The Brain has a Mind of its Own: Attachment, Neurobiology and the New Science of Psychotherapy

Jeremy Holmes writes in this 2020 book:

* Energy in FEP [Free Energy Principle] is not a physical phenomenon like heat, or electromagnetic radiation, but a superordinate explanatory category, akin to gravity (cf. Connolly & van Deventer, 2017), with both mental and physical connotations. FEP is a principle or framework for understanding the fundamentals of psychic life, conscious and unconscious, analogous, and not unrelated to, Freud’s pleasure and reality principles.
According to the FEP, the brain’s task is to select from, attend to, shape, and maintain homeostasis in the face of the streams of incoming neural energy from both its sense organs and its interoceptive and proprioceptive internal milieu. It does this by predicting , “top-down”, on the basis of previous experience, the likely meanings of this “bottom-up” input. These predictions follow the mathematics of the eighteenth-century cleric Thomas Bayes, and are thus known as “ Bayesian ”. The ever-changing discrepancies between prediction and sensation, between our generative models of the world and reality, activate Prediction Error Minimisation (PEM), in which the brain “instructs” itself to modify prior models of the world in the light of experience, whereby they become posteriors, and take actions which improve precision, clarify ambiguity, and align input with expectations.
From a psychotherapeutic viewpoint, interoceptions (i.e., bodily feelings) are especially important because they underpin affective life . In general, prediction errors – the discrepancies between what we want/expect and what our senses tell us is the case – are experienced as “bad” or painful, thereby motivating their minimisation. Conversely, when expectation and experience align, we feel “good” or happy. The psychological distress that brings people for psychotherapeutic help can be conceived as chronic states of unresolved prediction error. The aim of psychotherapy is to redress these by mobilising the capacity for action and model revision.
In FEP, energy is either free , or bound . Free energy reflects the ever-changing and potentially chaotic nature of the impact of the environment on the physical, psychological, and interpersonal self. Energy’s role is therefore ambiguous : it provides the vital information and sustenance needed for our evolutionarily derived tasks of adaptation, survival, and reproduction, and arguably forms the basis for creativity, but, unbound, can overwhelm the unprepared nervous system. The need to find and bind free energy is what motivates us, what “makes us tick”, what makes us exploit what we have, and explore and want to know more, and to think up better world models; failure to do so is demotivating, degenerating, and depressing.

* the brain needs to build up a repertoire of survival scenarios in order to match the likely risks with which its owner will be confronted in the course of a lifetime. Prediction error minimisation is continuously finessed against novelty so that we can go through the prior–posterior revision that improves adaptation. Since energy-binding is “rewarding” – via the dopaminergic system – PEM is powerfully motivating. When we are demotivated, or energy-minimising procedures are compromised by defective agency, chronic negative affective states presage psychological illness. Psychotherapy attempts to create the “duet for one” conditions where surprise becomes allowable and ultimately pleasurable – not least, as we shall see, in the form of healing tears.

* Prediction error minimising (PEM) will steer active inference – “Shall I look a little closer to make sure, or assume the worst and run away?” This describes a “secure attachment” response – a balanced attempt to match input with possible generative models of the world. In a UK park the stick is most unlikely to be a snake. In an Indian paddy field, it could well be.
A less secure response would be hyper activation, aka anxious attachment – “Treat all sticks as snakes” – or hypo activation, aka avoidant attachment – “Forget snakes, stick to sticks” – (cf. Mikulincer & Shaver, 2007). Each strategy may be understandable in the individual’s developmental context, but maladaptive in the present moment, the former leading towards anxiety/depression, the latter to potentially fatal risk-taking.

* A further useful, if oversimplified, distinction is between the worlds of people whose predictive dispositions classify them as orchids or dandelions (Boyce, 2019). The former (about 20 per cent in community samples) are hyper-responsive: in favourable environments they flourish, in unfavourable ones they fear they may wither and die. “Dandelions” are relatively insensitive to environmental influence, doing less well than “orchids” in good ones, but are somewhat impervious to unfavourable influences.

[LF: A friend says: I tell people I’m a sports car not a jeep. I’m not all weather, all conditions, reliable. I’m fickle, need maintenance, but if the road is clear, I can make up years of ground on people, but I will get caught in mud, rain. I’m not as resilient as normal people.”]

* Let’s turn to an everyday example of the Bayesian brain in action, hoping, with its help, to link PEM with some familiar psychoanalytic themes.
One fine spring morning, in the course of a daily run across agricultural land, I noticed that the farmer had recently sprayed weedkiller. There was an unpleasant sickly smell, eliciting slight nausea, bringing to mind a mild feeling of illness I’d had at a similar time the previous year. The next day, following the same course, the smell had gone, but I noted in my peripheral vision a dark flapping object. My first thought was that this was a bird, perhaps a crow, affected by yesterday’s poison. I turned my head to engage central vision , then approached to investigate further and if necessary rescue the corvid. The closer I came, however, the putative stricken bird revealed itself to be no more than a fragment of wind-blown black plastic, part of a discarded fertiliser bag.
This trivial incident illustrates a number of FEP and Bayesian principles.
• The stimulus was ambiguous , and therefore subject to high levels of error and potentially “free energy”.
• The “ prior ”, or meaning, attributed to this experience was based on selective sampling in peripheral vision together with recalled interoceptive nausea, leading to a top-down construct of “sick bird”. This memory-based construction of an ambiguous stimulus could be deemed as an example of transference .
• In order to resolve the ambiguity, free energy minimisation (FEM) was required, via a) action – turning my head and moving towards the flapping object in order to reduce “noise”, and increase the precision of sensory sampling, and
b) hypothesis revision – “The poison will have dispelled by today so it would be odd if the bird were still affected”.
• Active inference led to a stable “ posterior ”: a free energy-minimised representation of reality, external (“It’s only flapping plastic”) and internal (“No nausea; no illness”).
• Parsimony had generated two possibilities: sick bird, or plastic bag; the latter prevailed. Flap (surprise) became no-flap. Free energy was once more bound.

* Parents who are good at mentalising tend to have secure infants (Meins, Fernyhough, Fradley, & Tuckey, 2001). They readily put themselves in the child’s shoes, and can see that what from an adult perspective might seem trivial – a mother momentarily inaccessible – to a small child would feel dangerous and trigger abandonment anxiety. Parental mentalising – understanding and resonating with their infants’ affects – is initially non-verbal and implicit, communicated by facial expression, tone of voice, affiliative touch, swinging rhythms of soothing or stimulation. These embodied gestures present a model of the infant and her world from the caregiver’s perspective. This helps the child to integrate primary sensory signals into regularities of emotional and interpersonal meanings.

* People suffering with personality disorders are typically on a hair trigger for overwhelming anxiety (Allen, Fonagy, & Bateman, 2008). For them fast, rather than slow thinking is the norm (Kahneman, 2011). They are in the grip of perceptual distortion and ingrained prediction errors, driven by the need to bind free energy into a modicum of predictability, however dysfunctional. An early psychotherapeutic task is to re-establish a degree of biobehavioural synchrony. The patterns and rhythms of therapy help with this, as do the joint attention and affective mirroring typical of secure attachments. But the more disturbed the individual, the more problematic this is likely to be; such trust remains a fragile and fluctuating flower which varies from session to session and moment to moment within sessions. Identifying, understanding, and overcoming obstacles to biobehavioural synchrony are preconditions for effective psychotherapeutic work.

* PEM underwrites the psychiatric maxim that it is counterproductive to try to argue people out of their delusions. Acceptance of the validity of clients’ experience is a precondition for progress to a higher level of the PEM hierarchy where the possibility of illness can be jointly entertained, without clashing with lower-level faulty inferences.

If psychosis represents dysfunction in topdown prediction error minimisation, FEP sees autism as the converse. Here there is “enslavement to the senses” in which the inherent noisiness of sensory input is discounted. The capacity to contextualise and attend only to relevant input is offline, while mentalising – thinking about thinking – and hence subjecting sensory data to relevance criteria, is in abeyance.

* Psychopathology is conceived in the FEP framework as difficulties with active inference, due to a) impaired agency/action and/or b) failure of model revision in the light of experience. These result from:
1 Over-weighting top-down inference (psychosis)
2 Over-weighting interoceptions (somatisation disorders, depression) or exteroceptions (autism)
3 Paucity of priors (trauma)
4 Repression of interoceptions, making them unavailable for conscious-level PEM (depression)
5 Difficulties with recruiting others in duets for one, and hence go-it-alone maladaptive PEM procedures (substance abuse, suicidal acts, personality disorders)
6 Inappropriate complexity procedures: oversimplistic priors (personality disorders), or failure to reduce complexity and hence inhibition of action (OCD).

* If mental illnesses are diseases of social brains, then it is likely that evolution will have produced both natural and culturally mediated repair systems to reverse or mitigate them. In order to stave off entropy, living systems have evolved defences which help resist chaos, maintain structure, and enhance adaptation and survival (Connolly, 2018). In humans defences operate “all the way up”, from the cellular level of the immune system, through the interpersonal attachment dynamic, to societal structures, ranging from social care to tidal barriers and military hardware. These, like the systems they are designed to protect, are initially involuntary and automatic but, prosthetically enhanced, become goal-directed societal formations. We share our immune system with fellow mammals, but the epidemiology of high-density communities, urbanisation, and migration mean that we need vaccination and immunisation programmes to augment our capacity to outwit disease.

* From an FEP perspective, dreaming helps rework potentially traumatic free energy so that the associated terror can be “bound” and its concomitant mental pain made more tolerable and less disruptive. Top-down inferences, capable of linguistic representation, “bind” the energy associated with likely future fears – moving, changing, aging. By generating multivalent free-energy minimising priors they reduce prediction error. There is no escaping the emotional pain of loss, separation, and death, but if (to adapt Kipling) those “impostors” can be “met” with top-down priors, they will safeguard against, or at least postpone, entropic surprise.

* As one novice therapist breathlessly announced during her coffee break, “I’ve finally realised that transference is real: my first patient today described me as a hideous witch; the second said I was a beautiful angel.”

[LF: Reactions to Tony Fauci are often transference for people’s feelings about government bureaucratic power. Liberals generally approve while conservatives don’t.]

* For cooperation to be effective we need to take account of the viewpoints and motivations of others, and likewise to factor in our own psychology and how it will be perceived by our fellow co-operators. Again, this is typically implicit and below consciousness. It is a remarkable fact of urban life that opposing pedestrians in busy streets rarely bump into one another: walkers make unconscious predictions about the direction of travel of others and self and all runs smoothly.

* First, an “event” – for instance, a client’s outburst of explicit or covert anger triggered by a therapist’s holiday break, occupying no more than moments of clock time, may lead to extended collaborative reflection. Second, the discussion is likely to attend to “much previous” comparable interpersonal experience. The aim is to identify and modify both bottom-up and topdown procedures. Enhanced sensory sampling means that the client begins to tease out differences and the automatic assumptions these evoke, such as between a therapeutic break with a high probability of resumption, and a childhood history of being arbitrarily “dropped” by a divorcing parent. This scrutiny, if things go well, can lead to more realistic model revision about the nonirreversibility of losses.

* Avoidant clients, with intellectual defences, are both resistant to, and likely to benefit from encouragement to free associate. With their co-regulatory sensitive period reopened, they can explicitly think about repressed feelings and fears, which no longer have to be minimised for the sake of security. As in the Coan (2016) study, the potential for disruptive free energy associated with avoided feelings is mitigated by the therapist’s calm presence. Conversely, those with anxious attachment strategies typically feel overwhelmed by the uprush of interoceptive bodily feelings. Here the job of free association is to slow things down so they can be identified and subjected to rational top-down scrutiny (“Could we consider the possibility that that horrible stomach pain you get whenever your husband goes on a trip doesn’t necessarily spell gut cancer?”).

* Therapists help their clients to own their actions, and to become better able to differentiate those for which they are responsible from those in which they were victims, and thus to develop feelings of autonomy and enhanced control over their lives.

* From an FEP perspective, insecure attachments are vulnerability factors for psychological illness because they compromise active inference (Holmes & Slade, 2017). In the absence of an external or internal secure base, exploration, physical and psychological, is curtailed. This both limits the extent and range of sensory sampling of the environment, and the variety of priors or hypotheses available to account for them. Both the “breaking” (i.e., creative destruction) of existing priors and the “making” (i.e., creative construction) of new ones are inhibited (Holmes, 2010).
In anxious or “hyperactivating” attachment, agency tends to be absent or eroded. Rather than actively searching or changing their environment, sufferers remain passive in the face of loss, conflict, or trauma (Knox, 2010), a state famously described as “learned helplessness” (Maier & Seligman, 2016). Here the self is suffused with unmodulated affect. In terms of structure learning, commitment to the single prior of hopelessness (energy binding, but paralysing) – “Nothing I do will change anything” – precludes finding ways to live productively in the environment in which sufferers find themselves, and inhibits the testing of alternative hypotheses (“Maybe if I try this, things won’t be so bad after all”).

* Consider the tennis player John McEnroe’s famous challenge to the umpire: “ You cannot be serious ” in calling a ball “out” that the player was convinced was “in”. Pre-Hawk-Eye, an FEP-informed umpire might argue: “Neither of us can be absolutely certain whether that ball was on or off the line. Perceptions are inherently subject to error. Tennis balls fly faster than nerve impulses transmit. Your error minimisation procedure is informed by your interoceptive signals of desire to win the match. Your fury is an acting out of that desire, a manifestation of energy unbound. I on the other hand have no vested interest in who wins this game. My error minimisation procedure is fueled neither by amygdala-driven fear of failure, nor wish for dopaminergic reward. The rules of tennis require that you borrow my brain in inherently ambiguous circumstances such as this. My final and considered decision therefore is – the ball was out!”

The word conversation itself, with its implication of both togetherness (“con”) and opposition (“versus”), points to the essence of the psychotherapeutic project: acceptance – and challenge. The etymology of the word “conversation” includes the ideas of a) home, and b) sexual intercourse. Both bear on the work of psychotherapy. By providing a “home” – in attachment terms a secure base or holding environment – therapists offer the sense of safety and attention needed to explore and rework one’s deepest dispositions. “Sexuality” taken as a metaphor points to the arousal of intersubjectivity, mutuality, complementarity, and somatopsychic excitement which therapy must call into being if more complex psychic structures are to be generated.
If sex is a conversation, and conversation a form of sex, so too is attachment.

* Clients suffering from depression are conceptualised as being in the thrall of cognitive errors which dominate their affective world: “Everyone hates me”, “I am useless”, etc. These self-perpetuating, albeit spuriously parsimonious generalisations bind free energy, but also undermine agency. Passive helplessness pervades, interspersed with self-perpetuating depressive auto-denigration. CBT encourages its clients to see these negative views as “hypotheses”, in need of active testing through “experiments”. Action is thereby encouraged, with the hand-holding help of a therapist, enhancing sensory precision and reducing prediction error. When things go well, depressive priors begin to be revised in the light of experience: “Maybe I’m not such a failure as I thought I was”.
In psychoanalytic therapies the role of “action” is less explicit. But the very act of seeking help for psychological difficulties implies a degree of agency. Moreover, the process of “giving sorrow words” in the therapeutic setting, if seen in terms of speech acts, is agency-enhancing. By helping clients find the words to actualise their inchoate feelings, therapists help with the error minimisation that is, in Bernard’s terms, the condition of a free life. Expressing negative affect, with its concomitant enhancement of attention to interceptive detail, is especially conducive to reworking dysfunctional priors.

* To recall an incident from the author’s training, he was observing a young woman with anorexia nervosa being interviewed by a senior psychiatrist. In an interchange that lasted for five minutes or more, the following interchange was repeated over and over:
Interviewer: “Do you worry about your appearance?”

Patient: “Oh, no they’re fine, I never think about my parents” … etc., etc.

* The “conversational model” or “psychodynamic interpersonal therapy” (PIT) (Barkham, Guthrie, Hardy, & Margison, 2017) specifically foregrounds the therapeutic role of dialogue strategies. A prime therapeutic aim is to help clients develop a “feeling language” with which to give form to their inchoate sufferings. In the course of treatment patients learn to listen to their bodily sensations, and to find words with which to express, manage, and live with them. This corresponds with helping to overcome the interoception/generative model deficit, especially found in avoidant clients. Attention to the “minute particulars” of experience is emphasised, especially insofar as they are salient to the specific affordances of an individual’s life.

* An important psychotherapeutic implication of FEP is that in its mission to minimise surprise, as a proxy for entropy or free energy, the brain/mind can diminish or obliterate complexity and differentiation. The relational bedrock of psychotherapy helps clients tolerate and survive surprise , and so find new and more healthy ways of binding mental energy.


The binding of free energy equates to the resolution of prediction errors (i.e., surprise and uncertainty). Thus, the conversion of free into bound energy results from belief-updating to increase the accuracy–or decrease the complexity–associated with our beliefs about the world’s states of affairs.

In sum, Friston maintains that the brain’s main aim is to minimize “surprise”–as best it can.

Prediction error is minimized in two main ways:

Action, which reduces prediction errors by selectively sampling sensations that are the least surprising,6 thereby helping to approximate the organism to its environmental niche, or affordance (see below).
Perception. Changed perceptions follow from belief updating resulting in more reality-consonant predictions.

* We will touch on a number of key features of the analytic approach: free association, dreams, sexuality, reflective discourse, transference, and mentalising. All depend on “decoupling”–introducing a degree of “play” into the bottom-up/top-down surprise-minimizing articulations of everyday life (c.f., Holmes and Slade, 2017). In the presence of a modulating, moderating, affect-buffering therapist, surprise/energy unbound becomes tolerable and, when therapeutically scrutinized, extends the repertoire and range of a person’s counterfactual realities, i.e., priors. Built into this model is both “creativity” and “destruction,” in the sense that modification of error-prone priors entails their replacement with alternative hypotheses. The greater the range of prior hypotheses, the greater the opportunities for error-minimized binding and the less the need to resort to rigid, limited, or anachronistic priors, at the different levels of a hierarchy of generative models. This, in turn, enhances the adaptedness of the sufferer to their environment, including, via mentalising, the self. Part of the process makes the patient’s model more accurate by revised belief-formation, and part by complexity reduction, especially in relation to resolution of conflict and trauma.

* From a free energy perspective, psychological ill health implies simplistic top-down models, and/or restricted sensory sampling, while structured complexity, as opposed to chaos or rigidity, is a mark of psychological health. Psychotherapy aims to increase the repertoire of its subjects’ models of themselves and their environment. It is no mean task for analysts to challenge their patients, to break the mold of maladaptive energy binding, and to move psychic structures toward this augmented complexity.

* Analyzing transcripts of psychotherapy sessions, they show how the nature of therapeutic dialogue depends on the attachment status of both client and therapist. Securely attached clients–and therapists–engage in turn-taking “duets,” in which there is contact seeking, free exchange and modulation of affect and ideas. By contrast, insecurely attached people typically rebuff mutative speech acts. Their dialogue tends to be non-relational, with little affect-modulation, frequent backtracking, and repetitive interactive patterns.

The partial or occasionally total impasse created by these insecure speech patterns then becomes the focus of therapy. Painful affects–anxiety or misery–signal prediction errors, misalignment between wish and reality. But rather than leading to change, these become chronic and embedded. Psychotherapy mobilizes the active inference needed to resolve the impasse. The therapist enjoins the client to look at–mentalise–what is happening between them. Knowing that his or her hand is being metaphorically held, and that energy binding can be temporarily left to the therapist, the client can become more adventurous. In “duet for one” moments, initially fleetingly, therapist and client “sing” in ways that pertain to each and neither participant. Classical analytic geometry may encourage this–prone, in the absence of visual contact, patients take their analysts as part of themselves, drawing on the other’s “priors”–i.e., verbal “interpretations”–to widen the range of available top-down models of the world and its possibilities.


* We live in an entropic universe. Broken cups don’t spontaneously reassemble. Coffee cools once poured. Stars burn out. The exception is life itself. Quantum physicist Schrödinger coined the term ‘negentropy’ to describe how living matter, Canute-like for its lifetime, reverses this cosmic tide towards disorder and homogeneity.

The key to negentropy is homeostasis. As Bernard famously put it, the condition of a free life is the stability of the interior milieu – whether one is a unicellular amoeba or, like Schrödinger, a Nobel-prize winning primate. Homeostasis, and the more general processes of allostasis resist the forces of entropy, physiologically and behaviourally. Inherent in homeostasis are boundaries: cell membranes, the skin, the brain within its skull. Janus-like, homeostasis faces outwards towards the environment and inwards towards the milieu interieur. Temperature sensors in the skin tell us it’s a hot day; the sympathetic nervous system activates sweat glands, the brain tells us to fling off jumpers, move into the shade, etc., all in the service of resisting being entropically fried. Note that homeostats vary in ‘precision’ – some are highly sensitive, whereas others tolerate a great range of variation.

* The brain’s job is to counteract entropy and to maintain internal stability on behalf of the organism whose processes and behaviour it controls and directs; this applies, reflexively, to itself.

* Our sense organs, external and internal, are constantly bombarded by a vast range of stimuli from an ever-changing environment. To operate with maximum efficiency, the brain selects out the ‘meaning’ of its sensations, attending only to those that are relevant to its ‘affordances’ – its specific ecological niche – and especially to input that is anomalous or novel.

* On the basis of prior experience, the Bayesian brain continuously estimates the likelihood of future events. Probabilities are computed by comparing current states of affairs with past occurrences, estimating the extent of correspondence between them, factoring in the likelihood of errors in both memory and perception, and ending with a portion that represents that which cannot be predicted. This is ‘prediction error’, which must, in the service of negentropy, be minimised as far as is possible – prediction error minimisation or PEM.

The brain, ‘top-down’, uses Bayesian probabilities to clarify ‘bottom-up’ input, extero- and interocaptive: ‘My stomach is complaining, but it’s not surprising – I overdid it on the pudding, so it’s probably not cancer’; ‘I know that tune, I’ve heard it so many times – yes of course, it’s the Beatles’ Yellow Submarine’; ‘Is that a stick or a snake? Come on, no adders in city centres, probably safe to pick it up’.

* ‘Energy’ equates to information, albeit physically embodied in patterns of neuronal impulses, synaptic transmission (‘fire together, wire together’) and the neurohormonal environment. Prior models of the world, top-down, ‘bind’ incoming bottom-up information. Energy unbound, or prediction error, reflects novelty in need of binding – and so forestall the dangers of entropic chaos.

* Prediction error is minimised by ‘binding’ bottom-up energy (informational as well as physiological) by top-down generative models based on pre-existing patterns and concepts. Thus is order preserved, entropy eschewed. We know what we like and, mostly, see what we want and expect to see.

* But there will always be a discrepancy between our pre-existing models of the world and incoming sensations, an excess of energy that cannot be bound and will have to be passed onto the next level up of the hierarchy. Lockdown excepted, we don’t live huddled in ‘dark rooms’. The environment is constantly in flux; we need to explore as much as conserve – to find new sources of food, suitable mates, interest and excitement. Surprise, calibrated by the brain as the discrepancy between expectation and incoming sensation, is a proxy for free energy – and hence entropy. Surprise is both vital to survival but also potentially entropic, disruptive or even life-threatening. This represents the prediction error aforementioned. The brain minimises such surprise/error by whatever means possible.

At this point the role of affect becomes important. Free energy is aversive and can be thought of as representing mental pain. Conversely, ‘binding’ free energy is rewarding and therefore motivating. The role of affect, positive and negative, is to drive the free energy minimising processes.

* Given that incoming stimuli are inherently subject to error and imprecision, the brain increases precision by movement – approaching an ambiguous stimulus source, turning the head to use foveal rather than peripheral vision, switching lights on in order to see better, etc. Second, top-down model revision. Now we know what that vague shape really ‘is’ – a cat, clothes strewn on the floor, etc.: ‘Let’s listen more carefully. Oh, that’s not the Beatles at all, it’s the Beach Boys’. Third, and vitally in the case of social species such as our own, active inference is enhanced by recruiting help or ‘twogetherness’: ‘Did you hear something, or was I just imagining it?’; ‘You know about ’70s music – what was that group’s name?’. Friston & Frith call this ‘duets for one’ and have worked out the mathematics of such collaborative Markov blankets. Fourth, if all else fails, by choosing or fashioning environments that conform to the brain’s pre-existing models of the word: ‘I can’t stand modern music. Let’s go over to Classic FM’. This last aspect is captured by the psychoanalytic concept of ‘projective identification’, in which we shape our interpersonal world, often deleteriously, to conform with expectations: ‘You psychiatrists are all the same – never there when I need you’.

* Consider depression, typically triggered by loss, trauma or multiple setbacks. Adversity is widespread – poverty, inequality, racism – but not all succumb. To understand resilience, we need an illness model that encompasses not just events, but individuals’ responses to them. Attachment research shows that those who are securely attached are able to repair the inevitable ruptures to which all are prone, often through the typical sequence of protest, rage, grief and mourning. As children, securely attached people have had caregivers they could depend on to acknowledge their pain, tolerate protest and help them to move on. Repeated episodes of everyday rupture–repair cycles help build this resilience.

The free energy released by the rupture is bound by the child’s knowledge that help is at hand and that their epistemically trusted caregiver will provide a generative model to counteract the free energy associated with ruptures: ‘Don’t worry love, I’m just going to the loo, I’ll be back in a minute’. In the ‘still face’ paradigm, parents are asked to freeze their facial expression for 1 minute while talking or playing with their child. Securely attached children continue actively to try to re-engage with their caregivers in the confident expectation that they will be ‘back soon’. For insecurely attached children, by contrast, rather than rupture–repair, cycles of rupture–despair or rupture–disappear are the norm. Their caregivers have either themselves been overwhelmed by their child’s unhappiness and so despairingly abandon attempts to alleviate it; or repress the impact of the child’s mental pain and so ‘disappear’ emotionally. Both leave the child alone to find ways to bind the free energy the rupture evokes. When their caregiver’s face freezes they look away, become miserable and regressed, and often resort to self-soothing rituals such as rocking or emotional dissociation.

Such insecurely attached children are primed in later life for depression in response to loss or trauma or, in extreme cases, to developing post-traumatic stress disorder. The ingredients of free energy minimisation needed to maintain psychological equilibrium are for them problematic. Active inference is compromised. They tend to be passive rather than active. They stick with limited and simplistic and inflexible ‘top-down’ models such as ‘It’s no use trying to make things better, it never works’ or ‘Feelings are dangerous, best to keep them buried’. They find it hard to trust people and so can’t ‘borrow’ an intimate other’s brain with which to process feelings and build up alternative ways of viewing the world.

* The most commonly used therapy for depression, CBT, attempts to address these deficiencies. Therapists encourage patients actively to test their negative ‘hypotheses’ by looking more closely at their experiences and by exploring alternative top-down models to account for them (‘Maybe my boyfriend didn’t answer his phone because he’d run out of battery, not because he doesn’t love me’). But CBT has its limitations. ‘Treatment-resistant depression’ is common.15 People with personality disorders do badly with standard CBT, often refusing to engage or dropping out. The FEP provides explanations for this. From an FEP perspective, one way to minimise free energy is to gravitate towards or engender environments that confirm one’s view of the world, however negative. Depression relegates sufferers to emotionally impoverished relationships, stereotyped and simplistic top-down models, and thus becomes a self-fulfilling hypothesis, resistant to psychotherapeutic interventions. In addition, these negative top-down priors are ‘inferentially inert’, i.e. inaccessible for modification.

A degree of chaos/uncertainty/free energy needs to be tolerated before new generative models can evolve. Homeostatic imprecision needs to be tolerated for a while. The holding and ‘negative capability’ of the therapist’s ‘borrowed brain’ paves the way for a more complex, nuanced top-down reset. Given that people with personality disorders notoriously find it difficult to trust others, the brevity and defocus on the therapeutic relationship in standard CBT limits the scope for such fundamental change.

Moving from depression to an FEP perspective on trauma, the latter creates an overwhelming influx of free energy for which there are no available top-down models with which to bind it. Thoughts of cruelty, neglect and abuse remain in the realm of the unthinkable and are therefore ‘defended against’ by repression or dissociation. However, when jointly considered – under a shared Markov blanket – these bottom-up unprocessed experiences can be bound with the therapist’s encouragement and expertise into manageable narratives. However painful, they become less overwhelming, a source of new ways of thinking and psychic reorganisation.

* Free association taps into the mind’s normally unvoiced upward-welling stream of consciousness, counteracting the elusiveness of affect seen in the rupture–despair/disappear attachment pattern. This enables the range of top-down responses to be enhanced and aversive free energy minimised. At the top-down level, in a process comparable to the immune system’s lexicon of antigen-activated antibodies, dreaming is the means by which the mind generates a repertoire of narratives with which to bind the free energy which life’s vicissitudes engender. Transference analysis turns the spotlight on the limited varieties of top-down narratives that sufferers use in their dealings with intimate others to minimise free energy. The enigmatic ambiguity of therapists’ persona enables patients to experience, reconsider and extend the top-down assumptions with which they approach the world of intimate others.

* Psychoanalytic and attachment-derived mentalisation-based therapy (MBT) is now established as a highly effective therapy for borderline personality disorder, previously considered untreatable.Reference Bateman and Fonagy18 MBT leads to big reductions in medication use, suicide attempts, hospital admission and unemployment among people with borderline personality disorder, as compared with treatment as usual.

MBT is both practically and conceptually consistent with the FEM. It encourages patients (a) to identify the bottom-up feelings that fuel their self-injurious actions, (b) to pause and think of different ways of handling these, i.e. to tolerate a quantum of free energy with the help of the therapists’ ‘borrowed brain’ and (c) through mutual mentalising (therapist and patient together forming a neurobiological ‘bubble’) to generate more complex and adaptive models of the self and significant others. The result is manageable surprise: confounding sufferers’ negative assumptions about the world, becoming less overwhelmed by unbound affect (fewer ‘melt-downs’) and facilitating greater resilience.

About Luke Ford

I've written five books (see My work has been covered in the New York Times, the Los Angeles Times, and on 60 Minutes. I teach Alexander Technique in Beverly Hills (
This entry was posted in Attachment. Bookmark the permalink.