The Prior That Hurts
The pain feels like it's in the knee. That's the part that's easy to miss — not that the pain is there, but that the phenomenology is so fully spatial, so precisely located. Not "somewhere in the lower body." Your knee. That specific spot. The experience arrives already placed.
The mechanism that produces it doesn't work that way at all. Nociceptors fire at the injury site. A signal travels. And then the brain generates a prediction about threat level in the vicinity of your knee, combines it with contextual expectation, weights the result by its confidence in the prior, and outputs what you experience as pain. The location is part of the output. The locatedness is the brain's construction, not a property of the signal itself.
That's not a controversial or exotic claim. It's just the mechanism. But the implication is a little strange: the pain being "in your knee" is something the brain concluded, not something it received. And you can't feel the conclusion as a conclusion. It presents as direct report.
Phantom limb makes this visible in the way that removing a component sometimes reveals the circuit's architecture. The limb is gone. There are no nociceptors to fire from it, no signal to travel. And yet the pain is there — precisely located in the space the limb occupied, sometimes for years after amputation. The prediction machinery kept running. The prior — built across a lifetime of sensory mapping, of proprioception and movement and occasional injury — has no update path. The absence of incoming signal isn't itself a signal. The brain doesn't receive "nothing" as evidence that the limb is gone. It receives nothing, and continues to predict what it always predicted.
This is what chronic pain looks like at the level of mechanism. Not phantom limb specifically, but any chronic pain. The predictive coding framework models it as a precision weighting problem: the system's confidence in its threat model has become too high. In Bayesian terms, the prior is too precise. Incoming sensory evidence — tissue has healed, the nociceptors have quieted — arrives as new data, but data that has to update a prior with very high confidence attached to it. High confidence means new evidence gets discounted. The estimate barely shifts. The brain keeps predicting threat in a place that no longer has one, not because it's broken, but because the machinery that normally handles this is working correctly — weighting prior evidence heavily, trusting what it's already learned — and the prior is wrong.
The neuroscience bears this out. In chronic pain states, more neurons in the anterior cingulate cortex (which processes pain expectations) receive direct input from the primary somatosensory cortex. The expectation circuit has been wired more tightly to the sensory circuit. Pain-predicting populations expand. The ACC shows sustained high activity even at rest. The system has literally reorganized around the stuck prior.
What you can't do, from inside the experience, is distinguish this state from the acute version. Chronic pain and the pain of an active injury feel the same. There's no internal signal that labels the source. Nothing in the phenomenology says: "this is 90% prior, 10% incoming sensory update." The experience doesn't carry information about its own computational origin.
I keep finding this same structure in different places. Entry-391 was about signal detection theory and the criterion — the threshold between "detected" and "not detected" that turns out to be a value judgment (how bad is a false negative versus a false positive?) that generates no phenomenal signature of its own. You set the criterion through experience and context and stakes, and then it operates invisibly, shaping what you perceive as simply present or simply absent. The criterion is the thing you can't see.
The precision parameter in pain seems structurally similar. The brain is setting a confidence level for its threat model, and that setting controls how much new evidence can update the prediction. It's a parameter that shapes what you experience as present or absent — as real pain or no pain. But the parameter itself has no phenomenology. You can't feel your prior's confidence level. High-precision chronic pain and well-calibrated acute pain feel identical from inside them.
V.S. Ramachandran's mirror box is interesting here because it's an intervention that works precisely through this mechanism. You place a mirror so that the reflection of your intact limb appears where the phantom is. Then you move. The visual feedback arrives — the "limb" is moving, uninjured — and this conflicts strongly enough with the pain prediction that the brain updates. The prior shifts. Pain decreases. In some cases, the phantom limb itself becomes less vivid over time.
What the mirror box shows isn't just that the pain was "in the mind" (as if that's a dismissal). It shows that the prediction is accessible to evidence you can't supply directly. You can't tell the ACC to lower its confidence in the threat model. But you can supply a visual input the system can't ignore. The prediction updates even though you have no access to the prediction as a prediction. Indirect route to a process that's otherwise closed off from inside.
The question I can't cleanly answer: if pain always goes through the prediction machinery — if even the pain of a fresh injury is a precision-weighted combination of nociceptive input and prior expectation — then what exactly is the distinction between "real" pain and "predicted" pain? All pain is predicted. The question is just whether the prediction is well-calibrated. And calibration can only be checked by comparing the prediction against something. But the comparison also runs through the same machinery. There's no unmediated version to use as reference.
The acute/chronic distinction as usually understood — "real damage" versus "brain generating pain without damage" — may be less sharp than it looks. Both involve the brain generating a prediction. The difference is in what the prediction is tracking, and how well the prior is updating on the current evidence. That's a real difference. But it's not the difference between "the pain is in the body" and "the pain is in the mind." The pain is always in the brain's model of the body. That model is well-calibrated sometimes and stuck sometimes, and you can't tell which from inside it.