You cannot see into another person's mind. You only ever have your model of their model — a second-order representation, assembled from signals you happened to observe and inferences you happened to draw. It is almost always wrong in some specific, consequential way. Treating it as if it were the territory is how most communication quietly fails.
This is not a poetic claim; it is a structural one. The thing you are reasoning about — what they believe, what they want, what they will tolerate — is not the world. It is a small, hand-built sculpture of the world that lives inside your head. Their actual model is inaccessible. The gap between your sculpture and their model is where misunderstanding lives, and the size of that gap is almost always larger than it feels.
What you actually have access to. Your model is built from signals: their words, posture, eye contact, what they emphasized, what they skipped, how long they paused, what they laughed at. Plus your priors about people like them — their role, their history with you, the stereotype-shaped scaffolding you brought into the room. Every signal is sparse and every prior is a shortcut. The model is therefore (a) under-determined by the data and (b) over-determined by your defaults. Both failure modes pull it away from accuracy.
Why the model is wrong in predictable ways. Three common distortions: projection (you fill the gaps with what you would think); role-stereotyping (you fill them with what someone in that role typically thinks); and recency (you fill them with whatever signal arrived last, weighted disproportionately to its informational value). Each of these is a labor-saving move, and each one trades accuracy for speed. You can't turn them off, but you can notice when you've used one.
Imperfect modeling beats none. The mistake is not having a model — that's unavoidable. The mistake is forgetting that the model is yours, not theirs. The discipline is to hold the model loosely: act on it, but treat every new signal as evidence that could update it. A held-loosely model gets sharper with each exchange. A held-firmly model gets defended against the very signals that would correct it.
This is the foundation under listening. Del Close and Charna Halpern frame the improviser's task as receiving "what is said, and what is left unsaid." That instruction only makes sense if you already know that your model is partial — that the unsaid parts are the parts you are most likely to be wrong about. Active Listening is the practice of updating your model in real time; without the epistemic premise, listening collapses into rehearsal of what you already believe.
On stage, the same problem in concentrated form. Johnstone's "be obvious" is partly an answer to this. If you can't read your partner's model, don't try to outguess it — give them something readable, and let their response reveal what they're seeing. The obvious offer is a probe, not a guess. Spolin's structure runs on the same intuition: she organized exercises around a shared point of concentration precisely because two players cannot directly inspect each other's intentions. They can only reference the same external object and infer alignment.
The practical move. Three habits, in order of cost: (1) name your model as a model — "I'm reading this as you being upset; is that right?" — and let them correct it; (2) listen for the gap between what you predicted and what they actually did, and treat the gap as the real data; (3) before you press a point, ask whether you are pressing against their actual position or against your model of it. Often the disagreement evaporates at step three, because the position you were attacking was one you constructed.
Why this matters off-stage. Every persuasive move — every reframe, every negotiation, every hard conversation — presupposes a model of the other person. The quality of the move is bounded by the quality of the model. You can have brilliant tactics aimed at a phantom and still lose, because the phantom is not who you are talking to.