The interaction of cognitive and stimulus–response processes in the control of behaviour
Introduction
Do animals learn cognitions (i.e., knowledge about the world not tied to particular behaviour) or stimulus–response (S–R) connections (i.e., links between particular stimuli and particular responses)? Historically, some of psychology's most famous battles were fought over this bit of territory 96, 105, 230. However, in a seminal paper on learning, Hirsh [96]proposed that both cognitive and S–R processes coexist in an intact animal and that hippocampal lesions convert a cognitively functioning rat into something like a (S–R) automaton. Subsequently, Mishkin et al. ([142]p. 66) noted the important implication of Hirsh's argument that:
“...both sides in the great debate between behaviourism and cognitivism must ultimately be declared the winners, since the evidence from the study of amnesia demonstrates that both types of processes must be constantly present in normal behaviour.”
“If this radical resolution of that long and difficult debate proves correct, it will have enormous implications for both psychological and neurophysiological research.”
Hirsh's paper was published 23 years ago and its importance has been widely acknowledged. Different memory systems, characterised by something like S–R and cognitive forms, are now recognised 43, 196, 126, 201, 243, 244. Rather than competing theories, the terms S–R and cognitive now appear to describe processes that coexist in an animal. Although, within the study of learning and memory, multiple processes are acknowledged, the contention of the present paper is that the implications of these ideas for the control of behaviour have not been adequately explored. Hence the article reviews the literature with the aim of showing how a dual S–R and cognitive control can explain a variety of phenomena. The extensive literature on learning and memory will not be covered here, since a number of excellent reviews already exist 43, 196, 201. Learning and memory as such will be considered only in so far as they are directly relevant to the processes that underlie the control of behaviour. The paper suggests that the issues raised by Hirsh can provide a new link between psychology, ethology and neuroscience.
Contemporary behavioural scientists tend to favour cognitive theories whereas the S–R view is sometimes seen as being of historical interest only. I shall investigate which features of S–R theories are valid and, in terms of the control of behaviour, how they might be integrated within the dominant cognitive framework. For example, cognitive theories have difficulty with explaining such things as stereotypies and displacement activities. It will be argued that the apparently conflicting S–R and cognitive theories of behavioural control each describe a real and important aspect of the underlying processes.
Aside from historical differences amongst theorists, it must surely constitute an implicit assumption of almost all behavioural science that behaviour is under the joint control of: (a) external stimuli; and (b) internal cognitions and goals. However, we lack theoretical models that are able to suggest how the interaction occurs. What exactly is the role of stimuli in controlling behaviour? Are they simply a source of information to cognitive processes or a direct trigger to behaviour? How do stimulus and cognitive factors interact? In producing behaviour, under what circumstances are stimuli (external factor) and cognitions (knowledge stored in the nervous system) in competition, and when do they reinforce each other's effects?
In much of psychology, study of the causation of behaviour in terms of physically present stimuli and behaviour has come to take a poor nth place to cognition. As Spear and Isaacson ([200]p. 3) so aptly express it, an understanding of:
“....habitual, overlearned, elicited, short-latency behaviours that might even be characterised as species-specific—is of vast importance for our topic. This type of knowledge seems largely to have been ignored in theories of how humans process information, in favour of knowledge derived from language-mediated, `higher-order' processing”.
Or, as Turvey ([233]p. 211) similarly notes:
“...it is curious that theories of perception are rarely, if ever, constructed with reference to action. And, while theories of perception abound, theories of action are conspicuous by their absence. But it must necessarily be the case that, like warp and woof, perception and action are interwoven, and we are likely to lose perspective if we attend to one and neglect the other...”.
Although the situation might have improved slightly in the last 20 years (e.g., 72, 151), there is still a real deficiency of integrative theory.
In some ways analogous to the difference of emphasis within psychology, classical ethologists (e.g., 124, 216) developed theoretical models in which sign stimuli played a central role in triggering fixed action patterns whereas contemporary ethology places explanatory weight upon cognitive processes. This change in emphasis raises a number of issues. For instance, in accepting cognitive processes, are we to reject the older ethological models? If both types of model capture real aspects of behavioural control, then how do these aspects act in combination in determining behaviour? It will be suggested that the model can form a bridge both within ethology and between ethology and other behavioural sciences.
Traditionally, theories of learning and response production (whether S–R or cognitively orientated) were developed in parallel with theories of motivation (e.g., 19, 105, 124, 216, 230). Of late, the tendency has been towards compartmentalising these processes. The present paper is intended to break down these compartmental barriers. Therefore, the paper will address the issue of how the concept of motivation might be tied to a consideration of cognitive and S–R processes.
Many theorists in neuroscience, psychology and ethology have proposed hierarchical models of behavioural control, and indeed that of Gallistel [72]has powerfully influenced the present review. However, such models are usually discussed somewhat in isolation from other lines of theoretical development. This paper will argue that a consideration of S–R and cognitive controls inevitably has relevance to the topic of hierarchies. Cognitive and S–R processes imply controls at different levels in a behavioural hierarchy.
In proposing a joint cognitive and S–R control of behaviour, evidence will be reviewed which shows that the relative weighting of these processes as determinants of behaviour changes with: (a) development; (b) experience; and (c) pathology. In some cases, cognitive processes come to exert more control relative to S–R processes, as in the changes that accompany development. In other situations, S–R processes acquire more control relative to cognitive processes, as in habit formation and some cases of pathology where there is some loss of cognitive control.
Although there have been attempts to build bridges between ethology and psychology (e.g., 72, 75, 94, 227), a considerable gap still exists, even in the phenomena that form the topic of interest. The development of learning theory has always been largely within psychology, whereas sign-stimuli and displacement activities have fallen within the ethologists' domain, [216]and stereotypies and animal welfare are the responsibility of applied ethology 119, 135. The favourite explanatory terms of classical ethology, e.g., sign stimulus and fixed-action pattern [216], have never had a secure place in psychology. Yet the observations that gave rise to these early concepts are no less valid today. It is hoped that the model will serve to integrate ethological and psychological approaches.
Models that consider how (a) external, stimulus, and (b) internal, cognitive factors jointly determine control, usually address only a part of the system and relate to a restricted set of tasks such as the Stroop [34]. Workers in the rat [48]and human 164, 165areas make the occasional reference to phenomena that indicate modes of control generalizable across species. However, these areas of potential integration remain largely unexplored. The present article will propose a model with a potentially broad integrative power.
I hope to convince neuroscientists, psychologists (both of the human- and rat-orientated variety) and ethologists that it is worth looking over the traditional fences to see what is on the other side. Therefore, foolhardily, I stick my head above the parapet, and venture into such diverse fields as sign-stimuli in sticklebacks and schizophrenia in humans, using the S–R and cognitive distinction to look for common underlying processes of control. I cannot do justice to any given phenomenon, but I hope to broaden our perspective.
Section snippets
The basics of the model
Fig. 1a represents the most basic assumption of an S–R theory, applicable to psychological and ethological schools. A stimulus (S) is assumed to have a certain strength of tendency to produce a response (R); that is, S has a response eliciting potential, which can vary from zero to some maximum value. The strength of this tendency will depend upon innate factors and learning. In some cases, the links between S and R are relatively strong (e.g., the sign stimuli of classical ethology).
The
Developing the model
Fig. 1 summarises the most basic assumptions that underlie the present paper. Fig. 2 shows a development of the model. As one aspect of the control of behaviour, an array of external events (S1, S2,...) impinge upon the animal, shown as one set of inputs to Box C. As another aspect, there are: (a) cognitions, e.g., goals and expectations (Box A); and (b) physiological states, e.g., hunger (Box B). The cognitions activated depend to some extent upon the sensory events, but of course convey
Modulation of reflexes
The literature contains frequent references to a distinction between reflexes and non-reflexes, the latter described in such apparently exclusive terms as `motivational' [58], `purposive' system or `servomechanism' [73]. However, Fig. 2, which is applicable to both classes of process suggests that reflexes and non-reflexes lie at different points on a continuum of weighting between S–R and cognitive processes. On the one hand, reflexes are modulated by higher levels and, on the other hand,
Application to ethology
There are a number of phenomena in the ethological literature to which the model can be applied, and this section looks at a few of these. Although the phenomena have traditionally been studied by ethologists, it is clear that the explanatory model needed to explain them is one equally applicable to neuroscience and psychology.
Motivation
Motivation has been a topic of great concern to behavioural scientists taking various approaches, but fragmentation has often characterized the literature that has emerged. Implicit within the model proposed here, there is both flexibility of behaviour in enabling goals to be reached by various routes and more fixed, species-typical aspects of behaviour 19, 219. Motivation plays a role in both aspects of control, the model indicating outputs from motivation going both to S–R links and to the
Comparing species
Whereas a number of mammalian species seem similar in their ability to use S–R processes, differences are evident in their ability to perform tasks that require cognitive processes [142]. Passingham [159]compared rats and rhesus monkeys (Macaca mulatta) on a task in which an S1–R1, S2–R2 distinction was required, e.g., push a white door but pull a black door, and found that they did not differ on number of trials to attain criterion. However, the monkey is incomparably superior to the rat in
Human studies
So far, the review has mainly concerned the application of the model to non-humans. Though there is little contact between the two groups of researchers [75], in the present context similar considerations apply to human and non-human behaviour. In both cases, a joint cognitive and S–R model can be applied. This section will look at the possibility of synthesis through application of the model to humans, in particular to studies generally placed under the heading of `human performance'.
Adaptive considerations in shifting the weighting of control
From adaptive considerations, why should there be a switch from cognitive (controlled) to S–R (automatic) processing? Controlled processes are of limited capacity and generally relatively slow compared to automatic ones [77]. The capacity of automatic parallel processes would seem to be unlimited. The advantage of freeing a limited capacity system for more demanding tasks such as vigilance and association formation is not difficult to appreciate. There is also the bonus of speed.
If the
Learning and the control of behaviour
The role of learning has been implicit in much of the discussion so far. This section relates the model more explicitly to the body of knowledge entitled `learning theory'. It investigates the extent to which learning can be understood in terms of changes in cognitions, changes in S–R connections or both. The question of `what is learned?' has already been extensively investigated by learning theorists, and this section relates this discussion to the model. Theorists have also asked the related
Hierarchies of control
The notion of a weighting between internal and external factors in the control of behaviour, control being exerted at different levels and a shift of weighting with experience ([116]p. 191) leads logically to the notion of a hierarchy of behavioural control. This section will explore the relevance of the notion of hierarchy to the proposed model. That the controls of behaviour are organised in a hierarchy has had a distinguished history in behavioural science, though the literature on this
Consciousness
In the terms developed here, it is suggested that consciousness is a specialised aspect of cognitive control [9]and corresponds to a high-level in a hierarchy 9, 150, 173, 194. Much behaviour can be organised perfectly well at an unconscious level. Consciousness can be occupied with other things, and need only be alerted when things don't go according to plan ([92]p. 281), when novelty is involved or when we must overcome a strong habitual response or temptation [151].
The model fits some of the
Developmental effects
Either explicit or implicit within the literature on development is the idea of a changing of the locus of control of behaviour. Development is associated with gaining autonomy from sensory control [23]and acquisition of top-down control over behaviour that is organised at a lower level. Reflexes become integrated into cortical control [161]. Such control will perhaps most usually be inhibition, but excitation might also occur. The infant moves from dependence upon sensory stimuli (S–R
Brain regions and the model
It is desirable that ideas expressed in rather abstract terms can be related to actual neural structures (cf. [36]). This section looks at some neural structures whose function might be related to the S–R versus cognitive distinction. Ideally, we would like to be able to associate shifts in the mode of control with changes in different brain regions. Also, some behavioural abnormalities might be understood in part as a shift of weight between regions. However, a caution is in order: it will be
Neuropathology and psychopathology
Evidence points to certain pathology corresponding to a disruption of the normal pattern of joint S–R and cognitive control of behaviour, with S–R processes either: (a) adaptively taking over some of the function of lost cognitive controls; or (b) competitively and maladaptively assuming undue influence over the control of behaviour in the face of higher-level goals.
Comparison with other models
In proposing a distinction between cognitive and S–R controls of behaviour, the present model can be compared with earlier theories. Various authors have proposed dichotomies in the control of behaviour, e.g., sensory and non-sensory [91], declarative and procedural 47, 246, controlled and automatic [192], top-down (concept-driven) and bottom-up (data-driven) [86], conscious and unconscious [9], and rational and experiential [59]. The controlled and automatic dichotomy and the declarative vs
General discussion
The paper has demonstrated the wide application of a model of behavioural control involving S–R and cognitive features. The level of control exerted by these processes has been associated with the notion of behavioural hierarchy. According to the interpretation given here, a feature of behaviour is that the relative weighting of the S–R and cognitive processes changes as a function of: (a) development; (b) learning; and (c) malfunction. The fact that the same basic model has relevance in such a
Acknowledgements
I am very grateful to Kent Berridge, Anthony Dickinson and Piet Wiepkema and the referees for their most helpful comments on the manuscript. I am also grateful to Allison Richens and Yvonne Royals for their secretarial help.
References (248)
The functional organization of behaviour
Animal Behav.
(1976)- et al.
Endogenous opioids are involved in abnormal stereotyped behaviours of tethered sows
Neuropeptides
(1985) - et al.
Gating of motor cortex reflexes by prior instruction
Brain Res.
(1974) Frontal lobes
Curr. Opinion in Neurobiol.
(1993)The hippocampus and contextual retrieval of information from memory: A theory
Behavioural Biol.
(1974)Variations in the sensitivity of instrumental responding to reinforcer devaluation
Quart. J. Exp. Psychol.
(1982)Learning of movement sequences
Psychol. Bull.
(1984)- Allport, G.W. Pattern and Growth in Personality. London: Holt, Rinehardt and Winston;...
- Anderson, J.R. Language, Memory and Thought. Hillsdale: Lawrence Erlbaum;...
Acquisition of cognitive skill
Psychol. Rev.
(1982)