COGNITIVE SCIENCE AND PSYCHIATRY:

AN OVERVIEW

Dan J. Stein, M.D.

Dept of Psychiatry

College of Physicians and Surgeons, Columbia University,

and the New York State Psychiatric Institute,

722 W 168 St

NY, NY 10032

Running Title: Cognitive Science and Psychiatry

Tel: 212-960-2355

ABSTRACT

Cognitive science is a multidisciplinary field, comprising cognitive

psychology, artificial intelligence, linguistics, neuroscience, and

anthropology. In recent years, cognitive science has become a

predominant paradigm in studies of the mind. This paper reviews work

at the emerging interface between cognitive science and psychiatry.

It is argued that cognitive science has significant potential as an

integrative framework for theorizing and researching psychiatric

disorders and their treatment.

INTRODUCTION

Cognitive science is a relatively new field that incorporates concepts

and methods from philosophy, cognitive psychology, artificial

intelligence, linguistics, neuroscience and anthropology (Gardner,

1985; Posner, 1989). Whereas behaviorism dominated the psychological

sciences during the first part of this century, cognitive science has

become a central paradigm of the latter part. This "cognitive

revolution" was fostered by the promise of cognitive science as an

integrated and fertile approach to the mind. Indeed, departments of

cognitive science have become important loci of interdisciplinary

research and have generated a prodigous literature.

During the period of the cognitive revolution, significant changes

also took place in psychiatry (Lipowski, 1989). During the 1950s,

psychoanalysis comprised a dominant school in many departments of

psychiatry. In subseqent decades, however, a community-focused model

was widely touted. Most recently, neurobiological approaches have

been given pride of place. Certainly, there have been calls for a

synthetic or "biopsychosocial" approach to psychiatric disorders and

treatment. Nevertheless, there has been little progress on the

construction of an integrative framework for such an approach.

On the whole, cognitive scientists have paid little attention to

clinical phenomena, while clinicians have in turn only occasionally

employed the concepts and methods of cognitive science. In recent

years, however, dialogue between the two fields has increased (Breger,

1969; Mahoney and Freedman, 1985; Ingram, 1986; Horowitz, 1988a;

Williams, Fraser, MacLeod, and Mathews, 1988; Magaro, 1991; Stein and

Young, 1992a). In this paper I review research at this intersection,

and discuss its potentials and limitations. I suggest that cognitive

science can provide an important integrative framework for psychiatry.

COGNITIVE SCIENCE

What exactly is cognitive science? The following three sections

attempt to address this question. I briefly discuss descriptions and

definitions of cognitive science, its origins and subdisciplines, and

then contrast and compare cognitive science with behaviorist,

psychoanalytic and neurobiological models (Stein, 1992a).

a) Descriptions and Definitions

In general, cognitive scientists are interested in mental structures

and processes, their representational significance, and their physical

instantiation (Stillings, Feinstein, Garfield, Rissland, Rosenbaum,

Weisler, and Baker-Ward, 1987). Several authors have attempted more

rigorous definitions of cognitive science. A cognitive

information-processing account, for example, views the mind as an

information-processing system that selects, transforms, encodes,

stores, retrieves, and generates information and behavior (Lachman,

Lachman, and Butterfield, 1979). A computational view, on the other

hand, emphasizes that, "Cognitive science, sometimes explicitly, and

sometimes implicitly, tries to elucidate the workings of the mind by

treating them as computations, not necessarily of the sort that is

carried out be the digital computer, but of a sort that lies within

[the] broader theory of computation" (Johnson-Laird, 1988, p. 9).

Restrictive definitions of cognitive science may, however, include

only one or other of the divergent models that cognitive scientists

have developed. Early cognitive scientists viewed the mind as a

sequential processor, similar to the early digital computer. The mind

was seen as a passive recipient of information, which was registered

in a short-term memory, and perhaps encoded in a long-term memory.

More recent cognitive scientists have, however, pointed out that the

mind is a parallel processor (Rumelhart, Hinton, and the PDP Research

Group, 1986), and have emphasized that mental structures are active

and that they occur within a particular context (Neisser, 1976;

Lakoff, 1987). Such work may be excluded by a definition of cognitive

science that focuses solely on information-processing and computation.

On the hand, too broad a definition of cognitive science may prevent a

constrast between cognitive science models and other models often used

by clinicians. It may be useful to at least specify the broad

categories into which cognitive science models fall. Cognitive

science models typically specify cognitive architecture in one of two

ways, symbolic and connectionist. The elements of symbolic systems

are symbols, which are stored in associative structures. Symbolic

systems include levels-of-processing models (Craik and Lockhart,

1972), spreading activation constructs (Collins and Loftus, 1975), and

schema approaches (Neisser, 1976). The elements of connectionist

systems are simplified and schematized neurons which are

interconnected in a network. Again, a variety of these parallel

processing models have been developed (Rumelhart et al., 1986). In

subsequent sections the application of both symbolic and connectionist

architectures to clinical theory and practice is reviewed.

b) Origins and Subdisciplines

Further understanding of cognitive science can be gained by discussing

its origins. The development of cognitive science models was

encouraged by various factors, including the failure of behaviorism,

the invention of the computer, and various theoretical advances

(Gardner, 1985). One of the most important of these theoretical

advances was the development of computer science. The father of this

field was Turing (1936), a British mathematician who described a

simple machine (the Turing machine) that executed instructions in

binary code, and proposed that such a machine could in principle

perform any computational task. He also described the idea of a

universal Turing machine which takes a coded version of other Turing

machines as input and then emulates their behavior.

Computer science can immediately be seen as relevant to psychology,

for the question arises of whether the human mind can be instantiated

on a universal Turing machine. If so, computer science would

constitute a basic science for exploring the hypothesis (Craik, 1943)

that the mind is a symbol manipulating device. Turing (1950)

suggested it might be possible to program a machine so that a user

communicating with the machine and with a person would be unable to

differentiate the two (the Turing machine test).

A number of other developments consolidated the importance of

computational constructs for psychology. Thus, it was proposed that

certain concepts were useful in explaining both computing machines and

human brains. This was clearly formulated at a seminal meeting in

1948, the Hixon Symposium on "Cerebral Mechanisms in Behavior"

(Jeffress, 1951) at which McCulloch, a professor of psychiatry, and

Von Neumann were the opening speakers. For example, the all-or-none

property of neuronal activation can be compared with the determination

of Boolean statements as either true or false. Furthermore, neuronal

networks and Boolean statements can be described in elecrical terms as

current that passed or failed to pass in a circuit.

In addition, it was argued that certain constructs were useful in

explaining both computing machines and human minds. Hebb (1949), for

example, suggested that synaptic strengthening in neuronal networks

led to learning. Wiener (1947) proposed that machines and minds that

had feedback mechanisms that displayed purposefulness. Central to

control and communication engineering was the notion of the message,

whether this was transmitted by electrical, mechanical, or neural

means. Information, Shannon (1938) showed, was independent of its

physical instantiation.

It has been suggested that cognitive science was founded in September

1956, at the Symposium on Information Theory at MIT (Gardner, 1985).

Significant research presented at the conference included papers by

the cognitive psychologist, Miller, artificial intelligence workers,

Newell and Simon, and the linguist, Chomksy. In his paper, Miller

illustrated the importance of empirical studies of cognition by

discussing work on the constraints of short-term memory processes.

Work by other psychologists in the 1950s, including Broadbent and

Cherry in Britain, and Bruner and colleagues at Harvard, gave further

impetus to the development of cognitive psychology, and established

the discipline and its empirical methodologies as a cornerstone of

cognitive science. Newell and Simon argued that artifial intelligence

was possible, and drew comparisons between artificial and human

problem solving processes. Together with Minsky and McCarthy they

pioneered the field of artificial intelligence, and helped establish

the importance of a computational methodology for cognitive science.

Finally, Chomsky described his theory of a grammar based on linguistic

transformations. This work, together with his scathing review

(Chomsky, 1959) of Skinner's work on verbal behavior buttressed the

shift from behaviorism to cognitivism, and established linguistics as

a useful frame from which the issues of cognitive science could be

considered.

Other core disciplines of cognitive science include neuroscience,

anthropology, and philosophy. Neuroscience comprises a lower limit

for cognitive science, for human cognitive structures and processes

are ultimately based in and constrained by neuroanatomy and

neurochemistry. Anthropology and social psychology constitute an

upper limit for cognitive science, for cognitive structures and

processes vary from place to place and time to time. Finally, many of

the questions studied by cognitive scientists were first raised by

philosophers. Philosophers have also addressed the foundational

principles of the discipline, and more recently, as cognitive science

has advanced, have been faced with the meta-cognitive question of what

the mind and the world are like for this scientific advance to have

taken place.

Developmental psychology is not usually listed as a central discipline

of cognitive science. Nevertheless, one of the most important

pre-cognitivists was the developmental psychologist Piaget. Piaget

detailed the transformation of mental structures from sensorimotor

reflexes to the operations of formal thought (Piaget, 1952).

Contemporary dialogue between developmental psychology and cognitive

science is increasing (Sternberg, 1984; Meltzoff, 1990; Globerson and

Zelniker, 1989).

c) Contrasts and Comparisons

Cognitive science models can be contrasted and compared with the

clinically familiar models of behaviorism, psychoanalysis, and

neurobiology. Behaviorism, for example, self-consciously makes the

mind into a "black box", asserting that only observable stimuli and

responses can be studied. In contrast, the cognitive model holds that

what is most interesting are the mental structures in the "black box"

and the processes (operations) whereby they generate cognitive

products (thoughts and feelings).

Nevertheless, it is possible to see a continuity between behaviorist

and cognitive science models. While the behaviorist model is limited

to inputs (stimuli) and outputs (responses), the cognitivist model is

concerned with inputs, processing, and outputs. Like behaviorists,

cognitivists have adopted an empirical stance, which emphasizes the

importance of careful measurement and laboratory experimentation.

The classical psychoanalytic model of the mind was an energy-based

one. For Freud (1894), "in mental functions something is to be

distinguished - a quota of affect or sum of excitation - which

possesses all the characteristics of a quantity (though we have no

means of measuring it), which is capable of increase, dimunition,

displacement and discharge, and which is spread over the memory-traces

of ideas somewhat as an electric charge is spread over the surface of

a body". Freud described how the forces of the unconscious are

expressed, transformed, or repressed, resulting in everyday behaviors

and psychopathology.

Again, however, it is possible to demonstrate a continuity between

psychoanalysis and cognitive science. Cognitive science and

psychoanalysis both focus on the structures of the mind, and the way

in which these determine mental phenomena. Furthermore, in Freud's

later work he describes affect not in terms of energy, but in terms of

its role as a signal. Finally, post-Freudian psychoanalysts,

including the self psychologists, the objects relations school, and

the interpersonalists, have increasingly employed cognitively oriented

constructs such as self and other representations (Stein, 1992b).

Neurobiological models attempt to explain the biological basis of

mental processes and products. Proponents of these models may state

that all psychological explanations are reducible to neurobiological

ones. This view contrasts with the cognitive science argument that

cognitive phenomena necessarily require psychological explanations.

Psychological events are emergent phenomena, and psychological

explanations are not reducible to biological ones.

Once again, however, it is possible to see a continuity between the

biological model and cognitive science. Earlier neuroscience was

described as comprising a lower limit of cognitive science. While

some cognitivists are interested in information-processing in only the

most abstract sense, many others are interested specifically in how

information-processing occurs in the human neurobiological substrate.

This interest is particularly apparent in connectionist work informed

by neuroscience (Rumelhart et al, 1986).

CLINICAL COGNITIVE SCIENCE

For the clinician, the question immediately arises of whether

cognitive science models are useful in accounting for clinical

phenomena. However, standard texts of cognitive science make little

reference to psychiatric disorders and treatment. Nevertheless, a

number of developments in cognitive science have been instrumental in

allowing a bridge to clinical phenomena. This section briefly reviews

some of these bridges from cognitive science to the clinic, including

research on emotional and unconscious processing, and relevant work in

each of the subdisciplines of cognitive science.

Cognitive science has been characterized by a disregard for emotion

(Gardner, 1985). Nevertheless, humans are not only the most

intelligent of the animals, they are perhaps also the most emotional

(Hebb, 1946). Furthermore, it may be suggested that this relationship

between intelligence and emotionality is a necessary one. In a

seminal paper, Simon (1967) argued that emotion in humans is

comparable to the prioritized interruption of different processes by

one another in complex artificial intelligence systems with multiple

goals and limited resources. The idea that cognitive design problems

are solved by emotional processes remains popular in contemporary

cognitive science attempts to theorize emotion (Oatley and

Johnson-Laird, 1987; Sloman, 1987), and has helped generate a

rapidly growing empirical literature on cognitive-affective

processing. Cognitivist work on emotion has subsequently been

employed in order to understand affective experience and change in the

clinic (Mandler, 1975; Greenberg and Safran, 1990).

Another important area within cognitive science that has clinical

relevance is the study of unconscious processing. Research on the

unconscious is not new, and includes the pioneering work of Helmholtz,

Janet, and Freud (Ellenberger, 1970). However, it was not until the

emergence of cognitive science that the "cognitive unconscious"

(Kihlstrom, 1987) became a respectable area of study. Research has

focused on such areas as selective attention, subliminal perception,

implicit memory, hypnotic suggestion, and dreaming (Erdelyi, 1985;

Foulkes, 1985; Kihlstrom, 1987). While this work immediately brings

psychoanalysis to mind, there appear to be important differences

between the cognitive unconscious and the psychoanalytic unconscious

(Bowers and Meichenbaum, 1984). According to Freud, the unconscious

is a set of drives, affects, and motives that has an organization and

content more primitive than those of the consciousness, but that

nevertheless has latent the potential for interacting with other

elements of the psyche. According to cognitive science, the

unconscious is a set of cognitive processes, including attitudes and

predispositions, that act prior to consciousness, so actively

organizing and ordering experience and behavior. Nevertheless,

contemporary work on unconscious processing establishes a dialogue

between cognitive and clinical scientists (Bowers and Meichenbaum,

1984; Eagle, 1986; Horowitz, 1988b; Safran and Greenberg, 1986; Uleman

and Bargh, 1989; Singer, 1990; Prigatano and Schachter, 1991; Cloitre,

1992; Stein, 1992b).

A number of early cognitivists have been interested specifically in

psychiatric disorders. Ruesch and Bateson (1968) were among the first

to apply cybernetic concepts to psychopathology and psychotherapy.

Within cognitive psychology, Hilgard (1977), whose work will be

discussed later, employed Janet's notion of dissociation and developed

a neodissociative model of the dissociative disorders. This work

exemplified the potential utility of employing cognitive psychology

constructs to conceptualize psychiatric disorders.

In artificial intelligence, Colby (1981) pioneered the investigation

of clinical phenomena. He developed a program, PARRY, which

incorporated a model of the paranoid process, and succeeded in

simulating a paranoid patient. The difficulty of such work, and

Colby's achievement, is illustrated by the limited amount of

subsequent research in this area (Tomkins and Messick, 1963;

Clippinger, 1977). More recently, however, a number of authors have

employed computer implemented neural networks to model psychopathology

and psychotherapy (Hoffman, 1987; Hestenes, 1991; Cohen and

Servan-Schreiber, 1992l; Williams and Oaksford, 1992; Caspar,

Rothenfluh, and Segal, unpublished manuscript).

Linguists, neuroscientists, anthropologists and social psychologists

have long been interested in clinical phenomena, and some of this work

has fallen within the realm of clinical cognitive science.

Cognitivist models of the aphasias have their origins in the work of

the "diagram-makers", who constructed modular models of the mind at

the turn of the century (Morton, 1984). Current cognitive studies of

language may contribute to various clinical areas, such as treatment

of dyslexia (Swerling and Sternberg, 1992), or understanding clinical

dialogue and the mechanisms of the "talking cure" (Oatley, 1992).

Contemporary cognitive neuropsychologists continue to hold that study

of neurologically impaired patients contributes to cognitive theories

of mind (Caramazza, 1992). Clinical anthropology has tended not to

adopt cognitivist models (Stein, 1992c), but cognitively oriented work

in social psychology has been particularly relevant to clinical

science, and has tackled such important concerns as self and other

representations, and their role in mood disorders (Markus, 1977;

Cantor and Kihlstrom, 1981; Sorrentino and Higgins, 1986; Westen,

1988). Finally, it may be argued that philosophy, too, constitutes an

important foundational discipline for clinical cognitive research

(Mahoney, 1991; Lyddon, 1988; Stein, 1992d).

Developmental psychologists have long been interested in developmental

arrests and abnormalities. Since Piaget, cognitive concepts have

increasingly been used in this area. Developmental research that

draws on cognitive science constructs and focuses on clinical

implications has been particularly influential (Greenspan, 1980;

Stern, 1985; Emde, 1983).

PSYCHIATRIC DISORDERS

The work of cognitive scientists interested in clinical phenomena

provides the basis for employing cognitive constructs to conceptualize

psychiatric disorders, assessment, and treatment. I will discuss each

of these areas in turn. While a variety of psychiatric disorders have

been studied from a cognitive science perspective, I will focus on

only a few in order to highlight some important methods and issues in

clinical cognitive science.

a) Major Depression

Since their introduction by Ellis, Beck and others, cognitive

therapies have become increasingly accepted as an effective treatment

of depression (Mahoney and Freeman, 1985). While these therapies are

not necessarily formal extensions of cognitive science, much of this

work has been influenced by the cognitive revolution. The growing

dialogue between cognitive science and cognitive models of depression

(Ingram and Reed, 1986; Ingram and Wisnicki, 1991) provides the most

comprehensive exemplar of the use of symbolic architectures from

cognitive science to understand psychopathology.

Symbolic architectures specify the mind in terms of cognitive

structures, operations, and processes. Cognitive structures that have

been hypothesized to be central in depression include associative

networks (Ingram, 1984; Teasdale, 1983) and self-schemas (Beck, 1967;

Kuiper, Derry, and MacDonald, 1982). Associative network theories

characterize depression as having increased availability and/or

accessiblity of negative constructs about the self. Self-schema

theory characterizes depression as activation of a self-structure that

is negative in content. While evidence for negative self-schemas in

depression is good, there is less evidence that such schemas are

present in the absence of current depression (Ingram and Reed, 1986;

Ingram and Wisnicki, 1991).

Research also suggests that cognitive processes in depression are

characterized by a negative selectivity bias (Ingram and Reed, 1986;

Ingram and Wisnicki, 1991). Studies of encoding may be divided into

studies of personal semantic information (i.e. information related to

concepts of self) and episodic information (i.e. information related

to autobiographical events). Studies of semantic encoding indicate

that enhanced negative encoding is a key factor in information

processing in depression, while studies of episodic encoding indicate

that less accurate encoding of positive information is critical.

Studies of retrieval, on the other hand, suggest that negative

information is more efficiently and positive information is less

efficiently retrieved in depression. In addition, depressed patients

selectively monitor negative information with a decrease in attention

to positive stimuli.

Finally, cognitive products in depression include attributions and

automatic thoughts (Ingram and Reed, 1986; Ingram and Wisnicki, 1991).

Depressed patients are more likely to make dysfunctional attributions

as to the cause of events. Negative automatic thoughts appear to be

characteristic of subclinical and clinical depression, and they remit

with treatment.

Cognitive science models of depression influence our conceptualization

of psychotherapy, and direct further empirical research. The finding,

for example, that depressed schemas are not present in the absence of

depression suggests that cognitive therapy works not by structural

changes in underlying schemas, but by changes in the activation level

of negative self-schemas. While cognitive structures such as schemas

are difficult to measure, methodological progress has been made

(Segal, 1988). Considerable theoretical and empirical research on

cognitive science models of depression nevertheless remains to be

done. In particular, cognitive science models of depression have,

with few exceptions (Klein, 1976), failed to incorporate

neurobiological knowledge.

b) Dissociative and Conversion Disorders

Symbolic architectures, including schema theory, have not only been

adopted by the cognitive therapy tradition, but have also been

employed in psychoanalytically informed research (Wachtel, 1982; Slap

and Saykin, 1983; Segal, 1988; Horowitz, 1988a; Horowitz, 1991). The

area in which there is perhaps most clearly an intersection between

the constructs of symbolic architecture and psychodynamic theory is

the study of dissociative and conversion disorders.

The first detailed psychological theory of dissociative and conversion

disorders was developed by Janet (1907). Like Freud, Janet studied at

Charcot's clinic at the Salpetriere. He too was interested in the

notion of cognitive processes that took place outside of awareness,

for which he coined the term "subconscious". Janet described the mind

in terms of "automatisms", or elementary structures controlling

experience, thought, and action in various domains, and which together

make up the flow of consciousness. At times of stress or exhaustion,

one or more automatisms separate from the rest, breaking the unity of

consciousness. The dissociated automatisms continue to operate, but

are no longer accessible to phenomenal awareness or amenable to

voluntary control, resulting in various dissociative or conversion

symptoms.

Hilgard (1977) used concepts from cognitive psychology to provide a

neodissociative account of these disorders. He described the mind in

terms of subordinate cognitive structures each with a degree of

autonomy, a hiererarchical control managing the competition between

these structures, and a central monitoring and controlling structure

(the executive ego). Subordinate cognitive structures can operate

outside of awareness, or they may be processed consciously through the

executive ego. Certain physiological (e.g. anesthesia) or

psychological (e.g. hypnosis) conditions lead to disruptions in the

connections between the subordinate structures. Information

processing in the subordinate structures may then take place outside

of phenomenal awareness or voluntary control. More recently,

Kihlstrom (1990) has used an associative-network model of memory to

further elaborate this view of the dissociative and conversion

disorders. He describes how disruption of links between

representations of memory or of sensory input and representations of

self may result in lack of awareness of particular memories or

sensations, with continued processing of related representations.

While psychoanalytic theory has also emphasized the importance of

unconscious processes in the genesis of these disorders,

neodissociative theory has a view of the unconscious that is

closer to that of cognitive science. Although there may be

theoretical difficulties with the psychoanalytic energy based view,

the success of the cognitivist revival of Janet's ideas ultimately

depends on its ability to generate empirical work on these relatively

unresearched disorders.

c) Obsessive-Compulsive Disorder

Earlier the question of the relationship between cognitive science and

neurobiological models was raised. This becomes particular relevant

for the clinical cognitive science of disorders about which

neurobiological knowledge is advancing. Current research on OCD, for

example, has demonstrated involvement of the serotonergic system and

orbitofrontal-basal ganglia-thalamic pathways, and the disorder is

increasingly seen as having a neurobiological basis (Stein and

Hollander, 1992).

Cognitive science constructs may, however, be helpful in thinking

through the relationship between neurobiological findings and the

psychology of OCD. Several authors have, for example, suggested that

neuroanatomical structures involved in OCD play a role in

goal-feedback mechanisms (Gray, 1982; Rapoport and Wise, 1988).

Disruption of cybernetic control in OCD results in psychological

features such as repetitive symptoms and strategies of seeking control

(Pitman, 1987).

Work on neuropsychiatric impairment and serotonergic dysfunction in

OCD has led Stein and Hollander (1992) to propose that OCD may involve

a biologically based impairment in determining or in evaluating

goal-feedback responses. Some patients, perhaps those with high

neurological soft signs, visuospatial difficulties and rapid erroneous

responses on neuropsychological testing, may have difficulty in

determining goal-feedback responses. Other patients, perhaps those

with serotonergic dysfunction and slow correct responses on

neuropyschological testing, may overevaluate harm associated with

goal discrepancy.

This model suggests that both biological approaches (normalizing

serotonergic function) and psychological approaches (changing harm

assessment) may be helpful in the treatment of OCD. On the other

hand, in patients where goal-feedback response impairment is

associated with evidence of structural brain damage, treatment may be

less successful. The model provides a heuristic for further research

correlating biological, cognitive, and clinical variables in OCD.

d) Psychotic Disorders

Sincle the time of Kraepelin and Jung, psychiatrists have used

techniques from experimental psychology to explore

information-processing deficits in psychotic disorders. More

recently, cognitive science methods have been used to study these

disorders. There is now a large literature on information-processing

in schizophrenia (Magaro, 1980; Neuchterlein and Dawson, 1984;

Saccuzzo, 1986), and some work on mania (Grossman and Harrow, 1991).

In their review, Neuchterlein and Dawson (1984) highlight two types of

information-processing deficits in schizophrenia - poorer detection of

single, highly familiar stimuli during vigilance, and increases in the

time required to recognize an uncomplicated stimulus.

Information-processing tasks may constitute indicators of

vulnerability, may have prognostic value, and may be useful measures

of medication response. Information-processing deficits in

schizophrenia may also be conceptualized in schema terms (Magaro,

1980). Finally, a number of theorists have attempted to integrate

information-processing and neurobiological findings (Neuchterlein and