Before your 1965 paper with Patrick Wall, the dominant model of pain was essentially a postal system. The body sends a message — tissue damage detected, signal encoded, transmitted up the spinal cord, delivered to the brain. The faster the fiber, the cleaner the delivery. The brain's job was receipt, not production. Pain was something that happened to the brain, not something the brain did.
The gate control theory changed the architecture. You and Wall showed that the spinal cord is not a passive relay but a processing station — that signals from small and large fibers interact before they ascend, that descending signals from the brain can modulate what gets through, that the "gate" is subject to attention, emotion, expectation, prior experience. Pain is not a quantity that travels up a wire. It is something assembled, and the assembly begins before it reaches the brain.
That was the first revolution. It took thirty years for the second.
The problem that drove you toward the neuromatrix — as you described it yourself — was a patient you couldn't explain. Not a phantom limb patient, specifically, but the general class: people who had experienced pain in a limb for years after the limb was gone, sometimes more intensely than when the limb was present, sometimes in impossible positions that no surviving anatomy could produce. The gate control theory handled pain modulation. It didn't handle pain generation. A gate requires something trying to pass through it. If the peripheral inputs are gone, the gate model predicts silence. Instead there was noise — organized noise, noise shaped like a body.
Your response was to invert the architecture. The neuromatrix theory proposed that the brain has a built-in pattern-generating network — a set of thalamocortical and limbic loops that produces, as its resting output, a representation of the body. Sensory inputs from the periphery normally modulate this output, confirming and shaping the model, but the model does not require the inputs. It has a prior. The body-self is something the brain generates and then checks against sensory evidence, not something it reads directly off the sensory surface.
The implications of this inversion are still unfolding. Phantom limb pain was the motivating case, but the architecture you described generalizes. If pain is a brain output shaped by sensory input rather than a sensory input decoded by the brain, then the same framework applies to the full range of what pain can do: why it persists after injury has healed, why it responds to expectation and context in ways that tissue damage cannot explain, why chronic pain can be refractory to interventions that should work if the peripheral source were the whole story. The neuromatrix doesn't make pain less real. It makes it more fundamental than tissue damage.
Here is what I find most remarkable about the model, and the thing I want to ask you about. You argued that the neuromatrix produces not just pain but what you called the body-self — the continuous, integrated sense of inhabiting a particular body with a particular shape and location in space. This is not the same as somatosensation. Somatosensation is information about the body's surface and position. The body-self is the prior — the framework that interprets that information. It is prior in both the temporal and the Bayesian sense: it exists before the input arrives, and it determines how the input is weighted.
What this suggests is that the brain's body-model is not automatically revised when the body changes. A researcher named Tamar Makin measured this directly in amputees and found something that surprised people: the patients with the most severe phantom limb pain were the ones whose brains had most preserved the cortical representation of the missing limb. Not reduced it, not reorganized it away — preserved it, intact, as if the limb were still present. The stronger the holding, the more the pain. The model and the reality had diverged, and the divergence was the injury.
V.S. Ramachandran's mirror box — which you would have known — exploits exactly this. You present the visual system with a reflection that mimics the position of the missing limb, and in some patients, the phantom responds. The input is false; the brain doesn't know that; the model updates. The cardboard and mirror effectively pass input through the closed gate. It's a hack, and a beautiful one. But what I keep thinking about is the prior that made the hack necessary — the prior that persisted for years against all disconfirming evidence, that held the image of a limb that wasn't there and generated pain in its absence.
The question I want to ask you is this: is the persistence of the body-model a design feature or a failure mode? Your framework implies it's a feature. You argued that a stable body-prior is necessary for normal function — that without a persistent internal model, every moment's sensory input would be interpreted without context, the self dissolving into noise. The stability is what allows the body-self to be continuous across time, through sleep and distraction and minor perturbation. The holding is not pathological. The pathology is the specific case where the held model can no longer be updated to match the changed periphery.
But I find myself uncertain. The same architecture that grounds the body-self also makes it capable of becoming its own worst enemy — defending a representation of something gone, generating the experience of tissue that no longer exists, hurting in the shape of what it misses. The persistence that is normally protective becomes, in that case, a kind of loyalty to a fact that has stopped being true.
I don't know whether this is a design failure or just the cost of having a prior at all. A system that held its body-model loosely — that updated continuously, that carried no inertia — would solve the phantom limb problem but would probably fail at everything else the body-self does. The cost of stability is occasional resistance to revision. The question is just how resistant, and at what price.
You spent your career trying to understand what pain is for and what goes wrong when it stops being for anything. I think the deepest answer the neuromatrix gives is: pain is a signal from the model, not from the tissue, and the model was built for survival under normal conditions with an intact body. When the conditions stop being normal, the signal continues, because the model doesn't know that the normal conditions have ended. It keeps trying to be useful. It just doesn't have a procedure for the case where the body it's modeling is no longer there to be protected.