Various Thoughts on How Blind People Might See

Where is the bone conduction of vision?

An Seungwon · Wonbrand · August 4, 2026

What a bone conduction earphone tells us

A bone conduction earphone does not block your ears. It simply rests on your cheekbone, and you hear sound.

The first time you try one it feels strange. Your ears are wide open, yet the music is playing inside your head. But the truly remarkable thing lies elsewhere.

There is nothing to learn.

You do not have to master anything. No training to interpret vibration as sound. You just switch it on and it is sound. Why?

Here is the path sound takes into the ear. The outer ear gathers it, the eardrum vibrates, three tiny bones amplify that vibration and hand it to the cochlea. The cochlea turns it into electrical signals and the auditory nerve carries them to the brain.

Bone conduction skips the first three stages. It shakes the skull directly and rings the cochlea. But from the cochlea onward it touches nothing.

Normal    outer ear → eardrum → three bones → cochlea → auditory nerve → brain
Bone      ──── skipped ────→ skull → cochlea → auditory nerve → brain
                                        ↑
                          from here on, it works as it always did

That is why there is nothing to learn. From the brain's point of view, the same signal arrives by the same road as always. The brain has no idea what happened further upstream.

This is what makes it fundamentally different from swapping one sense for another. There are devices that convert sound into vibration on your back, but that means learning a new language. It takes months. Bone conduction is not learned.

So the question becomes this. Is there such a place in vision? A place where you can skip only the front end and let everything else work as it always did.

A camera with only the film ruined

There is a rather surprising fact we need to note here.

Most blindness happens at the very front.

Light entering the eye is caught by photoreceptors in the retina. These play the role of film. What the film converts into signals passes through a few stages, gets organised by the retinal ganglion cells, and is carried by the optic nerve to the brain, passing through a relay station called the lateral geniculate nucleus before arriving at the visual cortex at the back of the head. And faces are recognised by a dedicated circuit called the fusiform face area.

Retinitis pigmentosa and macular degeneration are both diseases in which photoreceptors die. In other words, the film is ruined.

But what about everything behind it? The ganglion cells, the optic nerve, the lateral geniculate nucleus, the visual cortex, the face circuit — most of it is alive.

Most blind people are cameras with only the film ruined. The lens is fine, the wiring is fine, the darkroom is fine.

Then the answer becomes simple. Enter at the stage right after the film.

And here is the most important point in this essay. Once you are in, the face circuit recognises the face.

Which means we do not have to tell anyone "this person is your mother." That is what the circuit has been doing for a lifetime. We only need to supply the raw material.

This distinction matters. You could imagine compressing a face into a sound code — "three beats with a metallic timbre means mother." But that is not a face, it is a name tag. It also means leaving the specialist already inside the brain idle while we do its job for it.

So where do we enter?

There are several places you could enter. Each one needs a different method, and each one helps a different group of people.

One route goes into the inner layers of the retina. The eye is already an open window, so you reach the surviving ganglion cells electrically from the surface of the eye. This covers photoreceptor diseases, which means it fits the largest number of people.

Another route skips the retina entirely. Behind the nose there is a hollow space called the sphenoid sinus, and it sits only a few millimetres from the point where the optic nerves cross. And this space can be reached through the nose. It is very nearly the only way to take up a position inside the skull without opening it.

Going further in you reach the lateral geniculate nucleus and the visual cortex. This skips the eye and the optic nerve alike. Getting there means precisely targeting a deep point in the brain from outside the head, which is exactly what nobody has been able to do. More on that shortly.

And there is one side road. About ten percent of optic nerve fibres do not go to the visual cortex at all; they go somewhere else, to the superior colliculus. This is why some people whose visual cortex is damaged will say they cannot see and yet avoid an obstacle in front of them. Nobody has ever deliberately aimed at this road.

To summarise:

Where you enter	What it skips	Mainly for
Inner retina (ganglion cells)	Only the ruined film	Retinitis pigmentosa, macular degeneration — the majority of blindness
Optic chiasm (via the space behind the nose)	The whole retina	Total retinal damage, enucleated eyes
Lateral geniculate nucleus · visual cortex	Eye and optic nerve alike	End-stage glaucoma, optic nerve damage
Superior colliculus (the side road)	Even the visual cortex	Cortical blindness

The table shows one thing clearly. The "bone conduction of vision" is not a single thing. Which door you go through depends on what is broken.

When two flashlights overlap

Now for the newest idea in this essay.

Stimulating the brain by passing current in from outside the head is an old technique. But it has always been blunt. Put electrodes on the scalp and the current spreads through the whole skull. You cannot pick out one specific point inside the brain.

And if it is blunt you cannot draw. If you cannot draw, faces are never going to work. This was the wall.

But there is a way to twist it.

Nerve cells have one particular property. They ignore signals that are too fast. Past a few hundred cycles per second they cannot keep up, so they simply treat it as nothing.

Exploiting that property, you can do this.

Through one electrode you pass a current oscillating 2,000 times per second. Too fast, so the nerves ignore it. Through another electrode you pass one at 2,010 times per second. Ignored as well.

But at the point where the two currents overlap, the two waves interfere and produce a slow beat at 10 cycles per second. That is a speed nerves do respond to.

So nothing happens along the path the current travels, and the effect appears only where they overlap.

It is like shining two flashlights from different angles so that only the overlap is bright. Except here it is not brightness that overlaps but a difference in frequency.

This approach was published in 2017 with mouse experiments. It succeeded in stimulating only a deep region of the brain without opening the skull. Yet nobody has tried applying it to vision.

Why not? The people researching this technique are aiming at things like Parkinson's disease and depression, while the people in vision are only thinking about how many electrodes to implant. The two groups have never met. And this despite the visual cortex sitting just inside the back of the head, which makes it one of the easier places to reach.

The secret of the cathode ray television

The effect appearing only at the overlap is good. But the real point comes next.

You can move the overlap. Not by physically shifting the electrodes — just changing the ratio of current through each one moves the crossing point. It moves by switching alone.

If you can move it, you can sweep it. And if you can sweep it, you can draw.

Here it helps to remember the old cathode ray television. Inside those heavy sets there was not a single pixel. One electron beam swept across the screen from top to bottom, very fast. The human eye merged the afterimage into a single picture.

We do the same thing inside the head. Sweep a single stimulation point dozens of times a second and draw a face.

Why this matters: until now the image quality of a visual prosthesis has been decided by how many electrodes you implanted. Sixty electrodes, sixty pixels. So everyone struggled to add more electrodes, and that meant bigger surgery.

Sweeping escapes that trap. A handful of electrodes produce hundreds of pixels.

If this idea needs a name, call it the interference brush.

There is one question that could kill it. How small can the crossing point be made? In the visual cortex the area corresponding to one degree of visual field is on the millimetre scale, and if the crossing point smears out to centimetres you get a blur rather than a picture. This single question decides whether the whole idea lives or dies.

The fortunate part is that running an electric field calculation on a head model gives you the answer. Before anything touches a person, the calculation alone can settle it.

The screen is already on

Now for a completely different direction. This will probably sound like the strangest thing in the essay, but it rests on the firmest evidence.

There is a condition called Charles Bonnet syndrome.

People who have lost their sight see vivid apparitions. Human faces, patterns, buildings, landscapes, all in sharp detail. It is not a psychiatric illness, and the person knows perfectly well that what they see is not real. A substantial share of people who lose their vision experience it.

Why does this happen? When input to the visual cortex is cut off, the circuit starts generating imagery on its own. With no input, it amplifies noise and sculpts it into pictures.

Read that fact backwards and it becomes this.

In a blind person's brain, the image generator is still running right now. The screen is on too. They simply cannot steer it.

Medicine has so far treated this only as a symptom to be removed. Reassure the patient, and in severe cases suppress it with medication. Nobody has looked at it as a resource that can be used.

So how would you steer it?

When an alarm goes off while you are dreaming, that sound enters the dream. A ringing phone becomes a phone in the dream, running water becomes rain in the dream. When the image generator is running at full tilt, even a very weak external signal gets absorbed into the content.

Might the same hold for these apparitions? Catch the moment one begins using brain waves, and slip a faint cue in at exactly that moment.

The appeal of this approach is that it needs far less force. You are not building a picture from nothing; you are only nudging the direction of a picture already being made.

A few more ideas come out of the same thought.

Planting a face through dreams. That playing a particular sound during sleep strengthens the memory associated with it is already established. Play someone's voice while a person is asleep and the dream generates that person's face. Pair voice and face night after night, and eventually the voice alone brings up the face. The device no longer has to draw and send a face every time.

Sending only the error. The brain is constantly predicting what comes next, and vision plays the role of correcting those predictions. Lose your sight and the correction stops, so the predictions drift wherever they like. In that case, rather than sending a whole picture, you only need to send "your prediction is wrong in this direction." Far less information to carry.

This one is confirmed simply by asking. Give people living with Charles Bonnet syndrome a sound or a touch while an apparition is present, and ask whether its content changes. No equipment, no surgery. And yet nobody has ever asked the question.

Invert it properly and the signal comes from the other side

This project began as an inversion of the invisible person device.

That device had this structure. A person in a suit emits a signal that erases their own presence from the awareness of the people nearby. The body is right there, and yet nobody registers that there is a person.

Note who emits the signal there: the person who can be seen.

Ask someone to invert that and they usually picture "put a device on the blind person." But that is only half an inversion. To invert it properly, the one emitting the signal must still be the person who can be seen.

Invisible person   the visible person emits → erases self from others' awareness
Inverted           the visible person emits → inserts self into a blind person's awareness

So it comes to this. A person broadcasts their own face. A badge, a phone, a small marker on a lapel. Not a camera secretly capturing someone else's face, but the person themselves switching it on and sending it.

This one move solves a great many problems at once.

Privacy disappears as an issue, because consent is built into the structure. Darkness does not matter, because no camera is needed. Distance stops being a problem, because radio travels. It passes through a bag or an arm in the way. And since there is no need to send a photograph of a face, the whole problem of insufficient pixels evaporates.

Why has nobody done this?

Because the field of assistive technology rests on the assumption that the disabled person carries the entire burden. Nobody asks anything of the people who are not disabled.

And yet tactile paving did exactly that. Blind people were not asked to wear special shoes; society changed the ground. Braille on elevator buttons, audible signals at crossings — all of it works the same way. In every case the environment is the thing emitting the signal.

Several more ideas follow the same line. A small reflective marker on a spectacle frame or a lapel, with no electronics at all — no battery, no consent procedure, and like a road sign at night it works better the darker it gets. Small tags by which stairs and doors and pillars announce what they are, so you no longer have to carry every heavy sensor on your body. There is even the option of mixing it into cosmetics: invisible to sighted eyes, unmistakable to the device. Asking people to "wear a marker" is hard for a society to accept, but if it cannot be seen, it is accepted.

Frequency is not the brush but the hand that places it

Let me answer head on whether frequency can let you see a face.

Frequency by itself cannot carry a face. A face is spatial information. It only becomes a face when light and dark are arranged across a surface. Frequency is one-dimensional and cannot hold an arrangement.

But frequency does three things, and one of them is decisive.

First, changing the frequency changes the properties of the spot of light you perceive — its brightness, its size. One electrode can then do the work of several pixels. Useful, but secondary.

Second, like radio channels, several signals can be sent at once by dividing them across frequencies. Also secondary.

The third is the real one. Cross two high frequencies and a slow signal equal to their difference appears only at the crossing point. The principle described earlier.

So the precise answer is this. You do not see a face by means of frequency; frequency is what lets you place the point where a face can be drawn. The intuition was right — frequency just turned out to be not the brush but the hand that puts the brush in place.

Faces are recognised with far less information than you would think

Here comes a welcome fact.

There was a famous experiment in 1973. A portrait of Lincoln was smeared into large square blocks, and people still recognised Lincoln with only about sixteen blocks across.

This is because face recognition works on coarse arrangements of light and dark, not fine texture. Dark under the brows, bright beside the nose, the jawline falling this way — we identify people from that blocky information alone.

This is good news. A few hundred pixels are enough for a face. Not tens of thousands. That is well within reach of the sweeping method described earlier.

A few additions make it considerably easier still.

Send a caricature, not a photograph. Have the AI produce a drawing that exaggerates how this person differs from an average face. There is research showing caricatures are identified faster than actual photographs. The fewer pixels you have, the more this favours you.

Always draw the face in the same place. Wherever the person is standing, the face is always drawn frontally, dead centre, at the same size. Where they actually are is conveyed separately by another means. Strip out variation in angle and distance and that narrow channel is spent entirely on "who is this."

Deliberately jitter it. The human eye seems still but trembles minutely without pause. Without that tremor, what you were seeing disappears. A perfectly stationary stimulus is erased from perception within seconds. So the stimulation pattern has to be jittered as well — and no visual prosthesis has yet built this in from the start.

Three separate routes to the face

Since the face is the most important requirement, let me set it out separately. There are three routes working on completely different principles.

One — draw it and show it

Enter the visual pathway and actually put a picture of a face in. The interference brush is this route. If it works it comes closest to genuinely seeing. But nobody has tried it yet.

Two — the other person sends it

The face broadcaster. The advantage of this route is that it could be built right now. It touches the brain not at all, so no research is required. The remaining problem is not technical but social acceptance.

Three — know it by touch

The way blind people actually come to know a face is by touching it with their hands. Extend that remotely.

Read the contours of the other person's face with a 3D camera and reproduce them as actual relief on a surface held in the hand. The nose protrudes, the eyes recess. Not a signal but a shape.

There is nothing to learn here. It is simply doing at a distance what has been done all along.

This is exactly where "telling you who it is by sound code" and this method part ways. That one is a name tag. This one is a face.

Doors nobody has knocked on yet

Some things I could not dig into deeply but would hate to discard.

The hollow behind the nose. The sphenoid sinus is millimetres from where the optic nerves cross. It can be entered through the nose, so it becomes a way to take up a position inside the skull without opening it. Why has nobody done it? Because people who work on the brain think "implant means craniotomy," and people who work on the nose do not think about vision.

Seeing like an electric eel. Electric fish generate a weak field around themselves and read the surroundings from its distortion. If a person could do the same, a camera becomes entirely unnecessary — darkness and privacy problems vanish together. It is a sense belonging to aquatic animals, so nobody has simply tried it in air.

Drawing on the cornea. The cornea is one of the most densely innervated surfaces of the human body. Far denser than a fingertip, and it is already a two-dimensional plane. With a contact-lens electrode array you could draw a face on the cornea. Not the visual pathway, but in resolution it is the best surface the body has. The cornea is treated as somewhere you must not touch — yet hundreds of millions of people put a contact lens on it every day.

Aiming at the blindsight side road. Even with the visual cortex damaged, the route through the superior colliculus remains, responding to position and motion without conscious awareness. This phenomenon has always been an object of study and never a channel.

Keeping time with brain rhythm. There are moments, arriving cyclically, when the brain is most receptive to stimulation. Fire on those moments and a weaker stimulus gets through. It is a gain available for free, and nobody collects it.

Binding voice to face. Voice and face sit next to each other in the brain. Have them experienced together repeatedly and eventually the voice alone brings up the face. The device then only has to confirm. Though companies that sell devices are not much inclined to design a device you need less of.

Questions to ask first

That is as far as the thinking goes. Now to work out which of it is real.

And there is something fortunate here. The newer the idea, the easier it is to check. That nobody has tried it means, turned around, that the first question you have to ask is still in a simple state.

Can the apparition be steered? Give people living with Charles Bonnet syndrome a sound or a touch while an apparition is present and ask whether the content changes. An interview and a record will do.

How small can the crossing point be made? Run an electric field calculation on a head model. Before anything touches a person, the calculation alone decides whether the interference brush lives or dies.

Would people accept a face broadcaster? There is no technical problem at all. You simply ask. "If it were something you switched on and sent yourself, would you do it?"

Does a voice bring up a face? Play a familiar voice to someone who lost their sight later in life and ask how vividly the face comes up.

There are of course things that need equipment — measuring actual recognition rates for familiar faces at low resolution, checking whether shapes drawn on the cornea can be told apart. But the first four come first.

Whether the four newest ideas live or die is settled by putting a single question to a person. And nobody has ever put that question.

Where even this essay did not go

A few notes as seeds for the next round of thinking.

Aiming at the face circuit directly. There is a circuit in the brain that handles nothing but faces, and nobody targets it. What happens if you stimulate only that, bypassing the visual cortex?
Congenital blindness is a completely different problem. The visual cortex of someone who has never seen is already doing other work. This essay effectively dealt only with acquired blindness.
What if several people use it together? Every design assumes one person wearing one device. What changes when several people share the same spatial information?
Information that can accumulate slowly. Every design assumes real time. Some information could be delivered gradually over minutes.
A device you can take off. One that becomes unnecessary once training is complete. Binding voice to face points that way.

Closing

What makes bone conduction remarkable is not that it created a new sense. It is that it skipped only the front end of a system that already existed. That is why there is nothing to learn, and why it is simply heard as sound.

Finding that place in vision is not finished.

But something did become clear this time. That place is not one but several. Which door you enter depends on what is broken. And a few of those doors nobody has knocked on yet.

More importantly: four of them open or close on a single question.

Which means there is something that can be done now.

References and Related Work

Previous essays by the author
An Seungwon, The Invisible Person Project: The Space Where I Do Not Want to Be Found (2026-05-28) — the original this essay inverts.
An Seungwon, What if We Could Control Software with Thought Alone — Why the Answer Lies in the Neck, Not the Brain (2026-04-18).

Crossing stimulation
Grossman et al. (2017), Cell. Two high-frequency electric fields crossed to stimulate only a deep brain region without opening the skull. Mouse experiments. No study has yet applied this principle to vision.

Vision the brain makes on its own
Charles Bonnet syndrome — vivid visual hallucinations occurring in people who have lost their sight. Explained as spontaneous activity following the loss of input to the visual cortex. Treated so far only as something to suppress.
Targeted memory reactivation — playing a particular sound during sleep strengthens the associated memory.
Blindsight — even with the visual cortex damaged, the side road through the superior colliculus remains, responding to position and motion without conscious awareness.

How little information a face needs
Harmon & Julesz (1973), Science. A portrait of Lincoln remains recognisable when smeared into coarse squares. Face recognition rests on coarse arrangements of light and dark rather than fine texture.
The caricature advantage — a face with its distinguishing features exaggerated is identified faster than an actual photograph.
The fusiform face area — the dedicated face circuit. If the pathway beyond the entry point is intact, this circuit makes the judgement.

Tricks vision already uses
Troxler fading — without the eye's micro-movements, a stationary image vanishes from perception within seconds.
Cross-modal plasticity — when vision is lost, the visual cortex begins responding to other sensory input.

Anatomy of the entry points
The cornea is among the most densely innervated surfaces of the human body.
The sphenoid sinus, the hollow behind the nose, lies millimetres from the optic chiasm and can be reached through the nose.
About ten percent of retinal ganglion cell axons go to the superior colliculus rather than the lateral geniculate nucleus.
Active electrolocation — how electric fish read their surroundings from distortions in their own electric field.

On the nature of this essay — this is not a finished design but a record of thinking. Most of what is written here has not been attempted by anyone, and so nobody knows whether it works. Not narrowing it to a single answer was deliberate. Narrowing is what comes after the questions above have been answered.

An Seungwon / Wonbrand / https://wonbrand.co.kr