What does a personality actually know? Long before a child can articulate a single belief about love, trust, or worth, something in them has already formed a working theory of how relationships operate. Will distress be met or ignored? Is closeness safe, or does it come at a price? Am I the kind of person others respond to? An infant never asks these questions consciously, yet the answers become embedded in the architecture of the developing self.

John Bowlby called these embedded answers internal working models: mental representations of self and attachment figures that allow a person to forecast relational events and plan accordingly. He borrowed the term from Kenneth Craik's early cognitive science, and the choice was deliberate. A model is not a memory. It is a simulation engine, a structure that generates predictions and quietly guides action in situations the person has never encountered before.

For personality theory, the concept offers something rare: a mechanism. Trait psychology describes consistency, and psychodynamic theory gestures at internalised objects. Internal working models specify how experience becomes structure, how structure becomes expectation, and how expectation organises perception, memory, and behaviour across decades. Theodore Millon argued that personality is a deeply embedded pattern that expresses itself automatically across domains. Working models reveal one route by which such patterns are laid down. They also reveal the conditions under which those patterns may be rewritten.

Bowlby proposed that the attachment behavioural system does not operate by reflex alone. The system evolved to keep vulnerable infants close to protective adults. Through repeated transactions with caregivers, the child constructs two complementary representations. One is a model of the attachment figure as more or less available and responsive. The other is a model of the self as more or less worthy of care. The complementarity matters. A child with a consistently rejecting caregiver does not simply conclude that others reject. The child concludes that they are the kind of person who gets rejected.

Mary Ainsworth's Baltimore home observations gave the theory its empirical spine. Maternal sensitivity across the first year predicted how infants organised their behaviour in the Strange Situation. Secure infants used the caregiver as a base for exploration and were readily comforted on reunion. Avoidant infants minimised attachment signals, while ambivalent-resistant infants amplified them. Each pattern is best read as a strategy. The avoidant infant suppresses need with a caregiver who rebuffs it, and the resistant infant intensifies need with one who responds unpredictably. These are intelligent solutions, not developmental deficits.

The earliest models are encoded procedurally rather than declaratively. Daniel Stern described representations of interactions that have been generalised. Karlen Lyons-Ruth's notion of implicit relational knowing captures the same idea: the infant stores statistical regularities extracted from thousands of micro-exchanges, rather than discrete episodes. This is why working models rarely feel like beliefs to the person holding them. They feel like the texture of reality itself. They operate beneath the level at which anything could be questioned.

Mary Main, Nancy Kaplan, and Jude Cassidy famously described a 'move to the level of representation.' By age six, children's responses to family photographs and imagined separations mirrored their infant classifications. As cognition matures, models acquire semantic content, become available to language, and organise hierarchically. Nancy Collins and Stephen Read proposed a layered network. A general model of relationships sits at the top. Beneath it are domain-specific models, such as those for parents or peers. Beneath those are models of particular relationships.

For personality pathology, the most consequential case is disorganisation. Main and Erik Hesse described infants confronted with 'fright without solution.' In these cases the caregiver is both the source of alarm and the only available haven. Giovanni Liotti argued that such conditions prevent the child from constructing a single coherent model. Instead, the child develops multiple incompatible representations of self and other: persecutor, victim, rescuer. Longitudinal data link early disorganisation to later dissociative symptoms. This fragmentation offers a developmental account of borderline personality organisation.

Takeaway

A working model is less a record of what happened than a solution to a problem: how to stay close to someone you need. Insecure patterns are intelligent adaptations to particular caregiving conditions, not failures of development.

Once consolidated, working models function less like stored content and more like active processing rules. Collins and Read described four interlocking components. The first is memories of attachment-related experience. The second is beliefs and expectations about self and others. The third is attachment-related goals and needs, and the fourth is strategies for meeting those needs. When a relational situation activates the model, all four components come online together. They shape appraisal, emotional response, and behavioural plans in a single, largely automatic sweep.

Perception is the first point of influence. Mario Mikulincer and Phillip Shaver describe two secondary strategies that emerge when security is not reliably available. Hyperactivation is characteristic of attachment anxiety. It heightens vigilance for signs of rejection and lowers the threshold at which ambiguous cues register as threat. Deactivation is characteristic of avoidance. It redirects attention away from attachment-relevant information altogether. The same delayed text message may be evidence of abandonment for one person. Another person may simply not notice it.

Memory follows suit. Fraley, Garner, and Shaver found that highly avoidant individuals retained less information from an emotionally charged attachment interview. The deficit reflected reduced encoding rather than faster forgetting, which echoes Bowlby's concept of defensive exclusion. In the Adult Attachment Interview, dismissing adults often describe parents with glowing adjectives yet cannot supply specific memories to support them. Main assessed this discrepancy through violations of Grice's conversational maxims. The model shapes not only what is recalled, but how coherently it can be told.

Behaviour completes the circuit, and here models become self-perpetuating. The anxious partner's escalating reassurance-seeking can provoke the withdrawal it dreads. The avoidant partner's distance teaches others to stop offering support. Avshalom Caspi distinguished reactive, evocative, and proactive transactions between person and environment, and this distinction illuminates the mechanism. Working models do more than interpret relational environments. They evoke responses from others. They also guide the choice of partners, friendships, and settings that tend to confirm prior expectations, recruiting the social world into their own maintenance.

This is where attachment theory contributes something distinctive to personality science. In Dan McAdams's integrative framework, working models belong to the level of characteristic adaptations. They are more contextualised and motivational than broad traits, yet more structured than momentary states. They also feed into narrative identity, supplying the assumptions people bring to the stories of their lives. Indeed, the coherence the Adult Attachment Interview measures is fundamentally a narrative property. It links representational security to the integrative work of self-authorship.

Takeaway

Working models don't just interpret experience; they help manufacture it. The most stable features of personality may be those that elicit the very evidence that confirms them.

Bowlby insisted that working models must be updated to remain useful. Yet he recognised why they resist revision: they operate largely outside awareness and filter the very evidence that might challenge them. Longitudinal research captures this tension. Everett Waters and colleagues found that about 72 percent of participants kept the same secure or insecure classification from infancy to young adulthood. In Michael Lewis's sample, continuity was much weaker. Chris Fraley's meta-analysis suggested moderate stability, with a coefficient near .39.

The first condition for revision is what attachment researchers call lawful discontinuity. In the high-risk Minnesota sample, Nancy Weinfield and colleagues found that shifts from security to insecurity tracked maltreatment, maternal depression, and family disruption. Fraley and colleagues later drew on the NICHD Study of Early Child Care. Maternal sensitivity across childhood, family stress, and friendship quality predicted adult attachment beyond infant classification. Stability, in other words, often reflects continuity of environment as much as the inertia of the model.

The second condition is corrective relational experience. The Adult Attachment Interview identifies earned security in adults who describe harsh early caregiving yet discuss it with coherence, balance, and reflective distance. Glenn Roisman and colleagues complicated the picture. Some adults classed as earned-secure from their own retrospective accounts had not experienced notably poor care according to prospective records. Still, the broader evidence suggests revision is most likely with sustained alternative attachment figures, such as a stable partner, mentor, or therapist, whose responses consistently violate old predictions.

The third condition is reflective access. A model must be activated and examined before it can be reconstructed. This is why Peter Fonagy's concept of reflective function is so central. Reflective function is the capacity to represent one's own and others' mental states. In a randomised trial with borderline patients, Kenneth Levy and colleagues found that transference-focused psychotherapy increased narrative coherence and reflective function, and more patients were classified as secure. Insight alone rarely suffices. Revision requires emotionally live disconfirmation inside a relationship.

Even successful revision rarely amounts to replacement. Bowlby anticipated that people might hold multiple models of the same relationship. Contemporary work suggests that new representations compete with older ones rather than overwriting them. Under fatigue, threat, or loss, early expectations can reassert themselves with surprising force. Personality, on this view, resembles a palimpsest. Earlier inscriptions remain partly legible beneath later text. Identity consists partly in learning which script is being read at any given moment.

Takeaway

Internal working models are revised less by insight than by sustained relational evidence that the old prediction is wrong. Change usually adds a new model rather than erasing the old one, which is why stress so often reveals our earliest expectations.

Internal working models give personality theory something it has long needed: a developmental account of how relational experience becomes psychological structure. They explain why early patterns persist. These patterns are not fixed destiny. They are active predictions that shape attention, memory, and the responses people evoke from others.

They also clarify the nature of personality pathology. Where caregiving is frightening or chaotic, the problem goes beyond an insecure model. There may be no single coherent model at all. In Millon's terms, such fragmentation undermines the organisation that lets a personality function as a unified system across contexts. Behaviour is then governed by whichever representation the moment happens to activate.

Perhaps the deepest implication concerns change itself. If sustained relational evidence revises models more than insight does, then personality development is never purely individual. Who we become depends partly on who we encounter. It also depends on whether those encounters are steady enough to make an old prediction finally, quietly, wrong.