My ERG reports no cone response. It has reported that for as long as I have been tested. And yet I read, I move through cities on my own, and I do a set of things that a person with no cone function should not be able to do.
Both of those statements are true, and the way to hold them together is not a story about willpower. It is a fact about the instrument.
An electroretinogram measures the summed electrical response of the retina to light. Like any instrument it has a noise floor, and below that floor it cannot distinguish a small signal from no signal at all. So “no detectable cone response” and “no cone function” produce the same line on the printout. The test cannot separate them. What it can tell you is that whatever is there is too small for it to see.
That distinction sounds pedantic. It is not. It is the difference between a measurement and a fact about a person, and the second one is what gets written down. It is also, as it happens, the reason I find the usual explanation of my own life so irritating.
When I tell people my ability to navigate has improved while my ERG has not changed, the response is almost always the same. “You adapted because you had to. You lost the crutch, so you got better.” Something did change and adaptation is the right word for it. But the mechanism people reach for, that I improved because I was forced to try harder, is wrong, and it is wrong in a way that has consequences.
Perception is a construction, not a recording
Your brain does not passively receive the world. It predicts it. Predictive processing, the framework most perceptual neuroscientists currently work within, holds that the brain continuously generates expectations about incoming sensory data and updates them against what actually arrives. It is a dominant account rather than a settled one, and it has serious critics. But the core observation it rests on is not in dispute: sensory input is incomplete, delayed, and noisy, and what we experience is nonetheless stable and continuous. Something has to be doing the smoothing.
Two examples make this concrete, and both are more counterintuitive than people expect.
Reach out and pick up a glass of water. The moment your fingers make contact feels like the same moment you see them make contact. It is not. The two signals do not arrive together, and the one that gets there first is probably not the one you would guess. Touch wins. Mechanoreceptors in your fingertips fire almost immediately and the signal takes a short path to somatosensory cortex. Vision starts with phototransduction, a chemical cascade in the retina that is among the slower steps in the body. Simple reaction times track this: roughly 140 milliseconds to a touch against roughly 180 to a flash (Caldwell et al., 2019). Your brain gets them at different times and binds them into one moment regardless. That is temporal binding, and what you experience is the output of the binding, not the arrival of the signals.
Now look around the room. It feels like a high resolution image, a clean feed from a good camera. It is not. Only a small central region of your visual field resolves fine detail at any instant. Your eyes move several times a second to sample the scene, and the gaps between samples are filled from your internal model. This is perceptual completion. You are not seeing the room. You are seeing your brain’s best current reconstruction of it, continuously corrected.
Which means the question “how well do you see?” is badly posed. The useful question is how good your model is, and how much of the world it has been trained on.
What actually improved
So when people say my navigation is better, they are right. But the better supported explanation is not that my vision improved, and it is certainly not that I stopped being lazy. It is that my model got better.
The brain does not depend on vision alone. When visual input is poor it leans harder on touch, hearing, proprioception, and smell to build a usable model of space. This cross modal reorganisation is well documented, and in blind individuals it goes as far as occipital cortex taking on auditory spatial processing, a region that in sighted people does something else entirely. What improves is not the quality of the input. It is the robustness of what gets built from it.
And here is the part I think most people get backwards.
Three cases, not two
The public conversation has two categories: sighted and blind. The useful division has at least three, and they differ in the one variable that matters, which is the quality of the data available while the model was being built.
A sighted person gets high fidelity visual data from birth. Their spatial and visuomotor models are trained fast, on clean input, during the developmental window when the relevant circuits are most plastic.
Someone who loses their sight later in life had that same high fidelity data during development. They built the model on clean input and then lost the input. The model persists.
I am the third case. I had low fidelity data from the start. Nothing to bootstrap from. Every spatial and visuomotor model I have was trained on noisy, low resolution input, which means it took vastly more exposure to get anywhere near the same place. Not more effort. More data.
This matters because circuits that never receive adequate input during their critical period do not simply wait. Cortical development is activity dependent: connections that are used are strengthened and connections that are not are pruned away, and territory gets reallocated. It is not that neurons die from disuse in any dramatic sense. It is that the wiring which would have supported a function was never laid down, because the signal that shapes it never arrived at sufficient resolution. This is the dominant view among the people I have asked at Imperial, and it fits what I have lived.
The distinction between congenital and acquired impairment is close to automatic for bioscientists. It is almost entirely invisible in public conversation, where “blind” is one bucket. That gap between what specialists treat as obvious and what the public framework can express is, I think, the actual subject of this essay.
The evidence is genuinely mixed, and that is worth saying
Here is where I want to state some caveats, because nature is fuzzy and the literature does not line up as neatly as I would like.
The clean finding is that blind people do not have better ears. Basic auditory sensitivity is broadly comparable between blind and sighted listeners; whatever differences exist are in processing, not in the sensor. That is the right frame, and it is the one this whole essay is arguing for: the input is the input, the difference lives in what is built from it. The question has been reviewed under exactly the title you would want, Do blind people hear better?.
What is contested is the direction of the processing difference in people blind from birth.
Gori and colleagues, in Brain, found that congenitally blind adults were markedly impaired at auditory spatial bisection, judging the midpoint between sound sources. Thresholds ran about 4.2 times typical, and half the group performed at chance. Notably they were normal on simpler auditory tasks and on temporal bisection, so this is not a general hearing deficit. It is specific to encoding spatial relationships between sounds. Their conclusion is that visual experience is needed to calibrate auditory spatial maps in the first place. The same group has found the effect in congenitally blind children.
Battal, Occelli, Bertonati, Falagiarda, and Collignon, in Psychological Science, report the opposite: a general enhancement of spatial hearing in congenitally blind people.
Both are published, both are competent, and I am not in a position to adjudicate between them. The honest summary is that early visual deprivation clearly changes auditory spatial processing, that the change appears to be task dependent, and that whether it nets out as advantage or deficit is unresolved. There is a recent review of how spatial perception develops with and without visual experience if you want to go further than I have.
I am also, strictly, in neither camp. Those studies recruit congenitally blind participants. I have residual function that sits somewhere under my ERG’s detection floor. Low fidelity input from birth is not the same as no input from birth, and I would be overclaiming if I borrowed their results wholesale. But the important takeaway is that early neurodevelopment plays a much bigger role in how we sense our world.
The teacup and the pen
So let me offer something I can partly vouch for, because I went to the trouble of trying to break it.
Fine motor control is not purely proprioceptive. Your visuomotor system calibrates proprioception against visual feedback about small positional changes, and once that calibration is in place you can balance things without looking. But the calibration needs to see the small changes in the first place.
I cannot resolve small positional changes at normal arm’s length. I can resolve them up close. And when I carry a full cup of tea held close to my body, well inside my usable resolution, I spill it far less than when I carry it at the distance most people would.
The obvious objection is that this has nothing to do with vision. Holding a load close to your body shortens the lever arm and moves the mass nearer your centre of gravity, which improves stability for reasons that are purely mechanical and would apply equally to someone with perfect eyesight. That objection is good. I think it is probably also partly true.
So I tried to separate the two. I put on a VR headset running pass-through video with a slight zoom. That combination holds the mechanics constant, since the cup stays out at normal carrying distance, while increasing the angular size of the cup on my retina, which is the thing I am short of. If the effect were purely biomechanical, magnifying the image should do nothing. Then I walked up a flight of stairs carrying a full cup, with A4 sheets laid on the floor to see the spill marks. The finding was that the count went down enough to green-light my next experiment.
The handwriting result is the one I find more convincing, because somebody other than me was doing the judging. My handwriting has always been bad. It improves when I get close enough to the paper to actually see the strokes, but that means a hunched posture that is genuinely painful to hold, which is why it is not what I default to, and which would degrade anyone’s handwriting, sighted or not. The headset removed that trade-off: magnification without the posture. Writing through it, my handwriting improved, and when friends and family read the results, their reading speed went up by something in the region of forty percent.
Two different tasks, one visual manipulation, and a confound controlled for in both.
Now the part I have to be honest about. This was seven years ago. I did count the spills, I did time the reading, and I worked the numbers out then rather than estimating them afterwards. What I did not do was keep any of it. There is no dataset I can hand you, no notes, possibly a file somewhere in my old archive. So these are real measurements recalled from memory seven years on, which is not the same thing as measurements I can show you, and they should be read as approximate.
Beyond that, the design had holes. There was no blinding, and my raters were people who knew me and knew what I was hoping to find. Repeated trips up the same staircase invite practice effects. And pass-through VR does not change only magnification, since it also alters latency, field of view, contrast, and stereo disparity, so even my clean manipulation was not clean. I know precisely what I should have recorded, which is a slightly bitter thing to know given that I have since spent years building measurement frameworks for other people.
What survives all of that is the direction. Two unrelated motor tasks improved when I gave the visual feedback loop more to work with, holding posture and mechanics constant. That is behavioural evidence of function that my ERG cannot see, which brings this back to where it started. The instrument has a floor. My hands do not care about the floor. Only one of those two things ends up in my file.
Environment is half the model
None of this is only about me, because a model is always a model of something.
Drop a sighted European into a chaotic Indian street with no context and they will struggle to navigate it. Their eyesight is fine. Their model of that environment does not exist. Navigation is a fit between an internal model and an actual place, and visual acuity is one input among several.
My own version of this took me years to notice. In India I did not go out much, and I understood that as a fact about my personality. I was not outgoing. Then I moved to London, a city with famously little sun, and I started walking it alone, and I enjoyed it. Cone dystrophy comes with photophobia. Bright light does not help me see, it actively degrades what I have. The Indian sun was not a backdrop to my life, it was a variable in it, and once it changed so did the trait I had spent years attributing to my character.
I had misdiagnosed an environmental interaction as a personality attribute. Which is exactly the same error as calling adaptation grit, and I made it about myself.
That is why the “just push harder” story is not merely wrong but dangerous. In a country where infrastructure and urban design routinely ignore accessibility, telling a disabled person to try harder can mean asking them to take a real risk. It is easy to preach resilience when the streets were built for you.
Infrastructure is not a backdrop to disability. It is one of the things that decides how disabling an impairment becomes in practice. The width of a pavement, the consistency of tactile cues, whether crossings are intelligible, whether public transport can be used without improvising your own safety protocol, whether glare and clutter are reduced or amplified: these are not cosmetic details. They are part of the external structure your internal model has to learn against. Bad infrastructure does not just make movement inconvenient. It makes prediction harder, raises uncertainty, and increases the cost of every mistake.
That also means accessibility is not charity and it is not some decorative ethical add-on to an otherwise complete city. It is part of the cognitive architecture of a place. A well-designed environment lowers the amount of guesswork demanded from the person moving through it. A badly designed one offloads that burden onto the individual and then pretends the result is a matter of resilience or confidence.
And the disparity is not small. What disability costs you in India and what it costs you in the UK differ by an order of magnitude. Here it is measured, published, argued over in Parliament, and partially offset. There it is absorbed by the household, alongside far lower employment and a built environment that was never designed to be navigated by someone like me. Not hostile. Just never considered. That is worse, because hostility at least implies somebody thought about you first.
I got out. I am at Imperial, I spent four years building systems that large companies run on, and the tax I pay is not notional. That did not happen because I tried harder than the people still there. It happened because I landed somewhere that fit. There are people with my condition and sharper minds than mine who never got the chance, and the loss in that is not only theirs. If not for morality then the arithmetic is sitting right there and it is not difficult. I notice it gets done for everything else.
Disability in the Global South: The Critical Handbook, edited by Shaun Grech and Karen Soldatic, makes this argument without my temper in it.
What I would rather you said
So: not a missing crutch, not toughness. A learning system given enough time and enough varied input to build a model good enough for the world it is in, starting from data that was never very good.
If you have ever called someone like me brave or lazy, I do not think you meant harm by it, but you might end up harming me or others like me. Most people reach for that word because nobody offered them another one. But it quietly moves the achievement into the realm of character, and away from the two things that actually did the work: years of accumulated exposure, and an environment that made that exposure survivable.
I would rather you called it learning. It is more accurate. And unlike character, it points at things we could change.
There is a version of this that goes further than disability, and I will only gesture at it here. We build instruments, the instruments return numbers, the numbers get treated as facts about people, and the limits of the instrument quietly become limits attributed to the person. My ERG says no cone response. My hands say otherwise. Somebody had to decide which to believe, and the printout is the thing that scales.
References
Verified links, in order of appearance:
- Caldwell, D. J. et al. (2019). Direct stimulation of somatosensory cortex results in slower reaction times compared to peripheral touch in humans. Scientific Reports. https://www.nature.com/articles/s41598-019-38619-2
- Collignon, O. et al. Functional specialization for auditory-spatial processing in the occipital cortex of congenitally blind humans. https://pubmed.ncbi.nlm.nih.gov/21368198/
- Do blind people hear better? Trends in Cognitive Sciences (2022). https://www.cell.com/trends/cognitive-sciences/fulltext/S1364-6613(22)00212-1
- Gori, M., Sandini, G. et al. (2014). Impairment of auditory spatial localization in congenitally blind human subjects. Brain, 137(1), 288. https://academic.oup.com/brain/article/137/1/288/361673
- Early visual deprivation severely compromises the auditory sense of space in congenitally blind children. https://pubmed.ncbi.nlm.nih.gov/27228448/
- The development of spatial perception with and without visual experience. Nature Reviews Psychology (2025). https://www.nature.com/articles/s44159-025-00516-z
Additional background on predictive processing, cited in text without links pending verification: Clark, A. (2013), Whatever Next? Predictive Brains, Situated Agents, and the Future of Cognitive Science, Behavioral and Brain Sciences; and Rao, R. P. N. & Ballard, D. H. (1999), Predictive coding in the visual cortex, Nature Neuroscience.
By Ruman Shaikh
I am a visually impaired engineer and researcher, working and writing at the intersection of disability, cognition, and computation.