The Psychological Unconscious

John F. Kihlstrom

 "The future of psychology belongs to your work."

William James to Sigmund Freud, 1909

(Taylor, 1999)

The doctrine of mentalism states that mental states are to actions as causes to effects. When classifying the mental states that are causally implicated in behavior, philosophers and psychologists have generally taken refuge in the threefold "trilogy of mind" proposed initially by Kant, and later adopted by the German and Scottish philosophers: cognition, emotion, and motivation (Hilgard, 1980b). At another level, there are qualia, distinctive states of mind associated with various sensory modalities -- what it is like (Nagel, 1974) to see rather than hear, or to smell rather than taste. Qualia also include the distinctive mental states associated with qualities within modality -- what it is like to see red as opposed to blue, to taste sweet as opposed to sour. However, we rarely experience disembodied reds and sweets: red and sweet are properties of things we encounter in our environment. Put another way, mental states are typically about something. Philosophers refer to this "aboutness" as intentionality, and many of them have argued that mental states are representational in that they are always about something. Indeed, Brentano (1874) went so far as to argue that intentionality is the mark of the mental: all mental states are intentional, and only mental states are intentional.

As psychology developed as an empirical science, research focused on those mental states that were accessible to consciousness. Thus, Wundt, Titchener, and other structuralists who founded the earliest psychological laboratories generally assumed that the mind is able to observe its own inner workings. Their research relied on the method of introspection, by which trained observers attempted to analyze their own percepts, memories, and thoughts into elementary sensations, images, and feelings. This line of scientific inquiry on conscious mental life was interrupted by the radical behaviorism of Watson and his followers, who argued that consciousness was nonexistent, epiphenomenal, or irrelevant to behavior. One of the most salutary by-products of the "cognitive revolution", and the subsequent development of an interdisciplinary cognitive science, has been the revival of interest in consciousness (Hilgard, 1977a, 1980a, 1987).

This is fine so far as it goes, but even the 19th century psychologists had recognized that the mental structures and processes underlying experience, thought, and action were not completely encompassed within the span of conscious awareness. That is to say, consciousness is not all there is to the mind. For example, Helmholtz concluded that conscious perception was the product of unconscious inferences based on the individual's knowledge of the world and memories of past experience. Somewhat later, Freud asserted that our conscious mental lives were determined by unconscious ideas, impulses, and emotions, as well as defense mechanisms unconsciously arrayed against them. Ever since that time, investigators have explored the cognitive and dynamic versions of the unconscious in separate, largely independent, lines of inquiry (Bornstein & Masling, 1998; Burston, 1986; Epstein, 1994; Shevrin, 1998; Shevrin, Bond, Brakel, Hertel, & Williams, 1996).

The dynamic unconscious is sometimes considered to be the intellectual property of psychodynamic approaches to personality and psychopathology that evolved beginning in the 19th century (Ellenberger, 1970; Macmillan, 1997), and especially the psychoanalytic tradition initiated by Sigmund Freud. As defined by Shevrin and Dickman (1980), the dynamic unconscious is psychological, meaning that the terms applied to conscious experience (perception, affect, motive, etc.) are also applicable to unconscious mental life; it is active, meaning that unconscious processes affect ongoing experience, thought, and action; and it is qualitatively different, meaning that unconscious processes are organized differently, and follow different procedural rules, than their conscious counterparts. In the clinical theory of psychoanalytic psychology, the psychological unconscious is manifest in the formation of symptoms -- bothersome ideas, impulses, and behaviors for which the patient cannot account, and over which he or she has no control. It should be noted that Shevrin and Dickman (1980) actually use the term psychological unconscious, as opposed to unconscious brain events and other physiological processes (see also Kihlstrom, 1984). Because their discussion is explicitly framed by psychoanalytic theory, it seems appropriate to introduce it here. However, it should be noted that except for certain implications concerning the drive-relatedness of unconscious processes, and the unconscious origins of dreams and symptoms, their treatment applies equally well to the cognitive and dynamic views of the psychological unconscious.

Based on his observations of hysterical patients, and his analysis of such phenomena as dreams, errors, and jokes, Freud (1900, Chapter 7) initially proposed a topographical division of the mind into three mental compartments, or "systems", which he called Cs, PCS, and Ucs. The system Cs, or conscious mind, contained those thoughts, feelings, motives, and actions of which we are phenomenally aware at the moment. Consciousness was explicitly likened to a sensory organ capable of perceiving other mental contents. The system PCS, by contrast, contained mental contents not currently in conscious awareness, but which were available to consciousness, and which could be brought into awareness under certain conditions. Finally, the system Ucs contained mental contents that are unavailable to consciousness -- that could not enter awareness under any circumstances. According to Freud, contents are exchanged between the systems PCS and Cs by virtue of cathexis -- by having attention paid to, or withdrawn from, them; contents residing in the system Ucs were kept out of (or expelled from) the system PCS by means of repression. As others (e.g., Erdelyi, 1985) have noted, this topographical model, with its spatial metaphors, may be read as an anticipation of modern multistore models of human information processing.

Freud maintained this account of the vicissitudes of consciousness for approximately two decades (Freud, 1912, 1915, 1915-1917), but then introduced a wholesale revision of his view, shifting from a topographical to a functional analysis of the mind (Freud, 1923, 1940). This new account postulated three different types of mental activity, rather than three different storage structures: the id, ego, and superego. The id was described as the seat of the instincts, which were expressed through either the automatic discharge of reflex action, or the hallucinatory wish-fulfillment of primary process thought. The ego is concerned with the external physical environment, and discovers reality by means of the logical operations of secondary process thought. The superego, similarly, is concerned with the constraints on instinctual expression imposed by the moral values of the external social environment.

Although it might seem natural to graft the topographical model onto the functional one, such a connection proved untenable. The id is strictly unconscious, and except in cases of psychosis can be known only through inference. By the same token, consciousness is necessarily a quality of the ego -- after all, the ego functions expressly to permit us to become aware of external reality. At the same time, however, the defense mechanisms are also part of the ego, and their operations are not accessible to consciousness; and since the ego cannot be conscious of all of external reality at once, some of its contents (and, correspondingly, of the superego) must necessarily be preconscious.

The problem of reconciling the two different divisions of the mind, topographic and functional, was not solved by Freud before he died. Nevertheless, his assignment of some nonconscious mental functions to the ego, in both its defensive and nondefensive spheres, initiated an important research tradition within post-Freudian psychoanalysis. Beginning with the work of Anna Freud, and especially in the hands of Heinz Hartmann, David Rapaport, and George Klein, psychoanalytic ego psychology focused on the nondefensive, reality-oriented tasks of the ego (Kihlstrom, 1988, 1994). The research of the ego psychologists dealt with conventional topics of perception, memory, and thinking, and in many respects it resembled that being performed elsewhere in academic laboratories. In other respects, however, their work was quite different: it favored prose over nonsense syllables as stimulus materials, for example, took images and dreams seriously, and emphasized the interplay of emotional, motivational, and cognitive processes. The tradition of psychoanalytic ego psychology was linked most closely with mainstream experimental psychology by the work of Bruner, Klein, and others on the "New Look" in perception an attendant research on such topics as subliminal perception, perceptual defense and vigilance, and repression-sensitization (Bruner & Klein, 1960; Erdelyi, 1974, 1985).

In the present context, the most important feature of psychoanalytic ego psychology is that it took seriously the question of the psychological unconscious, and of the relations between conscious and nonconscious mental processes, at a time when most academic psychologists had difficulty taking consciousness itself seriously. A sort of manifesto for this viewpoint has been offered by Shevrin and Dickman (1980), who review a number of studies of selective attention, subliminal perception, and event-related potentials in support of two broad propositions: (1) that the initial stage of human information processing is outside of consciousness, psychological in nature, active in its effect on consciousness, and operates on principles that are qualitatively different from those governing conscious cognition; and (2) that representation of a mental event in consciousness is jointly determined by stimulus, state, and motivational factors.

At the same time, it should be noted that while Shevrin and other ego psychologists locate their research and theorizing squarely within the Freudian tradition, little if any of their evidence bears directly on the propositions of classical psychoanalysis -- a point made by Shevrin and Dickman themselves (1980). In the first place, most work on selective attention and event-related potentials bears on mental states and processes that are preconscious, and do not address questions of nonconscious mental life -- adopting Freud's usage of these terms. For example, demonstrations of parallel processing at early stages of perception, while arguably evidence for qualitative differences between conscious and nonconscious cognition, do not perforce support a distinction between primary- and secondary-process thinking. Even research on perceptual defense and repression, while clearly relevant to the effects of emotion and motivation on cognition, rarely go beyond events that are merely unpleasant to tap the primitive sexual and aggressive contents that Freud attributed to the id --a criticism offered by Rapaport (1942) more than a half-century ago.

The undoubted success of ego-psychological research on preconscious (and even unconscious) mental life, while having its origins in neo-Freudian psychoanalysis, does not thereby support the essential propositions of psychoanalytic theory. This is because precisely the same propositions are offered, implicitly or explicitly, by cognitive theories that evolved independently of, and owe no intellectual allegiance to, the psychoanalytic tradition. Put another way: research on subliminal perception, motivated forgetting, and the like offers little support for the Freudian conception of nonconscious mental life because the propositions that have been tested are rarely unique to Freudian theory. Such support can only be provided by research that tests those hypotheses that are unique to Freudian theory -- for example, that unconscious contents are sexual and aggressive in nature, and that unconscious processes are primitive and irrational. Such experiments are hard to come by, and positive findings rarer still.

While Freud described the mechanism of the dynamic unconscious as one of repression, his intellectual rival Pierre Janet (1889, 1907) described the process as one of dissociation (actually his term was desagregation). Janet's work on hysteria was overshadowed by Freud's (Perry & Laurence, 1984), and his magnum opus Psychological Automatisms (1889) unfortunately has gone untranslated. For these reasons, Janet's theoretical ideas are known primarily through secondary sources (Ellenberger, 1970; Hilgard, 1977b), and only the briefest account of them can be given here.

Janet's theoretical work was predicated on Claude Bernard's paradigm of analysis followed by synthesis: the study of elementary psychological functions taken separately, and then the reconstruction of the whole mind based on knowledge of these parts. The elementary mental functions were labeled psychological automatisms: far from the elementary sensations, images, and feelings of the structuralists, they were construed as complex intelligent acts, adjusted to their circumstances, and accompanied by a rudimentary consciousness. Each automatism unites cognition, emotion, and motivation with action. Thus, automatisms resemble what some contemporary theorists (e.g., Anderson, 1983) would call productions (or production systems): condition-action units that are executed in response to appropriate contextual cues.

Janet held that under normal circumstances, all psychological automatisms were bound together into a single stream of consciousness: each accessible to introspection, and each susceptible to voluntary control. However, the occurrence of mental trauma, especially in a vulnerable individual, could result in the splitting off of one or more psychological automatisms from conscious monitoring and control. Under these circumstances, there would exist two or more streams of mental functioning (consciousness in James' broad sense), each processing inputs and outputs, but only one of which is accessible to phenomenal awareness and voluntary control. The dissociated automatisms constitute fixed ideas (idee fixe) which possess some degree of autonomy with respect to their development and effects on ongoing experience, thought, and action. The operation of these dissociated (as opposed to integrated or synthesized) psychological automatisms provides the mechanism for the major symptoms of hysteria: they produce the ideas, images, and behaviors that intrude, unbidden, on the stream of conscious thought and action; and their capacity to process information is responsible for the paradoxical ability of the hysterically blind or deaf to negotiate their environments successfully. Janet described these dissociated automatisms as subconscious as opposed to unconscious, and considered repression as just one possible mechanism for dissociation.

Janet's ideas were championed by the American psychologist Morton Prince (1906), and more recently by E.R. Hilgard (1977b), who proposed a "neodissociation" theory of divided consciousness (see also Kihlstrom, 1984, 1992a, 1998a). Neodissociation theory assumes that the mental apparatus consists of a set of cognitive structures similar to Janet's automatisms and Bartlett's (1932) schemata, which monitor, organize, and control both thought and action in various domains. Each of these structures can seek or avoid inputs or facilitate or inhibit outputs. The structures are arranged hierarchically, normally in communication with each other, and linked to a superordinate structure that provides for executive monitoring and control.

As the ultimate endpoint for all inputs, and the ultimate starting point for all outputs, the executive control structure provides the psychological basis for the phenomenal experiences of awareness and intentionality. However, certain conditions can alter the integration and organization of these structures, breaking the links between one or more subsystems, or between a subsystem and the executive. Such a situation would result in a condition of divided consciousness, in which percepts, thoughts, feelings, and actions are processed without being represented in phenomenal awareness. Such circumstances, of course, can lead to phenomena of implicit cognition, and to behaviors that are perceived as involuntary.

Whether in its original or updated forms, dissociation theory provides a rather different view of nonconscious mental functioning than psychoanalytic theory (Hilgard, 1977b; Kihlstrom, 1984, 1992a, 1998a). In the first place, dissociation theory holds that nonconscious mental contents are not necessarily restricted to primitive sexual and aggressive ideas and impulses, nor are nonconscious mental processes necessarily irrational, imagistic, or in any other way qualitatively different from conscious ones; they are simply not consciously accessible. In the second place, dissociation theory holds that the restriction of awareness need not be motivated by purposes of defense, nor need it necessarily have the effect of reducing conflict and anxiety; rather, it can occur simply as a consequence of particular psychological operations. While largely compatible with the principles of contemporary cognitive psychology, dissociation theory also offers a somewhat different perspective on the cognitive unconscious. Thus, nonconscious mental processes are not restricted to unconscious procedural knowledge, and nonconscious mental contents are not limited to unattended or degraded percepts and memories. These differences suggest that dissociative processes deserve more attention by both cognitive and clinical psychologists than they have received in the recent past.

Within 19th century academic psychology, perhaps the most forceful advocate of nonconscious mental life was William James (1890/1980; see also Hilgard, 1969; Kihlstrom & Tobias, 1989; Myers, 1986; Taylor, 1983, 1996). James held that mental states could be unconscious in at least two different senses. First, a mental event can be excluded from attention or consciousness: "We can neglect to attend to that which we nevertheless feel" (1890, 201, 455-458). These unattended, unconscious feelings are themselves mental states. Second, and more important, James drew on the clinical observations of cases of hysteria and multiple personality -- some made by others, some by himself (Taylor, 1983, 1996) -- to argue for a division of consciousness into primary and secondary (and, for that matter, tertiary and more) consciousnesses (sic), only one of which is accessible to phenomenal awareness at any point in time. To avoid possible oxymoron in the negation of consciousness, which was what really bothered him, James preferred to speak of co-conscious or subconscious mental states, rather than unconscious ones.

The radical behaviorists were no more interested in nonconscious than in conscious mental life, so empirical interest in the kinds of problems that interested Helmholtz and James, not to mention Freud, declined precipitously in the years after World War I. For almost half a century, interest in conscious and unconscious mental life was maintained by psychoanalysts, much in the same manner that Catholic monks preserved learning during the Dark Ages.

Serious theoretical interest in nonconscious mental life had to wait the triumph of the cognitive revolution (Hilgard, 1980a, 1987). For example, the classic multistore model of information processing, of the sort proposed by Atkinson and Shiffrin (1968), implicitly makes consciousness coterminous with attention and primary (short-term, working) memory. In this way, the model seems to identify nonconscious mental life with early, "preattentive" mental processes such as feature detection and pattern recognition, that occur prior to the formation of a mental representation of an event in primary memory. By regarding attention and rehearsal as prerequisites for a full-fledged cognitive analysis of an event, and by implicitly identifying consciousness with "higher" mental processes, the classic multistore model leaves little or no room for the psychological unconscious -- complex mental structures and processes that influence experience, thought, and action, but which are nevertheless inaccessible to phenomenal awareness. The giant step -- to the idea that mental states and processes could dynamically influence experience, thought, and action despite being inaccessible to phenomenal awareness and voluntary control -- required a wholesale revision of our concepts of attention and memory, as represented by research on automaticity and implicit memory.

Most research on unconscious mental life has focused on the cognitive corner of Hilgard's (1980b) trilogy of mind -- the cognitive unconscious (Kihlstrom, 1987, 1999a). The rediscovery of the unconscious (Kihlstrom, 1995b) began with comparisons between automatic and effortful mental processes (Kahneman & Triesman, 1984), and between explicit and implicit memory (Schacter, 1987), and it has continued with the extension of the explicit-implicit distinction into the domains of perception, learning, and thought. More recent developments, to be treated in later sections, have involves the extension of the explicit-implicit distinction further, to the domains of motivation (McClelland, Koestner, & Weinberger, 1989) and emotion (Kihlstrom, Mulvaney, Tobias, & Tobis, 1998).

The earliest information-processing theories of attention were based, to one degree or another, on the metaphor of the filter (for a review, see Pashler, 1997). Information which made it past the filter was available for "higher" information-processing activities, while information which did not make it past the filter was not. This same attentional filter was also the threshold which had to be crossed for information to be represented in phenomenal awareness. The filter theories of attention, in turn, raised questions about how permeable the attentional filter was, and how much information processing could occur preattentively. Later, the notion of an attentional filter was replaced by the notion of attentional capacity (for reviews, see Kahneman & Treisman, 1984; Logan, 1997; Pashler, 1997). Capacity theories begin with the assumption that human information-processing capacity is limited, and conclude that the ability to perform one or more task(s) depends both on the resources available and the resources required by the task(s) themselves. Whereas the filter models conceived of information processing as serial in nature, the capacity models implied that several tasks could be carried out simultaneously, so long as their attentional requirements did not exceed available resources.

The capacity view, in turn, led to a distinction between automatic and controlled processes (e.g., LaBerge & Samuels, 1974; Posner & Snyder, 1975; Schneider & Shiffrin, 1977). Automatic processes are inevitably engaged by the presentation of specific stimulus inputs, regardless of any intention on the part of the subject. Some automatic processes are innate, while others have been automatized through extensive practice -- a process of knowledge compilation through which knowledge is transformed from declarative (propositional) to procedural (rule) format (Anderson, 1982; see also Logan, 1988). But in either case, automatic processes are unconscious in the strict sense that they are inaccessible to phenomenal awareness under any circumstances.

The defining feature of an automatic process is that it is executed automatically in response to appropriate stimulus inputs. In this way, the notion of an automatic processes is tacitly modeled after the reflexes, taxes, and instincts (fixed action patterns) familiar from physiology and ethology, as well as the conditioned responses familiar from traditional learning theory (whether Pavlovian or Skinnerian). Of course, such a definition of automaticity is circular. Thus a second criterion, that automatic processes consume no attentional resources, seems to have been adopted in part to escape tautology, and perhaps because of anticipated difficulties in objectively measuring or controlling subjects' intentions. But it should be noted that, at base, the concept of automaticity does not require anything other than independence from intention. It is certainly possible to conceive of automatic processes that, once invoked by appropriate stimulus conditions, consume attentional resources -- just as a room heater, automatically activated by a thermostat, consumes electricity.

Hasher and Zacks (1979, 1984) offered additional criteria for defining a process as automatic. In their formulation, information is automatically processed if the following conditions hold:

The original and revisionist criteria for automaticity have formed the foundation for a number of interesting lines of research. However, it seems advisable to decouple Hasher & Zacks' (1979) additional criteria from the concept of automaticity, and treat the effects of such factors as training, age, and individual differences as empirical questions, as opposed to a priori assumptions.

A further quandary concerns the proper name for the opposite of automaticity. Posner and Snyder (1975) and Bargh (1984) contrast automatic processes with conscious ones; Schneider and Shiffrin (e.g., Shiffrin & Schnieder, 1984) with controlled; Logan (1980) and Bargh (1982) with attentional; and Hasher & Zacks (1979, 1984) with effortful. Each of these contrasts captures something about automaticity, but some seem to represent a priori theoretical commitments that should be expressed as empirical questions. It is not necessarily the case that automatic processes should be unconscious, and that intentional ones should be conscious; or even that automatic processes should consume no attentional resources. Thus, the implicit opposition of automatic, involuntary, unconscious, and effortless processes against those that are controlled, conscious, and effortful leads to a certain amount of uncertainty when classifying particular mental processes.

The concept of automaticity has played an increasingly powerful role in social psychology and personality (e.g., Bargh, 1984, 1989, 1994, 1997; Bargh & Barndollar, 1996; Devine, 1989; Fazio, Sanbonmatsu, Powell, & Kardes, 1986; Newman & Uleman, 1989; Pratto, 1994; Smith, 1994; Taylor & Fiske, 1978; Uleman, Newman, & Moskowitz, 1996; Wegner & Bargh, 1998; Wegner & Smart, 1997). The general argument is that some of the processes involved in social cognition, and some of the processes by which social cognitions are translated into social behavior, are executed automatically. Thus, it is generally accepted that attitudes, impressions, and other social judgments, as well as aggression, compliance, prejudice, and other social behaviors, are often mediated by automatic processes that operate outside phenomenal awareness and voluntary control (Kihlstrom, 1996c).

To some extent, what might be called the automaticity juggernaut within personality and social psychology seems to represent a reaction to a cognitive view of social interaction which seems, to some, to inappropriately emphasize conscious, rational, cognitive processes, at the expense of the unconscious, irrational, emotive, and conative. After all, the concept of automaticity is, at least tacitly, modeled on innate stimulus-response (S-R) connections such as reflexes, taxes, and instincts (fixed action patterns), as well as on those S-R connections acquired through the processes of classical and instrumental or operant conditioning. James (1890/1980), after all, discussed automaticity in relation to habit. Thus, in some respects, invocation of the concept of automaticity represents a reversion to earlier situationist views within social psychology (Berkowitz & Devine, 1995).

Bargh goes on to argue that most social behavior is indeed automatic in nature. In his view, social behavior occurs largely in response to environmental triggers, independent of the person's conscious intentions, beliefs, and choices; and it is preattentive, independent of the person's deployment of attention. Bargh's position is not classically Skinnerian, because he shares the central dogma of cognitive social psychology, which that social behavior is caused by the actor's internal mental representation of the situation, rather than the situation as it might be described objectively. But Bargh goes on to argue that this internal mental representation is itself constructed automatically, and preconsciously. Thus, by advocating "social ignition" over "social cognition", Bargh is able to maintain a superficial allegiance with cognitivism while at the same time harkening back to radical situationism. If the cognitive processes underlying social cognition and social behavior are largely automatic, then -- to put it bluntly -- not too much thought has gone into them.

At the same time, however, and somewhat ironically, the most recent developments in attention theory have been to undermine even the seemingly fundamental assumptions that automatic processes are independent of intention and of attentional capacity (Logan, 1997; Pashler, 1997). For example, very extensive practice with a task does reduce its demands for cognitive resources, as measured by central interference with decision making and response selection, but it does not eliminate these demands entirely. While practice improves the efficiency with which operations such as search and categorization can be performed, these improvements are highly stimulus-specific. Moreover, even such canonical demonstrations of automaticity as the Stroop color-word effect (MacLeod, 1991), are not entirely involuntary: the subject's task set makes a difference as to whether interference will occur.

Thus, while the concept of automaticity is intuitively appealing, and has proved extremely attractive to both cognitive and social psychologists, the empirical evidence generally fails to support the primary claims about automatic processes: that they are executed involuntarily and consume no cognitive resources. Of course, it is possible that alternative conceptualizations of automaticity will prove more viable than the those based on resource theories of attention. Thus, both Anderson (1982, 1992) and Logan (1988, 1992) have suggested that automaticity be reinterpreted in terms of the cognitive basis of task performance. For Logan, effortful performance relies on algorithms while automatic performance relies on memory. For Anderson, automaticity involves production rules encoded in procedural memory rather than facts recorded in consciously accessible declarative memory. In these theories, whether a mental process is performed unconsciously depends on less on its demands on cognitive capacity, or whether it is executed involuntarily, than on the format in which that process is represented. Proceduralization, rather than automaticity, might be a better framework for discussing unconscious processing in personality and social interaction (Smith, 1994).

While the procedural knowledge structures guiding thought and action may be unconscious, the declarative knowledge structures on which they operate are ordinarily thought to be available to conscious awareness. Thus, we generally assume that people notice and can describe the salient features of an object or event, even if they cannot articulate the way in which those features have been integrated to form certain judgments made about it. However, it is possible that cognitive procedures can operate on mental representations -- percepts, memories, and the like -- that are not themselves accessible to conscious awareness.

This is certainly the case with respect to memory. Consider, for example, studies of the amnesic syndrome resulting from bilateral damage to the medial temporal lobe (including the hippocampus) and diencephalon (including the mammillary bodies). These patients display a gross anterograde amnesia, meaning that they cannot remember events that occurred since the onset of the brain damage; other intellectual functions remain relatively intact. When they study a list of familiar words, and are asked to recall them shortly thereafter, they show gross impairments in memory compared to controls. But quite different results are obtained when they are asked to identify briefly presented words, or to complete a word stem or other fragment with a meaningful word. Not surprisingly, intact subjects show superior performance on trials where the correct response is a word that had appeared on the previously studied list, compared to those where the correct response is an entirely new word. This advantage of old over new items reflects a priming effect of the previous learning experience. However, amnesic subjects also show normal levels of priming, despite the fact that they cannot remember the words they studied (e.g., Warrington & Weiskrantz, 1974; for reviews, see Schacter, 1987; Shimamura, 1989; Squire & Knowlton, 1995). Priming is also spared in the retrograde and anterograde amnesia induced by electroconvulsive therapy for depression (Dorfman, Kihlstrom, Cork, & Misiaszek, 1995; Squire, Shimamura, & Graf, 1985).

In neurologically intact populations, studies of priming in tests involving lexical decision or word-identification (e.g., Jacoby & Dallas, 1981), or word-fragment completion (e.g., Tulving, Schacter, & Stark, 1982) have shown that the magnitude of the priming effect is essentially independent of the subject's ability to recognize the item as having been presented in a previous study session, and largely unaffected by experimental manipulations which enhance recall or recognition (for reviews, see Roediger & McDermott, 1993; Schacter, 1987). Along the same lines, Nelson (1978) showed significant savings in relearning paired-associates which subjects were unable to recall or even recognize from a previous learning experience. Priming and relearning effects such as these show that task performance may be affected by available memories of prior experiences, even though those experiences are not accessible to conscious recall.

On the basis of results such as these, Schacter (1987; Graf & Schacter, 1985) has drawn a distinction between explicit and implicit memory. Explicit memory involves the conscious reexperiencing of some aspect of the past, whereas implicit memory is revealed by a change in task performance that is attributable to information acquired during a prior episode. An increasingly large literature from both patient and nonpatient populations indicates that people can display implicit memory without having any conscious recollection of the experiential basis of the effect (Roediger & McDermott, 1993; Schacter, Chiu, & Ochser, 1993). Implicit memory is, in effect, unconscious memory: mental representations of past events influence current experience, thought, and action in the absence of, or independent of, conscious recollection of those events. The explicit-implicit distinction is tantamount to the distinction between direct and indirect memory (Johnson & Hasher, 1987; Richardson-Klavehn & Bjork, 1988), and between declarative and nondeclarative memory (Squire & Knowlton, 1995). However, the explicit-implicit distinction is preferable, because it centers on the question of the subjects' awareness of the memory in question, rather than their ability to verbalize it or the nature of the memory test (for a further discussion, see Kihlstrom, 1999a).

A study of memory in posthypnotic amnesia, initially construed as bearing on the distinction between episodic and semantic memory (Tulving, 1983), was one of the first to demonstrate a dissociation between explicit and implicit memory (Kihlstrom, 1980). In these experiments, hypnotized subjects memorized a list of unrelated words to a strict criterion of learning, and then received a suggestion that they would not be able to remember the words they had learned. On an initial test of free recall, highly hypnotizable subjects showed a very dense posthypnotic amnesia, remembering virtually none of the words they had previously memorized. Nevertheless, these amnesic subjects were significantly more likely to give list items as responses on a word-association test, compared to carefully matched control items -- a kind of priming effect. A similar priming effect was observed in a second experiment, in which subjects were asked to generate instances of various categories. Dorfman & Kihlstrom (1995) confirmed this priming effect in better designed experiments which compared free association to recall cued by semantic associates. Although the hypnotic subjects displayed implicit memory for their earlier experience, just as amnesic patients do, an important difference is that amnesic patients have not encoded these memories particularly well, as evidenced by the fact that there are no known circumstances under which they can display explicit memory for them. By contrast, posthypnotically amnesic subjects are able to recall their experiences perfectly following administration of prearranged reversibility cue. Thus, for hypnotic subjects the episodic memories remain available for conscious retrieval, by virtue of having been adequately encoded at the outset, even if they are temporarily inaccessible.

Studies of the memory disorders affecting patients with dissociative disorders such as psychogenic amnesia, psychogenic fugue, and multiple personality also reveal phenomena paralleling implicit memory (Kihlstrom, 1999b; Kihlstrom & Schacter, 1995; Schacter & Kihlstrom, 1999). For example, a rape victim studied by Gudjonsson (1979; Gudjonsson & Taylor, 1985) showed electrodermal responses to stimuli related to events that she could not remember. And a case of fugue was solved by asking the patient to randomly dial numbers on a telephone: she unknowingly dialed her mother, who subsequently provided an identification (Lyon, 1985). And in a case of amnesia following homosexual rape, the patient experienced an increase in subjective distress when presented with TAT cards depicting one person attacking another from behind (Kaszniak, Nussbaum, Berren, & Santiago, 1988; see also Schacter, Wang, Tulving, & Freedman, 1982). Finally, although a symmetrical or asymmetrical amnesia between personalities is commonly considered to be a cardinal symptom of multiple personality disorder, some inter-personality transfer may be observed on tests of implicit as opposed to explicit memory (Eich, Macaulay, Loewenstein, & Dihle, 1997; Ludwig, Brandsma, Wilbur, Bendfeldt, & Jameson, 1972; Nissen, Ross, Willingham, MacKenzie, & Schacter, 1989).

Somewhat surprisingly, implicit memory can be preserved, at least to some degree, in general anesthesia. The adequacy of general anesthesia is assessed, in large part, by the surgical patient's inability to remember his or her surgery (Rosen & Lunn, 1987). Although it is extremely rare for surgical patients to remember details of their operations, conversations among members of the medical team, and the like, there is some evidence for implicit as opposed to explicit memory for surgical events (for reviews, see Cork, Couture, & Kihlstrom, 1997; Merikle & Daneman, 1996). For example, Kihlstrom, Schacter, Cork, Hurt, & Behr (1990) gave anesthetized surgical patients auditory presentations of paired associates of the form ocean-water. After awakening, the patients were presented with the cue term and asked to give the response term which had been presented to them: not surprisingly, they were unable to do this. However, when retested under instructions to respond with the first word that came to mind, they showed a priming effect -- generating items from the presentation list at a higher rate than comparable items from a control list that had not been studied. Similar repetition priming effects have been shown in conscious sedation by propofol, which induces a profound anterograde amnesia without loss of consciousness (Cork, Friley, Heaton, Campbell, & Kihlstrom, 1996; for a review, see Polster, 1993).

Upon awakening, people rarely report any memory for events that transpired while they were asleep -- dreams and reveries, brief awakenings, episodes of sleepwalking or sleep talking, presentation of instructional materials, and the like (Bootzin, Kihlstrom, & Schacter, 1990; Ellman & Antrobus, 1991). In fact, such a lack of explicit memory is one of the subjective criteria by which sleep is diagnosed, and constitutes the main evidence against the efficacy of sleep learning (Aarons, 1976). Nevertheless, sleep learning has always been evaluated in terms of the person's conscious recollection of material presented during sleep, raising the question of whether implicit memory for sleep events might be preserved even if explicit memory is not. Despite some early hints of preserved implicit memory for sleep experiences (for a review, see Eich, 1990), the one formal experiment along these lines produced negative results (Wood, Bootzin, Kihlstrom, & Schacter, 1992). The apparent contradiction between the evidence for implicit memory in general anesthesia and the evidence against it in sleep remains to be resolved. However, it should be noted that the anesthesia experiment by Kihlstrom et al. (1990) differed from the sleep experiment by Wood et al. (1992) in terms of the type of implicit memory test involved. The implicit memory spared in anesthesia was repetition priming, while the implicit memory impaired in sleep was semantic priming. It is likely that preserved implicit memory in anesthesia does not extend to semantic priming, and it is possible that repetition priming might be preserved in sleep.

In this light, it should be noted that research on implicit memory, whether in neurologically impaired or intact subjects, has been dominated by studies of repetition priming effects of the sort observed in lexical decision, perceptual identification, stem-completion, and fragment-completion tasks (Roediger & McDermott, 1993). Performance on these tasks does not require any more than the processing of perceptual structure. However, there more to priming than repetition priming. For example, priming effects can be observed on free association and category generation tasks (e.g., Kihlstrom, 1980; Graf, Shimamura, & Squire, 1984), in the absence of explicit memory. Understanding the nature of implicit memory, when it is spared and when it is impaired, and the relations between implicit and explicit memory, requires research on these other expressions of unconscious memory.

Effects analogous to implicit memory may be observed in perception: Just as there are palpable effects on experience, thought, and action of past events that cannot be consciously remembered, so there appear to be similar effects of events that cannot be consciously perceived (Bornstein & Pittman, 1992; Dixon, 1971, 1981; Kihlstrom, 1996a; Kihlstrom, Barnhardt, & Tataryn, 1992).

Some of this research has an explicitly psychodynamic flavor. For example, Shevrin and his colleagues (Shevrin, 1998; Shevrin & Dickman, 1980; Shevrin et al., 1997) have studied the event-related potentials (ERPs) evoked in neurotic patients by words theoretically related to their complaints. By means of a tachistoscope, the words are presented too briefly to be consciously recognized: nevertheless, the patients appear to give differential ERPs to stimuli, depending on whether they are relevant to their complaints. In another line of research, Silverman and his colleagues (Silverman, 1976, 1983; Silverman, Lachmann, & Milich, 1982; Silverman & Weinberger, 1985) have reported that brief tachistoscopic presentations of "symbiotic" messages such as MOMMY AND I ARE ONE, to brief to be consciously recognized, can have effects on the behavior of both psychiatric patients and normal subjects. Shevrin's work focuses on the intensive study of carefully selected cases. Silverman's work has been somewhat more nomothetic in character, but his theoretical predictions have been controversial even among psychoanalysts, and his observations have proved difficult to replicate (Balay & Shevrin, 1988; but see Weinberger, 1989). Nevertheless, the two lines of research are obviously related to nonpsychodynamic work on implicit perception, and serve to illustrate the possibilities afforded by the marriage of psychodynamic theory with experimental method.

Ever since the first demonstration of subliminal perception, by Peirce and Jastrow (1884; see Kihlstrom, Barnhardt, & Tataryn, 1992), a variety of methodological critiques have sought to demonstrate that events cannot be analyzed for meaning unless they have been consciously identified and attended to (for a recent review, see Holender, 1986). Recently, however, a number of compelling demonstrations of preconscious processing have appeared in the literature. For example, Marcel (1983) employed a lexical decision task in which one stimulus word (the prime) is followed by another word (the target), and the subject has to decide whether the target is a meaningful word. Such judgments are facilitated when the prime is also a word, and especially when the prime and target are from the same taxonomic category; but most of these demonstrations have involved primes that were consciously detectable by the subject. Marcel followed his primes with masking stimuli, with the result that subjects were unable to reliably detect the primes. Nevertheless, such primes facilitated performance on the lexical decision task. Since semantic priming obviously requires some degree of semantic processing, it appears that meaning analyses can be performed on stimuli that are themselves outside of conscious awareness. Marcel's essential findings have since been confirmed by a number of investigators (e.g., Reingold & Merikle, 1989).

Despite persisting methodological critiques (e.g., Cheesman & Merikle, 1985, 1986; Erickson, 1960; Holender, 1986; Shanks & St. John, 1994), the available literature clearly supports the proposition that certain aspects of semantic processing can occur in the absence of conscious awareness. For example, Greenwald, Klinger, & Liu (1989) demonstrated the subliminal presentation of an emotionally valenced word (e.g., enemy) could facilitate evaluative judgments of affectively congruent targets (e.g., bad), even though subjects were unable to report the content or even the location of the prime (see also Draine & Greenwald, 1998; Greenwald, Klinger, & Schuh, 1995; Greenwald & Draine, 1998; Greenwald, Draine, & Abrams, 1996). Such a priming effect clearly requires that subjects be able to analyze the connotative meaning of the subliminal prime. Other studies have documented semantic processing of stimuli which, while not strictly subliminal, are processed preattentively by virtue of parafoveal or dichotic presentation (e.g., Bargh, 1992; Eich, 1984).

At the same time, there appear to be strict limits to the processing of subliminal and preattentive events (Greenwald, 1992). Thus, Greenwald and his colleagues have consistently failed to demonstrate evaluative priming effects of two-word phrases -- e.g., enemy loses, an evaluatively positive prime composed of two evaluatively negative words. Similarly, Merikle and his colleagues (Merikle & Reingold, 1992; Reingold & Merikle, 1993) have shown subliminal priming effects for familiar words, but not for pseudowords. Apparently, subliminal semantic processing is possible provided that the processing demands of the task are not too great. This general principle casts into doubt the studies of subliminal psychodynamic stimulation (Silverman, 1976; Silverman & Weinberger, 1985), which appear to require the subliminal analysis of complex verbal messages.

In subliminal perception, the stimulus is degraded by means of tachistoscopic presentation, or masking. In other instances, the stimulus, while not strictly subliminal, is degraded by virtue of presentation outside of the focus of attention -- in parafoveal segments of the visual field, for example, or over the unattended channel in dichotic listening experiments. However, there are other circumstances where perception without awareness occurs even though the environmental stimulus is not degraded in any sense. For example, Weiskrantz (1986, 1997) and his colleagues have reported a patient who had extensive damage to the striate cortex of the occipital lobes. Although the patient reported an inability to see, he was nonetheless able to respond appropriately to some visual stimuli -- a phenomenon called "blindsight" (for a review see Campion, Latto, & Smith, 1983, and commentaries). Similarly, patients with bilateral lesions to the mesial portions of the occipital and temporal cortex are unable to consciously recognize previously encountered faces as familiar -- a condition known as prosopagnosia. Nevertheless, there are now several reports indicating that prosopagnosic patients show differential behavioral responses to old and new faces (e.g., deHaan, Young, & Newcombe, 1987; Tranel & Damasio, 1985) -- a dissociation similar to the implicit memory seen in the amnesic syndrome. Similar phenomena have been observed in the visual neglect syndromes resulting from damage to the temporoparietal areas of the cerebral cortex (Bisiach, 1993; Rafal, 1998).

Because perception without awareness extends to cases beyond stimuli which are subliminal or unattended, Kihlstrom et al. (1992; see also Kihlstrom, 1996a) have argued for a distinction between explicit and implicit perception, paralleling the distinction between explicit and implicit memory. Explicit perception entails the subject's conscious perception of some object in the current environment, or the environment of the very recent past, as reflected in his or her ability to report the presence, location, form, identity, and/or activity of that object. Implicit memory refers to any change in the person's experience, thought, or action which is attributable to such an event, in the absence of (or independent of) conscious perception of that event. Research on subliminal and preattentive processing, and on neuropsychological syndromes such as blindsight and neglect, indicates that explicit and implicit perception can be dissociated, just as explicit and implicit memory can be. The term "implicit perception" captures a broader domain than is covered by the term "subliminal perception", because it covers the processing, outside of conscious awareness, of stimulus events which are clearly perceptible in terms of intensity, duration, and other characteristics. It also has the extra advantage of skirting the difficult psychophysical concept of the limen.

Further evidence that implicit perception goes beyond the subliminal and the preattentive are studies of the conversion disorders, once labeled "conversion hysteria" (for reviews, see Hilgard, 1977; Kihlstrom, 1992b, 1999b; Kihlstrom et al., 1992). For example, Hilgard (cited in Hilgard & Marquis, 1940) demonstrated that a patient with functional anesthesia and paralysis could acquire a conditioned finger-withdrawal response in the affected arm, while Brady and Lind (1961) showed that a functionally blind patient nonetheless displayed discriminative responses to visual stimulation. More recently, Sackeim, Nordlie, and Gur (1979) and Bryant and McConkey (1989c) have reported cases of visual conversion disorder in which choice behavior was influenced by visual cues even though the patients were unaware of the visual stimuli in question.

As with implicit memory, further evidence of implicit perception is found in hypnosis. For example, it may be suggested that the hypnotized subject cannot see a particular object in his or her visual field; or that he or she cannot hear what is being said. These suggestions obviously affect explicit perception of visual or auditory stimuli; but they do not necessarily affect implicit perception (Kihlstrom, 1998a). Thus, Hilgard (1977) noted that hypnotic suggestions for analgesia affect reports of experienced pain, but not physiological responses to the pain stimulus. Bryant and McConkey (1989a) found that the visual cues influenced the choice behavior of highly hypnotizable subjects who had received suggestions of total blindness; similarly, visual presentation of disambiguating cues influenced performance when hypnotically blind subjects were asked to spell aurally presented homophones (Bryant & McConkey, 1989b). Along the same lines, hypnotized subjects given suggestions for unilateral deafness nevertheless showed intrusions from the "deaf" ear in a dichotic listening task (Spanos, Jones, & Malfara, 1982).

Implicit perception effects are conceptually similar to subliminal memory effects, in that both reveal the impact on experience, thought, and action of events that are not accessible to conscious awareness. However, in contrast to implicit perception, the events contributing to implicit memory effects are clearly detectable by the subject at the time they occurred, attention was devoted to them, and they were at least momentarily represented in phenomenal awareness. Arguably, implicit memory should be reserved for those situations where a consciously perceived event is subsequently lost to conscious recollection, leaving implicit perception for instances (including, in principle, sleep and general anesthesia) where the stimulus was not consciously perceived in the first place. Because memory is the residual trace of perceptual activity, it stands to reason that implicit percepts can reveal themselves in memory -- even if it should turn out that implicit percepts produce only implicit memories. However, evidence for implicit perception and memory should not be taken as grounds for concluding that all current and past events, regardless of whether they are consciously attended, are encoded in memory and influence ongoing experience, thought, and action -- as implied, for example, by the specter of subliminal advertising (Moore, 1988), or subliminal persuasion (Greenwald, Spangenberg, Pratkanis, & Eskenazi, 1991; Merikle, 1988; Merikle & Skanes, 1992; Moore, 1995). To the contrary, a major task for future research is to discover the conditions under which implicit percepts and memories are formed, and those in which they are expressed.

Implicit perception and memory do not exhaust the domain of the psychological unconscious. Although the evidence is somewhat sparse, it appears we can also have implicit thought (Dorfman, Shames, & Kihlstrom, 1996; Kihlstrom, Shames, & Dorfman, 1996). Implicit thought is somewhat hard to define, but it is illustrated by some studies of problem solving by Bowers and his associates (Bowers, 1984, 1987; Bowers, Farvolden, & Marmigis, 1995; Bowers, Regehr, Balthazard, & Parker, 1990). In some of these experiments, subjects were presented with word triads patterned after those of the Remote Associates Test (RAT; Mednick, 1962), and instructed to think of a word that all three words had in common. Some of the triads were soluble, but others were not. In Bowers' adaptation of the RAT, subjects were presented with both kinds of triads simultaneously, as follows:

Triad A Triad B

PLAYING STILL

CREDIT PAGES

REPORT MUSIC

If they could not generate the solution to the soluble triad, they were forced to indicate which was soluble. Bowers et al. found that subjects could distinguish between soluble and insoluble triads, even though they could not solve the soluble triad.

These subjects in Bowers et al.'s (1990, 1995) experiments seemed to be responding to a "feeling of knowing" analogous to that observed in metamemory tasks (Metcalfe & Shimamura, 1994; Nelson, 1996; Reder, 1996). Their choices are clearly being guided by something that is neither a percept (the solution is not being presented to them) or a memory (the solution has not been presented in the past). For convenience, that "something" may be regarded as akin to a node in semantic memory, with the understanding that such a node can include perception-based representations such as images as well as meaning-based representations such as ideas and beliefs.

The implication of Bowers et al.'s (1990) experiments is activation spreads from nodes representing the RAT stimuli to nodes representing the RAT solutions, and that this latter activation can influence subjects' choice behavior, even though it does not rise to the level of conscious awareness. If this is the case, then we would expect unsolved RAT solutions to produce priming effects analogous to those observed in studies of implicit perception and memory. A series of lexical decision experiments by Shames (1994) indicates that they do (see also Dorfman et al., 1996; Kihlstrom et al., 1996). Interestingly, solved RAT items generally fail to produce priming, an outcome which Shames has interpreted as a kind of Zeigarnik effect: activation from unsolved problems persists, while activation from solved problems dissipates rapidly.

The kinds of effects observed by Bowers et al. (1990, 1995) and Shames (1994) seem relevant to the phenomena of intuition, incubation, and insight in problem solving (Dorfman et al., 1996; Kihlstrom et al., 1996). These phenomena have proved difficult to study under controlled laboratory conditions, and intuition has acquired an especially bad reputation as a source of error in human judgment (Nisbett & Ross, 1980). On the contrary, Bowers has argued that intuitions represent our tendency, as intelligent problem solvers, to go beyond the information given by a problem or a retrieval cue. As the way out of the closed cognitive loop of induction and deduction, intuitions are important elements in the creative process. In the present context, intuitions should be reconstrued as implicit thoughts -- gut feelings that we are correct, without knowing why, or even whether, we are right. Perhaps these implicit thoughts come into awareness through the process of incubation, culminating in insight -- the moment in which the solution to a problem, or some other thought, appears in conscious awareness.

Despite persisting questions, implicit perception and memory illustrate the cognitive unconscious, by showing perception and memory outside of phenomenal awareness. A rather different line of research has sought to document the conceptually related phenomenon of implicit learning -- as demonstrated by subjects' ability to use rules acquired through experience, in the absence of awareness of the rules themselves. In some ways, implicit learning is exemplified by language acquisition, where speakers acquire the ability to distinguish grammatical from ungrammatical utterances, even though they cannot articulate the grammatical rules underlying the judgments (Chomsky, 1980). Reber (1976, 1993) has attempted to model this process in the laboratory by developing artificial grammars whose rules control the construction of well-formed strings of letters. In Reber's procedure, subjects are asked to memorize a set of (perhaps) 20 grammatical letter strings (e.g., PVPXVPS or PTTTVPS). They are then tested with a number of new strings, some of which (e.g., PTTTTVPS) conform to the rule, while others (e.g., PTVPXVSP) do not. Reber has found that subjects are able to distinguish grammatical from nongrammatical letter strings at better than chance levels, even though none of them are able to give a full and accurate account of the grammatical rule which they have clearly induced from the study set.

Other investigators have produced similar sorts of demonstrations for comprehensive reviews, see Berry & Dienes, 1993; Seger, 1994). For example, Lewicki (1986) obtained effects similar to those of Reber (1993) in tasks involving social categorization and judgment. Subjects can learn to control the behavior of complex systems, without being able to specify the formula which relates inputs to outputs (e.g., Broadbent, FitzGerald, & Broadbent, 1986). Subjects can learn to predict the forthcoming events without being able to specify the sequential structure underlying those events (Nissen & Bullemer, 1987).

However, it should be noted that the interpretation of implicit learning in terms of the unconscious acquisition of knowledge remains somewhat controversial (Dulany, 1997; Shanks & St. John, 1994). In the first place, the subjects are by no means unconscious in the sense of being asleep or anesthetized. Nor is the learning experience inaccessible to conscious awareness in the same sense that the events implicated in implicit perception and implicit memory are. Even the claim that subjects that implicit learning are unaware of what they have learned is controversial. In the artificial grammar experiments, for example, the mere fact that subjects cannot articulate the Markov process by which grammatical strings were generated does not mean that they are unaware of what they have learned. Above-chance classification performance could well result from partial knowledge which is consciously accessible. The best that can be said, for now, is that the subjects in artificial grammar and sequence learning experiments experience themselves as behaving randomly, without an awareness of what they are doing.

As psychology shrugged off radical behaviorism, its renewed interest in conscious (and then unconscious) mental life was focused on cognition, and it treated cognition as cold and hard, conscious and deliberate. As the cognitive revolution developed, however, two trends emerged. On the one hand, as discussed above, cognitive psychology made increasing room for the cognitive unconscious, as reflected in the rise of research on automaticity and on implicit memory. On the other hand, largely under the influence of personality, social, and clinical psychology, the study of cognition expanded to include the hot and the wet -- reflected in increased interest in emotional and motivational influences on memory and other cognitive processes. This second trend seems to have eventuated in an "affective revolution", in which emotional life is studied in its own right, and not merely as a byproduct of cognitive processing. But this affective revolution, epitomized by the emergence of an interdisciplinary "affective science" (Ekman & Davidson, 1994) or "affective neuroscience" (Panskepp, 1991) modeled on cognitive science and cognitive neuroscience, seems to be focused on conscious feeling states. If we are ready to accept the notion of a cognitive unconscious, perhaps we are also ready to accept the notion of an emotional unconscious as well (Kihlstrom et al., 1999).

Of course, the idea of an emotional unconscious is not new. As we all know, Sigmund Freud argued that our conscious experience, thought, and action is shaped by emotional and motivational states of which we are unaware. All the classic Freudian defense mechanisms were designed to render us unaware of our true emotional states. However, in order to talk about the emotional unconscious we need not embrace the whole conceptual panoply of classical, or even neoFreudian, psychoanalysis -- we don't need the division of the mind into id, ego, and superego, the theory of infantile sexuality, the stages of psychosexual development, repression, or any of the rest of it. Modern research on cognition and the cognitive unconscious owes nothing whatsoever to Freud, and that is also the case with modern research on emotion and the emotional unconscious.

With respect to the emotional unconscious, the first thing to be noted is that conscious emotional responses can serve as expressions of implicit memory and perception, and perhaps implicit learning and thought as well. In both cases, the people in question are consciously aware of their feeling state, but are unconscious of the source of those emotions in their past or current experience.

On the memory side, brain-damaged, amnesic patients can acquire new emotional responses through experience, even though they cannot consciously remember the experiences themselves. For example, a study by Johnson and her colleagues exposed alcoholic Korsakoff syndrome patients, who suffer an anterograde amnesia as a result of bilateral damage to the diencephalon, to unfamiliar Korean melodies (Johnson, Kim, & Risse, 1985). Some melodies were played only once during the study phase, while others were played 5 or 10 times. Later, the patients were played these same melodies, along with other Korean melodies that were entirely new, and asked to indicate which they preferred. Both amnesic patients and control subjects preferred the old over the new melodies, reflecting what Zajonc (1968) has called the "mere exposure effect" (for a review, see Bornstein, 1989). However, the patients, being amnesic, showed greatly impaired levels of recognition: they liked what they heard, but they didn't know why.

And with respect to perception, we now know that intact subjects can show mere exposure effects on preference judgments even though the exposures were so degraded as to be consciously imperceptible. A case in point is a study by Kunst-Wilson and Zajonc (1980), involving tachistoscopic presentations of drawings of irregular polygons. The subjects in this case were neurologically intact, but the exposures were so brief that they were not consciously perceived by the subjects, as confirmed by a later recognition test. Nevertheless, the subjects showed the mere exposure effect: the more subliminal presentations the stimuli received, the more the subjects liked them (see also Bornstein, 1992). The subjects liked what they saw, but they didn't know why (see also Seamon, Brody, & Kauff, 1983; Seamon, Marsh, & Brody, 1984).

Bornstein and his colleagues have extended the subliminal mere exposure effect found with neutral stimuli to faces: not only did subjects show more positive attitudes toward people depicted in tachistoscopically presented photographs, but they also interacted more positively with these same individuals when they later encountered them in a contrived social interaction (Bornstein, Leone, & Galley, 1987). Zajonc (1980) has used these results to claim that affective responses are independent of, and perhaps even prior to, cognitive processing. However, Mandler and his colleagues showed that mere exposure, outside of awareness, also increased ratings of brightness, darkness, and disliking (Mandler, Nakamura, & Van Zandt, 1987). This finding suggests that the preference effect of Kunst-Wilson and Zajonc (1980) seems to be a specific instantiation of a more general principle that activation of an internal representation of an object affects judgment about any relevant dimension that object (Mandler et al., 1987), and does not support specific claims concerning the priority of affect (Lazarus, 1984; Zajonc, 1984). On the other hand, Seamon and his colleagues have recently reported subliminal exposure effects on liking and disliking judgments, but not on judgments of lightness and darkness (Seamon, McKenna, & Binder, 1998). This discrepancy in the literature remains to be resolved.

Nevertheless, unconscious effects on preference judgments, and other emotional responses, set the stage for other analyses of unconscious influences on personality and social interaction (Cantor & Kihlstrom, 1987; Kihlstrom & Cantor, 1999). Some early research along these lines was reported by Nisbett and Wilson (1977; see also Wilson, 1985; Wilson & Stone, 1985), who argued that people largely lack introspective access to the actual determinants of their judgments and other behaviors (for critiques, see Bowers, 1984; Smith & Miller, 1978). More recently, research by Lewicki (1986; Lewicki & Hill, 1987) has shown that presumably affect-laden information about the features of social stimuli (and the covariations among them) can be acquired through implicit learning, and influence behavior even though it is stored in a form that is inaccessible to conscious awareness.

This much is pretty clear from the available research, although more needs to be done in both arenas. In particular, the acquisition of emotional responses by amnesic patients has not been studied much since Johnson et al.'s (1985) original work. However, in view of the ongoing debate over recovered memories of trauma, it is important to enter a strong cautionary note. The recovered memory literature frequently distinguishes between a conscious "recall memory" and an unconscious "feeling memory", the latter term referring to an emotional response to a current situation which is triggered by an unconscious memory of past trauma (e.g., Frederickson, 1993). The notion of a feeling memory is a throwback to the prepsychoanalytic notion of Breuer and Freud (1893-1895/1955, p. 7) that "hysterics suffer... from reminiscences", and finds some support in experimental demonstrations of emotion as an expression of implicit memory. But there is an important difference: the experimental literature provides independent corroboration of the past emotion-eliciting event -- information which may be rarely available in clinical practice. Nevertheless, clinicians who embrace the concept of recovered memory may inappropriately infer a history of prior trauma from the patient's current emotional state, in the absence of any independent corroborative evidence. This is, of course, a mistake -- the logical mistake of "affirming the consequent" -- a mistake which may lead patients to reconstruct distorted or false memories of their past (Kihlstrom, 1996b, 199b). While there is no question that implicit memories of trauma can, in principle, affect a person's current experience, thought, and action, in the absence of independent, objective, corroboration there is no scientific basis for inferring the past from current emotional symptoms (Kihlstrom, 1997b).

Let us now turn to the other side of the emotional unconscious -- the proposition that there is a formal distinction between two expressions of emotion, explicit and implicit. Paralleling the usage of these descriptors in the domain of the cognitive unconscious, explicit emotion refers to the person's conscious awareness of an emotion, feeling, or mood state; implicit emotion, by contrast, refers to changes in experience, thought, or action that are attributable to one's emotional state, independent of his or her conscious awareness of that state. In terms of measurement, explicit emotion tasks require the subject to reflect on, and report, his or her conscious feeling states; implicit emotion tasks do not.

The inspiration for this idea comes from Lang's (1968) multiple-systems theory of emotion. According to Lang, every emotional response consists of three components: verbal-cognitive, corresponding to a subjective feeling state such as fear; overt motor, corresponding to a behavioral response such as escape or avoidance; and covert physiological, corresponding to a change in some autonomic index such as skin conductance or heart rate. We usually think of these three components or systems as covarying together: when people feel afraid, their heart rates go up and they avoid the fear stimulus. When their fear is reduced, heart rate and avoidance decrease as well. However, Lang has proposed that these three systems are partially independent, so that under some conditions they can move in quite different directions. Rachman and Hodgson (1974; Hodgson & Rachman, 1974) picked up on Lang's theme and applied the term desynchrony to cases where one component of emotional response is dissociated from the others (for critical reviews of desynchrony, see Hugdahl, 1981; Turpin, 1991; Zinbarg, 1998).

Given that both Lang (1968) and Rachman and Hodgson (1974) were writing from a tradition of behavior therapy, which emphasizes objective measurement, it is perhaps natural that they were most interested in desynchronies between the behavioral and physiological components of emotion. In the present context, however, interest adheres only to cases that represent the emotional analog of the explicit-implicit distinction in memory: where the subjective component of an emotion, the conscious feeling state, is absent, while the behavioral and/or physiological components persist outside of phenomenal awareness.

One illustration of the dissociation between explicit and implicit emotion comes from hypnosis: the phenomenon of hypnotic analgesia (Hilgard & Hilgard, 1975). Following appropriate suggestions, many highly hypnotizable subjects report feeling no pain when exposed to normally painful stimulation. Although hypnotic analgesia may be construed as a special case of sensory anesthesia, there is more to pain than mere sensation. In addition to sensory pain, which provides information about the location and severity of an irritation or injury, there is an explicitly emotional component of suffering which is not present in the other skin senses, such as touch and temperature (Melzack & Torgerson, 1971). Sensory pain and suffering are dissociable in terms of subjects' pain ratings, and they appear to be mediated by different brain systems. The important thing, however, is that hypnotic analgesia can eliminate both sensory pain and emotional suffering.

Hilgard (1977) has proposed that hypnotic analgesia is mediated by an amnesia-like dissociative barrier which blocks the subject's conscious perception and awareness of pain, but does not impair the registration of the pain stimulus outside of conscious awareness. Accordingly, we would expect to see evidence of this registration in implicit indices of pain that do not require conscious awareness. And this is just what we find: hypnotic analgesia has little effect on physiological responses to pain stimulation (Hilgard & Hilgard, 1975). The diminution of sensory pain and suffering, along with the preservation of physiological responses to the pain stimulus, is exactly the sort of desynchrony we are seeking between explicit and implicit emotion.

Apparently, dissociations between emotional awareness and physiology are found quite commonly in the anxiety disorders. For example, cardiology clinics frequently encounter patients who complain of tachycardia, but who have no other signs of coronary arrest. It turns out that these patients are not having heart attacks at all. Instead, they are having panic attacks, even though they experience no subjective fear (aside from distress over the heart symptom itself). This syndrome even has a name: fearless panic attacks (Beitman, Mukerji, Russell, & Grafing, 1993). The patient is showing all the physiological signs of fear, but doesn't experience fear itself.

The emotional deficits associated with schizophrenia also have a flavor of desynchrony. Thus, flat affect refers to a deficit in the behavioral expression or display of emotion, which may not extend to subjective experience or physiology. Interestingly, Kring and Neale (1996) have shown that schizophrenics express less emotion than normals in response to emotional film clips, even though they do not differ in terms of self-reports of emotional experience. Of course, this is the opposite of the dissociation of interest in this chapter, because the conscious experience of emotion is not impaired. However, the dissociation of interest may well be found in anhedonia, another feature of schizophrenia (and a dimension of normal personality as well; see Chapman, Chapman, & Raulin, 1976): a deficit in the conscious experience of positive emotion which leaves the behavioral or physiological expressions of emotion unimpaired.

A whole host of individual differences in emotional experience and expression which involve just this form of desynchrony. For example, Weinberger, Schwartz, and Davidson (1979) were interested in a group of subjects, labeled repressors, who reported low levels of trait anxiety, as shown on the Taylor Manifest Anxiety Scale but high levels of defensiveness on the Marlowe-Crowne Social Desirability Scale. Despite their general denial of distress, Weinberger et al. found, and Asendorf and Scherer (1983) subsequently confirmed, that these subjects showed increased response latencies and elevated levels of physiological response when asked to read sexual and aggressive verbal phrases. Perhaps people with a repressive coping style have a talent for desynchrony: they may not experience high levels of stress, even though their physiology is churning away anxiously. Unfortunately, despite the development of new measures of repressive tendency (Weinberger, 1997), there has been no followup of this early evidence of implicit emotion.

Other individual difference variables may also be relevant to the emotional unconscious. For example, in alexithymia, which people have difficulty describing their emotional states, or in discriminating one emotional state from another (Taylor & Taylor, 1997). The restricted emotional life of alexithymics seems at least superficially similar to repressive coping style, and it may well be that they have "no words for moods" because they are not aware of their emotional feelings in the first place. Interestingly, alexithymia is a prominent feature among neurological patients who have undergone hemispheric commissurotomy, suggesting that the division in awareness includes an inability to communicate, via language centers of the left hemisphere, emotion arising from centers in the right hemisphere. The question is whether we can find evidence for implicit emotion in these patients, and in neurologically intact alexithymics as well, in terms of behavioral or physiological indices (Lane, Ahern, Schwartz, & Kaszniak, 1997).

On the developmental side, Lane and his colleagues has proposed that there are five levels of emotional awareness, determined by the manner in which the individual's emotional states are organized (e.g., Lane & Schwartz, 1987, 1992). At the lowest level, the person is aware only of bodily sensations; at the next higher level, the person is also aware of action tendencies; but at neither level is the person aware of emotional feelings as such. This awareness begins to emerge only at the emotional counterpart of Piaget's "preoperational" level, and progresses to awareness of emotional blends and opposites, and finally to awareness of subtle nuances of emotion. According to Lane, emotional development does not necessarily parallel cognitive development, such that some adults, firmly ensconced in formal operations with respect to their cognitive abilities, may have only primitive, sensory-motor emotional reactions -- in other words, no emotional awareness at all. People operating a low levels of emotional awareness, then, might be expected to show a dissociation between explicit emotion, consciously experienced, and implicit emotion, expressed in behavior or physiology.

Turning from personality to social psychology, Greenwald and Banaji (1995) have recently applied the explicit-implicit distinction to the concept of attitude. This is interesting because, as Thurstone noted long ago, emotion is central to social attitudes: they are affective dispositions to favor or oppose certain individuals, groups, or policies, and they are measured on dimensions which have affective connotations: pro and anti, like and dislike, positive and negative, and so on. Classical social psychology assumes that people are aware of their attitudes, which is why attitudes are typically assessed by self-report scales. However, Greenwald and Banaji have suggested that people may possess positive and negative implicit attitudes about themselves and other people, which can affect ongoing social behavior outside of conscious awareness.

A particularly provocative demonstration that explicit and implicit attitudes can be dissociated is provided by Greenwald and Schuh's (1995) analysis of reference citation practices by social scientists engaged in prejudice research. When authors' names are classified into categories of Jewish, non-Jewish (presumably Gentile) and other, Greenwald & Schuh found that authors were 40% more likely to cite colleagues from their own ethnic category. Accepting the stereotype of the average prejudice researcher as liberal and unprejudiced, it seems that ethnic prejudice can influence citation behavior, even if it does not creep into conscious awareness.

An experimental demonstration of implicit attitudes is provided by a series of studies of the "false fame" effect by Banaji and her colleagues. For example, Banaji and Greenwald (1995) asked their subjects to study a list of names of famous and nonfamous men and women. Later, the subjects were presented with a longer list of famous and nonfamous names, including those studied previously, and asked to identify which were famous. Jacoby and his colleagues have found that subjects tend to falsely identify previously studied nonfamous names as famous -- a kind of priming effect roughly analogous to Zajonc's mere exposure effect. But in Banaji's study, the false fame effect (Jacoby, Kelley, Brown, & Jasechko, 1989) was bigger for male than for female names. Apparently, the average subject was more likely to associate achievement with males than with females revealing that they hold negative stereotypes about women.

The notion of implicit attitudes is provocative, but it is one thing to demonstrate the implicit effect of attitudes on tasks that do not require conscious awareness of those attitudes, and something else to demonstrate that explicit and implicit attitudes are actually dissociable. We might like to think that the average social scientist is unprejudiced, and the average college student doesn't embrace gender stereotypes; but that might not be true, and if we actually asked them, they might well own up to their bias and bigotry. Recently, Wittenbrink, Judd, and Park (1997) performed a formal comparison of explicit and implicit racial attitudes. Their white subjects showed priming in a lexical decision task when positive trait words were preceded by the prime white, and negative traits were preceded by the prime black. This is evidence of implicit racial prejudice, but the magnitude of the race-specific priming effect was correlated with scores on a questionnaire measure of racial prejudice. So, there wasn't really any dissociation. Implicit measures may be very useful in studies of attitudes and prejudice, but researchers need to actually test for explicit-implicit dissociations before we accept implicit attitudes as evidence of an emotional unconscious whose contents are different from those which are accessible to phenomenal awareness.

In light of the earlier discussion of "feeling memories", more should be said at this point about the logic of inferring unconscious emotions. We recognize priming effects as evidence of implicit memory because we can trace them to specific objectively observable events, and we can objectively trace the relationship between the prime and the target. Put another way, we can identify an implicit expression of memory because we know what happened to the subject in the past. But by the same logic, in order to identify an implicit expression of emotion, we have to know what emotional state the subject should be experiencing -- which emotional state is being represented, and expressed, outside of conscious awareness.

In principle, this can be accomplished by considering the convergence of four operations on a hypothetical emotional construct (Campbell & Fiske, 1959; Garner, Hake, & Eriksen, 1956; Kihlstrom, 1984; Kihlstrom et al., 1999). First, we need to know the adequate stimulus for emotion -- that is, we need to have a set of stimuli which, under ordinary circumstances, reliably elicit particular emotions in normal subjects. Next, we need to have reliable, valid measures of the subjective, behavioral, and physiological responses to the emotion stimulus. With these in hand, we can investigate the failures of convergence which underlie the concept of desynchrony. For example, if in the presence of a known fear stimulus, a person shows the behavioral and physiological components of fear, such as appropriate facial expressions and elevated heart rate, without the subjective experience of fear, we can say that we have demonstrated a dissociation between explicit and implicit emotion. Of course, such an inference crucially depends on our being secure in the knowledge that particular environmental stimuli typically evoke particular emotional states, and that these emotional states are typically accompanied by particular patterns of behavioral response and physiological activity.

Still and all, at least in principle, the emotional unconscious has two different aspects. On the one hand, we may be unaware of the percepts, memories, and thoughts which give rise to our emotional feelings. In this case, emotion serves as an implicit expression of perception, memory, and thought. On the other hand, we may be aware of what we are perceiving, remembering, or thinking, but unaware of the emotions instigated by these cognitions. In this case, behavioral and physiological changes serve as implicit expressions of emotion.

Interestingly, both aspects of the emotional unconscious are anticipated in the neuropsychological model of fear recently offered by LeDoux (1995). LeDoux proposes that fear stimuli are processed by the amygdala, which in turn generates appropriate behavioral, autonomic, and endocrine responses. Cortical arousal, feedback of somatic and visceral information, and information about the fear stimulus are then integrated in working memory to generate the subjective experience of being afraid. Thus, if for some reason the eliciting stimulus is not represented in working memory, the person will experience fear without being aware of the fear stimulus -- in this case, emotion will serve as an implicit expression of perception or memory, as described earlier. Alternatively, a disconnection between the amygdala and the cerebral cortex can produce a dissociation between explicit and implicit emotion: the person will behave in a fearful manner without feeling fear or anxiety.

The fact that a popular and powerful neuropsychological model of emotion can produce both aspects of the emotional unconscious is, in my view, warrant to pursue the matter further. Still, it must be remembered that LeDoux's model is based almost entirely on animal research on fear. It would be useful to have know more about dissociable neural systems for emotions other than fear, and to have positive evidence of implicit emotion in humans, who can talk to us about their conscious experiences. For the present, the experimental and clinical evidence for a dissociation between explicit and implicit emotion is not yet convincing, and the methodological requirements for such demonstrations have not yet been fully met. But while the hypothesis of unconscious emotional states has not yet garnered convincing support, it can no longer be rejected out of hand. If we are willing to speak of implicit percepts, memories, and thoughts that are dissociated from their explicit counterparts, then we must be willing to speak of implicit emotions in the same terms.

And if we are willing to speak of implicit emotions, we must also be prepared to speak of implicit motives. Although Kant asserted that feeling and desire (along with knowledge) were irreducible faculties of mind (Hilgard, 1980b), emotion and motivation are often closely linked -- as evidenced, for example, in the very title of the journal Motivation and Emotion (Kleinginna & Kleinginna, 1981a, 1981b). Often, one of these terms is defined, at least in part, by the other. Thus, the motivational states which energize, direct, and select behavior often have a hedonic quality of pleasure or unpleasure to them, while emotions can have the same drive or incentive functions traditionally ascribed to motives. Buck (1985) defined emotion as the mechanism by which we read out information concerning motivational systems. Viewed from this perspective, emotions might be construed are consciously accessible while motives may be unconscious. But we have already determined that, at least in principle, emotions can be unconscious too; and certainly, many of our motivational states are accessible to phenomenal awareness. Thus we are returned to the question: can motivational states be unconscious?

Put more precisely, can we observe dissociations between explicit and implicit expressions of motivation? In parallel with the cognitive and emotional cases, we must entertain a formal distinction between two expressions of motivation, explicit and implicit. Explicit motivation can be defined as the conscious representation of a conative state, or the desire to engage in some particular activity, as represented by craving for food, yearning for love, and the like. By contrast, implicit motivation refers to changes in experience, thought, or action that are attributable to one's motivational state, independent of his or her conscious awareness of that state. In terms of measurement, explicit motivation tasks require the subject to reflect on, and report, his or her conscious desires; implicit motivation tasks do not. Closing the parallel with emotion, we might hypothesize that behavioral or physiological signs of motivation can be dissociated from conscious desires.

On one level this is obviously true. Consider, for example, such basic biological motives as hunger and thirst. The neuroendocrine systems that maintain homeostasis are largely subcortical, and homeostatic behaviors such as eating and drinking may be observed in animals whose nervous systems are probably not complex enough to support introspective phenomenal awareness (Rosenzweig, Leiman, and Breedlove, 1996). These same primitive homeostatic mechanisms exist in humans, who in addition possess the symbolic capacity for conscious representation of motivational states. Therefore, it might be possible, at least in principle, for a person to eat or drink, in (implicit) response to changes in levels of cell fluids or blood sugar, without (explicitly) feeling hungry or thirsty.

In a trivial sense, of course, even causal observation of everyday life clearly indicates that people are perfectly capable of eating and drinking when they are not hungry or thirsty. H.M. and other amnesic patients will continue to eat food that is offered to them, long after their caloric needs have been satisfied (Rozin, Dow, Moscovitch, & Rajaram, 1998). Eating and drinking can occur in the absence of conscious hunger and thirst because eating and drinking are not just homeostatic behaviors: they are also social behaviors, engaged in for social purposes. But social eating and drinking, as described, do not represent unconscious motivation: they represent behaviors whose motivational sources lie elsewhere than in homeostatic self-regulation -- sources which may well be accessible to conscious awareness.

A better example of genuinely unconscious motivation may be found in the phenomenon of posthypnotic suggestion -- a phenomenon which, since the time of Freud, has served as a prime examine of unconsciously motivated behavior. (Ellenberger, 1970). In posthypnotic suggestion, a subject carries out, after hypnosis has been terminated, a response which was suggested while he or she was hypnotized (Barnier & McConkey, 1996; Sheehan & Orne, 1968). The administration of a posthypnotic suggestion is typically accompanied by a suggestion for posthypnotic amnesia -- a temporary inability to remember, after hypnosis, the events and experiences which transpired while the subject was hypnotized. In this instance, which is the classic case, the subject may recognize that he or she is doing something unusual, but does not know why. Although posthypnotic responses appear to break into the stream of the subject's behavior, and to be outside volitional control, it should be understood that they are not automatic in the traditional, technical sense of that term. For example, their execution is not invariant across environmental conditions (Spanos, Menary, Brett, Cross, & Ahmed, 1987). Moreover, their execution consumes attentional resources (Hoyt & Kihlstrom, 1988). Posthypnotic responses are not experienced as involuntary because they are automatic, and thus outside the subject's conscious control, but because their motivational origins, in the hypnotist's suggestions, are outside the subjects's conscious awareness (Kihlstrom, 1992a).

Some evidence for the claim that posthypnotic suggestions are outside the subject's awareness is provided by an experiment by Bowers (1975) in which subjects received suggestions designed to alter their preferences for painting style. After establishing subjects' baseline preference for portraits or landscapes (Phase 1), they were hypnotized and received a suggestion to prefer the opposite (i.e., landscapes or portraits). This suggestion was covered by amnesia, and the subjects were brought out of hypnosis. In the course of testing their response to the posthypnotic suggestion (Phase 2), half the subjects were subjected to a verbal-conditioning procedure in which the experimenter additionally shaped their preferences by means of appropriate contingencies of reinforcement. In Phase 3, when the posthypnotic suggestion was canceled and reinforcement contingency discontinued, the subjects who received both the suggestion and the reinforcement reverted to their baseline preferences. By contrast, those who received only the suggestion maintained their new bias. In line with social-psychological theories of attribution and self-perception, Bowers concluded that the subjects in the suggestion-plus-reinforcement group, aware of the not-very-subtle experimental contingencies in Phase 2, discounted that behavior and changed it accordingly as soon as the contingencies were changed. By contrast, the subjects in the suggestion-alone group, unaware of the suggestion, attributed their choices in Phase 2 to their intrinsic preferences, and maintained them accordingly.

In the recent history of psychology, the notion of implicit motivation was first articulated by McClelland, Koestner, and Weinberger (1989) -- interestingly, without any reference to the already-emerging concept of explicit memory (Schacter, 1987). For McClelland et al., explicit motives are self-attributed: the person is aware of the motive, can reflect on it, and an report it in interviews or on personality questionnaires. Implicit motives, by contrast, are inferred from the person's performance on such exercises as the Thematic Apperception Test (TAT). As such, the distinction between explicit and implicit motives is an extension of McClelland's (1980) earlier distinction between respondent and operant motive measures.

Of course, it has long been known that motives as assessed by "projective" instruments such as the TAT, even when investigators employ reliable coding schemes, do not correlate with nominally the same motives as assessed by "objective" instruments such as the Personality Research Form (PRF; Jackson, 1974). Rather than taking such empirical findings as a reason for abandoning picture-story and other projective measurements, McClelland et al. (1989) conclude that implicit and self-attributed motives influence different classes of behavior, and that they respond to different types of influence (e.g., Bornstein, 1998). Implicit motives are more strongly related to long-term behavioral trends, while self-attributed motives are more strongly related to immediate choices. Self-attributed motives are more strongly linked to normative goals than are their implicit counterparts. Self-attributed motives are aroused by extrinsic social demands, while implicit motives are aroused by intrinsic task incentives.

In other words, McClelland et al. postulate two dissociable motive systems, one explicit and the other implicit, just as Schacter (1987) and Squire (Squire & Knowlton, 1995) postulate two dissociable memory systems or Lang (1968) postulates multiple, dissociable, components of emotion. The low correlations between TAT and PRF motive scores, far from reflecting the poor psychometric properties of the TAT (or, for that matter, the PRF), instead reflect the dissociability of the underlying motivational systems which these measures respectively tap. One of these motive systems is accessible to conscious awareness; the other is not, and influences the individual's experience, thought, and action unconsciously. By virtue of implicit motives, people engage in goal-oriented behavior without being aware of what their motives or goals are. Or, at least, that is the hypothesis.

A rather different perspective on the motivational unconscious is offered by Bargh (1997; Bargh & Barndollar, 1996), as part of his general embrace of the concept of automaticity. According to the traditional, folk-psychological model of motivation, the person consciously selects some intended behavior in order to achieve some goal, and then deliberately executes that behavior. While it is commonly accepted that some skilled goal-directed behaviors are executed automatically and unconsciously, much like a concert pianist plays an arpeggio, Bargh also automates the process of goal-selection -- the selection of the music, not just the touch of fingers to keys. According to this auto-motive model, by virtue of having been frequently and consistently chosen in particular situation, goals and motives (these terms are essentially interchangeable) themselves can be automatically and unconsciously invoked by environmental events. Once activated, goal-oriented behaviors can be executed outside of awareness as well.

It should be noted, however, that while the implicit motives discussed by McClelland et al. (1989) are themselves inaccessible to conscious awareness (at least on hypothesis), Bargh's (1997) auto-motive model only asserts that the person's motives are selected automatically, in the absence of conscious intention or choice. It does not necessarily follow that the person is not aware of the motives themselves. Thus, it may very well be that achievement or affiliation goals may be primed by events in the current or past environment (Bargh, Gollwitzer, & Barndollar, 1996; cited in Bargh, 1997); but these goals themselves may well be represented in the person's conscious awareness. In the absence of evidence that the motives themselves are inaccessible to phenomenal awareness, the automatically activated motives envisioned by Bargh are probably better construed as motivational expressions of implicit perception or memory, rather than as implicit motives.

The wide variety of clinical and experimental studies summarized here, conducted in a wide variety of domains and with many different types of subjects, provide evidence for several different aspects of the psychological unconscious. In the first place, there is ample evidence that certain mental procedures, if not strictly automatic, operate unconsciously in the sense that we have no direct introspective awareness of them: they can be known only indirectly, by inference. With respect to the cognitive contents on which these processes operate, there is also ample evidence for implicit memories and implicit percepts, which influence experience, thought, and action independently of, and even in the absence of, conscious perception or recollection. There is also more tentative evidence for implicit thoughts, supporting the experience of intuition in creative problem solving, and for implicit learning processes. There is evidence that emotional responses, in the form of consciously experienced feeling states, can occur as expressions of implicit perception and memory, if not as products of implicit learning and thought as well. Similarly, motivational states can be activated automatically; further research may establish that motives, like emotions, can serve as expressions of implicit cognition. Moreover, setting the cognitive unconscious aside, there are good theoretical reasons to suspect that implicit emotional and motivational states can themselves affect expressive and goal-directed behavior outside of conscious awareness.

Lately, there has been some tendency to claim that these findings prove that Freud was right after all (Bornstein & Masling, 1998; Erdelyi, 1985, 1996; Shevrin et al., 1996). For example, Westen (1998a, 1998b) after performing a review not unlike the present one, has concluded that "[t]he notion of unconscious processes is not psychoanalytic voodoo, and it is not the fantasy of muddle-headed clinicians. It is not only clinically indispensable, but it is good science" (p. 35). True enough, so far as it goes, but Westen ignores the fact that none of the literature he has reviewed bears on the particular view of unconscious mental life offered by Freud. The fact that amnesic patients show priming effects on word-stem completion tasks, and can acquire positive and negative emotional responses to other people, without having any conscious recollection of the experiences responsible for these effects, cannot be offered in support of a theory that attributes conscious behavior to repressed sexual and aggressive urges. None of the experiments reviewed involve sexual or aggressive contents, none of their results imply defensive acts of repression, and none of their results support hermeneutic methods of interpreting manifest contents in terms of latent contents. To say that this body of research supports psychoanalytic theory is to make what the philosopher Gilbert Ryle called a category mistake.

It is true, as noted earlier, that the notion of a psychological unconscious, which long predated Freud (Ellenberger, 1970; Klein, 1977; Whyte, 1960), was conserved by scientific and clinical psychoanalysts throughout the dark days of behaviorism, and into the early days of the cognitive revolution, when cognitive psychologists were preoccupied by manifestations of conscious mental life, such as attention, short-term memory, and mental imagery. However, the revival of research on the psychological unconscious, in the late 1970s and early 1980s, was essentially independent of psychoanalysis. With the possible exception of Silverman's (1976) work on subliminal symbiotic stimulation, modern laboratory research provides no support for the psychoanalytic view of unconscious mental life. That line of research, in turn, directly contradicts the overwhelming conclusion from carefully controlled empirical research that subliminal and other forms of preattentive processing is analytically limited -- too limited to permit the analysis of symbiotic stimuli.

One response to this state of affairs is to argue that psychoanalytic theory itself has evolved since Freud, and that it is therefore unfair to bind psychoanalysis so tightly to the Freudian vision of repressed infantile sexual and aggressive urges, symbolically represented in dreams, errors, and symptoms, and revealed on the couch through free association. Westen (1998b) himself recently attempted this gambit, arguing that critics of psychoanalysis attack an archaic, obsolete version of psychodynamic theory, and ignore more recent developments such as ego psychology and object relations theory. But, to borrow the language of the Vietnam War, this destroys the village in order to save it. Culturally, the 20th century has been the century of Sigmund Freud, not the century of Heinz Kohut or Melanie Klein. Freud's legacy is not to be assessed in terms of ideas which emerged since Freud died, but rather in terms of the ideas propounded by Freud himself through the 24 volumes of his collected works. Chief among these, as Bornstein and Masling note at the very beginning of this book, is a particular view of unconscious mental life -- a view which, to date, has found little or no support in empirical science.

More positively, the studies reviewed here indicate that consciousness is not to be identified with any particular perceptual-cognitive functions such as discriminative response to stimulation, perception, memory, or the higher mental processes involved in judgment or problem-solving. All of these functions can proceed outside of phenomenal awareness. Rather, consciousness is an experiential quality that may accompany any of these functions. The fact of conscious awareness may have particular consequences for psychological function -- it seems necessary for voluntary control, for example, as well as communicating one's mental states to others, and for sponsored teaching. But it is not necessary for many forms of complex psychological functioning. Moreover, these findings suggest a taxonomy of nonconscious mental structures and processes constituting the domain of the psychological unconscious.

There are, within the domain of procedural knowledge, a number of complex processes that are unconscious in the proper sense -- unavailable to introspection, in principle, under any circumstances. By virtue of routinization (or perhaps because they are innate), such procedures operate on declarative knowledge without either conscious intent or conscious awareness, in order to construct the person's ongoing experience, thought, and action. Execution of these mental processes, which can be known only indirectly through inference, is inevitable and consumes no attentional capacity. They may be described as unconscious in the strict sense of that term -- in short, they comprise the unconscious proper.

In principle, declarative knowledge is available to phenomenal awareness, and can be known directly through introspection or retrospection. However, it is now clear that procedural knowledge can interact with, and utilize, declarative knowledge that is not itself accessible to conscious awareness. Many phenomena of implicit perception, memory, and thought, especially those associated with degraded stimulus processing, suggest a category of preconscious declarative knowledge structures. Unlike truly unconscious procedural knowledge, these aspects of declarative knowledge would be available to awareness under ordinary circumstances. Although activated to some degree by current or prior perceptual-cognitive activity, and thus able to influence ongoing experience, thought, and action, they do not cross the threshold required for representation in working memory, and thus for conscious awareness. These representations, which underlie the phenomena of implicit perception and memory, reside on the fringes of consciousness and changed circumstances could render them consciously accessible -- at least in principle.

Finally, the phenomena of hypnosis, hysteria, and related states seem to exemplify a category of subconscious declarative knowledge. These mental representations, fully activated by perceptual inputs or acts of thought, above the threshold ordinarily required for representation in working memory, and available to introspection and retrospection under some circumstances, seem nevertheless dissociated from phenomenal awareness (Hilgard, 1977). Dissociative phenomena are of theoretical interest because they do not comfortably classify as either unconscious or preconscious. They are not limited to innate or routinized procedural knowledge; their execution is not automatic in the traditional sense, because it is sensitive to context and consumes cognitive capacity. The stimulus input has not been degraded in any way, and the resulting memory traces are fully encoded and available for explicit retrieval. From the point of view of activation notions of consciousness, these phenomena are theoretically interesting because they indicate that high levels of activation, supported by the active deployment of attention and complex mental processing, while presumably necessary for residence in working memory, are not sufficient for conscious awareness.

What, then, is required in order to achieve conscious awareness? At a psychological level of analysis, it seems that conscious awareness requires that a mental representation of an event be connected with some mental representation of the self as agent or experiencer of that event (Kihlstrom, 1993, 1997a). In his discussion of the stream of consciousness, James (1890) wrote that "the first fact for... psychologists is that thinking of some sort goes on" (p. 219). He also wrote, immediately thereafter, that "thought tends to personal form" (p. 220) -- that is, every thought (by which James meant every conscious mental state) is part of a personal consciousness:

In other words, an episode of ongoing experience, thought, and action becomes conscious if, and only if, a link is made between the mental representation of the event itself and some mental representation of the self as the agent or experiencer of that event. This mental representation of self (Kihlstrom & Klein, 1994, 1997), including one's internal cognitive, affective, and conative environment, resides in working memory, as a memory structure, along with coexisting representations of the current external environment (Anderson, 1983). Both self and context representations are necessary for the construction of a full-fledged conscious perception -- which, following James, always seems to take the following form:

I see (or hear, smell, taste, etc.) this, now.

And since memory is the residual trace of perceptual activity, these elements are necessary for the reconstruction of a full-fledged conscious recollections as well.

Within a generic associative network theory of knowledge representation (e.g., Anderson, 1983), an episode of experience is represented by one node connecting three others: an event node, containing a raw description of an event; a context node, specifying the spatial and temporal (and perhaps emotional and motivational) context in which the event occurred; and a self node, indicating the person as the agent or the experiencer of the event. Conscious recollection of such an event occurs only when the representation of the self is retrieved along with some other information about the event. The inability to retrieve the links among all three types of propositions accounts for some of the peculiarities in conscious memory (Kihlstrom, 1984, 1993, 1997a; Kihlstrom & Tobias, 1989; Reed, 1988). What unites the various phenomena of the cognitive unconscious -- unconscious procedural knowledge and the various forms of implicit perception, memory, and thought comprising preconscious or subsconscious declarative knowledge -- is that the link to self either does not get forged in the first place, or else it is subsequently lost. Thus, Claparede (1911/1951; see also Kihlstrom, 1995a) wrote of the amnesic syndrome:

As the 19th century turned into the 20th, the psychological unconscious was much in the air, but little was known about its nature and limits. When James said to Freud, "The future of psychology belongs to your work", he was referring to unconscious mental life in general, rather than Freud's particular conception of it. That work was suspended during the heyday of behaviorism, but it is in full swing now as the 20th century turns into the 21st. The success and vigor of research on unconscious mental life is clear to almost everyone. Of course, some doubters and sceptics remain -- egged on no doubt by the excessive claims of some theorists, and some clinicians, who retain a Romantic view of the unconscious as all-powerful and all-knowing. This work promises much to the personality and social psychologists of the future. A full century since the publication of Janet's Psychological Automatisms, and James (1890) Principles, and six decades since the death of Freud, and 50 years since the birth of the New Look, the study of nonconscious life has been completely revolutionized. For the first time, contemporary cognitive psychology has begun to offer a clear theoretical framework for studying the relations between conscious and nonconscious mental life. Along with the development of a new class of psychological theories has come a new set of observations, derived from sophisticated new experimental paradigms, including research in cognitive neuropsychology. Thus far, this body of research has revealed a view of nonconscious mental life that is more extensive than the unconscious inference of Helmholtz, but also quite different -- kindler, gentler, and more rational -- from the seething unconscious of Freud.

Still and all, it should be recognized that almost all of the work to date has been done within the confines of cognitive psychology and cognitive neuropsychology, with relatively little attention paid to unconscious emotional and motivational life, or to the role of unconscious processes in personality and social interaction. Thus, it would seem that an important agenda item over the near term would be the deliberate adoption by personality and social psychologists of the concepts and principles that have served their cognitive colleagues so well, and the systematic extension of research on the psychological unconscious beyond words and polygons to people and actions, and beyond implicit cognition to implicit emotion and implicit motivation.

Author Notes

The point of view represented herein is based in part on research supported by Grants #MH-35856 and #MH-44739 from the National Institute of Mental Health.

I thank Jennifer Beer, Michael Kim, Andres Martinez, Lillian Park, Katherine Shobe, and Heidi Wenk for their comments at various stages in the preparation of this chapter.

References