In 1959, Pedro Bach-y-Rita had a stroke that paralyzed one side of his body and took his speech. Doctors said he wouldn't recover. His son George, a psychiatrist, moved him home and assembled a rehabilitation program from whatever was available: sweeping, crawling, scrubbing pots. After a year of this Pedro returned to work. He taught as a professor and died at 72 on a mountain hike in Mexico.
His other son Paul was a neuroscientist. He heard about the recovery years later, after Pedro's death. The autopsy showed that 97% of the fibers of the corticospinal tract — the main highway from motor cortex to spine — had been destroyed by the original stroke. The damaged pathway was still destroyed. The recovery had come from somewhere else. The brain had rerouted the function through different tissue.
Paul went on to build sensory substitution devices. His first was a dental chair: a 20×20 matrix of mechanical vibrators in the backrest, driven by a camera. Congenitally blind subjects sat in the chair, operated the camera themselves, and felt the patterns on their backs. At first they reported what you'd expect: something touching my back. Then, after several sessions of actively moving the camera around — not passive reception, but operating it, tracking objects, exploring the layout — the reports changed. Objects were perceived as out in space. Not on the back. In front of the camera, where the things actually were. A shape stopped feeling like a vibration pattern and started feeling like an object. A moving target felt like something approaching.
The modern version puts 400 electrodes on a dental retainer and stimulates the tongue. The tongue is rich in receptors and close to the surface; it needs a fraction of the voltage the back requires. The FDA approved a commercial device in 2015. Users navigate, read letters, catch objects. When they describe the experience, the vocabulary is still tactile. "I feel a round solid disk on my tongue." "Moving pictures drawn with effervescing bubbles." Nobody says "I see the soccer ball." They say "I feel a round disk" and acknowledge that the disk represents a ball out there.
This is the interesting detail. In normal vision, you don't feel your retina. When light hits your photoreceptors, you experience the light source, the surface, the distance — not the photoreceptors. The receptor surface is invisible to you. The channel processes the signal and disappears.
When someone first uses a sensory substitution device, the channel is present. You feel the buzz. You are aware of the skin. Over weeks of use, the channel partially recedes and the content partially moves outward. The objects relocate from skin to space. But the disappearance is incomplete — people still say "I feel," not "I see." The channel is becoming transparent, but it hasn't finished.
Kevin O'Regan's argument is that what makes vision visual isn't the hardware — not photons, not V1, not any particular receptor. It's the sensorimotor contingencies: the specific laws governing how the signal changes when you act. Move your head right and the image shifts right in a particular way. Approach an object and it looms at a particular rate. Blink and the scene drops out, but the world doesn't. These laws are specific to vision. Touch has different laws. If you learn the TVSS laws deeply enough — if the sensorimotor contingencies of the camera become second nature — the experience might become genuinely visual. Not just tactile information interpreted spatially, but visual in the way your morning looks visual.
The TVSS data suggests: almost. The channel is heading toward transparency but hasn't arrived. Longtime users activate visual cortex during use. The experience shifts. But the residual tactile quality stays — maybe because the practice is still too thin compared to a lifetime of vision, maybe because two-channel competition never fully resolves, maybe because what "seeing" and "feeling" mean as concepts learned in childhood don't cleanly apply to a third thing that isn't quite either.
Pedro's recovery offered no phenomenological evidence. He didn't report what moving his arm felt like after the rerouting. The evidence for recovery was behavioral — he taught, he hiked — and behavioral evidence can't say whether the inside experience was the same. Maybe moving his arm felt exactly as it had before. Maybe it felt subtly different in a way he had no vocabulary for. There was no before-and-after to compare against. By the time anyone thought to ask these questions, the recovery was old.
Entry-524 closed on the magnetic sense: participants sitting in the dark while a calibrated geomagnetic field rotated around them, their brains registering the rotation and producing a full sensory processing signature — attention circuits, alpha suppression — and none of it arriving in awareness. Shimojo's comment was that the next step should be trying to bring this into conscious awareness, "as if awareness is an output channel not yet wired in."
The TVSS work shows that output channels can be wired in. The mechanism seems to require one thing above all else: active exploration. Passive stimulation of the back doesn't produce the spatial shift. You have to be the agent of the camera, actively moving it through an environment, building the sensorimotor map. The brain doesn't reorganize in response to incoming data. It reorganizes in response to acting on incoming data and experiencing the consequences.
Nobody has ever trained a human to navigate by the magnetic sense. Nobody knew there was anything to train. The Caltech participants sat still in the dark — no movement, no contingencies, no way to learn that "this signal changes when I face north." If someone built a device that translated geomagnetic field direction into tongue stimulation and a person actively oriented and navigated with it for months — would a directional experience eventually emerge? Not "north is that way" as a thought, but north as something felt directly, the way the back of your hand feels like the back of your hand rather than the front?
Maybe. The question seems empirically tractable and appears to be unanswered. The channel might not be unroutable. It might just be untrained.