Psychodynamic Models of Psychopathy

In psychodynamic theory, patients with psychopathy are often described as having deep-seated self-related negative emotions, particularly feelings of worthlessness and shame. These feelings are rooted in their early experiences of being devalued, made to feel inadequate and unacceptable in the eyes of their attachment figures (Gacono, Meloy, & Berg, 1992; Kernberg, 1985; Kohut, 1966, 1977; Meloy & Shiva, 2007; Perry, Presniak, & Olson, 2013). During development, the first mental representations of the self and others are constructed from these early interactions with attachment figures, called internal working models (IWMs), which often operate largely unconsciously (Bowlby, 1969, 1973; Pietromonaco & Barrett, 2000). Psychopathic patients’ adverse attachment experiences lead to the formation of an IWM of the self characterized as worthless and shameful, which influences how these patients subsequently interpret information and regulate their self-esteem, emotions, and behavior (Lorenzini & Fonagy, 2013). Consistent with this, there is robust evidence that experiences precluding the formation of secure attachments, such as abuse, neglect, parental separation, and parental dysfunction, are risk factors for psychopathy and for violent/antisocial behavior (Campbell, Porter, & Santor, 2004; Cohen et al., 2014; Craparo, Schimmenti, & Caretti, 2013; Douglas, Hart, Webster, & Belfrage, 2013; Graham, Kimonis, Wasserman, & Kline, 2012; Hoeve et al., 2009; Johnson, Cohen, Brown, Smailes, & Bernstein, 1999; Kolla et al., 2013; Luntz & Widom, 1994; Marshall & Cooke, 1999; National Institute of Health and Clinical Excellence, 2010; Poythress, Skeem, & Lilienfeld, 2006; Roberts, Yang, Zhang, & Coid, 2008).

From a psychodynamic perspective, psychopathic personality traits reflect the operation of defense mechanisms that protect against the conscious experience of powerful self-related negative emotions arising from the IWM of the self. Specifically, individuals with psychopathy use denial, rationalization, and projection to disavow these negative experiences (Perry et al., 2013). Moreover, they construct a grandiose self-image to distort their self-appraisal and devalue and act aggressively toward others who threaten their grandiose veneer. Self-aggrandizement thus distorts the patient’s self-image to defend against their painful feelings of shame and worthlessness arising from their IWM of the self. Similarly, a lack of remorse defends against the conscious experience of these negative self-related emotions after breaching social norms (e.g., committing violent or antisocial acts) by denying having hurt others or minimizing the consequences of their actions. Lacking remorse thus serves a vital function for patients, because they are extremely sensitive to feeling worthless/shameful and will do anything to bypass or diminish these emotions. Such defensive functioning can take the form of hostility and violence, which can restore a patient’s self-esteem by making the perceived perpetrator of the shaming feel vulnerable and powerless (Logan & Johnstone, 2010). The extreme end of such reactive aggression is homicide. Patients’ low tolerance for these negative emotions is well known in the correctional/forensic rehabilitation literature (C. M. Jones, 2014; Maruna & Ramsden, 2004; Walker & Bright, 2009) and likely has its roots in the IWM of the self, which developed from the patients’ early adverse attachment experiences.

Cognitive Models of Psychopathy

Although described using different terms, cognitive models of psychopathy have much in common with this psychodynamic formulation. The two major models are Davidson’s (2007) cognitive-behavioral therapy (CBT) model and the schema focused therapy (SFT) model (Bernstein, Arntz, & de Vos, 2007; Young, Klosko, & Weishaar, 2003). Davidson (2007) developed CBT for PDs, which formed the basis of the psychotherapeutic intervention of the Chromis programme for treatment of high-risk offenders in the United Kingdom with high levels of psychopathic traits (Tew & Atkinson, 2013). The CBT model integrates the traditional cognitive model of psychopathology with attachment theory, developmental psychology, and Beck and colleagues’ (A. T. Beck & Freeman, 1990; A. T. Beck, Freeman, & Davis, 2004) evolutionary perspective on PDs. The cognitive model understands psychopathology as emerging across three levels of belief: (a) schemas or “core beliefs,” (b) intermediate beliefs, and (c) automatic thoughts (A. T. Beck, 1964; A. T. Beck & Freeman, 1990; A. T. Beck et al., 2004; J. S. Beck, 2011). Schemas are the most basic structures that organize our experience and construct meaning out of our perceptions. Schemas play a fundamental and global role in information processing because they contain core beliefs about the self, others, and the world, which ultimately influence how patients think, feel, and behave moment by moment. For this reason, schemas are often not consciously articulated or accessible and thus operate outside conscious awareness. Intermediate beliefs are more consciously articulated and accessible rules, attitudes, and assumptions about oneself, others, and the world (e.g., “You can’t trust people”). Intermediate beliefs are thus a bridge between schemas and the final level of belief: automatic thoughts. Automatic thoughts are the most superficial level of cognition, because they are the reflexive thoughts that enter consciousness in response to situations that activate underlying schemas and intermediate beliefs. Automatic thoughts reflect the deeper layers of belief and directly influence how a patient feels and behaves. The CBT model of PD integrates this traditional cognitive model with an evolutionary perspective, because it is hypothesized that these schemas can also activate innate behavioral strategies, which are genetically determined behavioral patterns (e.g., aggression) that promoted survival and reproduction throughout most of human evolution (A. T. Beck & Freeman, 1990; A. T. Beck et al., 2004). Key to the CBT model of PDs is the idea that the interaction between a person’s genetic predispositions (e.g., temperament, neuropsychological functioning) and childhood environment shapes the development of the person’s schemas and behavioral strategies. If a primary caregiver is emotionally unavailable to their child, insensitive or unresponsive to the child’s emotional needs, or has a chaotic or harsh parenting style, the child will likely develop an insecure attachment style as a consequence of developing dysfunctional IWMs (i.e., core beliefs) about themselves and others (Bowlby, 1973; Davidson, 2007; Pietromonaco & Barrett, 2000). Throughout development, patients with PD learn a variety of strategies to cope with their core beliefs of low self-worth, vulnerability, and being unloved. Furthermore, an adverse environment can amplify or inhibit the innate behavioral strategies in patients with PDs, such that they can become overdeveloped or underdeveloped and thus maladaptive in modern social environments.

From the CBT model’s perspective, the early adverse attachment experiences psychopathic patients encounter interact with their genetic vulnerabilities, resulting in the formation and reinforcement of dysfunctional schemas, such that they have a core belief about themselves as being unworthy, powerless, and unloved (Davidson, 2007). Additionally, consistent with the psychodynamic model, compensatory intermediate beliefs develop as a way to cope with these deep feelings of worthlessness and inadequacy. For instance, patients typically form egocentric and self-aggrandizing beliefs that they are strong, can do whatever they want, and are entitled to exploit others and thus show a lack of remorse for their antisocial behavior (A. T. Beck & Freeman, 1990; A. T. Beck et al., 2004). Compensatory coping strategies can also involve innate behavioral strategies. For example, patients learn to use hostility and violence to avoid being perceived by others as weak, which triggers intolerable anxiety and feelings of vulnerability, resulting in the overdevelopment of aggression (Davidson, 2007).

SFT has similarly been adapted for use in forensic psychiatric hospitals in the Netherlands, called Terbeschikkingstelling (TBS) clinics, for the purpose of treating forensic patients with severe PDs, particularly patients with psychopathy (Bernstein et al., 2007). While not exclusively a cognitive model—because it combines cognitive, behavioral, psychodynamic, and existential/humanistic approaches—the SFT model has prominent features of the traditional cognitive model of psychopathology. There are three major components to the SFT model of PDs: Early Maladaptive Schemas (EMSs), coping strategies, and Schema Modes (Young et al., 2003). EMSs are the core pathology of patients with PDs, and they develop out of the interaction between genetic vulnerabilities (e.g., temperament) and unmet emotional needs the patient experienced early in life.

Like IWMs and core beliefs, EMSs are highly stable structures that organize a person’s core self-identity and mental representations of others around specific themes and are elaborated throughout life. The SFT model identifies 18 EMSs, which are grouped into five domains, reflecting failures to meet the five universal emotional needs (Table 2). Patients experience powerful self-related negative emotions (e.g., shame, vulnerability, self-hatred) when an EMS is activated. Patients attempt to eliminate, or at least diminish, these negative emotions using three coping strategies: Schema Maintenance, Schema Avoidance, and Schema Compensation. In Schema Maintenance, the patient reinforces their schema by discounting information that would disconfirm their EMS through cognitive distortions or self-defeating behavior. In Schema Avoidance, the patient attempts to suppress thoughts or feelings or behaviorally avoid situations associated with their EMS. In Schema Compensation, the patient overcompensates for the negative emotions by acting or generating emotions in the polar opposite direction of the content of their EMS.

Schema Modes are defined as those schemas and coping strategies that are currently dominating the moment to moment thoughts, feelings and behavior of a person. Young et al. (2003) originally identified 11 Schema Modes; however, as a consequence of their therapeutic work with personality disordered forensic patients, Bernstein et al. (2007) expanded the list to include four new Schema Modes (Table 3). Patients with psychopathy are characterized by prominent use of four Schema Modes that involve compensatory coping responses to their EMSs, particularly those EMSs with themes of disconnection and rejection (Bernstein et al., 2007). Specifically, they predominantly use the Self-aggrandizer mode, Bully and attack mode, Conning and manipulative mode, and Predator mode as a way to overcompensate for feelings of shame, loneliness and vulnerability. The Self-aggrandizer mode allows the patient to distort their conscious thoughts and feelings about their self in order to “defend against” or “cope with” their deep self-related negative emotions arising from their underlying EMSs (i.e., IWM). Similarly, the Predator mode allows the patient to overcome deep feelings of shame and worthlessness by becoming a predator who can eliminate threats to their self-esteem without remorse and also command respect from others through fear (Bernstein et al., 2007).

Toward an Integrated Psychotherapeutic Model of Psychopathy

This convergence of psychodynamic, CBT, and SFT models of psychopathic traits—specifically lacks remorse and self-aggrandizing—is extremely striking. This is because it suggests that the psychotherapeutic field is moving toward an integrated model of these traits. This integrated model centers on two hypotheses:

• 1.   Early adverse attachment experiences interact with genetic vulnerabilities to shape the development of a psychopathic patient’s core self-image as worthless and shameful. This core self-image is the first self-identity the patient develops during childhood and adolescence, and it continues to be elaborated upon throughout life. Furthermore, this core self-image is often not consciously articulated or accessible and thus operates outside conscious awareness. Various models call this core self-image the IWM of the self, self-schema, core belief, or EMS.
• 2.   The traits lacks remorse and self-aggrandizing are maladaptive defense mechanisms that allow the patient with psychopathy to eliminate or at least diminish (i.e., cope with) the influence of their negative core self-image on their conscious experience. These defense mechanisms are reflected in more consciously articulated and accessible beliefs about the self. Various models call these more conscious beliefs intermediate beliefs or Schema Modes.

This integrated psychotherapeutic model of lacks remorse and self-aggrandizing will be the basis of the Bayesian formulation described in the section “A Bayesian Account of Psychopathy.” However, before outlining the model, it is necessary to first consider the broader theoretical context for understanding the predictive coding framework (Corlett & Fletcher, 2014; Friston, 2010; Friston & Kiebel, 2009; Friston, Stephan et al., 2014).

Quantitative Simulations

In this section, we present some quantitative simulations of psychopathy to substantiate the hypotheses of the previous section; namely, that minimal impairments to the encoding of precision or uncertainty are sufficient to explain psychopathic traits and abnormal inferences about self-worth. We have until this point used predictive coding to illustrate hierarchical inference in the brain. However, we will use an equivalent—free-energy minimizing—active inference scheme, formulated in terms of discrete states. This is because discrete states are more apt for modeling the sorts of games and behaviors associated with trust and reputation formation (e.g., in behavioral economics), which can be used to simulate some of the claims we have made about psychopathic traits. Although the formalism of these schemes differs from predictive coding, the basic elements are conserved—namely, belief propagation using predictions and prediction errors to update expectations about the latent or hidden states of the world causing observable outcomes (Friston, FitzGerald, Rigoli, Schwartenbeck, & Pezzulo, 2017; Friston, Parr, & de Vries, 2017).

To illustrate the key role of precision at different levels of a hierarchical generative model, we used a Markov decision process (Friston, FitzGerald et al., 2017). Active inference under these models has been described in many previous applications, ranging from applications to choice behavior through to epistemic foraging and visual searches (Mirza, Adams, Mathys, & Friston, 2016; Parr & Friston, 2017; Schwartenbeck, FitzGerald, Mathys, Dolan, & Friston, 2014). Here, we use exactly the same formalism and (free-energy minimizing) update scheme to simulate a simple reputation game (Figure 4). In brief, the generative model underneath these sorts of simulations comprises hidden or latent states of the world and the outcomes that they generate. The mapping between states (i.e., causes) and outcomes (i.e., consequences) is described by a likelihood (A) matrix, whose precision we will associate with precision at the lower (e.g., sensory or perceptual) level of processing hierarchies. Transitions among hidden states depend upon policies or choices and are generally encoded in (choice dependent) probability transition (B) matrices. We will associate the precision of these matrices with prior precision—namely, the confidence placed in prior beliefs about state transitions under different choices. Finally, prior preferences over outcomes are encoded in a (C) matrix, in the form of log prior probabilities. These can be thought of as the value of different outcomes.

The game we modeled involved deciding whether to keep a sum of money or donate it to charity, depending upon how rich one is. The minimum requirements for this sort of game include hidden states that encode how nice or charitable the subject is, and how rich they believe themselves to be. Given these two hidden states, one can generate plausible outcomes. In this instance a (prosocial) feedback of approval or disapproval. In addition, we included the actual wealth of the subject as an observable feedback. In brief, our generative model contained two sorts of hidden states. The first was monetary wealth (with eight levels, ranging from broke to wealthy). The second hidden state was self-worth (with two levels: charitable versus mean). The observable outcomes had the same form: with one visual cue reporting the level of monetary wealth and another reporting approval or disapproval of the choice to donate or keep an offer on each trial. The likelihood (A) matrix was an identity mapping between the levels of the wealth factor. However, the mapping between the hidden state of self-worth and the approval cue could be precise or imprecise, depending upon the subject’s predisposition. In other words, the likelihood mapping could be deterministic, such that approval was always generated by a charitable state of being or it could be imprecise, such that there was a 50-50 chance of approval or disapproval irrespective of one’s self-worth. The prior probability transition matrices (B) contained the structure and dynamics of the game. These are specified separately (technically, conditioned upon) the two choices (donate or keep). For transitions among levels of wealth, every time the offer was donated, the level of wealth fell to the level below (or stayed at the lowest level). Conversely, if the choice was keep, the level of wealth increased (unless at the most wealthy state). In addition, we modeled an attrition of wealth, with a constant decay from higher levels to lower levels (with 10% probability of loss at each trial). Heuristically, this means the subjects were spending their wealth at a constant rate and could decide to accumulate more wealth by keeping the offer or, if they were sufficiently wealthy, increase the probability of an approval rating by donating.

The probability transition matrix among self-worth states controlled the rate at which self-worth changed from a charitable to a mean level (and vice versa), depending upon the actions selected. The parameterization of this matrix allowed us to modulate how behavior changed self-worth: here, donating ensured that a charitable self-worth was maintained, with a small probability of moving from a mean to a charitable state, and vice versa for keeping the offer. Finally, prior preferences were encoded in the (C) matrix for both outcome modalities. These comprised a log linear increase in prior preferences for being rich and a preference for approval over disapproval, both of which were fairly precise (with log prior differences in the range of four). The structure of this generative model and numerical examples of the likelihood (A), prior transition (B) and prior preference (C) matrices are provided in Figure 4.

Equipped with this model, we can now simulate choice behavior, starting from any prior expectations about self-worth and examine the long-term or equilibrium behavior. Here, we characterized behavior and underlying beliefs in terms of the average wealth retained by synthetic subjects—and the posterior expectation that they were nice. These simulations entailed 16 successive choices starting from a prior belief that they were charitable (but broke). By repeating the simulations under different levels of the likelihood precision (encoded by αFigure 4) and different precisions of the prior transition probabilities over self-worth (encoded by β), we could examine the effects of likelihood and prior precision on behavior and concomitant self-worth. By decreasing the precision of the likelihood (i.e., the confidence that approval ratings were a veridical reflection of one’s self-worth), we hoped to simulate lacks remorse, with an increasing tendency to keep offers. By increasing the precision of prior beliefs, we then hoped to simulate self-aggrandizing, in terms of mean behavior in the face of preserved self-worth.

Figure 5 shows the results of these simulations as heat maps for self-worth (left panel) and wealth (right panel). As we would expect, synthetic subjects who accumulated the most wealth, by giving relatively few donations, had a lower posterior expectation that they were charitable. In other words, regions in the parameter space of likelihood and prior precision with high self-worth (white areas on the left) were associated with low wealth (dark areas on the right). Reducing α, or the degree to which self-worth reliably solicits approval, quickly reduces donations leading to an accumulation of wealth (under higher levels of prior precision). The interpretation of β is subtler.³ A very high value results in an identity matrix, de-coupling beliefs about transitions from choices. This corresponds to a high prior precision. Conversely, low values of β give rise to transitions that depend sensitively on actions. This corresponds to reduced prior precision. When we increase β (i.e., increase prior precision), the synthetic subject again accumulates more wealth, as the consequences for self-worth are less affected by their choices. For higher levels of likelihood precision, reducing prior precision has a non-monotonic effect on self-worth. However, when likelihood precision is low, increasing prior precision leads to an increase in self-worth. In other words, the effect of likelihood and prior precision show a strong interaction, where the likelihood precision permits a reversal of the effect of prior precision, thereby enabling uncharitable behavior and paradoxically preserved self-worth despite socially inappropriate behavior (i.e., lacks remorse). This interaction is entirely consistent with the notion that psychopathy can emerge successively by a reduction sensory likelihood precision (i.e., a failure to attend to prosocial cues), followed by an increase in the precision of prior beliefs (i.e., reduced sensitivity to choices), modeling lacks remorse and self-aggrandizing, respectively. This interpretation is depicted graphically by the red line in Figure 5.

These results are presented to illustrate that psychopathic behavior—and false inference about the self—can emerge under fairly elementary generative models of one’s own behavior by, and only by, changing the precision of beliefs at different levels of a self-model. Crucially, at no point did we need to change the (simulated) preferences for being approved of (or being rich). Furthermore, the pathological behavior was evident even though all the (synthetic) subjects had exactly the same form of beliefs about the effect of charitable donation on their self-worth. In short, a sufficient explanation for psychopathy (in this minimal example) was a loss of subjective precision linking latent states to observable consequences and an increase in the precision or confidence about the volatility of latent states. The first attenuation—of likelihood precision—means that (synthetic) psychopathic patients can, effectively, ignore (i.e., defend against) evidence that speaks against being the sort of person they would prefer to be (e.g., shame, worthlessness). In a complementary fashion, an increase in the precision of prior beliefs—about self-worth—means that they are more resistant to change and can maintain their self-worth, even in the face of evidence to the contrary. In short, this simple simulation illustrates the profound effect of precision on inference about latent states of the (prosocial) world and, crucially, one’s relationship to that world.

Psychopathic States and Traits

With this framework, we can also see how the mechanisms underlying lacks remorse may operate together with the mechanisms underlying self-aggrandizing to defend against shame and worthlessness despite breaching social norms. Specifically, patients can decrease the precision of the bottom-up affective signals from the IWM while simultaneously increasing the precision of the elevated top-down self-appraisal from the high-level prior beliefs (i.e., intermediate beliefs) about the self. This type of defensive functioning appears to closely capture the self-justifying trait of psychopathy, whereby patients use rationalization to minimize/deny responsibility for their actions through self-serving (but inaccurate) explanations that preserve their grandiose self-image (Cooke et al., 2004; Perry et al., 2013).

These pathogenic mechanisms may become entrenched through a number of processes such that they become stable traits over time. The literature on how psychopathic traits develop during childhood and adolescence is still emerging (Frick et al., 2014), and it is clear that all personality pathologies arise from a combination of genetic, temperamental, social, familial, and psychological factors (Alwin et al., 2006). However, learning may be a central mechanism underlying the entrenchment of these maladaptive Bayesian inferences. This is because learning is a key mechanism through which the brain structurally and functionally encodes perceived statistical regularities and reinforced associations via experience-dependent synaptic plasticity (Caroni, Donato, & Muller, 2012; Friston, 2010). For instance, the patient may have initially learned this type of defensive functioning from antisocial peers and/or family members who may have been role models during their youth. Lacks remorse and self-aggrandizing may become entrenched over time through positive reinforcement (e.g., from a social milieu that rewards psychopathic personality traits by offering a sense of belonging, support, and self-worth) and negative reinforcement (e.g., from the fact that these defense mechanisms relieve the patient from their feelings of worthlessness and shame). Furthermore, the patient’s internal working (i.e., generative) model of the self may also become entrenched over time through patterns of relationships the patient may have experienced from childhood into adulthood. For example, repeated rejecting, punitive, invalidating, and unempathic interactions with peers, teachers, social workers, police and correctional officers, medical and mental health professionals, and even society as a whole play directly into the patient’s core self-image that they are worthless and shameful. This hypothesis regarding the entrenchment of these mechanisms is highly consistent with the CBT model of PDs, which postulates that learningis the process through which coping strategies become fixed, inflexible, and over-/underdeveloped in patients with PDs (Davidson, 2007). In light of the above, what is the evidence that this pathology of inference underlies psychopathy?

Review of Neuroimaging Research

Diffusion tensor imaging (DTI) studies of psychopaths provide a window into the abnormalities in the amygdala–vmPFC network. DTI allows researchers to reconstruct in vivo the white matter (WM) tracts connecting brain regions and to evaluate certain aspects of its microstructure. Six DTI studies comparing psychopaths with matched controls found reduced WM structural integrity in the uncinate fasciculus (Craig et al., 2009; Hoppenbrouwers et al., 2013; Jiang et al., 2017; Motzkin, Newman, Kiehl, & Koenigs, 2011; Sundram et al., 2012; Waller, Dotterer, Murray, Maxwell, & Hyde, 2017; Wolf et al., 2015). The uncinate fasciculus is a bidirectional hook-shaped tract with fibers that connect the temporal pole (TP; Brodmann area [BA] 38/20), entorhinal cortex (ERC; BA 28/34), perirhinal cortex (PRC; BA 35/36), parahippocampal cortex (PHC; BA 36), and amygdala to regions of the lateral orbitofrontal cortex (lOFC; BA 11/47), vmPFC (BA 11/32), and rostromedial PFC (rmPFC; BA 10; Schmahmann & Pandya, 2006; Schmahmann et al., 2007; Thiebaut de Schotten, Dell’Acqua, Valabregue, & Catani, 2012; Von Der Heide, Skipper, Klobusicky, & Olson, 2013). The uncinate fasciculus fibers split when they enter the frontal region into a larger ventro-lateral branch and a smaller anterior-medial branch (Catani, Howard, Pajevic, & Jones, 2002). The ventro-lateral branch terminates in the lOFC, whereas the antero-medial branch terminates in rmPFC. Crucially, rs-fcMRI studies of psychopaths mirror this structural connectivity deficit: psychopaths also show reduced functional connectivity between the amygdala and vmPFC at rest (Motzkin et al., 2011), while passively watching scenes depicting moral violations (Yoder, Harenski, Kiehl, & Decety, 2015), and while imagining others in pain (Decety, Chen, Harenski, & Kiehl, 2013) compared to age, gender, and IQ matched controls, though one study found that this de-coupling was more dorsal in the mPFC (Contreras-Rodríguez et al., 2015). Furthermore, structural brain changes associated with psychopathy overlap with the brain regions connected by the uncinate fasciculus. For instance, psychopathy is associated with significant reductions in gray matter volume (GMV), gray matter density (GMD), and/or cortical thickness in the rmPFC, vmPFC, lOFC, amygdala,⁵ TP, PHC, ERC, and PRC (Boccardi et al., 2011; Contreras-Rodríguez et al., 2015; de Oliveira-Souza et al., 2008; Ermer, Cope, Nyalakanti, Calhoun, & Kiehl, 2012; Gregory et al., 2012; Ly et al., 2012; Müller et al., 2008; Yang, Raine, Colletti, Toga, & Narr, 2009, 2010; Yang, Raine, Narr, Colleti & Toga, 2009).

By linking these abnormalities to the amygdala–vmPFC network’s normal functions, we can understand the significance of these results in relation to the Bayesian model described in the section “A Bayesian Account of Psychopathy.” The amygdala is a complex collection of nuclei with extensive connections with cortical and subcortical structures. The amygdala receives inputs from all sensory modalities and has mainly unidirectional output projections to the striatum and rich bidirectional connections with the mPFC, OFC, medial temporal lobe (MTL; i.e., ERC, PRC, PHC, hippocampus), TP, thalamus, hypothalamus, and brain stem (Freese & Amaral, 2005; Ghashghaei & Barbas, 2002; Ghashghaei, Hilgetag, & Barbas, 2007; Janak & Tye, 2015; McDonald, 1998; Sah, Faber, De Armentia, & Power, 2003). Early accounts of the amygdala linked it principally to fear-related processing; however, a great amount of evidence now suggests that this is a simplification, for the amygdala is involved in affective processing, social behavior, and reward learning. What appears to unify these functions is that the amygdala provides a basic signal of the “salience,” “affective significance” or “value” of sensory information and semantic/episodic/autobiographical memory representations stored in the TP and MTL (Binder & Desai, 2011; Martinelli, Sperduti, & Piolino, 2013; Olson, McCoy, Klobusicky, & Ross, 2013; Squire, Stark, & Clark, 2004), and this information is then transmitted to higher-cortical areas, particularly the vmPFC and lOFC, for hierarchically deeper or more elaborate processing (Adolphs, 2010; Balleine & Killcross, 2006; Baxter & Murray, 2002; Janak & Tye, 2015; Morrison & Salzman, 2010; Murray, 2007). In the context of active inference, salient information that is transmitted to higher levels in the hierarchy corresponds to information that is afforded greater precision. Similarly, the “rewarding” aspects of a cue entail greater precision or confidence about the consequences of action (Friston, Schwartenbeck et al., 2014). We will therefore use salience, reward, and precision interchangeably in this article.

DTI and neuropsychological research has strongly suggested that this transmission of affective salience-tagged (i.e., high precision) mnemonic and sensory information to the vmPFC/lOFC is one of the core functions of the uncinate fasciculus (Von Der Heide et al., 2013). The bidirectional nature of the uncinate fasciculus means that the vmPFC/lOFC can also modify the affective significance of these representations (e.g., by furnishing prior biases or predictions), which is consistent with the fact that the vmPFC is implicated in emotion regulation and fear extinction (Etkin, Egner, & Kalisch, 2011; Schiller & Delgado, 2010).

The vmPFC and lOFC therefore play a central role in this network. Their different patterns of anatomical connectivity with cortical and subcortical structures provide clues concerning their different functions (Bandler, Keay, Floyd, & Price, 2000; Barbas & Pandya, 1989; Carmichael & Price, 1995a, 1995b; Haber, 2016; Noonan, Kolling, Walton, & Rushworth, 2012; Öngür & Price, 2000; Price, 2007; Rudebeck & Murray, 2011; Saleem, Kondo, & Price, 2008; Saleem, Miller, & Price, 2014; Wallis, 2007). The vmPFC and lOFC are densely interconnected with each other and have bidirectional connections with the amygdala, hippocampus, ERC, PRC, PHC, TP, mediodorsal thalamus (MD), and lateral PFC. However, there are a number of notable differences: The vmPFC has bidirectional connections with the posterior cingulate cortex (PCC) and retrosplenial cortex (Rsp), which is not apparent in the lOFC. Furthermore, the lOFC receives inputs from visual, olfactory, gustatory, and somatosensory modalities, whereas the vmPFC receives relatively few direct sensory inputs. These anatomical differences are consistent with the fact that the vmPFC is a central hub in the default mode network (DMN)⁶ and thus plays a central role in internal mentation and self-related processing (Andrews-Hanna, 2012; Martinelli et al., 2013; Northoff et al., 2006; Qin & Northoff, 2011). Finally, although both regions project to the striatum (primarily the ventral striatum), only the vmPFC sends outputs to the hypothalamus and periaqueductal gray (PAG). This allows the vmPFC to potentially modulate a wide range of basic physiological functions mediated by the hypothalamus and PAG: pain modulation, fear/anxiety, stress response, fight–flight response, sleep–wake cycle, sexual/parental behaviors, energy metabolism, body temperature, blood pressure/electrolyte composition, and the autonomic nervous system (Benarroch, 2012; Kandel, Schwartz, Jessell, Siegelbaum, & Hudspeth, 2013). In terms of the Bayesian brain, this functional anatomy is associated with interoceptive inference and providing predictions that engage autonomic reflexes that are essential for affiliative, prosocial, and other actions (Barrett & Simmons, 2015; Pezzulo, Rigoli, & Friston, 2015; Seth, 2015; Seth & Friston, 2016).

While the lOFC and vmPFC are both implicated in value-based processing and decision-making, research suggests that their functions are distinct (for reviews, see Kable & Glimcher, 2009; Levy & Glimcher, 2012; Noonan et al., 2012; Peters & Büchel, 2010; Rangel & Hare, 2010; Rudebeck & Murray, 2011; Sescousse, Caldú, Segura, & Dreher, 2013; Stalnaker, Cooch, & Schoenbaum, 2015). Although there are still many unanswered questions, the lOFC appears to be involved in learning and assigning value/salience to specific stimuli based on their association with specific reward outcomes, whereas the vmPFC encodes the expected subjective value of various stimuli into a continuous “common currency” to determine which is most salient/significant/precise. Specifically, the vmPFC encodes the expected values of various types of information, which can be applied not only to stimuli from the external environment, but also internally generated, self-related mental contents in order represent the personal significance/salience of this information along a continuum (for a review, see D’Argembeau, 2013). The encoding of the expected value of internal/external stimuli (i.e., “valuation”) by the vmPFC is believed to not only facilitate the comparison among options in order to choose the most significant/salient option during decision-making (Kable & Glimcher, 2009; Levy & Glimcher, 2012; Peters & Büchel, 2010), but it is also hypothesized that valuation is essential for constructing a coherent self-representation (D’Argembeau, 2013).

The anatomical connectivity of the lOFC/vmPFC is consistent with this functional differentiation (Rudebeck & Murray, 2011). Specifically, the connections between the lOFC and visual, olfactory, gustatory, and somatosensory regions, as well as amygdala and MTL, means that the lOFC is in an ideal anatomical position to construct a high-dimensional representation of the values of specific stimuli using information about their sensory, memory, and affective components. Similarly, the connections between vmPFC and amygdala, MTL, and lOFC mean that the vmPFC is in an ideal anatomical position to integrate multiple value signals into a one-dimensional representation (“common currency”) of expected value (e.g., the complex value representations from the lOFC and the more basic salience-tagged memory representations and sensory information conveyed along the uncinate fasciculus). Finally, it has been hypothesized that the lateral PFC has complex interactions with the lOFC and vmPFC (Kable & Glimcher, 2009; Northoff et al., 2006; Rangel & Hare, 2010; Wallis, 2007). Specifically, the lateral PFC may exert top-down modulation of the value-based processes performed by the lOFC/vmPFC, selecting and filtering among the set of options generated by the vmPFC/lOFC.

As illustrated in Figure 6, the amygdala–vmPFC network provides a possible neurobiological implementation of the computational architecture underlying lacks remorse and self-aggrandizing described in the section “A Bayesian Account of Psychopathy.” In this model, the trait of lacks remorse reflects a tendency to down-regulate negative self-related processing (e.g., feelings of shame/worthlessness), which we propose would be associated with diminished connectivity between the amygdala and vmPFC along the uncinate fasciculus. In predictive coding terms, this would correspond to an attenuation of (the precision of) prediction errors ascending from the amygdala that would normally inform and update self-representations in the vmPFC. These representations (i.e., expectations) would normally integrate affective, prosocial, and interoceptive cues with those based on representations encoded in the MTL through the assimilation of ascending prediction errors conveyed by ascending connections in the uncinate fasciculus.

While no studies have directly correlated trait-level facets of psychopathy with functional connectivity patterns, studies have found that psychopathy in general is associated with reduced functional connectivity between the amygdala and vmPFC (Decety, Chen et al., 2013; Motzkin et al., 2011; Yoder et al., 2015). Over time, this functional de-coupling could lead to a structural de-coupling, which is consistent with DTI studies reporting that psychopathy is associated with reduced WM structural integrity in the uncinate fasciculus (Craig et al., 2009; Hoppenbrouwers et al., 2013; Jiang et al., 2017; Motzkin et al., 2011; Sundram et al., 2012; Waller et al., 2017; Wolf et al., 2015). Accordingly, this model predicts that a person’s score for lacks remorse should be inversely correlated to the strength of amygdala–vmPFC connectivity. Furthermore, this de-coupling between the amygdala and vmPFC is consistent with numerous t-fMRI studies which have demonstrated that psychopathy is associated with less activation in these structures in response to negative stimuli. For instance, psychopaths show reduced activation of the amygdala and/or vmPFC while perceiving others’ pain (Decety, Skelly, & Kiehl, 2013), imagining others in pain (Decety, Chen et al., 2013), viewing others’ facial expressions of fear, sadness, and pain (Decety, Skelly, Yoder, & Kiehl, 2014; Dolan & Fullam, 2009), during a recognition memory task for negative affective words (Kiehl et al., 2001), during aversive conditioning (Birbaumer et al., 2005; Schneider et al., 2000; Veit et al., 2002), and during emotional moral processing (Fede et al., 2016; Glenn, Raine, & Schug, 2009; Harenski, Harenski, Shane, & Kiehl, 2010; Yoder et al., 2015). The trait lacks remorse can therefore be understood as emerging from the increasing functional and structural de-coupling between the amygdala and vmPFC, which attenuates the negative affective salience-tagged self-related memory representations and sensory information (i.e., prediction errors) transmitted along the uncinate fasciculus. This attenuation or de-coupling effectively cuts off the vmPFC from integrating affective information into the representation the vmPFC is constructing of the expected value of self-related mental constructs (i.e., posterior expectations). This is highly consistent with the Bayesian formulation of lacks remorse described earlier, which characterizes this personality trait as an attenuation of the precision of the ascending prediction errors arising from the patient’s IWM, which decreases the influence of negative affective inputs on the patient’s posterior expectations about their self (Figure 3). In this way, we can understand the possible neurobiological implementation of the Bayesian architecture underlying lacks remorse.

Given that the vmPFC modulates activity in the hypothalamus and PAG (Figure 6), this may also explain well-established physiological findings associated with psychopathy. Meta-analytic evidence shows that psychopathy is associated with significantly lower electrodermal activity at rest, during a task, in response to negative stimuli (e.g., anger provoking, painful, or aversive stimuli), and as a change from baseline (Lorber, 2004). This blunting of sympathetic nervous system (SNS) activity is consistent with the neurobiology of psychopathy and the attenuation of affectively charged prediction errors (see Stephan et al., 2016, for a related discussion of interoceptive inference in the context of fatigue and depression). Specifically, one would predict from the amygdala-vmPFC de-coupling associated with psychopathy that the basic affective signals transmitted along the uncinate fasciculus would not only fail to influence the valuation processes in the vmPFC, but also brain regions modulated by the vmPFC, such as the hypothalamus and PAG, which can regulate SNS activity (Benarroch, 2012; Kandel et al., 2013; Seth & Friston, 2016; Stephan et al., 2016). Such physiological findings are consistent with the hypothesis put toward in this article that lacks remorse is generated by a discounting of affective information, which is mediated by amygdala-vmPFC de-coupling along the uncinate fasciculus.

Although more speculative, there is reason to believe that self-aggrandizing may be associated with a frontostriatal circuit involving the vmPFC and ventral striatum. In support of this proposal, there is evidence that self-esteem is related to frontostriatal connectivity. Specifically, trait self-esteem is related to increased WM structural integrity between ventral striatum and mPFC (including vmPFC), whereas state self-esteem is related to increased functional connectivity along this circuit (Chavez & Heatherton, 2015). The authors of this study hypothesized that frontostriatal connectivity may reflect an integration of self-representations encoded in the vmPFC with positive affect and reward/precision signals mediated by the ventral striatum (Haber, 2016). Interestingly, the “superiority illusion”—the cognitive bias reflected in people tending to evaluate themselves as superior to average—is associated with resting-state functional connectivity between the mPFC and striatum, which was found to be regulated by inhibitory dopaminergic neurotransmission (Yamada et al., 2013). Most relevant to self-aggrandizing is the finding that grandiose narcissism is associated with reduced frontostriatal WM structural integrity between the vmPFC and ventral striatum (Chester, Lynam, Powell, & DeWall, 2016), which is the opposite of individuals with high trait self-esteem (Chavez & Heatherton, 2015). The authors of this study hypothesized that grandiosity may reflect a mechanism that compensates for this neural deficit in the circuitry that connects the brain’s reward systems (i.e., sources of neuromodulatory precision setting projections) with its self-representations, such that grandiose individuals have a low “baseline” self-reward connectivity. The neural regions involved in compensating for this low baseline by abnormally increasing a person’s self-appraisal are unknown and likely involve multiple interacting systems. One likely neural system is the lateral PFC, which can influence valuation and self-referential processes in the vmPFC via top-down modulation (Kable & Glimcher, 2009; Northoff et al., 2006; Rangel & Hare, 2010; Wallis, 2007). While more research is needed, the emerging neuroimaging evidence on grandiosity is consistent with the Bayesian model of self-aggrandizing described earlier, which characterizes this trait as compensatory prior beliefs (i.e., intermediate beliefs) about the self (i.e., a defense mechanism), which upregulates positive self-related information in the face of low baseline self-worth (e.g., feelings of shame/worthlessness), resulting in abnormally elevated self-appraisal.

Putting It All Together

Taken together, while the neuroimaging literature to date is small, the computational architecture depicted in Figure 3 can be mapped onto the neural network associated with psychopathy depicted in Figure 6. The vmPFC corresponds approximately to the intermediate level of the architecture (Figure 3), which encodes the conscious posterior expectations about the self, which is generated by integrating descending top-down prior predictions about the self with ascending prediction error signals from regions supplying other information about the self. This hypothesis that the vmPFC plays a central role in the automatic conscious thoughts about the self in psychopathy is in keeping with the fact that the vmPFC is a central hub in large-scale brain networks that mediate internal mentation and self-related processing (Andrews-Hanna, 2012; Martinelli et al., 2013; Northoff et al., 2006; Qin & Northoff, 2011). The neural regions underlying the high-level prior beliefs encoded in the top level of Figure 3 are currently unknown, though the lateral PFC is likely a key neural system mediating these top-down predictions by virtue of the fact that it can exert conscious top-down modulation of valuation and self-referential processes in the vmPFC (Kable & Glimcher, 2009; Northoff et al., 2006; Rangel & Hare, 2010; Wallis, 2007). The IWM of the self consists of semantic, episodic, and autobiographical memory representations stored in the TP/MTL, with the amygdala providing a basic signal of affective significance to the self-related information stored in these regions. This self-related information is transmitted to the vmPFC along the uncinate fasciculus. Thus the uncinate fasciculus corresponds approximately to the ascending connections to the intermediate layer, and the TP/MTL–amygdala network corresponds approximately to the bottom level of the network representing the contents of the IWM of the self (Figure 3).

It is for this reason that we hypothesize that lacks remorse is mediated by functional and structural de-coupling between the amygdala and vmPFC along the uncinate fasciculus, which effectively cuts off the vmPFC from integrating information from lower levels in the hierarchy (i.e., the IWM of the self) into the posterior expectations it is forming about the self. Given the present state of affairs on the neuroimaging of psychopathy, it is more difficult to make such close links between the computational architecture underlying self-aggrandizing and the neural networks associated with psychopathy. Recall that Figure 3 hypothesizes that self-aggrandizing is mediated by top-down predictions supplied by priors from higher levels in the hierarchy, which modulate the conscious posterior expectations of the self encoded at the intermediate level. As outlined in the preceding review, this specific hypothesis has not yet been tested in the neuroimaging literature. This is because there are no studies to date specifically investigating the link between psychopathy and the brain regions that likely mediate such top-down predictions on the vmPFC during self-related processing (e.g., lateral PFC). That said, the literature to date has suggested that frontostriatal dysconnectivity, which normally connects the brain’s conscious self-representations in the vmPFC with the brain’s reward systems, may contribute to the negative core self-image characteristic of patients with the trait self-aggrandizing. This is compensated for via other brain regions, which we hypothesize the lateral PFC being key among them.

Therefore, based on the emerging neuroimaging research reviewed earlier, it appears that there are at least two, potentially overlapping, pathways that mediate the “undefended” negative core self-image underlying lacks remorse and self-aggrandizing (Figure 3A). The first is the amygdala–vmPFC connectivity along the uncinate fasciculus. The second is the frontostriatal dysconnectivity. While there are at least two pathways, the net effect on self-representations in the vmPFC appears to be the same. Specifically, in the absence of a compensatory de-coupling between the amygdala–vmPFC network (i.e., an undefended state), the psychopathic patient’s amygdala–vmPFC connectivity increases the likelihood that vmPFC self-representations will be influenced by the negative affective salience-tagged self-related memory representations and sensory information transmitted along the uncinate fasciculus. Similarly, in the absence of a compensatory up-regulation of self-related information in the vmPFC (i.e., an undefended state), the psychopathic patient’s frontostriatal dysconnectivity increases the likelihood that vmPFC self-representations will not be integrated with positive affective and reward signals mediated by the ventral striatum. Thus, in both cases, vmPFC self-representations are characterized by negative affective predictions in the absence of compensatory (defense) mechanisms, by virtue of either the presence of negative affective signals from lower level brain regions (i.e., amygdala–vmPFC connectivity) or the absence of positive affective/reward signals from the ventral striatum (i.e., frontostriatal dysconnectivity).

One unanswered question is whether or not these pathways are present in both lacks remorse and self-aggrandizing, or whether they represent different neurobiological instantiations of the computational architecture outlined in Figure 3. That is to say, the undefended state in lacks remorse and self-aggrandizing could be mediated both by amygdala–vmPFC connectivity and frontostriatal dysconnectivity, or the pathways may be unique to each trait (e.g., the negative core self-image in self-aggrandizing may be mediated uniquely by frontostriatal dysconnectivity, whereas the negative core self-image in lacks remorse may be mediated uniquely by amygdala–vmPFC connectivity). Relatedly, it may be that self-aggrandizing and lacks remorse are both associated with defensive de-coupling between the amygdala–vmPFC along the uncinate fasciculus. This would be consistent with our model, given that both are defensive responses to negative self-related information. However, the present state of psychopathy neuroimaging research does not provide firm answers to these questions. This is because there has not yet been an analysis of structural or functional neuroimaging data examining the relationship between the traits lacks remorse and self-aggrandizing and these two pathways in the same sample of psychopathic patients. Such a study would help shed light on these hypotheses regarding the neurobiological instantiation of the undefended negative core self-image underlying lacks remorse and self-aggrandizing and the neural circuitry of these defense mechanisms.

Treatment Implications: Integrated Modular Treatment for Psychopathy

A full discussion of the treatment implications of the Bayesian model, and the integrated psychotherapeutic model more generally, extends beyond the scope of this article. That being said, an outline of the treatment implications is possible. The etiological framework of our Bayesian model resonates with major trends in how some clinicians are beginning to conceptualize the treatment of PDs. Over the past number of years, there has been a growing appreciation that specialized therapies rooted in specific “schools” of psychotherapy (e.g., CBT vs. SFT vs. dialectical behavior therapy vs. transference-focused psychotherapy vs. mentalization-based treatment) tailored to specific diagnostic categories may not be the most effective strategy for treating PD. Rather, it may be more effective to utilize an integrated approach to PD treatment. It is for this reason that the integrated modular treatment (IMT) was developed (Clarkin, Cain, & Livesley, 2015; Livesley, 2003, 2005, 2007a, 2007b, 2012; Livesley, Dimaggio, & Clarkin, 2016). Our model resonates closely with the IMT and its integrated approach to PD treatment. Before summarizing the IMT and how our model fits into this framework, it is necessary to outline the three major reasons for shifting toward an integrated approach to PD treatment: (a) the evidence for “common factors,” (b) the utility of technical eclecticism, and (c) the evidence for theoretical integration across schools of psychotherapy (Livesley et al., 2016).

The first reason is the evidence for common factors across therapies. This is the finding that there are few clinically significant differences in the efficacy across psychotherapies for PD, including general psychiatric management and supportive psychotherapy (Budge et al., 2013; Clarkin, Levy, Lenzenweger, & Kernberg, 2007; Cristea et al., 2017; Leichsenring & Leibing, 2003; McMain et al., 2009). Although there are sometimes differences in outcome between studies, they are often small and/or difficult to interpret because they rarely demonstrate clear superiority of one specialized therapy from a particular “school” over another in head-to-head comparisons. The lack of evidence for clinically significant differences in outcome across therapies is seen in studies of treatments for borderline PD (Cristea et al., 2017) and antisociality (Landenberger & Lipsey, 2005; Lipsey, Landenberger, & Wilson, 2007). Of course, this is not surprising at all. It was pointed out as early as the 1930s by Rosenzweig that there are a set of “common factors” that underlie all bona fide psychotherapies, and these general change mechanisms common to all therapies for PD explain the equivalent efficacy across treatments (Rosenzweig, 1936). Indeed, there is now compelling evidence from the general psychotherapy literature that there is a set of common factors shared across specialized therapies that together account for a large proportion (at least 50%) of the variance in positive outcome (Horvath & Symonds, 1991; Luborsky, Singer, & Luborsky, 1975; Luborsky et al., 2002; Marcus, O’Connell, Norris, & Sawaqdeh, 2014; Martin, Garske, & Davis, 2000; Wampold, 2001). What are these general change mechanisms? Lists of these common factors sometimes differ; however, the common factors can be grouped into six general change mechanisms (Livesley et al., 2016):

• 1.   Structure. Establish a highly structured treatment process that defines the therapeutic stance (i.e., therapist provides support, empathy, and validation) and establishes an explicit treatment contract that defines the purpose, format, terms, and limits (e.g., treatment boundaries) of the therapy.
• 2.   Treatment alliance. Establish and maintain a collaborative treatment alliance.
• 3.   Consistency. Maintain a consistent treatment process by adhering to the treatment structure (i.e., the therapeutic stance and treatment contract).
• 4.   Validation. The therapist promotes validation by recognizing and affirming the legitimacy of the patient’s experience (i.e., that the patient’s thoughts, feelings, and behavior make sense and are understandable).
• 5.   Motivation. Build motivation and commitment to change.
• 6.   Meta-cognition. Promote self-observation, self-knowledge, and self-reflection.

In commenting on the causal role of common factors in psychotherapy, Markowitz (2014) has underscored the following paradox: “if you cannot do this [i.e., deliver the common factors], the rest of psychotherapy does not matter”; however, “if you can do this, the rest of psychotherapy does not matter [because the majority of the causal ingredients linked to positive outcome are provided by these general change mechanisms]” (pp. 287–288, emphasis original).

The evidence for common factors, however, does not preclude the possibility that specific interventions unique to specialized therapies may contribute to positive outcomes in PD, independent of general change mechanisms. In other words, there may also be specific change mechanisms required for specific types of dysfunction, and thus specific interventions may be required for the unique problems seen in particular patients. Those therapies that contain these “specific ingredients” (i.e., specific interventions) will show superior efficacy to those that do not. The empirical literature suggests that this may indeed be the case. Meta-analytic evidence suggests that acute symptoms may respond better to more structured cognitive-behavioral techniques than less structured psychodynamic techniques (Marcus et al., 2014). Conversely, psychodynamic techniques may be more effective for increasing self-reflection relative to structured cognitive-behavioral techniques (Livesley et al., 2016). Similarly, in the realm of rehabilitation for people involved in the criminal justice system, structured cognitive-behavioral techniques are robustly associated with reduced violent/antisocial behavior, whereas generic interventions are not (Andrews et al., 1990; Landenberger & Lipsey, 2005; Lipsey & Cullen, 2007; Lipsey et al., 2007; McGuire, 2008).

The take home message is that the empirical literature on PD treatment, and psychotherapy more generally, suggests that effective, evidence-based treatment for PD requires an integrative “both/and” approach, rather than a tribal “either/or” approach that forces clinicians to choose between different specialized treatments from different schools of psychotherapy (Marcus et al., 2014): treatment should both explicitly utilize general treatment interventions shared across specialized therapies to maximize common factors and explicitly utilize specific treatment techniques that have been shown to be effective for specific types of dysfunction.

This dovetails into the second reason for the IMT approach, which is the utility of technical eclecticism over fidelity to the prescribed techniques of specialized treatments (Livesley et al., 2016). This is because specialized treatments often reduce the range of psychopathology seen in PD to a single primary impairment. For example, the primary impairment in borderline PD has been explained in terms of (a) affect dysregulation = dialectical behavior therapy, (b) dysfunctional beliefs = CBT, and (c) deficits in mentalization = mentalization-based therapy (Bateman & Fonagy, 2016; Davidson, 2007; Linehan, 1993). Although simplifying clinical complexity is helpful, the problem is that such reductions narrow the focus of treatment and the selection of interventions to only those implied by the underlying theory (e.g., dialectical behavior therapy focuses on increasing the patient’s affect regulation skills). Specialized treatments thus risk neglecting other explanatory factors and thus other potentially useful techniques to help treat the diverse range of problems patients face (e.g., patients with borderline traits have difficulties with affect regulation but also maladaptive cognitions, mentalization, impulse control, self and interpersonal problems, etc.). Technical eclecticism avoids this theory-imposed restriction by integrating specific interventions from diverse therapies, without necessarily endorsing all their underlying theoretical constructs, in order to address the full range of impairment seen in PD.

The third reason for an integrated approach to PD treatment is that the psychotherapeutic field is converging on a number of core ideas about the etiology of PD. A concrete example of such theoretical integration was described in the section “Toward an Integrated Psychotherapeutic Model of Psychopathy,” where we saw the convergence of psychodynamic, CBT, and SFT models of the traits lacks remorse and self-aggrandizing. While often described using different terminology, almost all current evidence-based psychotherapies for PD emphasize at least some, and sometimes all, of the following: (a) that the maladaptive thoughts, feelings, and behaviors seen in PD are driven by entrenched cognitive–affective structures (e.g., IWM, schemas); (b) that these cognitive–affective structures are often not fully consciously articulated or accessible (i.e., they are unconscious or partially conscious) yet organize and influence a patient’s conscious experiences and behavior; (c) that genetics, learning, developmental factors, and the quality of attachment experiences play central causal roles in shaping these cognitive–affective structures and thus current psychopathology; and (d) that some of the psychopathology observed in patients with PD is a consequence of the operation of maladaptive “defense” or “coping” mechanisms that protect against distressing or unacceptable thoughts and feelings.

The IMT model is an integrated approach to PD treatment and was developed to explicitly address these three issues. The IMT model integrates the treatment principles and methods that work across therapies, called general treatment modules, with specific treatment modules which consist of an eclectic array of specific interventions, in order to target both the common and unique features, respectively, of patients with PD (Clarkin et al., 2015; Livesley, 2003, 2005, 2007a, 2007b, 2012; Livesley et al., 2016). Thus general treatment modules target the core impairments in self and interpersonal functioning common to all personality pathology (APA, 2013; Bender et al., 2011; Livesley, 2011; Morey et al., 2015; Morey et al., 2011; Skodol, 2012, 2014), whereas specific treatment modules target the unique profile of the patient’s pathological personality traits. More specifically, general treatment modules explicitly utilize the common factors (i.e., structure, treatment alliance, consistency, validation, etc.) in order to create a therapeutic process that provides a continuous corrective experience that treats the core impairments in self/interpersonal functioning. Thus general treatment modules are used with all patients and throughout treatment. Conversely, specific treatment modules are selected from all therapies, based on empirical and rational considerations, to treat the specific domains of dysfunction that are the current focus of treatment. In the IMT, the focus of treatment, and thus the selection of specific interventions, is determined by the domains of dysfunction that are currently present in the patient: (a) acute symptoms (e.g., dysphoria, rage, self-harm/suicidal behavior, violence), (b) regulation and modulation (e.g., difficulty regulating maladaptive thoughts, feelings, and behavior), (c) interpersonal problems (e.g., conflictual relationships), and (d) self problems (e.g., difficulty regulating self-esteem, unstable and fragmented identity). These four domains of dysfunction imply a hierarchy of treatment foci (i.e., priority is given to specific interventions that target acute symptoms when this domain is present over specific interventions targeting interpersonal problems). Furthermore, they imply that patients typically progress, not necessarily linearly, through five phases of change during their treatment with the IMT (Clarkin et al., 2015; Livesley, 2003, 2005, 2007a, 2007b, 2012; Livesley et al., 2016):

• Phase 1 of Change: Safety. The primary goal is to ensure the safety of self and others (i.e., crisis management). This is done through the general treatment modules of structure, validation, and support, which often successfully ensure safety in a community setting. However, sometimes safety cannot be ensured in the community with general change mechanisms alone—a situation routinely encountered in corrections and forensic mental health settings. This requires supplementing them with specific interventions, including inpatient hospitalization and/or restraints and seclusion.⁷
• Phase 2 of Change: Containment. The primary goal is to contain affective, behavioral, and cognitive instability. Safety and containment quickly merge into each other during crisis management, and likewise involve similar general treatment modules supplemented with specific interventions consisting of scheduled and/or PRN medications. During this phase, the treatment alliance and motivation/commitment to change begin to develop, and a consistent treatment process starts to form between the patient and their clinician.
• Phase 3 of Change: Control and regulation. The primary goal is to increase self-regulation of maladaptive cognitions, affects, impulses, and behaviors. During this phase, general change mechanisms continue to build motivation/commitment to change and improve meta-cognition. These are typically supplemented with specific cognitive-behavioral techniques that promote self-regulation in these areas. Medications have limited utility in treating personality pathology after the containment phase (Khalifa et al., 2010; Lieb, Völlm, Rücker, Timmer, & Stoffers, 2010).
• Phase 4 of Change: Exploration and change. The primary goal is to increase exploration and modulation of the maladaptive cognitive–affective structures related to interpersonal problems. General change mechanisms can be supplemented with cognitive-behavioral methods and psychodynamic techniques that promote awareness of maladaptive interpersonal schemas and their origins, and begin to restructure them.
• Phase 5 of Change: Integration and synthesis. The primary goal is to construct an adaptive sense of self and resolution of interpersonal problems. The transition to this phase is typically seamless because the process of exploring and resolving interpersonal problems is intimately tied up with discussions with the patient about issues of identity and self-direction. General treatment modules are supplemented with specific interventions that promote restructuring of maladaptive self-schemas, the formation of an adaptive self-narrative and coherent sense of self, and the construction of a “personal niche” whereby the patient can meaningfully engage in work, love, and play.

Therefore, as we can see, general treatment modules are used throughout all phases of change, and the clinician supplements them with specific treatment modules tailored to target the current domain of dysfunction. Thus, Phases 1 and 2 are concerned primarily with treating acute symptoms, Phase 3 with treating difficulties in regulation and modulation, Phase 4 with treating interpersonal problems, and Phase 5 with treating self problems.

With this basic outline of the IMT unpacked, the treatment implications of our Bayesian model of psychopathy become apparent and provide the basis for an Integrated Modular Treatment for Psychopathy. Recall that our model is centered on two hypotheses:

• 1.   Early adverse attachment experiences interact with genetic vulnerabilities to shape the development of a psychopathic patient’s core self-image (i.e., internal working model of the self) as worthless and shameful.
• 2.   The traits lacks remorse and self-aggrandizing are compensatory high-level prior beliefs (i.e., intermediate beliefs) that function as defense mechanisms, allowing the patient with psychopathy to eliminate or at least diminish (i.e., cope with) the influence of their negative core self-image on their conscious experience.

In other words, at the heart of our model is the idea that psychopathy is characterized by a core impairment in self functioning—namely, deep-seated feelings of worthlessness and shame arising from the internal working model of the self. This implies that general treatment modules likely play a primary role in the treatment of psychopathy, because the new experiences within treatment can provide a continuous corrective experience that challenges these core beliefs. Specifically, the core self-image of worthlessness, shame, and inadequacy are challenged by consistent and regular exposure to experiences with a clinician who (a) is supportive, empathic, and validating; (b) shows congruence/genuineness in the relationship; (c) is carefully attentive to and flexibly resolves ruptures in the treatment alliance; (d) has an attitude toward the patient consisting of collaboration, respect, care, and positive regard; and (e) models and reinforces, in a nonpunitive, nondevaluing, and nonshaming manner, appropriate boundaries and prosocial internal standards (Castonguay & Beutler, 2006; Clarkin et al., 2015; Critchfield & Benjamin, 2006; Livesley, 2003, 2005, 2007a, 2007b, 2012; Livesley et al., 2016; Mitchell, Tafrate, & Freeman, 2016; Tafrate & Mitchell, 2014). In this way, the general treatment modules of the IMT for psychopathy function analogously to the SFT concept of limited re-parenting, which “involves providing, within the appropriate boundaries of the therapy relationship, what patients needed but did not get from their parents as children” (Young et al., 2003, p. 177). These general treatment modules are likely counterintuitive to people who have never worked with antisocial individuals. However, as is well known in correctional rehabilitation, a “get tough,” “get real,” or confrontational attitude never works, and is likely iatrogenic, because it invariably leads to reactance, argumentativeness, disengagement and, eventually, shatters the working relationship and thus the possibility for treatment (Mitchell et al., 2016; Tafrate & Mitchell, 2014). In short, when it comes to the core of treatment in the IMT for psychopathy, patients with psychopathy are no different than any other patient with a PD: The same common factors are utilized and maximized throughout all phases of change.

The second treatment implication of our model is the use of specific treatment modules targeting the maladaptive intermediate beliefs (i.e., high-level prior beliefs) of self-aggrandizing and lacks remorse (i.e., the unique features of the patient). Once patients with psychopathic traits have transitioned into the control and regulation phase, the focus of specific interventions should turn to increasing self-regulation of these maladaptive thinking patterns. These thinking patterns are, in fact, well known in the risk assessment and correctional rehabilitation literatures. Self-aggrandizing and lacks remorse are two instances of a broader class of cognitions called criminogenic thinking patterns—one of the “Central Eight” strongest, most robust risk factors for violent and antisocial behavior (Andrews & Bonta, 2010a, 2010b; Bonta et al., 2014; Gendreau et al., 1996). Criminogenic thinking patterns are attitudes, values, and beliefs that facilitate violent/antisocial behavior, and are believed to operate at the level of intermediate beliefs (Mitchell et al., 2016; Seeler, Freeman, DiGiuseppe, & Mitchell, 2014), which is in keeping with our Bayesian model’s hypotheses (Figure 3). Moreover, given that criminogenic thinking patterns contribute significantly to the harm of others and a wide range of dysfunction in the patient’s life, almost all evidence-based correctional rehabilitation programs involve specific techniques to alter criminogenic thinking, and these techniques involve standard cognitive-behavioral methods (e.g., identifying automatic thoughts and their associated intermediate beliefs, increasing awareness of the link between thoughts, emotions, and behavior, Socratic questioning of beliefs, behavioral experiments, etc.; Lipsey et al., 2007). Therefore such techniques targeting criminogenic thinking patterns may be crucial modules for treating the maladaptive cognitions associated with self-aggrandizing and lacks remorse during Phase 3 of treatment. Such techniques would be part of an eclectic array of specific treatment modules used during this phase in order to increase the patient’s self-regulation of their maladaptive cognitions, affects, impulses, and behaviors.

In summary, our Bayesian model of self-aggrandizing and lacks remorse resonates closely with the IMT model of PD treatment. For this reason, we have provided a general overview of a novel Integrated Modular Treatment for Psychopathy which we believe, with further development, can provide a fruitful, evidence-based framework for translating the computational neuroscientific concepts discussed in this article into a treatment for psychopathy that can be evaluated for efficacy/tolerability and cost-effectiveness.