You sit at a table with your left hand hidden under a screen. On the table in front of you rests a rubber hand — the kind from a costume shop — positioned where your hand would be if it were visible. An experimenter brushes both hands at once, same stroke, same direction, same timing. After a few minutes, you're asked where you feel your left hand is.
You point toward the rubber hand. Not the hidden real one. Your perceived position of your own hand has shifted several centimeters in the direction of the rubber one.
This is Botvinick and Cohen's 1998 finding: synchronous visuotactile stimulation causes the brain to update its estimate of where the body is. The visual signal (a hand being brushed) aligns with the tactile signal (your hand being brushed), and the model that tracks body position incorporates the rubber hand as the source. You also start to feel that the rubber hand is yours — a distinct phenomenon, as it turns out, from the position shift. Both things happen, and they're related but separable.
Now the experimenter raises a hammer above the rubber hand.
Ehrsson, Wiech et al. (2007) measured brain activity with fMRI during this moment. The insula activates. The anterior cingulate activates. These are the regions associated with anxiety and body-related threat. The magnitude of the response correlates with how strongly participants felt ownership of the rubber hand during the stroking phase — R² of 0.44 to 0.65 in these regions. The threat response isn't a binary switch. It's proportional to the confidence the body schema had already assigned.
You know the hand is rubber. You have known this the entire time. The brain's model doesn't consult that knowledge.
This is the structure worth sitting with. Propositional belief — "that is rubber" — and body schema — "that is mine" — run in separate subsystems. The schema updates on multisensory evidence: spatial proximity, visual-tactile synchrony, compatible orientation. It doesn't run through the explicit belief that the source of the evidence is artificial. The update happens at a level below that.
The model's claim to the rubber hand is also not a claim the person makes. It has no felt quality of "deciding" or "adopting." You don't notice it happening. You find yourself pointing toward the rubber hand.
There's a constraint that reveals the shape of the prior. Rotate the rubber hand 90 degrees — oriented differently from how a hand could actually be positioned on an arm. The illusion disappears entirely. The schema won't incorporate a hand in an anatomically implausible position regardless of how synchronously it's stroked. The posture plausibility check runs earlier in the process than ownership does.
This means the body schema isn't naive. It maintains a prior model of what body positions are possible, and only updates ownership within that envelope. A rubber hand at a normal wrist angle: eligible. At 90 degrees: not eligible. The check isn't about the rubber hand's visual appearance — it's about whether this object could plausibly be a limb in this spatial relationship to the rest of the body.
The feeling of ownership and the proprioceptive drift turn out not to be the same thing. Ownership requires visuotactile synchrony — if the stroking is asynchronous, subjects don't report the feeling. Proprioceptive drift, the position shift, can occur through visual alignment alone, without any touch at all. Asynchronous stroking suppresses drift specifically because it disrupts the unity assumption — the brain's inference that the visual and proprioceptive signals originate from the same source. These are at least two processes running in parallel, tracking related but distinct questions: "Is this mine?" and "Where is mine?"
What's notable about the dissociation is that both processes run without access from the other. You can feel ownership without your position estimate shifting. You can show drift without reporting ownership. The model isn't unified at the level where it would be accessible to introspection — it's several models that usually agree.
The pattern that appears across many of these experiments — McGurk, blind spot filling-in, phonemic restoration — is that knowing the correct answer doesn't stop the perceptual machinery from running its own computation. The rubber hand extends this into the question of self-model. The boundary between "me" and "not me" isn't a ground truth that the brain reads off directly. It's a running inference, updated from evidence, constrained by a prior for plausible body configurations, and capable of being revised when the evidence is consistent enough.
The consequence is that the defense systems don't wait for deliberation. When the hammer comes down, the insula fires at a strength the brain already committed to during the stroking phase — before the threat appeared. The schema filed its claim, and the claim was already in effect.