motor program; and (2) is selected for this function by instrumental reinforcement mediated by the connectivity of the neurons that make up the set (Rolls & Williams 1987). Within this overall function, the role played by the caudate subsystem is that of encoding the specific content (in terms of relationships between stimuli, responses and reinforcement) of successive steps in the program. The complementary role played by the accumbens subsystem is that of (1) switching between steps in the motor program; and (2), in interaction with the SHS, monitoring the smooth running of the motor program in terms of progress towards the intended goal. In more detail, the specific hypotheses are as follows.
1. The caudate system, by way of its connections with sensory and motor cortices, encodes the specific sensorimotor content of each step in a motor program (e.g. for a rat, turn left at a junction in a maze).
2. The accumbens system operates in tandem with the caudate system so as to permit switching from one step to the next in a program. In addition, there are outputs to exploratory behavioural systems (via the superior colliculus and mesencephalic locomotor region) that are activated in response to novel stimuli.
3. Both the establishment of the sequence of steps that makes up a given motor program, and the subsequent orderly (i.e., goal-directed) running of the program, are guided by the projection to n. accumbens from the amygdala; this projection conveys information concerning cue-reinforcement associations (Rolls & Williams 1987).
4. The septohippocampal system is responsible for checking whether the outcome of a particular motor step matches the expected outcome; this information is transmitted to n.accumbens by the projection from the subiculum. For 'match' a topographically localised signal is sent to the accumbens, permitting switching to the next step in the motor program. For 'mismatch' a generalised subicular input to accumbens interrupts all accumbens function relating to motor steps, enabling instead the activation of exploratory behaviour outputs.
5. The activities of the caudate, accumbens, and septohipppocampal systems are coordinated and kept in step with one another by the prefrontal cortex, acting by way of its interconnections, respectively, with (a) the cortical components of the caudate system, (b) n. accumbens, dorsomedial thalamus and amygdala, and (c) the entorhinal and cingulate cortices.
6. Timing is coordinated between the septohippocampal monitoring system and the basal ganglia motor programming system; given the assumption that time is quantized in the SHS by the theta rhythm (Gray, 1982a), corresponding to an 'instant' of about a tenth of a second, this must also be the duration of a step in the motor program.
The structure of this part of the overall model is obviously similar to the part outlined in section 1. There is therefore no need to repeat points made already concerning the light, if any, that the model is able in and of itself to throw upon the relationship between brain-and-behaviour and consciousness. As we have seen, the fundamental questions raised by this relationship remain opaque to such models. However, application of the model to the cognitive abnormalities of acute schizophrenia throws this opacity into particularly stark relief (Gray 1993).
The theory of the neuropsychology of schizophrenia that my colleagues and I have developed (Gray et al. 1991a; 1991b; Schmajuk et al., under revision) is intended to span the complete range of explanation from a malfunction in the brain to the psychological symptoms of the condition (Figure 7). It integrates four levels of description: (1) a structural abnormality in the brain (specifically, in the limbic forebrain, affecting the hippocampal formation, amygdala, and temporal and frontal neocortex) gives rise to (2) a functional neurochemical abnormality in the brain (specifically, relative hyperactivity of transmission in the ascending mesolimbic dopaminergic pathway), and this in turn (3) disrupts a cognitive process (specifically, the integration of past regularities of experience with current stimulus recognition, learning and action), and so produces (4) the positive symptoms characteristic of acute schizophrenic psychosis (Figure 8). Notice that, if the explicandum (step 4) in this chain were susceptible of definition in ordinary biological terms (e.g., a failure in thermoregulation), this type of integrative neuroscientific explanation is familiar and poses no theoretical or philosophical problems.
--------------------------
Figures 7 and 8 about here
--------------------------
Unlike thermoregularion, however, most positive (Crow 1980) symptoms of schizophrenia (e.g., auditory hallucinations, delusional beliefs, enhanced sensory awareness, difficulties in the focussing of attention, etc; see Freedman, 1974, for a distillation of autobiographical accounts) are necessarily linked to conscious experience. So the question arises: how do we get from the first, clearly neuroscientific, steps (1 and 2) in this chain of explanation to the fourth, which apparently belongs to a different universe of discourse? The answer lies in step 3. This utilises the same ambiguity that we noted at the end of the previous section: a 'weakening of the influence of stored memories of regularities of previous input on current perception', proposed by Hemsley (1987) to underlie the positive symptoms of schizophrenia, describes a process that can be treated either as conscious or not; and this ambiguity is deliberately left unresolved (Gray 1993). This failure to disambiguate is not due to oversight or insufficient analysis; it is a general, and probably (given current understanding) inevitable, feature of all similar contemporary attempts to appply neuroscientific concepts to the explanation of conscious phenomena.
Nor are such theories rendered useless by their inherent ambiguity. On the contrary, it has, for example, been possible to derive from our model of schizophrenia, and the earlier hypotheses (Solomon et al. 1981; Weiner et al. 1981) of which this model is an elaboration, a variety of detailed predictions, and to test them successfully in experiments with both animal and human subjects, utilising both behavioural and neuroscientific data. Much of the evidence in support of the theory derives from two key behavioural paradigms, latent inhibition and the Kamin blocking effect, both first studied in experimental animals and also demonstrable with human subjects. Here, I shall consider only latent inhibition (Lubow 1973; 1989). This is an extremely simple phenomenon. If a potential conditioned stimulus (CS) (e.g., a tone or white noise) is presented to the subject a number of times (typically, 20-40) without consequence, this preexposure retards subsequent learning when the stimulus does have consequence (e.g., it now predicts the occurrence of a second stimulus: footshock if the subject is a rat, a counter increment for human subjects). In Hemsley's (1987) terms, a past regularity (CS with no consequence) adversely affects the current learning of a new association. If, therefore, one were unable normally to integrate current learning with such past regularities, latent inhibition should be blocked, i.e., learning to a preexposed CS should resemble learning to a novel CS.
A number of experimental manipulations have been shown to influence latent inhibition in ways predicted by the theory (references in Gray et al. 1991a unless others given). Thus, the indirect dopamine agonist and psychotomimetic, amphetamine, blocks latent inhibition in both rats and human volunteers, showing in both cases an inverse dependence upon dose (N. S. Gray et al. 1992a); and this effect is reversed by dopamine receptor antagonists with anti-psychotic action. Given on their own, such anti-psychotic drugs increase latent inhibition in both animal (Dunn et al. 1993) and human (Williams et al. 1994) subjects. The model Gray et al. (1991a; 1991b) model attributes the effect of amphetamine on latent inhibition to the release of dopamine specifically in n. accumbens (Figure 8); intense dopaminergic activity in this structure is thought to interrupt motor programming and substitute exploratory behaviour (Kelly et al. 1975). This prediction is supported by several observations. First, latent inhibition is abolished also by nicotine (Joseph et al. 1993), a drug which causes dopamine release in n. accumbens but not in the caudate-putamen (Brazell et al. 1990). Second, a CS paired with footshock elicits conditioned dopamine release in n. accumbens, and this response is blocked if the CS is first preexposed, paralleling behavioural latent inhibition (Young et al. 1993). Third, amphetamine injected directly into n. accumbens, but not into the caudate-putamen, has been reported to block latent inhibition by Solomon & Staton (1982), although Killcross & Robbins (1993) failed to replicate this finding. Fourth, excitotoxic lesions of the shell of n. accumbens also blocks latent inhibition (Tai et al., in press). Fifth, destruction of the dopaminergic afferents to the n. accumbens (by local injection of the catecholamine-specific neurotoxin, 6- hydroxydopamine) has the same effect as systemic administration of dopamine-receptor antagonists, that is, this lesion potentiates latent inhibition (Peters et al., in preparation).
The Gray et al. (1991a) model also predicts that latent inhibition should be affected by manipulation of the input from the hippocampal formation, via the subiculum, to n. accumbens. This prediction too is supported by several lines of observation. First, latent inhibition is abolished after large hippocampal- system lesions and after destruction of the serotonergic innervation of the hippocampus (Cassaday et al. 1993a); and it loses its normal dependence upon context after more selective hippocampal lesions (Good & Honey 1993). Second, section of the pathway connecting the subiculum to n. accumbens also blocks latent inhibition; and this effect is reversed by administration of the dopamine-receptor antagonist, haloperidol, consistent with the hypothesis that section of the subiculo-accumbens projection affects behaviour by functionally increasing dopaminergic transmission in n. accumbens (Tarrasch et al. 1992). Third, the abolition of latent inhibition seen after large hippocampal lesions is also reversed by systemic haloperidol (Schmajuk & Christiansen, in preparation). Fourth, Cassaday et al. (1993b) were able to block latent inhibition by systemic administration of certain serotonergic agonists but not others; and Boulenguez et al. (1994) have demonstrated, after intrasubicular injection of these compounds, that those which block latent inhibition, but not those without this effect, give rise to dopamine release in n. accumbens. Fifth, work in J. N. P. Rawlins' (Yee et al., submitted) laboratory has shown that lesions in the retrohippocampal region (entorhinal cortex plus subicular area) also block latent inhibition and that this effect too is reversed by systemic haloperidol. (Rawlins' work also shows that there is a projection to n. accumbens from the entorhinal cortex as well as from the subiculum; and that, for blockade of latent inhibition, destruction of the subicular region alone is insufficient. Thus it may be necessary to reformulate the critical output station from the hippocampal formation to the accumbens system as the 'retrohippocampal' rather than the 'subicular' area. This change of detail does not affect the arguments pursued here.)
With few exceptions (Killcross & Robbins 1993), then, these observations are consistent with the hypothesis that latent inhibition is disrupted by enhanced dopaminergic transmission in n. accumbens, whether this is primary or secondary to structural damage in the hippocampal formation. Furthermore, the limited human data suggest that the same processes that underlie latent inhibition in the rat do so also in Man, although of course the specific details of the experimental procedures differ widely in the two cases. Finally, and critically, as initially reported by Baruch et al. (1988a) and replicated by N. S. Gray et al. (1992b) and M. Geyer (pers. comm.), latent inhibition is absent in the early stages of an acute schizophrenic episode. Medication with anti-psychotic drugs normalises latent inhibition over a period of several weeks, the time it also normally takes for these drugs to reduce psychotic symptoms. In unmedicated patients normalisation of latent inhibition is more drawn out, taking over a year (N. S. Gray et al., in press). What is not yet clear is the exact relationship between disrupted latent inhibition and the occurrence of positive psychotic symptoms. Whereas in the Baruch et al. (1988a) study this relationship was close, N. S. Gray et al. (1992b) selected chronic schizophrenics in whom, in spite of neuroleptic treatment, the positive-symptom score remained as high as in the acute patients; nonetheless, the acute but not the chronic patients showed loss of latent inhibition. Thus, loss of latent inhibition appears to be a more sensitive index of changes in dopaminergic transmission (as indicated also by the ability of oral amphetamine to disrupt latent inhibition in normal volunteers; N. S. Gray et al., 1992a) than are positive psychotic symptoms. This is an issue that requires further research.
It is important to note that acute schizophrenics in the preexposed condition of the latent inhibition paradigm learn the association faster than do preexposed normal controls. This eliminates the possibility that the aberrant cognitive performance of the patients is an artefact due to interference from some aspect of their illness (hearing voices in the head, insufficient motivation, unwillingness to cooperate, etc). Patient groups other than schizophrenics have not so far shown comparable changes, although data are limited to agoraphobics (Baruch 1988) and children with attention deficit disorder (Lubow & Josman 1994). However, normal individuals with high scores on questionnaire measures of schizotypy have reduced latent inhibition compared to individuals low on schizotypy (Baruch et al. 1988b; Lipp & Vaitl 1992; Lubow et al. 1992). Thus, latent inhibition is affected by both state (increased dopaminergic transmission due to drug administration in normal subjects, or during the acute psychotic period in patients) and trait (schizotypy) factors, as indeed is the occurrence of schizophrenic symptoms.
A possible inference from these experiments is that the conscious experience of the amphetamine-treated rat shares important features with that of amphetamine-treated human beings and schizophrenic patients (Gray 1993). The form of the theory these experiments have been designed to test, combined with the nature of the schizophrenic symptoms that it attempts to explain, strongly implies that the effect of amphetamine in blocking latent inhibition in human subjects should be understood as reflecting a change in conscious experience (taking the form, roughly speaking, that the drug enhances the salience in consciousness of the preexposed CS; cf the descriptions, summarised by Freedman 1974, of feelings of 'enhanced sensory awareness' in schizophrenics' autobiographical accounts of their illness). But similar effects require similar explanations, especially if they are encompassed by the same theory based upon the same evidence. It should therefore follow that, if the blockade of latent inhibition by amphetamine in human subjects reflects a change in conscious experience, so it does in the rat.
Since it is still disputed whether animals other than Man have conscious experiences of any kind, the existence of a body of data and theory from which such inferences may be drawn, even if they remain tentative, is no mean achievement. Nonetheless, a point made in connection with the similar discussion of the application of the model to anxiety (section 1) remains valid: to the extent that models of this kind rely on processes whose status -- conscious or otherwise -- is left ambiguous, to that very same extent they necessarily fail to contribute to a resolution of the substantive problem of consciousness, viz, what difference is created by some neural processes achieving consciousness while others do not?
3. A new hypothesis
Although, as we have seen, application of the comparator model to both anxiety and schizophrenia has had implications for the general study of consciousness, this was not the primary purpose for which the model was developed. In this section, in contrast, I propose a new hypothesis, based upon the comparator model, which aims explicitly to provide an account of a number of specific features of conscious experience in general. Given the extensive previous elaboration of the comparator model, outlined above and in earlier publications (especially Gray 1982a and Gray et al. 1991a), the hypothesis is (to a first approximation) simply stated: the contents of consciousness consist of the outputs of the subicular comparator. This hypothesis is not simply an extension of the application of the model to anxiety or schizophrenia, nor does it seek support in the same data as did these earlier applications. Rather, it represents a new insight (remembering, of course, that insights can be false as well as true): a range of well-known features of conscious experience appear to fall out naturally from the hypothesis. In the next section we consider what these features are. First, however, the hypothesis needs to be stated in more detail.
It will be recalled (section 1) that the subicular area is the region, within the model, in which predicted states of the world (arriving from the Papez loop) are compared to actual (perceived) states of the world (arriving from the entorhinal cortex, part of the temporal lobe). Each such comparison (occurring on the order of once every hundred milliseconds) results in a decision of 'match' or 'mismatch', the two types of decision having different consequences for the outputs distributed from the subicular area to other parts of the brain. From an informational point of view, the contents of the neural messages that are submitted to the process of comparison are each equivalent to a multi-modal and highly elaborated perceptual description (e.g., 'a red scented rose, in a glass vase in the corner of the room, rustling in a draft near an open door'). But these equivalences are not, as it were, written in the neural code carrying the information; they derive, rather, from the relevant sensory inputs, motor programs, potential behavioural outputs and links to associative memory stores that jointly give rise to the neural activity that reaches the subicular area from the neocortex and from the Papez loop.
Exactly how the process of comparison is made is unknown; but one can envisage sets of neurons wired up in such a way that they constitute 'and' or 'or' gates, etc, and which will automatically ensure that the result of the prediction initiated around the Papez loop one 'instant' (one hundred milliseconds) ago enters the appropriate set of gates in the same instant as the next current perceived state of the world arriving from the entorhinal cortex. Once the match/mismatch decision is made in the subicular comparator, this decision is sent on to other structures. A match decision is passed on as a topographically discrete message to the nucleus accumbens, where it confirms the successful outcome of the last step in the motor program and permits the input to the accumbens from the amygdala to trigger initiation of the next step in the program. A mismatch decision is passed on in two ways: as a message to the cingulate cortex, where it brings the current motor program to a halt; and as a generalised message to the nucleus accumbens, which triggers exploratory behaviour. In either case (match or mismatch), a message is transmitted back to memory stores in the temporal lobe, indicating either that a stored associative regularity has been confirmed or that memory stores need to be updated (Gray, 1982a, p. 283; Gray & Rawlins 1986).
From one point of view, the outputs of this comparison process are simple binary decisions: match or mismatch. Clearly, however, the contents of consciousness contain much more than this. Thus the proposal is that the informational equivalences (or some portion of them) that correspond to neural activity in the subicular area (and/or in the associated circuits and regions from which that region receives, and to which it sends, neural messages) are jointly instantiated into conscious experience; and that they are marked as 'familiar/expected' or 'novel/unexpected', depending upon the outcome of the comparison process. Now, the bracketed phrases in the previous sentence indicate several theoretical alternatives that present themselves at this juncture.
The first alternative is that conscious experience is tightly linked to activity in a highly restricted region of the brain corresponding to the point at which the binary match/mismatch decision is made (ex hypothesi, the subicular area). As pointed out by B. Libet (pers. comm., June 17, 1993), this proposal should predict that "destruction of the subicular area should abolish the contents of consciousness" . There is no evidence to support this prediction. (Nor, it should be added, is there evidence implicating other localised brain regions in the requisite manner. To be sure, there are regions in the mid-brain whose destruction abolishes consciousness. However, damage of this kind has much wider effects than merely eliminating the properties of subjective experience. For example, such damage also eliminates all waking behaviour, much of which appears not to depend upon conscious experience.) It is not clear, however, how fatal Libet's objection is. The subicular area, and even more the whole hippocampal system, has a complex spatial organisation within the human brain and, when it suffers damage, this is in consequence usually only partial. It is difficult to destroy these regions totally even in experimental animals. Supposing nonetheless that we succeed in doing so, we would still not know how to detect altered consciousness (as distinct from altered learning, memory etc) in animals -- precisely because we lack a transparent theory of the relationship between consciousness and brain-and-behaviour.
A second alternative would involve neural feedback from the subicular area to those regions from which the activity feeding into its match/mismatch decision has been derived, such feedback occurring at the time that each binary decision is made. However, if feedback of this kind is necessary for the entry into consciousness of activity in the regions concerned, then this proposal would appear to make the same prediction (apparently false but, as noted above, neither adequately tested nor probably as yet testable) with regard to the effects of subicular damage as the first. There is a further possible objection to this alternative. If all the informational equivalences to which subicular activity corresponds, and the inputs that give rise to them, are jointly instantiated into consciousness, then conscious experience should contain more than is found there. Recall that the subicular comparison process requires inputs from all sensory modalities, from motor programming systems, from descriptions of potential behavioural outputs, and from associative memory stores. But, as argued persuasively and in detail by Jackendoff (1987), one is typically conscious of only the perceptual components of this conglomeration, not e.g. of the processes of motor programming or of memory-guided processes of selection or semantic interpretation. Thus, one would need to suppose that only some of the systems linked to the subicular comparison process receive what one might call 'consciousness-instigating' feedback -- a qualification that would appear to weaken this second alternative.
Nonetheless, it is the second alternative that I adopt. I am prompted to do so by a further feature of consciousness, pointed out by Jackendoff (1987, p. 51): "the fact that experience is sharply differentiated by modality. There is no mistaking visual awareness, for example, for auditory awareness, or either of these for tactile awareness." It would be consistent with this feature of consciousness that at least part of the neural activity that gives rise to conscious experience should remain closely tied to the different perceptual systems themselves.
However, the "disunity of awareness", as Jackendoff (loc. cit.) calls this phenomenon, must be considered together with its complement: the fact that the various contents of consciousness, though each of irreducibly different modalities, are nonetheless somehow combined into a single whole. As Searle (1993) puts it: "we never just have, for example, a pain in the elbow, a feeling of warmth, or an experience of seeing something red, we have them all occurring simultaneously as part of one unified conscious experience."
This unity of awareness has traditionally been one of the arguments used to support the notion that conscious experience is linked to activity in one, privileged part of the brain. The prototypical hypothesis of this kind is that of Descartes, who linked mental and spiritual life to the pineal gland. Accordingly, Dennett and Kinsbourne (1992; Dennett 1991) refer, disparagingly, to such a privileged locus in the brain as the "Cartesian theatre." As part of his argument against theories that invoke a Cartesian theatre (such as the one proposed here), Kinsbourne (1993, p. 45) states that "there is no area in the brain that receives inputs from all sensory sources." However, this statement is incorrect. Just such a sensory convergence occurs onto the region of the temporal cortex that projects into the hippocampal formation and in the hippocampus itself; and single-unit recording from neurons in the latter structure demonstrate responses to a wide variety of multimodal environmental features (for review, see Gray 1982a; Eichenbaum et al. 1994). (A second part of Kinsbourne's argument against the Cartesian theatre is also incorrect: "Which representations, then, contribute to awareness? None are specially designated for this purpose. Any one is capable of doing so, should it become entrained with the dominant neuronal action pattern in the cortex" (loc. cit., p. 46). However, as noted above, and argued in detail by Jackendoff (1987), it is very largely perceptual material that enters consciousness, to the exclusion of other potential sources of material, even when these dominate behaviour, as e.g. in the case when one is executing a complex and fast- moving motor skill.)
How, then, can one accomodate both Jackendoff's point about the disunity of awareness and the equally striking phenomenal experience of unity in awareness (O'Keefe 1985, p. 90; Searle 1993)? An account capable of this reconciliation would attribute unity to activity in the hippocampal system (as discussed in more detail below), upon which all perceptual systems converge, and disunity to additional, linked activity in the perceptual systems themselves. Thus, the hypothesis can now be restated as follows: the contents of consciousness consist of activity in the subicular comparator, together with feedback from the comparator to those sets of neurons in perceptual systems which have just provided input to the comparator in respect of the current process of comparison.
Adoption of this second alternative of the 'comparator' hypothesis gives rise to two further theoretical options. Should we regard 'feedback' from the comparator to perceptual systems: (1) as merely flagging the activity present in the latter as 'expected/familiar' or 'unexpected/novel'; or (2) as contributing in a more nuanced manner to the description of the perceived world that finally enters consciousness?
An advantage of choosing option (2) is that the argument pursued here then joins hands with those that have been advanced, with much success, by such 'constructivist' theorists as Neisser (1976) and Jackendoff (1987). Indeed, the present approach in general shares much with Neisser's (1976) concept of "the perceptual cycle." As applied to the visual case, he states this concept as follows. "The cognitive structures crucial for vision are the anticipatory schemata that prepare the perceiver to accept certain kinds of information rather than others and thus control the activity of looking. Because we can see only what we know how to look for, it is these schemata (together with the information actually available) that determine what will be perceived" (loc. cit., p. 20). Neisser's contrast of schemata with "information actually available" corresponds, within the present approach, to that between predictions and descriptions of the current perceived state of the world, the former resulting from processing in the subicular comparator circuits, the latter from processing in cortical perceptual systems.
Jackendoff (1987) takes this type of analysis further, concentrating in particular on the question: what is the level at which informational structures enter consciousness? His closely-argued answer to this question takes the form of an "intermediate-level" theory of consciousness. Roughly speaking, this theory holds that the contents of consciousness reflect informational structures that are derived from a combination (within each modality of perception) of bottom-up and top-down processing. As Jackendoff shows, one is not normally (and perhaps never) aware of sensation unaffected by conceptual interpretation, nor of pure conceptual structure, but only of an admixture of the two that optimizes the fit between them. In line with the intermediate-level theory (Jackendoff 1987, p. 298), "feedback" from the subicular comparator to cortical perceptual systems will be understood here to mean that this feedback is actively used (together with inputs originating at sensory surfaces) in the construction of an optimal fit between bottom-up and top-down processing. In such a constructional process, the flagging of some parts of the current informational structures as "novel" and others as "familiar" would be expected to play a critical role in indicating the currently needed adjustments between predicted and actual perceived states of the world.
Up to this point, Jackendoff's analysis (with which the present argument now converges) has been applied to modality- specific processing. He then goes on to ask (loc. cit., p. 300): "How is it that entities detected in multiple modalities can be experienced as unified? For instance, when I look at something and handle it at the same time, how can I experience it as the same object, if my awareness is disunified into visual and haptic modalities? The answer comes from the character of processing. The visual and haptic representations that support awareness of the object are each in registration with 3D model [in Marr's 1982 sense] and conceptual structures that encode the shape, identity and category of the object. If it so happens that they are in registration with the same 3D model and conceptual structure, then the two modalities will be understood and experienced as simultaneous manifestations of the same object." Within the present theory, the conceptual structure that performs this unifying function is part of the information that circulates in the subicular comparator system, where it aids the making of predictions that are then fed back to modality-specific perceptual systems.
I conclude this section by considering how the new hypothesis relates to the known anatomy and physiology of the hippocampal system, and to previous hypotheses concerning the functions of this system.
As already noted, it is consistent with the hypothesis that (largely via the entorhinal cortex in the temporal lobe) the hippocampal formation both receives inputs from all modalities, after extensive prior processing at all levels of the cortical sensory systems, and projects back to these systems (for reviews, see O'Keefe & Nadel 1978; Gray 1982a; Eichenbaum et al. 1994). The hippocampal formation also receives afferents from the prefrontal and cingulate cortices that are capable of delivering requisite information concerning ongoing motor programs; afferents from the amygdala and from ascending monoamine systems able to deliver information concerning motivational and reinforcing events; afferents from the temporal lobe able to convey information from memory stores; and, in virtually all these cases, there are relatively direct routes by which the hippocampal formation is able to feed back to the structures that project to it (Gray 1982a). A key feature of the anatomical connections of the hippocampal formation, and the subicular region in particular, is its location on Papez' (1937) circuit: thus, efferents from the subiculum descend to the mammillary bodies and thalamus, and then re-ascend to the cingulate cortex and back to the subiculum. Finally, afferents from the prefrontal cortex constitute a likely route by which, in Man, cortical language systems can influence hippocampal processing (Gray 1982a, chapters 13 and 14).
Also consistent with the hypothesis is the fact that neurons in the hippocampus respond to a wide diversity of multimodal environmental features (for reviews, see O'Keefe & Nadel 1978; Gray 1982a; Eichenbaum et al. 1994). It is particularly striking that these firing repertoires are flexible and rapidly disposable: that is to say, individual neurons pick up on whatever regularities the experimenter chooses to put into the animal's environment, do so quickly, and can respond to one feature in one environment but to a quite different one in a second environment (for review, see Gray 1982a). These features are what one might expect of a structure with a close relationship to conscious experience. Also consistent with the hypothesis is the capacity of hippocampal neurons to respond differentially depending upon the familiarity or novelty of environmental stimuli or combinations of environmental stimuli (O'Keefe & Nadel 1978; Gray 1982a). The important role played by novelty and match/mismatch decisions in conscious experience has been stressed by many writers (Baars 1988, p. 181; Mandler 1984; for a recent experimental example, see Johnston et al. 1990).
As pointed out by Baars (loc. cit.), any analysis of novelty/familiarity requires the concept of context: "all understandable novelty exists within a relatively stable context that is not novel." A number of theories of hippocampal function have allotted to it just such a role in the analysis of context (Hirsh 1974; Winocur 1981). In O'Keefe & Nadel's (1978) theory of hippocampal function in animals, the relevant context is specifically spatial (although, in applying their theory to Man, these authors propose the more general function of "cognitive mapping"). The key role played by spatial frameworks in conscious experience is at once apparent to introspection. The experimental evidence for a hippocampal role in the analysis of context -- a role that is not confined to spatial contexts even in animals -- is abundant (for reviews, see Gray 1982a; Eichenbaum et al. 1994). In general, this evidence suggests that the hippocampus is necessary for normal associations to be formed between focal stimulus events and the context in which they occur (for a recent example, see Good & Honey 1993).
In the light of Jackendoff's (1987) intermediate-level theory of consciousness, a hippocampal function of this kind achieves particular significance. As noted above, his analysis shows that the perceptual contents of consciousness are constructs derived from an interaction between bottom-up sensory processing, on the one hand, and top-down conceptual processing, on the other, neither of which on their own enters consciousness. Both he and Baars (1988, p. 227) give as an example of this kind of interaction the well-known tip-of-the- tongue phenomenon, in which the context primes a well-defined gap which, however, lacks clearly defined conscious properties until it is filled by a word recognised as being 'correct'. Putting this kind of analysis together with the neuro- psychological argument pursued here, we may propose that Jackendoff's level of conceptual processing is identical to the contextual analysis that has been attributed to the hippocampal system (Hirsh 1974; Winocur 1981; O'Keefe & Nadel 1978); that the specification of context (and therefore, conceptual structure) forms part of the prediction of the next perceived state of the world computed by the Papez loop for entry into the subicular comparator; that this is brought together with current multimodal perceptual information in the comparator; and that, finally, the outputs of this comparison are fed back to the perceptual systems in the manner discussed above.
4. The Level of Analysis Occupied by the New Hypothesis
Note that the only events initially seen as constituting the various processes performed by the comparator system are trains of neural impulses, synaptic transmissions, and the like. The informational language that may also be used to describe these events does not necessarily imply that there are additional events beyond the neural ones; it simply constitutes an alternative, and heuristically useful, way of talking about the same neural events. (For a discussion of this issue, see Searle 1993 and Marsh 1993, pp. 162-164.) An example from a different domain may reinforce this point.
The genetic code, made up of sequences of the four DNA bases, may be treated as having both syntax (including, e.g., segmentation into triplets of bases) and semantics (the aminoacids that are specified by particular triplets, and the peptides etc. specified by higher-order combinations of aminoacids). No-one supposes, however, that strings of DNA bases consist of more than one level of physical reality -- the syntax and semantics do not constitute levels additional to the physicochemical level. But this is not to say that talk of syntax and semantics in this case is just a way of talking. On the contrary: making use of the syntax and semantics to 'read' the genetic code delivers rich insights into biological reality that purely chemical analysis cannot provide. There are several features of this example which appear to have parallels in the action of the brain.
First, the applicability of both the physicochemical and the syntactic levels of discourse to the genetic code is possible because the laws of physics and chemistry can be satisfied while still leaving open different combinatorial options (Polanyi & Prosch 1975). Thus, each correct syntactic structure is instantiated by its own specific physicochemical process, but the occurrence of that rather than another process is not fully determined only by physicochemical laws. Pylyshyn (1980; 1984) has stressed the relevance of this principle to the case of the relationship of brain function to computational processes. As he puts it: "the formal syntactic structure of particular occurrences (tokens) of symbolic expressions corresponds to real physical differences in the system, differences that affect the relevant features of the system's behavior" (Pylyshyn 1984, p. 74; and see also pp. 62- 69).
Second, the presence in the genetic code of a semantic level is due to selection by consequences. Thus, the reason that the construction of particular aminoacids and higher- order biological entities are specified by particular strings of DNA lies in the Darwinian survival value that has been conferred on the ancestors of the relevant organisms by the possession of just those entities. In the same way, it appears likely, the reason the brain generates particular sequences of neural (computational) processes lies in the ontogenetic survival value (successful adaptation to the environment) conferred on the organism by those sequences; and the semantics of such processes can be regarded as lying in the types of adaptation to which they give rise. (This principle is, of course, a major positive legacy from the Behaviourist tradition in psychology.)
Third, although there is only one level of physical reality represented by strings of DNA bases, no full account can be given of the particular strings that exist by appeal only to the laws of physics and chemistry. Such an account needs also to appeal to biological laws, in particular, those of Darwinian survival and Mendelian genetics. Both for this reason and because, as noted above, the laws of physics and chemistry leave open options for combining DNA bases, there is a real, but un-mysterious, sense in which genetics cannot be reduced to chemistry (Polanyi & Prosch 1975). The parallel case in respect of brain function can be stated particularly clearly by using the example of language. "Anything that is going on in the brain to produce syntax or semantics is part of neurophysiology; there are nerve cells doing all the things that nerve cells have to do. Then, however, one must ask: why are the nerve cells doing those things and not others? The answer to that question is not in terms of neurophysiology, it is in terms of constraints that are required for communication between individuals, because that's what speech is all about. The constraints on speech between individuals include a level of syntax, because without syntax you don't have the informational combinatorial capacities that you need for language, and a level of semantics, because without that you don't have shared referents. Neither the semantic nor the syntactic properties that are necessary are properties of neurophysiological events; they are properties of the communication system" (Gray in Marsh 1993, p. 163; and, for a more general discussion, see Pylyshyn 1984, pp. 62-69).
Up until this point, then, nothing has been said that requires the postulation of other than physicochemical events taking place in and between nerve cells; even though a full account of why these events take the specific form they do requires biological laws as well as those of physics and chemistry. The hypothesis proposed here, however, goes one stage further to consider the level of conscious experience. As noted in Section 3, this hypothesis holds that the successive contents of consciousness consist in the successive results of the match/mismatch process that takes place once per 100-millisecond instant in the subiculum, followed by feedback to the perceptual systems whose outputs have contributed to that process. It is only at this point in our theory construction that we postulate a second set of events - - conscious events -- as occurring besides, and in some way as yet unknown linked to, the neural events that constitute the subicular comparison process. Note that, as argued above, it appears possible in principle to give a perfectly good materialist account of brain events, whether considered under a physicochemical or a syntactic (computational) or semantic description (Pylyshyn 1980; 1984), without recourse to the hypothesis of consciousness. Thus, the main reason for bringing consciousness into our account is the same as Hilary's reason for climbing Mount Everest: because it is there (as datum).
In a penetrating discussion of the appropriate levels at which one should describe brain events, Pylyshyn (1980; 1984) introduces a further useful concept: that of "cognitive penetrability." This term is used to distinguish between "phenomena that can be explained functionally and those that must be explained by appeal to semantically interpreted representations"; the former are said to belong to the brain's "functional architecture", whereas the latter are "cognitive processes" sensu stricto (Pylyshyn 1984, p. 130).
The hallmark of "cognitively penetrable" -- commonly also described as "intentional" (Searle 1983) -- processes is that "the relation between environmental events and behavior can be radically, yet systematically, varied by a wide range of conditions that need have no more in common than that they provide certain information" (Pylyshyn 1980, p. 120). Another way of putting this distinction is to contrast a representational system for which the semantically interpreted content of representations plays a causal role in the unfolding of events, with one for which such representational content does not play a causal role. Clearly, while it seems legitimate to interpret portion X of the genetic code as representing the particular protein, A, for which it codes, and while it is also legitimate to suppose that the value to organisms of possessing protein A has played a causal role in determining the existence in the genome of portion X of the genetic code, it makes no sense at all to suppose that a semantically interpreted representation by X of A plays a causal role in the production by the cell of the protein. It would seem, therefore, that the existence of cognitively penetrable processes (which can hardly be in doubt, at least for our own species; Pylyshyn 1984) requires a level of analysis that goes beyond those considered earlier in this section.
The question therefore arises: is this further level identical to the level at which conscious events enter the system? Pylyshyn's (1984, p. 265) answer to this question is explicit: the distinction between functional architecture and cognitively penetrable processes "cuts across the conscious- unconscious distinction." However, he offers little support for this assertion.
Contrary to Pylyshyn's view, it would seem parsimonious to suppose that, beyond the level of neural processes (capable of analysis in neurophysiological, syntactic and semantic terms, as indicated above, but nonetheless consisting of only one set of events), there is (at most) only one other level that need be taken into account. We know that the level of conscious events exists (while leaving open the question whether these will eventually turn out to be identical to the neural events with which they are associated, as supposed by mind-brain identity theorists; Borst 1970; Gray 1971). Thus, on grounds of parsimony we should, so long as we can, suppose that the level of intentionality (cognitive penetrability) is the same as that of conscious experience.
This assumption is aided, in the case of the argument pursued here, by the fact that the contents of consciousness postulated by the new hypothesis have exactly the kinds of property required by an intentional analysis. That is to say, they consist of multimodal perceptual descriptions derived by comparison between predicted and actual perceived states of the world. Thus, they are constructs whose relationship to states in the real environment is indirect, and therefore capable of referential slippage; although, of course, such constructs must in general bear a reasonably close relationship to states in the real environment, since they derive from a process of selection by consequences in the manner outlined above. Indeed, one way of considering the new hypothesis is as providing an explanation for the existence of intentionality (a suggestion made also by Emrich 1992). If that (tentative) claim were to turn out to be correct, the theory developed here could be seen both as constituting a possible account of some of the features of consciousness-as- datum, and as proposing a hypothesis about the nature of conscious events able to explain certain features of behaviour (viz, those grouped by Pylyshyn under the rubric of 'cognitive penetrability').
5. Testing the hypothesis against features of conscious experience
This section considers a number of features of conscious experience which appear to find a natural explanation in the light of the new hypothesis; and some other features which do not fit the hypothesis so well.
5.1 Conscious experience is closely linked to current action plans. This feature of consciousness flows naturally from the model: the current motor program forms part of the data used to construct the next prediction; and the output of the match/mismatch decision is fed back to the motor program either to permit continuation or to bring it to a halt (Gray et al. 1991a; Gray, in press).
5.2 One is conscious of the outputs of motor programs, not of the program itself. It is a fact of common experience that we are not normally aware of either the planning or the execution of movements as such, but only of the end-points (which may include kinaesthetic and proprioceptive feedback making up the 'feel' of the movement) that constitute the successive sub-goals of these movements (cf. Lashley 1956; O'Keefe 1985, p. 69; Jackendoff 1987, p. 45). Consider, for example, running towards and kicking a football; or articulating and speaking a sentence. As Velmans (in Marsh 1993, p. 121) points out, we are aware even of what we are saying only after we have said it; and the sub-vocal speech that constitutes much of the process we call 'thinking' consists (for me, at least) of the hearing of words -- i.e., outputs, clearly, of a linguistic motor program -- in my head. Since the postulated comparator is designed precisely to compare actual with expected outputs of motor programs, it follows naturally from the hypothesis that such outputs constitute the contents of consciousness.
5.3 Consciousness occurs too late to affect the outcomes of the processes to which it is apparently linked. Velmans (1991) has reviewed a range of human information processes that are normally accompanied by conscious awareness (the analysis and selection of stimuli, learning and memory, and the production of voluntary responses, including those requiring planning and creativity). He has marshalled evidence that, in each of these cases, the relevant conscious events follow the information processes to which they are related. Libet's (1985; 1993; Libet et al. 1991) well-known experiments, on the delays between (i) neural events in certain brain regions and (ii) the transitions to conscious awareness of either sensation or volition related to these neural events, give rise to the same conclusion. Such a conclusion is to be expected if a monitoring system, of the kind proposed here, forms the basis of the contents of consciousness. However, there is a discrepancy between the duration of the delay suggested by Libet's experiments (about 500 ms) and the delay that would be expected from the comparator model: given that time is quantised into units about 100 ms long, the average delay should be about 50 ms. A possible account of such extra-long durations would postulate an additional delay prior to the initiation of the comparison process, linked perhaps to the unusual nature of Libet's experimental stimuli.
5.4 What about pain? The arguments of 5.2 and 5.3 treat consciousness as a monitoring process (see also Weiskrantz 1988). But Humphrey (in Marsh 1993, p. 166) has objected, "what has a stab of pain got to do with monitoring?" A possible answer to this objection is that a system which is concerned with predicting what should happen next, and keeping check that motor programs are proceeding 'according to plan', is one that must also have an interrupt mechanism for times when things are not going according to plan. Pain can perhaps be construed as just such an interrupt mechanism -- and a particularly powerful one. It shares with other features of consciousness that it occurs too late actually to affect responses to the noxious stimulus that gives rise to it (the hand is withdrawn from the flame before the pain is felt). Thus the information it provides must, like other forms of mismatch, gain such functional utility as it possesses from effects on future repetitions of the same motor program. In this sense, therefore, pain is a form of monitoring: it is telling you that the motor program you have just exercised is faulty and needs modification.
5.5 Conscious events occur on a time-scale that is about two orders of magnitude slower than that of the neural events that underlie them. The expected duration of a content of consciousness on the comparator model is the duration of a quantised time unit, i.e. about 100 milliseconds. This time unit arose from considerations of the neuronal circuitry proposed to mediate the comparator function, and especially the duration of a wave of the hippocampal theta rhythm (Gray 1982a; 1982b); but it is about the right order of magnitude for the duration of conscious events (for a summary, see Baars 1988, pp. 96-97). However, the comparator model also supposes that time is indeed quantised into these units, in order to permit comparison between corresponding neuronal descriptions of actual and expected events. Conscious experience, in contrast, does not appear to come in separate chunks in this way.
5.6 Relative to real time, apparent time is 'smeared' in conscious experience. The discussion engendered by Libet's (1985; 1993) experiments has clarified the difference between the real-time sequencing of events and the sequencing of events in experienced time (Dennett & Kinsbourne 1992; Dennett 1991; Nagel in Marsh 1993, pp. 144-145). The hypothesis proposed here implies that what enters consciousness will reflect the prediction made by the comparator circuitry as to the next likely perceived event, and whether that prediction is sufficiently well verified or not by the actual perceived event. It follows, therefore, from the hypothesis that experienced time will be based upon these predictions and their contents, and not directly upon the sequencing of events in real time. It is also clear that the precision with which experienced events can be dated with respect to their relative time of occurrence is much coarser (Nagel in Marsh 1993, loc. cit.) than the precision that might be expected from the time- scale of occurrence of the neuronal events that underlie consciousness. The quantisation of time into 100-millisecond instants by the comparator model implies just this result: that events occurring within a single instant have no definite time sequence with respect to each other; and that any apparent time sequence within such an instant is determined by the informational content of the inputs to the comparator which determine the contents of consciousness, rather than by the objective timing of events occurring at sensory surfaces.
5.7 Conscious experience has a spatio-temporal unity that is quite unlike the patterns of neuronal events that underlie it. There are two distinct, though probably related issues raised by this contrast. One -- the so-called 'binding problem' -- lies at the level of neuronal events. These are distributed widely in both time (on the neuronal millisecond scale) and brain location: what holds together the set that makes up a particular content of consciousness? The other concerns the perceptual features that drive the firing of subsets of neurons in particular brain locations. These features each only reflect aspects (e.g., in the visual domain, colour or motion; Zeki 1978) of the fully experienced percept: what holds the separate perceptual features together? A possible answer to both questions is suggested by the comparator model, namely: that the temporally and spatially distributed neuronal events that code the separate perceptual features contributing to an overall perceptual experience are themselves non-conscious; that they combine to cause the entry of a particular prediction into the comparator circuits; that this prediction includes the specification of spatial and temporal coordinates referred to the outside world; and that it is this prediction which constitutes the basis of the eventual conscious experience. If this line of argument is correct, the binding problem becomes more tractable, since the relationship between neuronal events and conscious experience is determined by the conjoint inputs into just one brain region, i.e., the subicular area. The outputs from this region, redistributed (see Section 3) to the sites of origin of its own inputs, then activate the detailed features that constitute the contents of consciousness.
5.8 Conscious experience consists of a series of constructions based upon a model of some aspect of the world rather than direct reflections of inputs from the world (Oatley 1988). Dennett & Kinsbourne (1992; Dennett 1991; Kinsbourne 1993) have recently discussed a number of perceptual phenomena which provide dramatic evidence for the essentially constructive nature of consciousness. This feature of conscious experience is clearly built into the comparator model, since each content of consciousness depends upon the predicted state of the world that enters the comparator (see Section 3). Matching is conducted upon a limited set of attributes, and only these can enter the contents of consciousness. Furthermore, 'match' decisions, clearly, can be incorrect, in terms of what is actually happening in the world, even if sensory systems have correctly described the perceptual world; but it is these incorrect 'match' decisions that will constitute the contents of consciousness.
5.9 There are 'multiple drafts' of conscious experience. Dennett & Kinsbourne (1992; Dennett 1991; Kinsbourne 1993) have marshalled data (e.g., from Geldard & Sherrick's 1972 'cutaneous rabbit' experiment) indicating that initial perceptual input may give rise to alternative conscious experiences depending upon later perceptual input. In principle, phenomena of this kind are compatible with the comparator model, since it could be argued that the potential multiple drafts compete with each other non-consciously until the victor enters the comparator circuits (i.e., provides a prediction as to the next expected state of the world), thereby constituting a basis for eventual conscious experience. However, as in Libet's experiments (see 5.6), the time-scale of the cutaneous rabbit is too slow for the comparator model: determination of the contents of consciousness can take up to a second or so (Dennett & Kinsbourne 1992, p. 186) rather than the tenth of a second suggested as the maximum by the comparator model. Dennett & Kinsbourne (1992) have interpreted such data as showing the impossibility of a 'Cartesian theatre', that is, a single, unified (even if ramified, as discussed in 5.7 and Section 3) locus in the brain at which neuronal activity becomes conscious. In distinction to their position, the comparator model preserves the notion of a Cartesian theatre, consisting in the comparator circuitry. Thus, the phenomena cited by Dennett & Kinsbourne as supporting their position, and especially the time-scale over which these phenomena operate, constitute an important objection to the comparator model. The account of Libet's experimental results (in terms of a delay prior to the onset of the comparison process), tentatively proposed at the end of Section 5.3, cannot apply in this case, since the critical interval is interposed between different stimuli that give rise to the conscious experience.
5.10 Consciousness is highly selective. This well-known aspect of conscious experience flows directly from the comparator model. This specifies (Gray 1982a, 1982b) two routes to selectivity. First, inputs to the subicular comparator from sensory systems are determined by selective processes guided by the content of the last output from the comparator and the currently active prediction; in this way, continuity is provided to link successive events related, e.g., by a common motor program. Second, any event that has a previous association with punishment or nonreward or any novel event is given priority for processing over others; this selectivity is accomplished within the hippocampal formation, a critical role being played by the monoaminergic afferents to this structure from the locus coeruleus and raphe nuclei.
5.11 Neuronal events operate in parallel; consciousness operates serially. This feature of consciousness too flows directly from the model, which supposes a succession of match/mismatch decisions by the comparator.
5.12 Conscious experience is closely linked to memory, both in that current experience can be distorted by past experience and in that episodic memory is a major criterion by which we infer that a conscious experience has indeed occurred. Again, this feature of consciousness flows directly from the model. The formation of a prediction is explicitly based upon a combination of current input and stored regularities involving similar stimuli and/or responses; and the outcome of the current match/mismatch decision is used to update memory stores (Gray 1982a; 1982b; Gray & Rawlins 1986).
5.13 In cases of sensory neglect not only is there lack of consciousness of the neglected stimuli, there is also no awareness that anything is missing (Jeannerod 1987). As pointed out by Kinsbourne (in Marsh 1993, p. 258), this can be explained if consciousness is based upon a comparator system. Suppose that the damage to the brain underlying the neglect has removed, not only the capacity to detect the neglected input, but also the capacity to set up the prediction that such an input is to be expected: no prediction, no mismatch, and so no awareness of anything missing.
5.14 Novelty and familiarity. The comparator model distinguishes between match and mismatch decisions, corresponding (if the decisions are correct) to novel or familiar events. Does this distinction imply a similar distinction between two modes of experience? It is clear that novel events have privileged access to conscious experience (see 5.10). But it is equally clear that these are not the only events that have access to consciousness.
Consider, for example, the case of listening to a familiar piece of music: without conscious experience, this would be a rather pointless exercise. This is a particularly interesting example: my introspections reveal a definite predictive process, in which the next expected note is often generated in my head just before it enters again through my ears. There also appears to be a connection between the occurrence of this kind of predictive process and the ability to remember the music. The first time one hears a new piece of sufficient complexity it is difficult either to grasp its musical structure or to recall it. With repetitition, it becomes both more comprehensible during listening and more easily recalled. Both these changes appear to be related to the increasing capacity to predict the next event in the musical sequence. Completely unpredicted events, in contrast, while they at once gain access to consciousness, are often difficult to recall (cf Baars 1988, p. 181). This relationship between familiarity and memorability is to be expected on the comparator model: a 'match' decision automatically provides a list of attributes on which the matching has been made and which can now be confirmed in memory stores.
Thus both match and mismatch outcomes of the comparator can enter conscious experience (each perhaps, as noted above, with slightly different properties); and mismatch outcomes appear always do so. A difficulty arises, however, when we ask the question: do all match outcomes register in consciousness? Consider a well-practiced motor program (the usual example is driving while simultaneously carrying on a conversation; e.g., O'Keefe 1985). The comparator system must continue to operate during the execution of such a program, since any mismatch (e.g., the front offside wheel crossing over the middle of the road) is rapidly detected. Yet everyone is familiar with the sudden realisation that one has just driven several miles, apparently without having paid any conscious attention to the process of driving. Schneider and Shiffrin (1977) coined the term 'automatic processing' for this kind of capacity, suggesting that it arises as a result of prolonged practice on a task. The present hypothesis has no principled way of accounting for the difference between such apparently unconscious 'match' outcomes and the conscious variety that occur in the music example.
5.15 The contents of consciousness are multimodal (cf. O'Keefe 1985, p. 90). This follows directly from the multimodal nature of the sensory information that reaches the hippocampal formation from the neocortex via the temporal lobe (see the discussion of the unity of awareness in Section 3).
5.16 What about simple single stimuli? "The idea that conscious contents arise only in connection with match- mismatch decisions of a comparator seems difficult to apply to simple cases of awareness, for example that of a light touch at a finger" (B. Libet, pers. comm., June 17 1993). This objection can perhaps be met by the proposal that the onset of any stimulus event immediately gives rise to a predictive process, based upon the most relevant previous experiences of similar stimulus events, as to the continuation (with or without change) of that same event. We have recently found inclusion of a postulate along these lines to be of value in formulating a mathematical model (Schmajuk et al., under revision) of latent inhibition, a phenomenon that is central to application of the general neuropsychological model to schizophrenia (see Section 2).
5.17 Anxiety and schizophrenia. Since the new hypothesis is an extension of previous models of anxiety and schizophrenia (as outlined in Sections 1 and 2), it cannot directly be counted in its favour that it is applicable to these conditions. It is nonetheless worth indicating just how certain features of the subjectively experienced symptoms of anxiety and psychosis fit with the hypothesis.
A well-known feature of anxiety is that anxiogenic stimuli dominate conscious experience to the exclusion of other experiences, giving rise e.g. to worry, rumination and failure to concentrate on other matters in hand. This feature of anxiety follows naturally from the priority given to threat stimuli within the comparator system (Gray 1982a; Gray 1982b; Section 1, above).
The model of schizophrenia (Gray et al. 1991a; 1991b) within which the new hypothesis is embedded treats positive psychotic symptoms as arising because stimuli that ought, if processing were proceeding normally, to be treated as 'expected/familiar' are in fact treated as 'unexpected/novel'. This argument leads to a natural account of the way in which apparently trivial stimuli are able to force themselves upon the awareness of the schizophrenic (Hemsley 1987, 1993). Disruption in the Kamin blocking effect, which has many of the same credentials as a model of positive psychotic symptoms as does disrupted latent inhibition (Jones et al., 1992), provides an equally natural account of the manner in which schizophrenics form spurious delusional associations (Hemsley, 1993), although not of the specific contents of such delusions. Frith (1987; Frith & Done 1989) has proposed a related hypothesis, according to which the neural circuitry normally responsible for tagging actions as 'willed' (in our model, the link from prefrontal cortex to entorhinal cortex) is faulty in schizophrenia, giving rise to a variety of symptoms (including verbal auditory hallucinations) in which the patient experiences his own acts as alien. A further interesting possibility emerges from the new hypothesis, as formulated here. Matussek (1952, p. 92) describes a patient who was aware of: "a lack of continuity of his perceptions in both space and over time. He saw the environment only in fragments. There was no appreciation of the whole. He saw only details against a meaningless background." This is the kind of phenomenology that one might expect from a breakdown in the capacity of the comparator system (making use of contextually derived conceptual structure) to hold in register with each other (Jackendoff 1987, p. 300; Section 3 above) the different inputs that arrive there simultaneously.
There are also a number of more general observations in psychopathology that support a role for temporal-lobe structures in subjective experience, including e.g. links between abnormal temporal-lobe function and epileptic 'auras', between lesions and disruption of episodic memory, and between electrical stimulation and forced remembering (Lishman 1987). Detailed consideration of these phenomena, however, would take us too far afield.
The arguments considered in 5.1-5.17 all apply the new hypothesis to existing data or to generally available introspective evidence. The hypothesis would, of course, be much more valuable if it led to new predictions open to experimental test. The fact that I am unable to generate such predictions may indicate a basic weakness in the hypothesis relative to other hypotheses that are possible in the current state of knowledge; or it may indicate a general difficulty while we lack a more general, transparent theory of consciousness (Nagel in Marsh 1993, p. 4).
6. Limitations of the hypothesis
In this final section I wish to consider what kind of advance the new hypothesis would represent, assuming that it turns out to be essentially correct. Even though this eventuality has a low probability, the exercise is worthwhile, I believe, since it can be used to consider anew the features that a truly successful theory of consciousness will need to possess. To re-iterate material in the Introduction, such a theory would need to explain: (1) how subjective experiences evolved; (2) how they confer survival value on organisms possessing them (the evolutionary questions); (3) how they arise out of brain events; and (4) how they alter behaviour (the mechanistic questions) (Gray 1971, p. 251).
From the biological point of view, it is the last of these questions which is in many ways crucial. If we knew how consciousness alters the behaviour that it accompanies, we would be able to see what survival value (question 2) consciousness confers, and so how it might have evolved (question 1). Unfortunately, our hypothesis has little to offer in this respect. The fault lies, however, more in the data than in the hypothesis itself. As noted in section 5.3, these suggest that most important psychological functions are completed before consciousness has time to have anything to do about them (Velmans 1991). This feature of the data is nicely accomodated by a hypothesis that links the contents of consciousness to the outcomes of a monitoring process. But the fit between hypothesis and data still leaves a void: if consciousness is a product of Darwinian evolution, it must confer survival value and therefore it must affect behaviour.
Rather than abandon this biological perspective, we would surely wish to continue to search for such a behavioural function. One suggestion arises from the original aim of the comparator hypothesis, which was to provide a neuro- psychological account of anxiety (Gray 1982a; 1982b). From that point of view, the role of the comparator was seen as the identification of motor programs as faulty (in that they lead to unpredicted events, to punishment or to the failure to obtain reward), followed by a search for an alternative, more successful program. If consciousness is tightly linked to the operation of the comparator, as our hypothesis suggests, then perhaps it is here that we should seek for its function: not in transactions with the environment as they actually happen, but in the modification of such transactions for future use. To be sure, this proposal flies in the face of a stubborn intuition that consciousness is to do with the here and now. It is difficult, for example, to suppose that the only handicap associated with the lack of full visual experience in Weiskrantz's (1986) 'blindsight' cases is a failure to adapt to future possibilities of visual input into the scotoma. Nonetheless, stubborn intuitions have proved to be wrong before; so this possibility is perhaps worth following up. An alternative approach to the evolutionary questions (1 and 2, above) would be to start on a taxonomic quest, by classifying, for example, species in terms of the development of the particular neuronal machinery (hippocampus, subiculum etc) proposed by the hypothesis as underlying the formation of the contents of consciousness; and, in parallel, in terms of the behaviour patterns that depend upon the integrity of these structures. Such an exercise might in turn suggest possible answers to the evolutionary questions. But note that absence of the relevant neuronal structures could not be taken to indicate absence of consciousness (since, in other cases,it is known that quite different structures can achieve the same functions; compare, e.g., the mammalian retina with insects' compound eyes).
In these ways, then, while the hypothesis does not fare at all well by the standards of questions 1, 2 and 4, it may nonetheless have some limited heuristic value. But the question to which it was primarily addressed was the third: how does subjective experience arise out of brain events? How does it fare by this standard?
The hypothesis does, I think, go beyond brute correlation: that is, beyond the mere statement that conscious experience is related to brain events of such-and-such a kind in such-and-such a place. It does so by proposing ways in which specific features of the relevant brain events give rise to specific features of the contents of consciousness. Note, however, that the critical features of the proposed brain events from which the relevant derivations are drawn are only secondarily neuronal features (impulses travelling into and out from the subiculum, for example). The critical features, rather, are specified in terms of information-processing activities: prediction, comparison, match-mismatch decisions, etc. It is from these notions that it is possible to offer an account of the fact, for example, that one is conscious of the outcomes of motor programs, not of the motor program itself (section 5.2). Thus it is a neuropsychological hypothesis that is proposed, not a purely neural one. In this way, therefore, the hypothesis is perhaps just as congenial to those, such as Dennett (1992), who seek to explain (or explain away) consciousness in functionalist (information-processing) terms as to those, such as Searle (1980; 1993), who expect that we shall one day be able to point to specific features of the physical activity of the brain as giving rise to consciousness.
It is just this neutrality between these two points of view which exposes the limitations of the hypothesis. This can be seen as setting up a three-way set of equivalences: activity in a particular neural circuit (hippocampus, subiculum, etc) = information-processing of a particular kind (the comparator function) = generation of the contents of consciousness. But the hypothesis has nothing to say about the following questions. (i) Suppose we changed the circuitry while retaining the information processing: would the contents of consciousness remain the same? (ii) Suppose we changed the information processes performed by the circuitry: would the same neuronal activity still generate conscious contents? (iii) Suppose the answers to questions (i) and (ii) are that one must preserve both the neural machinery and the information-processing functions in order to generate conscious experience: why does this particular combination produce any kind of conscious experience (as distinct from the particular contents of consciousness, for which the hypothesis does begin to offer a genuine account) rather than none?
By normal scientific standards, then, it would appear that we are still a long way from having a transparent theory (Nagel in Marsh 1993, p. 4), i.e. one able to predict the occurrence of conscious events, given relevant facts about behaviour and/or brain events (whether the latter are conceived in neuronal terms, in information-processing terms, or both). Nor does it appear likely that such a theory will emerge simply from the gathering of new facts of the same kind (useful as this exercise is likely to be for the eventual construction of a successful theory). The new hypothesis proposed here does, however, take us a small step forward, in that it attempts to account for the form taken by specific features of the contents of consciousness. How can we go beyond this? If it is correct (Marsh 1993, pp. 77-78) that we are waiting for a new kind of theory, then it is likely to be as difficult to answer this question now as it would have been to think about hunting for bosons in 1900.
Amsel, A. (1962). Frustrative nonreward in partial reinforcement and discrimination learning: some recent history and a theoretical extension. Psychological Review 69:306-28.
Amsel, A. (1992). Frustration theory. Cambridge University Press.
Baars, B.J. (1988). A Cognitive Theory of Consciousness. Cambridge University Press.
Baruch, I., Hemsley, D.R. & Gray, J.A. (1988). Differential performance of acute and chronic schizophrenics in a latent inhibition task. Journal of Nervous and Mental Disease 176:598-606.
Borst, C.V. (ed.) (1970). The mind-body identity theory. St Martin.
Boulenguez, P, Joseph, M H, Gray, J A & Mitchell, S N (1994). Dopamine release in the nucleus accumbens after systemic and intrahippocampal injections of 5HT-1 agonists which differentially affect attention in the rat. British Journal of Pharmacology, 111, Suppl, 153P.
Brazell, M.P., Mitchell, S.N., Joseph, M.H. & Gray, J.A. (1990). Acue administration of nicotine increases the in vivo extracellular levels of dopamine, 3,4-dihydroxyphenylacetic acid and ascorbic acid preferentially in the nucleus accumbens of the rats: comparison with caudate-putamen. Neuropharmacology 29:1177-1185.
Cassaday, H.J., Mitchell, S.N., Williams, J.H. & Gray, J.A. (1993a). 5,7-Dihydroxytryptamine lesions in the fornix- fimbria attenuate latent inhibition. Behavioral & Neural Biology 59: 194-207.
Cassaday, H.J., Hodges, H. & Gray, J.A. (1993b). The effects of ritanserin, RU 24969 and 8-OH-DPAT on latent inhibition in the rat. Journal of Psychopharmacology 7: 63-71.
Cooper, L.A. & Shepard, R.N. (1973). Chronometric studies of the rotation of mental images. In: Visual information processing ed. W.G. Chase. Academic Press.
Crow, T.J. (1980). Positive and negative schizophrenic symptoms and the role of dopamine. British Journal of Psychiatry 137:383-86.
Dennett, D.C. (1991). Consciousness explained. Cambridge University Press.
Dennett, D.C. & Kinsbourne, M. (1992). Time and the observer: the where and when of consciousness in the brain. Behavioral and Brain Sciences 15:183-247.
Deutsch, J.A. (1964). The structural basis of behaviour. Cambridge University Press.
Dunn, L.A., Atwater, G.E. & Kilts, C.D. (1993). Effects of antipsychotic drugs on latent inhibition: sensitivity and specificity of an animal behavioural model of clinical drug action. Psychopharmacology 112: 315-323.
Edelman, G.M. (1989). The remembered present. Basic Books.
Eichenbaum, H., Otto, T. & Cohen, N.J. (1994). Two functional components of the hippocampal memory system. Behavioral and Brain Sciences in press.
Emrich, H.M. (1992). Subjectivity, error correction capacity and the pathogenesis of delusions of reference. In: Phenomenology Language & Schizophrenia, eds. M. Spitzer, F.A. Uehlein, M.A. Schwartz & C. Mundt. Springer Verlag.
Freedman, B.J. (1974). The subjective experience of perceptual and cognitive disturbance in schizophrenia. Archives of General Psychiatry 30:333-40.
Frith, C.D. (1987). The positive and negative symptoms of schizophrenia reflect impairments in the perception and initiation of action. Psychological Medicine 17:631-48.
Frith, C.D. & Done, D.J. (1989). Experiences of alien control in schizophrenia reflect a disorder in the central monitoring of action. Psychological Medicine 19: 359-363.
Geldard, F.A. & Sherrick, C.E. (1972). "'The Cutaneous Rabbit': A perceptual illusion." Science 178:178-179.
Good, M. & Honey, R. (1993). Selective hippocampus lesions abolish contextual specificity of latent inhibition and conditioning. Behavioural Neuroscience 107: 23-33.
Gray, J.A. (1971). The mind-brain identity theory as a scientific hypothesis. Philosophical Quarterly 21: 247-252.
Gray, J.A. (1975). Elements of a two-process theory of learning. Academic Press.
Gray, J.A. (1977). Drug effects on fear and frustration: possible limbic site of action of minor tranquillizers. In: Handbook of Psychopharmacology vol. 8, ed. L.L. Iversen, S.D. Iversen & S.H. Snyder. Plenum Press.
Gray, J.A. (1982a). The neuropsychology of anxiety: An enquiry into the functions of the septo-hippocampal system. Oxford University Press.
Gray, J.A. (1982b). Precis of 'The neuropsychology of anxiety: An enquiry into the functions of the septo- hippocampal system.' Behavioral and Brain Sciences 5:469-484.
Gray, J.A. (1984). The hippocampus as an interface between cognition and emotion. In: Animal cognition ed. H.L. Roitblat, T.G. Bever & H.S. Terrace. Erlbaum.
Gray, J.A. (1987). The mind-brain identity theory as a scientific hypothesis: a second look. In: Mindwaves ed. C. Blakemore & S. Greenfield. Blackwell.
Gray, J.A. (1993). Consciousness, schizophrenia and scientific theory. Experimental and Theoretical Studies of Consciousness ed. J. Marsh. Ciba Foundation Symposium 174. John Wiley & Sons.
Gray, J.A. (in press). The neuropsychology of anxiety and schizophrenia. In: The Cognitive Neurosciences ed. M.J. Gazzaniga. MIT Press.
Gray, J.A., Feldon, J., Rawlins, J.N.P., Hemsley, D.R. & Smith, A.D. (1991a). The neuropsychology of schizophrenia. Behavioral and Brain Sciences 14:1-20.
Gray, J.A., Hemsley, D.R., Feldon, J., Gray, N.S. & Rawlins, J.N.P. (1991b). Schiz bits: misses, mysteries and hits. Behavioral and Brain Sciences 14:56-84.
Gray, J.A. & Rawlins, J.N.P. (1986). Comparator and buffer memory: an attempt to integrate two models of hippocampal function. In: The hippocampus vol. 4. ed. R.L. Isaacson & K.H. Pribram. Plenum Press.
Gray, N.S., Hemsley, D.R. & Gray, J.A. (1992a). Abolition of latent inhibition in acute, but not chronic, schizophrenics. Neurology, Psychiatry and Brain Research 1:83-89.
Gray, N.S., Pickering, A.D., Hemsley, D.R., Dawling S. & Gray, J.A. (1992b). Abolition of latent inhibition by a single 5 mg dose of amphetamine in man. Psychopharmacology 107:425-30.
Gray, N.S., Pilowsky, L.S., Gray, J.A. & Kerwin, R.W. (in press). Latent inhibition in drug naive schizophrenics: relationship to duration of illness and dopamine D2 binding using SPET. Schizophrenia Research.
Groves, P.M. (1983). A theory of the functional organization of the neostriatum and the neostriatal control of voluntary movement. Brain Research Reviews 5:109-32.
Hemsley, D.R. (1987). An experimental psychological model for schizophrenia. In: Search for the causes of schizophrenia ed H.Hafner, W.F. Gattaz & W. Janzavik. Springer Verlag.
Hemsley, D R (1993). A simple (or simplistic?) cognitive model for schizophrenia. Behaviour Research and Therapy, 31, 633-645.
Hirsh, R. (1974). The hippocampus and contextual retrieval of informaton from memory: a theory. Behavioural Biology 12: 421-444.
Jackendoff, R. (1987). Consciousness and the Computational Mind. MIT Press.
Jeannerod, M. (ed.) (1987). Neurophysiological and neuropsychological aspects of spatial neglect. Amsterdam: North Holland.
Johnston, W.A., Hawley, K.J., Plewe, S.H., Elliott, J.M.G. & De Witt, M.J. (1990). Attention capture by novel stimuli. Journal of Experimental Psychology, 119: 397-411.
Jones, S.H., Gray, J.A. & Hemsley, D.R. (1992). Loss of the Kamin blocking effect in acute but not chronic schizophrenics. Biological Psychiatry 32:739-55.
Kelly, P.K., Seviour, P.W. & Iversen, S.D. (1975). Amphetamine and apomorphine resonses in the rat following 6- OHDA lesions of the nucleus accumbens septi and corpus striatum. Brain Research 94: 507-522.
Killcross, A.S. & Robbins, T.W. (1993). Differential effects of intra-accumbens and systemic amphetamine on latent inhibition using an on-baseline, within-subject conditioned suppression paradigm. Psychopharmacology 110: 479-489.
Kinsbourne, M. (1993). Integrated cortical field model of consciousness. In: Experimental and theoretical studies of consciousness ed. J. Marsh. Ciba Foundation Symposium 174. John Wiley & Sons.
Lashley, K. (1956). Cerebral organization and behavior. In: The brain and human behavior ed. H. Solomon, S. Cobb & W. Penfield. Williams & Wilkins.
Libet, B. (1985). Unconscious cerebral initiative and the role of conscious will in voluntary action. Behavioral and Brain Sciences 8:529-66.
Libet, B. (1993). The neural time factor in conscious and unconscious events. In: Experimental and theoretical studies of consciousness ed. J. Marsh. Ciba Foundation Symposium 174. John Wiley & Sons.
Libet, B., Pearl, D.K., Morledge, D.E., Gleason, C.A., Hosobuchi, Y. & Barbaro, N.M. (1991). Control of the transition from sensory detection to sensory awareness in man by the duration of a thalamic stimulus. Brain 114:1731-57.
Lipp, O.V. & Vaitl, D. (1992). Latent inhibition in human Pavlovian differential conditioning: effect of additional stimulation after preexposure and relation to schizotypal traits. Personality and Individual Differences, 13: 1003- 1012.
Lishman, A. (1987). Organic Psychiatry. Blackwell.
Lubow, R.E. (1973). Latent inhibition. Psychological Bulletin 79:398-407.
Lubow, R.E. (1989). Latent inhibition and conditioned attention theory. Cambridge University Press.
Lubow, R.E. & Josman, Z.E. (1994). Latent inhibition deficits in hyperactive children. Journal of Child Psychiatry and Psychology (in press).
Lubow, R.E., Ingberg-Sachs, Y., Zalstein-Orda, N. & Gewirtz, J.C. (1992). Latent inhibition in low and high "psychotic- prone" normal subjects. Personality and Individual Differences 13: 563-572.
Mandler, G. (1984). Mind and Body. W.W. Norton & Co.
Marcel, A.J. & Bisiach, E. (eds.) (1988). Consciousness in contemporary science. Clarendon Press.
Marr, D. (1982). Vision. Freeman.
Marsh, J. (ed.) (1993). Experimental and theoretical studies of consciousness. Ciba Foundation Symposium 174. John Wiley & Sons.
Matussek, P. (1952). Studies in delusional perception. Psychiatrie und Zeitschrift Neurologie 189:279-318. (trans. 1987 in: The clinical roots of the schizophrenia concept, ed. J. Cutting & M. Shepherd. Cambridge University Press)
Meehl, P.E. (1966). The compleat autocerebroscopist: a though-experiment on Professor Feigl's Mind-Body Identity Thesis. In: Mind, Matter and Method eds. P.K. Feyerabend & G. Maxwell. University of Minnesota Press.
Mowrer, O.H. (1960). Learning and Behavior. John Wiley & Sons.
Neisser, U. (1976). Cognition and Reality. Freeman.
Newsome, W.T. & Salzman, C.D. (1993). The neuronal basis of motion perception. In: Experimental and theoretical studies of consciousness ed. J Marsh. Ciba Foundation Symposium 174. John Wiley & Sons.
Oatley, K. (1988). On changing one's mind: a possible function of consciousness. In: Consciousness in Contemporary Science ed. A.J. Marcel & E. Bisiach. Clarendon Press.
O'Keefe, J. (1985). Is consciousness the gateway to the hippocampal cognitive map? A speculative essay on the neural basis of mind. In: Brain and Mind ed. D.A. Oakley. Methuen.
Papez, J.W. (1937). A proposed mechanism of emotion. Archives of Neurology Psychiatry 38: 725-743.
Penney, J.B. & Young, A.B. (1981). GABA as the pallidothalamic neurotransmitter. Implications for basal ganglia function. Brain Research 207:195-99.
Peters, S. L., Grigoryan, G. A., Joseph, M. H., Hodges, H. and Gray J. A. Facilitation of latent inhibition by 6-OHDA lesions of the nucleus accumbens (in prep).
Polanyi, M. & Prosch, H. (1975). Meaning. University of Chicago Press.
Pylyshin, Z.W. (1980). Cognition and computation: issues in the foundations of cognitive science. Behavioral and Brain Sciences 3: 111-132.
Pylyshyn, Z.W. (1986). Computation and Cognition. MIT Press.
Rawlins, J.N.P. (1985). Associations across time: The hippocampus as a temporary memory store. Behavioral and Brain Sciences 8:479-528.
Rolls, E.T. & Williams, G.V. (1987). Sensory and movement related neuronal activity in different regions of the primate striatum. In: Basal ganglia and behavior: Sensory aspects and motor functioning ed. J.S. Schneider & T.I. Kidsky. Hans Huber.
Schmajuk, N. A. and Christiansen, B. Haloperidol reinstates latent inhibition impaired by hippocampal lesions (in prep).
Schmajuk, N.A., Lam, Y-W. & Gray, J.A. (under revision). Latent inhibition: A neural network approach. Behavioral and Brain Sciences.
Schneider, W. & Shiffrin, R.M. (1977). Controlled and automatic human information processing. I. Detection, search and attention. Psychological Review 84:1-66.
Searle, J. (1980). Minds, brains and programs. Behavioral and Brain Sciences 3:417-57,
Searle, J. (1983). Intentionality. Cambridge University Press.
Searle, J. (1993). The problem of consciousness. In: Experimental and theoretical studies of consciousness ed. J. Marsh. Ciba Foundation Symposium 174. John Wiley & Sons.
Solomon, P.R., Crider, A., Winkelman, J.W., Turi, A., Kamer, R.M. & Kaplan, L.J. (1981). Disrupted latent inhibition in the rat with chronic amphetamine or haloperidol-induced supersensitivity: Relationship to schizophrenic attention disorder. Biological Psychiatry.
Swerdlow, N.R. & Koob, G.F. (1987). Dopamine, schizophrenia, mania and depression: Toward a unified hypothesis of cortico- striato-pallidothalamic function. Behavioral and Brain Sciences 10:197-245.
Tai, C-T, Cassaday, H J, Feldon, J & Rawlins, J N P. Both electrolytic and excitotoxic lesions of nucleus accumbens disrupt latent inhibition in rats. Behavioral and Neural Biology, in press.
Velmans, M. (1991). Is human information processing conscious? Behavioral and Brain Sciences 14:651-726.
Weiner, I., Lubow, R.E. & Feldon, J. (1981). Chronic amphetamine and latent inhibition. Behavioural Brain Research 2:285-86.
Weiskrantz, L. (1986). Blindsight: a case study and implications. Oxford: Oxford University Press.
Weiskrantz, L. (1988). Some contributions of neuropsychology of vision and memory to the problem of consciousness. In: Consciousness in Contemporary Science ed. A.J. Marcel & E. Bisiach. Clarendon Press.
Williams, J.H., Wellman, N.A., Rawlins, J.N.P., Geaney, D.P., Cowen, P.J. & Feldon, J. (1994). Haloperidol increases latent inhibition in high schizotypal subjects. Abstract of paper given at 6th Winter Workshop on Schizophrenia, Les Diablerets, Schizophrenia Research (in press).
Winocur, G. (1981). The amnesic syndrome: a deficit in cue utilization. In: Human Memory and Amnesia ed. L. Cermak. Erlbaum.
Yee, B K, Feldon, J & Rawlins, J N P. Latent inhibition in rats is abolished by NMDA-induced neuronal loss in the retrohippocampal region but this lesion effect can be prevented by systemic haloperidol treatment. Journal of Neuroscience, submitted.
Young, A.M.J., Joseph, M.H. & Gray, J.A. (1993). Latent inhibition of conditioned dopamine release in rat nucleus accumbens. Neuroscience 54:5-9.
Zeki, S.M. (1978). Functional specialization in the visual cortex of the rhesus monkey. Nature 274:423-8.
Zohar, D. (1990). The Quantum Self. Bloomsbury.
Figure Legends [Figures only available in hard copy]
Figure 1. The Behavioural Inhibition System (BIS) as defined by its inputs and outputs.
Figure 2. The Septohippocampal System: The three major building blocks are shown in heavy print: HF, the hippocampal formation, made up of the entorhinal cortex, EC, the dentate gyrus, DG, 3, CA 1, and the subicular area, SUB: SA, the septal area, containing the medial and lateral septal areas, MSA and LSA; and the Papez circuit, which receives projections from and returns them to the subicular area via the mammillary bodies, MB, anteroventral thalamus, AVT, and cingulate cortex, CING. Other structures shown are the hypothalamus, HYP, the locus coeruleus, LC, the raphe nuclei, RAP, and the prefrontal cortex, PFC. Arrows show direction of projection; the projection form SUB to MSA lacks anatomical confirmation. Words in lower case show postulated functions; beh.inhib., behavioural inhibition. From Gray (1982b).
Figure 3. Information processing required for the comparator function of the septohippocampal system.
Figure 4. The Behavioural Approach (BAS) as defined by its inputs and outputs.
Figure 5. Above The Caudate Motor System: non-limbic cortico- striato-pallido-thalamic-midbrain circuitry. MCX: motor and sensorimotor cortex. VA/VL: ventral anterior and ventrolateral thalamic nuclei. CP: caudate-putamen (dorsal striatum). DP: dorsal pallidum. SN: Substantia nigra.
Below The Accumbens Motor System: Limbic Cortic-striato- pallido-thalamic-midbrain circuitry. LCX: limbic cortex, including prefrontal and cingulate areas. DM: dorsomedial thalamic nucleus. NAC: nucleus accumbens (ventral striatum). VP: ventral pallidum. A 10: dopaminergic nucleus A 10 in the ventral tegmental area.
GLU, CABA and DA: the neurotramsmitters, glutamate, gamma- aminobutyric acid and dopamine. + -: excitation and inhibition. I, II, III: feedback loops, the first two positive, the third negative. Based on Swerlow and Koob (1987).
Figure 6. The basal ganglia and their connections with the limbic system. Structures: SMC = sensorimotor cortex; PFC = prefrontal cortex; EC = entorhinal cortex; SHS = Septohippocampal system; Subic = subicular area; Amyg = amygdala; VA/VL = nucleus (N.) ventralis anterior and ventralis lateralis thalami; VM = N. ventralis medialis thalami; DM = dorsalis medialis thalami; DP = dorsal pallidum; VP = ventral pallidum; CP = caudate-putamen; N. Acc = N. accumbens; SNpr = substantia nigra, pars reticulata; SNpc = substantia nigra, pars compacta; A 10 =A 10 in ventral tegmental area; SC = superior colliculus; PPN = penduculopantine nucleus. Transimitters: GLU = glutamate; DA = dopamine; GABA = gamma-aminobutyric acid. From Gray et al. (1991a).
Figure 7. An integrative theory (Gray et al. 1991a) of positive schizophrenic symptoms (top), seen as arising from a structural abnormality in the brain (bottom), which gives rise to a functional neurochemical abnormality, and hence to an abnormality in cognitive processing.
Figure 8. A schematic summary of the theory of schizophrenia proposed by Gray et al. (1991a). (A) The abnormality of cognitive processing consists of a failure to integrate past regularities of experience with the current control of perception and action. (B) This reflects a dysfunctional connection between the limbic forebrain and the basal ganglia. (C) The specific pathway carrying the dysfunctional connection is from the subiculum (in the limbic forebrain) to the nucleus accumbens (in the basal ganglia). (D) The computing functions thus disrupted are the passage of information from a comparator system, utilizing stored traces of past regularities (limbic forebrain), to a motor programming system (located in the basal ganglia) controlling perception and action.