Internal Working Models

A simple explanation

You carry, somewhere below thought, two answers you do not remember writing. One is the answer to am I worth being cared for? The other is the answer to can the people around me be relied upon when it matters? You did not arrive at these answers by reasoning. You absorbed them, in the first years of your life, from how the people closest to you responded — or did not respond — when you needed them. The two answers, sitting underneath everything, are what John Bowlby called your internal working models.

A model of the self. A model of the other. Built early, held quietly, and consulted automatically every time another person enters the room.

An everyday example

A friend cancels plans on short notice. The cancellation is genuine — a work crisis, an apology, an offer to reschedule. The information in the message is small and clean. But the meaning your system assigns to it is not the message. It is the message read through the model.

One person reads: something came up; we'll find another evening. The model of the other says people are generally available; the model of the self says I am generally worth showing up for. The cancellation lands as a fact, not a verdict.

Another person reads: I knew it; people pull away; I shouldn't have looked forward to it. The model of the other says people withdraw when it matters; the model of the self says I am the kind of person who is left. The cancellation lands as confirmation — of something that was already there before the message arrived.

Same text. Two different inner worlds.

What is an internal working model in attachment theory?

Bowlby's insight was that attachment is not just a behaviour but a representation. The infant who repeatedly experiences a parent as responsive does not only learn to seek that parent in distress — they construct a mental map in which help comes when needed and I am the kind of being for whom help comes. That map then runs in the background of every later relational encounter, not as conscious belief but as a pre-set interpretation, the way a base layer in a piece of software shapes everything that runs on top of it.

The model has two interlocking faces. The model of self answers questions about lovability, worth, and competence to elicit care: am I the kind of person who gets met? The model of other answers questions about availability and reliability: will the people I matter to be there when I need them? The two are not independent. A model of the self as unworthy tends to predict a model of the other as withholding (because the self-as-unworthy model expects to be left). A model of the other as unreliable tends to bend the model of the self toward over- or under-reliance. The pair is acquired together and runs together.

How do internal working models form?

The early years of life are a vast act of pattern extraction. The infant cannot interpret a single episode of distress and response; what they can do, over thousands of repetitions, is extract the shape of what happens when need arises. Was the cry usually met? Was the response calm or anxious or angry? Was contact warm and then sustained, or warm and then withdrawn? Did the caregiver return to baseline after a rupture, or did they need the child to manage the caregiver's state first?

From this vast accumulation, a model precipitates. It is computationally efficient: rather than treating each new encounter as a fresh problem, the system stores a default expectation and applies it. Efficiency is the gift; rigidity is the cost.

The Belonging System, working at the Bowlby layer, is the calibrator. Its job is not to decide whether any given person is safe — that would be prohibitively expensive — but to set the prior against which each new relational signal is read. The model is that prior.

Why do attachment styles persist into adulthood?

Because the model is consulted before the person is. Adult relational encounters do not arrive at a blank interpreter; they arrive at a system that already has a hypothesis about what is about to happen. The hypothesis colours what is noticed, what is dismissed, what is remembered. A cancellation read through a people withdraw model is logged as further evidence; a cancellation read through a people are reliable model is logged as an isolated event. Over time, the model accumulates its own confirming data — not because reality is rigged but because the reader is.

This is the mechanism that makes attachment style stable. The behaviour does not persist because of some internal stubbornness; it persists because the model is doing exactly what models do, which is to apply a stored interpretation to ambiguous data. Anxious-preoccupied, avoidant-dismissing, disorganised — each is a behavioural signature of a particular model running in real time, generating relational moves that tend to elicit the responses the model predicts.

The behavioral loop

The model in motion across a single encounter, then across years:

1. Signal — a relational event arrives: a text, a glance, a silence, a return.
2. Model consultation — before conscious interpretation, the model supplies a default reading. The default is shaped by self-model and other-model together.
3. Behaviour — the response is generated from the reading, not from the raw signal. The response often makes the predicted outcome more likely (the anxious reach that triggers withdrawal; the avoidant pull-back that reads as disinterest).
4. Confirmation — the resulting outcome is logged as further evidence for the model. Ambiguous cases are read in the model's favour. Clear contradictions are explained away or filed as exceptions.
5. Reinforcement — the model is slightly more confident than before. The next signal is interpreted slightly faster through the same template.
6. Style stabilisation — across years, the model becomes the relational style. The model and the style are the same object viewed from inside and outside.

Emotional drivers

The strongest drivers of model-confirmation are not beliefs but felt senses. The anxious-preoccupied system carries a low-grade ambient signal that closeness is precarious; the avoidant system carries a low-grade ambient signal that closeness is costly; the disorganised system carries both at once, in alternation. These felt senses are not conclusions; they are the texture of the model, and they precede thought.

Beneath the felt senses sits the original loss the model is protecting against. The model of self as unworthy is not a self-attack — it is, originally, a way of making sense of a caregiver who could not meet the need. If I am unworthy, the unreliability is explicable. The model preserves the relationship at the cost of the self. Decades later the relationship is gone and the cost is still being paid.

What your nervous system does

The model lives in the autonomic patterning that the early relationship installed. Cues of relational uncertainty — a delayed reply, an ambiguous tone, an unexpected closeness — produce a small autonomic response before the cortex has interpreted the cue. The body is already moving toward its characteristic pattern (reach, withdraw, freeze) when the conscious mind catches up and supplies the reasons.

This is why the model feels true. It does not arrive as a belief to be examined; it arrives as a body-state to be inhabited. The cortex then narrates the body-state, and the narration sounds like an objective reading of the situation when it is in fact a reading of the body's stored prediction.

The DojoWell interpretation

In MDT terms, the internal working model is the Belonging System's calibration memory — the stored interpretation of what relationships are like that is applied to each new relational moment. The System's job is to protect connection: to read whether the room is safe enough to belong in, whether the bid is likely to be met, whether the other person can be trusted with the weight of need. To do that job in real time, it cannot start fresh in every encounter. It consults the model.

When the model was built in conditions of consistent, attuned responsiveness, the System's defaults match most current rooms, and reading the room costs little. Density across relational life runs high. When the model was built in conditions of inconsistency, neglect, intrusion, or frightening behaviour, the System's defaults are protective against a past room the present one does not necessarily contain. The cost is paid in misread invitations, pre-empted closeness, and the slow accumulation of confirming evidence for a model that no longer fits — the residue of identity_fragmentation, in which the self the model predicts and the self lived into are at quiet odds.

The substitution mechanism, at the level of the model, is confirmation over fresh reading. The System, denied a coherent template, accepts a stable but stale one. The substitute is real protection against a vanished threat. The original — the room as it is now, with the people in it as they are now — is the bid the model cannot quite let in.

Revising the model is the work of earned security: sustained corrective relational experience, often supported by therapy, in which the present room is allowed to write new data over the old template. The deposit is delayed. The numerator climbs slowly, over years, as residue lifts and the System learns to read this room instead of the first one. Density — eventually — rises toward what the system was always reaching for.

The framework's contribution is not to invent the model. Bowlby did that. The contribution is to name the model explicitly so the work of revising it can be done deliberately — and to place it inside an equation that makes both its cost and its revisability legible.

Can internal working models change?

Yes — and the evidence for this is one of the more hopeful findings in attachment research. The path is not insight alone. Insight names the model; it does not, by itself, replace it. The model is revised through repeated relational experience in which the new pattern is large enough, consistent enough, and trusted enough for the system to update.

The most reliable conditions for revision are a long-running secure relationship — a partner, a close friend, a therapist — in which the model's predictions are quietly, repeatedly, gently disconfirmed. Not in a single dramatic correction; the model is robust to single events. In thousands of small ones, until the prior itself shifts.

This is the earned-secure path. The model that was built in a less-than-secure early environment is revised, in adulthood, through a sustained current environment that the system eventually trusts enough to learn from. The original model does not disappear; it is overlaid by a newer, more accurate one. Under stress the older model can still surface — which is why earned security is not the same as never-fragile security — but the dominant prior, in most rooms, becomes the new one.

How do I know what my internal working model is?

You do not need to introspect for the answer. The model is more visible in the patterns of your relational life than in your reports about yourself.

Three observations are usually enough. First: what is the default reading you supply to ambiguous relational signals? A delayed reply, an unreturned look, a friend's distraction — what story does your system reach for first? Second: what is the move you make under relational uncertainty? Do you reach, withdraw, manage the other person's state, freeze? The move is the model in action. Third: what is the pattern your relationships repeat? Not the specifics but the shape. The shape is the model's signature.

The point is not to label yourself. It is to make the model legible enough that the System's reading can be paused, in moments that matter, long enough for the present room to be read on its own terms.

Practical steps

1. Name the prior, not the situation. In an ambiguous relational moment, ask not what is happening? but what is my system already predicting? The prior is the model. Naming it pauses it.
2. Distinguish the model from the person. The person in the room is not the person the model was built against. Most relational disappointments come from reading the present through the past. The discipline is to let the present supply its own data.
3. Find one secure relationship and let it teach. Earned security is not built alone. A long-running, attuned, consistent relationship — partner, friend, therapist — is the workshop. The model revises slowly, in the presence of patient, repeated counter-evidence.
4. Notice the body's prediction before the mind's narration. The autonomic shift arrives first. Catching the body-state lets you hold the cortex's narration more lightly. The narration is the model speaking; the body is what the model is protecting.
5. Track confirmation bias gently. The model logs evidence for itself. Once a month, take one relational episode and ask whether the model's reading was the only available one. Most of the time it was not.
6. Be patient with the revision. Models built over years are revised over years. The Belonging System is conservative for good evolutionary reasons. The deposit on this work is large and delayed.

Reflection questions

• What is your default reading of an ambiguous relational signal — a delayed reply, an unreturned look, a friend's distraction? Where did that reading come from?
• What is the relational pattern, across your adult life, that has repeated? What does its shape suggest about the prior your system is supplying?
• Is there a relationship in your life that is quietly, consistently disconfirming an older model? What has it been depositing that the model has not yet integrated?
• Where in your relational life are you reading the present room through the data of a room you no longer live in?

Frequently Asked Questions