Implications of Research in Cognitive Neuroscience for Psychodynamic Psychotherapy

The first systematic approach to learning emerged from the laboratories of Pavlov, Skinner, and hundreds of other researchers who studied what came to be known as classical and operant conditioning. For much of the first half of the twentieth century, researchers from a behavioral tradition argued that the most complex behaviors reflect a handful of learning mechanisms shared by humans and other animals that could be understood without reference to internal mental processes. The animal learns in classical conditioning to produce a relatively automatic response when a previously neutral stimulus (the conditioned stimulus) is repeatedly paired with a stimulus that innately (prior to learning) produces a similar response (a conditioned response). The best known example occurred in Pavlov’s experiments, in which dogs learned to salivate at the sound of a tone that tended to precede presentation of meat. The animal learns in operant conditioning to associate certain behaviors with consequences—reinforcers and punishers—that increase or decrease the likelihood of the behavior recurring. In general, classical conditioning tends to involve involuntary reactions, whereas operant conditioning involves voluntary behaviors that a person or animal performs or inhibits to obtain or avoid rewarding or aversive consequences.

The understanding of classical and operant conditioning led, in the 1950s and 1960s, to the development of behavior therapies aimed at altering conditioned emotional responses and maladaptive behaviors. Classically conditioned emotional responses are involved in many forms of psychopathology, particularly in anxiety disorders (e.g., in the startle responses and intense anxiety and autonomic reactivity that occur when patients with posttraumatic stress disorder encounter ‘triggers’ that resemble in some way those present during a traumatic event). Some of the earliest behavioral treatments emerged directly from research on classical conditioning processes, as researchers and clinicians developed exposure techniques to try to break associative links between stimuli (or imagined stimuli, as in flashbacks of traumatic events) and intense negative feelings, particularly fear and anxiety (Wolpe, 1958). Exposure means presenting the person with the feared stimulus and preventing him or her from escaping the initial feelings of anxiety or panic. Over time, if the person cannot escape exposure, the intense emotional reaction irrationally associated with an objectively nonthreatening stimulus will generally wane if not extinguish entirely. Using exposure to treat conditioned emotional responses in anxiety disorders has been demonstrated to be quite efficacious (Barlow, 2002).

Behavior therapists similarly learned to use principles of operant conditioning to treat a range of problems, such as maladaptive parenting strategies that fostered rather than curtailed aggression. Most behavioral treatments use both classical and operant principles to promote behavior change. For example, effective treatments of anxiety disorders tend to address not only classically conditioned emotional responses hut also the avoidance mechanisms patients develop through operant conditioning to escape frightening experiences (e.g., agoraphobic avoidance of situations associated with panic attacks). Thus, behavior therapists typically combine exposure to threatening stimuli (aimed at extinguishing a classically conditioned response) with response prevention (preventing the patient from escaping the feared stimulus and hence extinguishing a response learned via operant conditioning).

Although highly productive, the behaviorist enterprise ultimately ran aground as the dominant perspective in experimental psychology as researchers increasingly recognized anomalies that could not be understood without reference to mental processes. Buoyed by developments in artificial intelligence (and the development of high-speed computers), cognitive science began to displace behaviorism in a scientific revolution that began in the late 1950s (see Robins et al., 1999). Based on the metaphor of the mind as a computer, researchers developed a serial processing model of cognition—that is, a model in which information passes sequentially (serially) through a series of three memory stores (Atkinson and Shiffrin, 1968). This three-stage model, now sometimes called the ‘modal model’ (Healy and McNamara, 1996), provided the theoretical basis for cognitive research for 30 years.

According to this model, following a brief initial stage of sensory registration that retains information for a fraction of a second, information is held in short-term memory, which can maintain roughly seven pieces of information in consciousness for about 30 seconds (Miller, 1956). (The move to 10-digit local phone numbers in the US in the 1990s, necessitated by the proliferation of fax and computer lines, has posed a challenge to the limits of human short-term memory.) The next stage is long-term memory, from which information, if properly processed (e.g., memorized in a way that is meaningful), is retrieved as needed into short-term memory. Although information may remain in long-term memory indefinitely, in general, the more frequently and recently information has been used, the easier it is to retrieve.

Researchers offered a number of theories and metaphors to describe the way information is stored in long-term memory. One emphasizes associative networks: pieces of information are associatively connected with one another, so that activating one node (unit of information) on a network spreads activation to related nodes. Suppose a participant in an experiment is presented (‘primed’) with the word bird and subsequently asked to press a button as soon as she recognizes each of a series of words. With priming, she will respond more quickly to the word ‘robin’ than to the word ‘butter.’ The reason is that bird and robin are located along the same network of associations, so that activating one spreads activation to the other. Another way researchers have described the organization of memory is in terms of schemas, patterns of thought that guide perception and memory. Thus, if an eye witness to an accident is asked how quickly a car smashed into another car, she is likely to estimate a higher speed than if asked how quickly the car was going when it hit the other car, because ‘smash’ activates a schema that implies high impact (Loftus et al., 1975).

This information-processing model offers a general view not only of memory but of thinking—that is, of the processes by which people manipulate remembered information to solve problems. According to the model that dominated the field for 30 years (and remains the foundation of many cognitive models of thought and decision making, with some caveats; see Markman and Gentner, 2000), when people want to make a decision, they use short-term memory to maintain current information, retrieve relevant information from long-term memory, and perform various operations on the information held there (Newell and Simon, 1972; Klahr and Simon, 2001). Thus, problem solving involves parsing a problem into an initial state (how things currently are), a goal state, and potential operators that might transform the initial state into the goal state.

This way of thinking about cognition provided the zeitgeist within which cognitive approaches to psychotherapy developed in the 1960s (e.g., Ellis, 1962; A. T. Beck, 1967, 1995). Early cognitive models of therapy tended to presume a serial model of cognition, in which people feel and act based on the thoughts that come into consciousness (or on ‘automatic thoughts’ that lie just outside the periphery of awareness but can be readily retrieved with proper cueing). An important goal of these therapies is to change dysfunctional attitudes, views of the self, and things people say to themselves that are associated empirically with negative mood states such as depression and dysfunctional behaviors such as bulimic binge-purge cycles. Although the information processing models of the 1960s and 1970s were relatively silent about the kinds of classical and operant learning processes studied by behaviorists, in clinical practice by the late 1970s cognitive-behavioral approaches began to emerge that integrated behavioral techniques with cognitive strategies designed to change dysfunctional thinking patterns.

In the last decade the modal model has undergone considerable evolution in four interrelated respects, which probably constitute more of a revolution than an evolution in thinking. The first change is a shift away from a serial processing model. In the modal model, stages of memory storage and retrieval occur sequentially, one at a time, with most of the ‘real’ work of cognition done by bringing information into short-term memory. Contemporary researchers, however, recognize that most processing occurs outside of awareness, as the brain processes multiple pieces of information in parallel. Serial processing, in this view, is the task of a specialized memory system, working memory (a construct that evolved from the construct of short-term memory, referring to a ‘work space’ in which the individual can consciously manipulate information; see Baddeley, 1995; Richardson, 1996).

A second and related shift is from conceiving of memory as involving ‘stores’ (places where memories are ‘kept’) to a view of memory and cognition as involving multiple circuits or systems. For example, when a person sees an object, cortical circuits involving the occipital and lower (inferior) temporal lobes are involved in breaking it into component parts and comparing it with familiar objects, and a second circuit running from the occipital lobes through the upper (superior) temporal and parietal lobes attempts to pinpoint its location in space. The person is never aware of using different circuits to identify an image and locate it in space, because both circuits are part of a broader neutral circuit that integrates the information—and does this so quickly that the person has no phenomenological experience of anything other than the immediate recognition of having seen a squirrel running across the road. This conception of memory systems is bolstered by research showing that memory for episodes (e.g., remembering what happened yesterday), memory about the emotional meaning of stimuli (e.g., whether something has consistently been associated with pain), memory for procedures (e.g., playing a piece on the piano), and working memory constitute neurologically distinct memory systems. For example, memory for episodes requires an intact hippocampus, but a person with hippocampal damage can still associate a stimulus with an emotional response, even though he may have no memory for having ever encountered it (e.g., Bechara et al., 1994). Working memory, in contrast, is readily disrupted by lesions to the dorsolateral prefrontal cortex, which is involved in deliberate conscious thinking and decision making.

A third major shift has occurred with the recognition of the existence of two ways that memory can be expressed, either explicitly (via conscious recall or recognition) or implicitly (in behavior, independent of conscious control). Explicit memory refers to conscious memory for ideas, facts, and episodes. Implicit memory refers to memory that is observable in behavior but is not consciously brought to mind (Roediger, 1990; Schacter, 1992, 1998). One kind of implicit memory is procedural memory, which refers to ‘how to’ knowledge of procedures or skills, such as how close to stand to another person or how to respond when someone reaches out his or her hand for a handshake. Another kind of implicit memory involves associative memory. For example, priming subjects with an infrequently used word such as syncopate among a long list of words renders them more likely a week later to respond with the correct word when asked to fill in the missing letters of the word fragment, S-----ATE. This occurs even though they may lack any conscious recollection of whether syncopate was on the list a week earlier (Tulving et al., 1982). Essentially, the network of associations still has some residual activation, leading to memory expressed in behavior but not in conscious recollection.

A fourth shift involves a change in metaphor. Cognitive psychologists in the late 1950s and early 1960s saw in the computer a powerful metaphor for the human mind. Today, cognitive scientists are turning to a different metaphor: mind as brain. In this view, memory is not so much a matter of ‘storing’ something somewhere in the brain and later retrieving it (as in a computer file) than a process by which an experience activates a set of neurons distributed throughout the brain that can in turn be reactivated by similar experiences or efforts at recollection. In this view, memory is simply a potential for reactivation of a set of neurons that together constitute a representation. The notion of using the brain as a metaphor for the mind may seem today obvious if not tautological; however, metaphors of mind have tended to follow understanding in other domains, particularly in the physical sciences. Freud, for example, certainly knew that what he referred to as mental processes occur through the actions of brain processes, and he developed some complex models of neural excitation that appear today to be remarkably prescient in multiple respects (Freud, 1966; Pribram and Gill, 1976; Westen, 1998). However, because knowledge of the brain was so primitive, be turned to metaphors from physics to explain how mental ‘dynamics’ function. Similarly, the information processing theorists of the 1960s through 1980s tended to draw their inspiration from computer technology. Not until knowledge of the brain expanded exponentially in the last two decades did brain processes become potential metaphors for mental processes.

The notion of mind as brain is central to connectionist, or parallel distributed processing, models of perception, memory, and thinking (Rumelhart et al., 1986; Kunda and Thagard, 1996; Smith, 1998). Connectionist models suggest that most information processing occurs in parallel, outside of awareness, as multiple components of a thought, memory, or perception are processed simultaneously. Representations are distributed throughout the brain over many sets of neurons processing different aspects of a thought, perception, or memory, rather than ‘located’ in any particular part of the brain. Knowledge lies in the connections among these neural units or nodes, which, like neurons, can either inhibit or activate each other. Cognitive activity involves a process of constraint satisfaction, in which the brain simultaneously and unconsciously processes multiple features of a stimulus, attended to by different nodes or sets of nodes in a network that provides constraints on the conclusions that can be drawn. The brain then draws the best tentative conclusion it can based on the available data. In other words, it equilibrates to the solution that provides the best ‘fit’ to the data. Thus, if a patient is crying, the clinicians interpretation of that crying as tears of pain or joy will depend on auditory and semantic cues processed simultaneously (in parallel).

Connectionist models have the advantage of building in a way of modeling both the chronic ways people tend to process information and moment to moment changes in the way they view important people and experiences in their lives (see Barsalou, 1999). One of the virtues of connectionist models is their suggestion that representations, such as a person’s representations of significant others, are not static. Rather, the representation of a significant other activated at any given point depends on the context. Thus, the same person can represent his wife as impossible to deal with at one time but a source of loving support at another, depending on aspects of his ‘wife network’ activated by the current situation, his feeling state, and so forth. At the same time, the chronic activation of a way of seeing something or someone—that is, the frequent activation of a set of neurons representing some aspect of that person—will create an attractor state, a pattern of neural firing that is readily activated under particular circumstances. Thus, a patient with a critical parent may be ‘primed’ to hear his therapist’s comments as criticisms because a network representing self-being-criticized-by-parental-figure is an attractor state that ‘attracts’ the brain to this interpretation. In this view, then, a representation is not something ‘stored’ in the brain. It is a potential for reactivation of a set of neural units that have been activated together in the past. Activating part of that network may reproduce much of the original experience (as in an episodic memory, e.g., of a time the parent was critical, or more directly in a flashback in posttraumatic stress disorder) or may influence the way the person interprets current experiences.

Fundamental to all psychoanalytic forms of treatment is the effort to map the idiosyncratic associative networks that may be relevant to the patient’s sources of distress. The goal, as first enunciated by Freud, was to give the patient more freedom to make conscious, explicit choices. Indeed, in describing the process of open-ended, long-term therapy to patients, it can be very useful to offer a simple explanation such as the following:

The existence of unconscious or implicit networks—which tend to he resistant to change because they reflect longstanding regularities in the person’s experience and allow him or her to navigate the world in ways that feel predictable (even if sometimes rigid, inaccurate, or otherwise maladaptive)—provides perhaps the best empirical justification for long-term therapies. Deeply engrained views of the self, others—called ‘internal working models of relationships’ in research on attachment (see, e.g., Bowlby, 1973; Main et al., 1985; Fonagy et al., 2002) and object representations in theory and research on interpersonal functioning more broadly in psychoanalysis (see, e.g., Greenberg and Mitchell, 1983; Westen, 1990, 1991; Blatt et al., 1997)—may take months or years to identify in their various manifestations. The same is true of problematic ways of regulating emotions (defenses) that are triggered automatically and may lead to a cascade of internal and interpersonal events.

For example, patients with prominent passive-aggressive features are often unaware of both their anger and the ways they put other people in uncomfortable positions—which in turn lead others to become angry at or avoid them. This, in turn, makes the patient more angry and passive-aggressive. Consciously, these patients view themselves as helpless victims of indifferent or mean-spirited others; unconsciously, they provoke precisely the behavior that makes them feel mistreated. Breaking into these kinds of self-sustaining spirals—into what Wachtel (1997) calls cyclical psychodynamics—can take a long time, because the patient cannot report them. Such dynamics may become most apparent—and most workable as a treatment issue—when they show up in the therapeutic relationship (Luborsky and Crits-Christoph, 1998).

Thus, contemporary research in cognitive neuroscience corroborates some central psychoanalytic assumptions that have been the source of tremendous controversy for a century. At the same time, this research also poses some important challenges for psychodynamic psychotherapy and suggests potential refinements in theory and technique (see Westen, 2002; Westen and Gabbard, 2002a; Westen and Gabbard, 2002b; Gabbard and Westen, 2003). For example, research in cognitive neuroscience suggests precisely why the psychoanalytic practice of exploring patients’ associations to symptoms, feelings, or events—asking them what comes to mind—can often be very useful: people cannot report on their implicit networks, and they typically invent plausible but often inaccurate explanations if called upon to do so (e.g., when asked, ‘Why do you think you felt that way?’; Nisbett and Wilson, 1977). On the other hand, this same body of research suggests limits to free association as a therapeutic technique, on two grounds.

First, although free association can be essential in exploring implicit networks, it may do very little to illuminate or alter explicit (conscious) beliefs, procedures, or ways of behaving that operate through the action of different neural networks. As argued below, with limited therapeutic time (even for patients treated more than once a week), attention to implicit processes inherently comes at the expense of attention to explicit processes, which can also wreak havoc on a person’s quality of life, and nothing guarantees that even emotionally important change in implicit expectations, motives, feelings, or conflicts will alter conscious habits of thought or behavior that have attained functional autonomy over years or decades of use. This recognition is precisely what led Aaron Beck (1976) to develop cognitive therapy for depression.

Second, research in cognitive science suggests that what is on a person’s (unconscious) mind at any time is a joint function of what is chronically on his mind (much of which is likely, in fact, to reflect concerns forged in childhood) and what is recently on his mind, which may or may not be related to the concerns that brought the patient to treatment. In other words, the particular associations that emerge in any analytic hour if the patient follows what Freud called the ‘fundamental rule’ of psychoanalysis (namely, to say whatever comes to mind) may or may not prove useful to explore, depending on what has been activated recently in and out of the consulting room. Any given set of associations reflects some combination of clinically meaningful signal and clinically less meaningful noise, and one cannot always distinguish the two. Over time one would expect important material to be reflected repeatedly in the patient’s associations, as chronically activated networks influence the patient’s thought, feeling, and behavior in the treatment. However, waiting for important material to emerge, particularly in the context of therapeutic interventions (particularly interpretations) that shape subsequent associations, is likely to be an inefficient process.

Patients can also avoid doing things associated with anxiety, such as allowing themselves to fall in love because doing so is associated with anxiety or fear of rejection. As a result, some of the most important networks may never be activated to the extent necessary for useful exploratory work until the patient actually exposes herself to the feared situation. Under such circumstances, the therapist may do well to encourage the patient to approach what she fears, to alter the patient’s associative networks and/or to bring material to the fore in the patient’s associative networks that are most important in maintaining maladaptive patterns. Freud himself noted that people do not get free of their fears unless they confront them, and he practiced a much more active mode of therapy than practiced by subsequent generations of analysts for many years. As many psychoanalytically oriented clinicians now recognize, good treatment probably requires a balance of exploration and exhortation at the service of further exploration and behavioral change (see Wachtel, 1997; Gabbard and Westen, 2003).

One useful way to employ associative techniques in once- or twice-weekly psychotherapy, where one does not have the luxury to explore whatever associations come to mind at any given time, is what might be called directed free association, in which the therapist targets particular thoughts, feelings, or memories for further associative work (see Westen, 2000b).

For example, one patient had a pattern of becoming excited about some plan (e.g., spending the evening out with friends) but then finding himself depressed and unable to imagine that he would really enjoy doing it. (For a sophisticated cognitive-dynamic explanation of the way patients shift between such ‘states of mind,’ see Horowitz, 1979.) As a result, the patient experienced few pleasures in life. After the pattern became clear, I routinely asked him to imagine as vividly as possible what he initially thought and felt when he was excited about the plan or to picture the moment he found himself feeling depressed and uninterested in pursuing it. I would ask him to walk me through the episode or image moment by moment or scene by scene, taking associations along the way, much as Freud would have explored a dream, encouraging him to report whatever he felt at the time and whatever thoughts, feelings, images, or memories emerged as he pictured the experience. Doing so led to a series of associations and memories in which he wished for something that subsequently fell through and his corresponding fear of hoping for enjoyment, as well as survivor guilt around a mentally retarded sister who could never have such pleasures and toward whom he felt a mixture of love and (largely unacknowledged) resentment.

Data from the cognitive neurosciences may also help shed new light on psychoanalytic constructs such as transference. The connectionist notion of representations as potentials for reactivation—as sets of neurons that have been activated in the past and are hence more readily activated as a unit in the future—offers a mechanism to explain the long-held psychoanalytic position that patients are likely to express important conflicts, defenses, motives, and interpersonal patterns in their relationship with the therapist (Westen and Gabbard, 2002b). To the extent that the therapeutic situation or relationship matches prototypes from the past, it is likely to activate similar responses (for empirical evidence, see Andersen and Baum, 1994; Luborsky and Crits-Christoph, 1998). It should therefore not be surprising if important relational patterns emerge in a relationship in which the patient self-discloses and becomes attached in an intimate but asymmetrical relationship with another who is trying to be helpful, nurturant, and attentive primarily to his or her needs. Inherent in the cognitive situation of the therapeutic relationship is the likelihood that the therapist will be experienced as an authority figure, an attachment figure, or an object of love or affection, which renders exploration of the therapeutic relationship of particular use if the patient is presenting with problems that include interpersonal components (which is nearly always the case).

Important dynamics are likely to emerge relatively quickly and persistently in patients with rigid maladaptive patterns of interpersonal functioning, cognition, and emotion regulation. For example, narcissistic patients tend to manifest particular patterns when interacting with their therapists regardless of the therapist’s theoretical understanding or technical approach. Empirically, clinicians of all therapeutic orientations report that narcissistic patients need excessive admiration from them, vacillate between idealizing and devaluing them, and need to be special to the therapist at the same time as being sadistic and hostile and feeling criticized by the therapist (Bradley et al., in press). Correspondingly, therapists of all theoretical orientations report similar countertransference reactions to their narcissistic patients: They tend to feel annoyed, manipulated, used, criticized and as if they are ‘walking on eggshells’ with the patient, and correspondingly frequently fight their impulses to be sadistic themselves or to drift off during sessions (Betan et al., in press).