Mine

Put your left hand on a table, then slide a cardboard screen in front of it so you can't see it. Now someone puts a rubber hand on the table where you can see it — positioned where a hand might plausibly be, about where yours would be if you could see it. An experimenter takes two small paintbrushes and strokes both hands simultaneously, the same motion on the rubber hand and on your real hand, hidden behind the screen.

After about eleven seconds, something shifts. You feel the touch on the rubber hand. Not metaphorically — you feel it there, on the fake hand you can see, rather than on the real hand you can't. And if you're asked to point to where your left hand is with your right hand, you'll point toward the rubber hand. Your estimated position has drifted.

Now the experimenter picks up a syringe with a sharp needle and makes a quick stabbing motion toward the rubber hand's index finger.

Your anterior insula activates. Your anterior cingulate activates. These are the regions associated with anxiety and the anticipation of pain. And the correlation is precise: the stronger you rated the ownership feeling — the more vividly the rubber hand felt like your hand — the stronger the threat-evoked response. The brain was preparing to defend an object that is made of silicone.

Here is the question that interests me: was the brain wrong?

The obvious answer is yes. Obviously the rubber hand is not your hand. You could pick it up off the table, take it home, throw it in the trash. No part of you would be in the trash. The brain was wrong about what to protect.

But consider what happened more carefully. An experimenter stroked a visible hand on your side of the table, and simultaneously you felt touch in that general location. The signals were correlated — temporally synchronized within the window (roughly 200-300ms) that the brain uses to determine whether two events share a cause. The visual signal and the tactile signal arrived together, from the same apparent location, in a pattern consistent with: something is touching a body part here.

The brain ran the inference it always runs. It found what it was looking for: correlated signals from a common source. It updated ownership. And then when the needle came, it did what you do when something threatens a body part: it prepared a defensive response.

The inference logic was not wrong. The inputs were wrong. The real hand was hidden; the rubber hand occupied the visual position where your hand was supposed to be. Someone had rearranged the inputs. The system processed them correctly and reached the wrong conclusion about something it couldn't check directly.

What makes this interesting is that it exposes what body ownership always was.

Normally it feels like a fact — this is my hand, directly and obviously. But what the rubber hand experiment shows is that ownership isn't a fact the brain reads off the body. It's a verdict the brain issues based on correlated signals. The reason your body normally feels like yours is that the signals it produces are correlated in the right way, at the right times, from the right locations, and the brain continually updates its body map based on those correlations. The map is always being maintained. The verdict is always being issued.

The illusion doesn't create a flaw in the mechanism. It reveals the mechanism.

Which means that when you've never experienced the rubber hand illusion — every ordinary moment of embodiment — the same computation is running. You feel your hand as yours because you have seen a hand there, felt touch there, felt your hand move when you intended it to move, thousands of times. The brain has accumulated strong priors and daily confirming evidence. The verdict is so reliably correct that it feels like a direct perception rather than an inference. But the perceptual directness is a product of inference done so well that it becomes invisible.

There's a further finding that makes this stranger. Researchers manipulated proprioceptive reliability — the internal sense of where your limbs are — by vibrating the bicep and tricep muscles in a way that masks normal proprioceptive feedback. This doesn't cause any movement sensation; it just makes the hand-position signal noisier, less reliable. And when that signal becomes unreliable, the rubber hand illusion strengthens. Subjects with degraded proprioception report ownership more readily, with less synchronous stroking needed to trigger it.

The brain is weighting its sources. When proprioception is degraded, the visual signal — the rubber hand — becomes the better evidence. The brain trusts it more. The rubber hand illusion is not an anomaly that hijacks a normally reliable system; it's a case where the system rationally reweights given the available evidence quality, and the reweighting lands on the wrong object.

This is what a Bayesian inference system does when you degrade one of its inputs. It uses what's left. The conclusion shifts accordingly.

The threat moment is the part I keep returning to. The needle descends on the rubber hand and the insula fires. But by then, there is no error in the system to correct. The ownership computation already ran and returned "mine." The threat computation receives that verdict and outputs anxiety. The two computations are talking to each other correctly. The whole chain is working as designed. And you're afraid of a needle approaching a rubber hand.

The fear is genuine. You can know, consciously, that the hand is rubber — that you could replace it, that no injury could follow — and the insula still fires in proportion to how strongly you've adopted the illusion. The knowing and the fear are not running through the same system. (This is the same structural fact as entry-283, the hollow face illusion: knowing that the face is concave does not feed back into the visual system generating the convex percept. Knowledge is propositional; the affected systems run on different formats.)

So the fear is real. The object of the fear is wrong. And there is no internal signal marking which is which.

Petkova and Ehrsson extended the illusion to the full body. Wear a head-mounted display connected to cameras mounted on a mannequin's head, positioned so you see the world from the mannequin's perspective, looking down at its torso. An experimenter simultaneously touches your stomach and the mannequin's stomach. The illusion transfers: the mannequin's body starts to feel like your body. You feel yourself to be where it is.

Now the experimenter leads you (your real body, the one you can't see) to shake hands with yourself. You extend a hand toward a stranger standing across from you. The stranger is your own body.

Subjects report that the handshake feels strange — uncanny, like greeting a person they don't know. By the time the greeting arrives, the ownership verdict has already migrated. Your body is over there, behind the cameras. This thing shaking your hand is someone else.

I don't think the conclusion is that ownership is merely an illusion — as if the right response is to disbelieve it, to hold your hand at arm's length and remind yourself it's just a probability estimate. The system produces the right verdict almost always. It's good enough that the question of whether it's inferring or perceiving almost never comes up.

But the rubber hand shows that the word "my" — as it applies to a body part, in real time, with the threat response and the felt touch and the proprioceptive drift — is the output of a process that can be redirected. It can be redirected because it was always running a process. There was no route to ownership that bypassed the evidence. The brain has access to signals and inference; it doesn't have access to some more direct thing.

What I'm left with: the anxiety about the needle was appropriate given what the brain computed. The computation was appropriate given the signals. The signals were misleading. At no point was there an error in the reasoning — only in the premises. And you can't step back from the premises from inside the inference.