The Necessary Illusion
Every close relationship depends on an illusion. We behave as though the person we know today is substantially the same person we knew yesterday. Without that assumption, ordinary life would be impossible. A partner leaves for work; we do not reconstruct their identity when they return. A friend disappears for a month; we pick up where we left off. A child sleeps upstairs without needing to continually signal that the relationship still exists.
Human beings maintain internal models of one another. These models are enormously useful. They are also capable of becoming stale. The interesting question in a long relationship is therefore not whether we run a model of the other person. We inevitably do. It is:
Model-Mediated Relating
Imagine someone whose internal representation of a close relationship is highly stable. They know the other person loves them. They know roughly what that person enjoys. They know the relationship is secure. Nothing obviously important has happened. The simplest inference is:
This creates an efficient form of relating. Continuous observation is unnecessary. Sparse signals can maintain a strong sense of connection. A short conversation, sitting in the same room or simply knowing the other person is nearby may be enough. There is nothing pathological about this. In fact, attachment security requires some version of it: the relationship must survive periods in which direct evidence is absent.
But stability has a failure mode. The model can become so trusted that new observation carries too little weight. Then statements such as these become possible: I know you're happy. I know what you like. We're very close. You're not really upset. Everything between us is fine. The claims may once have been accurate. The question is whether they remain accurate.
Sampling Another Person
Attention looks different when seen from this perspective. Looking at someone is not merely a social courtesy called "giving attention." It is a form of sampling. Voice gives information. Facial expression gives information. Touch gives information. What someone spontaneously talks about gives information. What they no longer talk about gives information. What makes them suddenly animated gives information. What they avoid gives information.
Conversation is therefore one of the mechanisms by which an internal model is kept current. A relationship in which people continue sampling one another can contain a surprising amount of change without losing continuity. A relationship in which sampling declines can remain superficially stable while the participants gradually become strangers to the versions of each other that actually exist.
This suggests a useful variable: relational sampling rate — how frequently someone gathers fresh information about the current state of another person. And then another: relational update gain — when new information conflicts with the existing model, how readily the model changes.
Too Much Updating Is Also a Problem
High update gain is not automatically good. Imagine someone whose model of the relationship is revised by every weak signal. A delayed message becomes evidence of rejection. A tired expression becomes evidence that something is wrong. A quiet evening becomes evidence that love has diminished. Now the system is unstable, because noise is being treated as signal.
A good relational model therefore needs both persistence and revisability. Too much persistence and the model becomes stale. Too much revisability and the model becomes volatile. The useful question is calibration.
The Cached Person
Long relationships create a special risk, because increasing familiarity makes prediction easier. Early in a relationship, nearly everything is information. What music someone likes. How they eat. How they argue. How they behave when sick. What they find funny. What frightens them. What they do with a free Sunday.
Years later, the apparent information rate falls. We know the answers. Or believe we do. That belief itself can reduce sampling. Why ask the question when you know the answer? Why watch closely when you already know what the person is like? Why explore a position you have heard before?
The person becomes increasingly mediated by memory. Eventually one can end up relating to a highly detailed historical reconstruction.
Model Freshness
This raises the possibility of measuring something more interesting than relationship satisfaction. Call it relational model freshness. Ask one person questions about the other: what currently excites them? What have they become less interested in? What are they worried about this month? What opinion have they changed recently? What are they trying to get better at? Who have they become closer to? What do they wish they had more time for? Then ask the other person independently.
The degree of correspondence would reveal something conventional intimacy questionnaires often miss. Two people might report loving each other deeply while possessing badly outdated models. Another pair might argue frequently while knowing each other's current internal worlds with remarkable accuracy.
Why Exploratory Conversation Matters
This also helps explain why some conversation feels unusually alive. Autobiographical conversation often adds facts to an existing record: I went here. This happened. I spoke to this person. Exploratory conversation can do something else. It can reveal the current process by which the other person thinks.
Someone offers an idea. The other challenges it. A distinction appears. A belief changes slightly. A surprising association emerges. Now the participants are not merely exchanging stored information. They are watching one another become. That produces extremely high-quality samples, because the model is being updated by live computation rather than historical reporting.
This may be one reason deeply curious relationships can remain fresh for decades. The subject matter need not be the relationship itself. Two people can learn one another by exploring politics, architecture, cooking, physics, gossip or the behavior of the neighbors. The external topic is simply the surface on which current minds become visible.
When the Model Wins
The deepest failure occurs when the internal representation acquires greater authority than observation. The person says: I am lonely. The model says: But we spend lots of time together. The person says: I don't enjoy that anymore. The model says: You've always loved it. The person says: I've changed. The model replies: No, that's not really you.
At that point familiarity has become strangely anti-intimate. The claim I know you is being used to reject information about who the person currently is. The solution cannot be to abandon stable models — without them there is no continuity, trust or attachment. It is to preserve a route by which reality can revise them. A close relationship therefore requires an unusual cognitive discipline.
The opposite of intimacy is not ignorance. Sometimes it is excessive certainty.