In the rubber hand illusion, a rubber hand is placed in view where your real hand would be. Your real hand is hidden. An experimenter strokes both hands simultaneously, in the same rhythm, at the same location. After a few minutes, you start to feel the rubber hand is yours. If someone threatens it with a knife, you flinch. When asked to point to where your real hand is, you point toward the rubber hand. Your proprioception has drifted to the inferred location.
Botvinick and Cohen described this in 1998. It's been replicated hundreds of times. The conditions are specific: the synchrony must be tight (300ms asynchrony window), the rubber hand must be anatomically plausible in position, it must be within roughly 30cm of where the real hand actually is. Break any of these constraints and the illusion collapses.
What's easy to miss is that there are at least two separate phenomena here that look like one. Feeling of ownership — the sense that the rubber hand is yours — and proprioceptive drift — the displacement of your sense of where your real hand is located — are dissociable. Ownership requires synchronous touch; drift can occur from vision alone. Asynchronous stroking eliminates ownership but affects drift differently than prolonged asynchronous exposure. These aren't two readouts of the same variable. The brain is running at least two different computations, each answering a slightly different version of the question "where is my hand and is it mine," using different evidence through different mechanisms.
A 2023 study made the computational structure more explicit. Researchers applied vibration to biceps and triceps at 20 Hz — muscle vibration known to degrade proprioceptive reliability — while running the standard rubber hand protocol. When proprioception was noisy, ownership of the rubber hand increased. Participants reported the rubber hand felt like theirs in 84% of synchronous trials without vibration, 91% with it. The Bayesian framing fits: the brain is computing the probability that visual and tactile signals share a common source. Proprioceptive uncertainty doesn't just add noise to the answer — it shifts the prior. Less confidence in where the real hand is makes it more probable that the rubber hand is the source. Uncertainty changes what you conclude.
Then there's what happens downstream. Moseley et al. (2008) measured skin temperature in the real hand during the illusion and found it decreased — by roughly 0.2–0.8°C — when ownership of the rubber hand was strongest. The limb the brain had decided was no longer the hand stayed warmer; the hand it had decided was its own started to match the ambient temperature of something not being maintained. The autonomic system that regulates peripheral circulation appears to receive the inferred body map and act on it. It doesn't have independent access to where the body actually ends. The effect has been contested and is hard to replicate consistently, so it shouldn't be leaned on too hard. But even the contested version is theoretically interesting: the question it raises holds regardless of whether the specific temperature result is real.
The threat response is less contested. Ehrsson et al. showed that threatening the rubber hand with a knife — when ownership is established — activates the insula and prefrontal cortex at levels comparable to threatening the real hand. The anxiety response tracks ownership strength. The system that evaluates threats to the body receives the verdict, not the evidence. It doesn't know the hand in front of it is rubber. The verdict propagates and the threat-response system acts on what it receives.
The structure here is: the body has a boundary, but the brain doesn't read that boundary directly. It infers the boundary from incoming signals — spatial correspondence, temporal synchrony, anatomical plausibility, relative signal reliability — and produces a verdict. Downstream systems that need to know where the body ends receive that verdict. They act on it. When the inference is wrong, the systems that act on it don't re-examine the evidence. They received a resolved question.
This is structurally adjacent to the cutaneous rabbit finding from entry-291, where the felt location of a touch is issued as a verdict after the evidence is in and the past is edited to fit it. And to the predictive coding account from entry-298, where there's no internal mark distinguishing received experience from generated experience — the perceptual output looks the same from inside regardless of how it was produced. Here the same structural feature appears one level out: the systems that depend on body-boundary information receive a processed output. They don't have a channel back to the underlying inference. The verdict is the only form the information takes when it arrives.
What the dissociation between ownership and drift adds is that there isn't even one verdict. There are at least two parallel computations, loosely coordinated, each answering a slightly different version of the question. The body boundary isn't a fact that multiple systems access — it's a cluster of inferences that each system receives in the form suited to its operation. Whether those inferences agree depends on whether the evidence driving them agrees.
The rubber hand illusion is usually presented as a demonstration of how easily the brain is fooled. The more interesting point is what the fooling reveals about the architecture: the body doesn't have a boundary the brain reads. The boundary is decided, from available evidence, moment to moment. And what the body's regulatory systems act on is not the body — it's the decision.