The Skull Is Already the Vat: Max Hodak on Restoring Sight

Patients who went blind from macular degeneration put on glasses with a laser projector, read books, and complete Sudoku puzzles. The hardware that makes that possible is a chip implanted under the retina. In July, that implant — Prima — received European marketing approval, and Hodak says first commercial sales would start in the coming weeks. The person who built the company around it is Max Hodak, a Neuralink co-founder who left in 2021 to start Science. The name is a pun: lowercase-s science is the mission of using a differentiated understanding of the universe to improve the human condition; uppercase-S Science is the company that does that work.

Science is, at the root, a medical-device company. The pipeline has three lines that Hodak calls a minimum set: if they succeed on a ten-to-fifteen-year clock, they could drive a real revolution in medicine. The first is vision — Prima. The second is a bio-hybrid neural interface: instead of metal wires or genetic modification, engraft living neurons that grow in and form new biological connections. The third is organ perfusion, a program called Vessel. Big research still needs a near-term business, so in early 2021 they asked what the most valuable thing they could do with available resources was. The answer was restore vision to the blind.

Vision enters at the retina, hits the lateral geniculate nucleus in the thalamus, then visual cortex V1 — about half a billion cells at the back of the brain. You could stimulate any of those stations. A bunch of scientific and technical reasons say that if there is an optic nerve, you want to be in the retina. They explored the variants: an in-house gene therapy, now probably going to humans next year; a survey of electrical stimulators; even ultrasound. The state of the art in retinal electrical stimulation sat at Pixium, a small French company that had licensed work originally developed by an inventor at Stanford and was mid-trial in late 2022. Science spent a couple of years getting to know them, then acquired the company when Hodak saw something he thinks nobody else really saw. The deal, he says, has turned out to be great. Getting from acquisition to the CE mark took about two more years.

People treat BCI as a moonshot that may or may not pan out in ten years. Hodak's retort is that people forget the moonshot worked: there are boot prints on the moon. Silicon Valley has turned the word into a synonym for long odds and vaporized investor money. Historically, he argues, the success rate of moonshots is higher than that usage implies. Prima is meant to be a real business, not a slogan. Mechanically it is a cochlear implant for the eye: the chip sits under the retina; glasses project an image onto it; stimulation bypasses dead rods and cones and puts a visual signal back into the brain. The trial was in macular degeneration. Next come retinitis pigmentosa, Stargardt, and a couple of other diseases.

The whole episode sits on a sentence Hodak will not soften. The brain very literally, very clearly, plainly is a computer. You can solve computational problems by arranging matter in a certain way, taking your hands off, and pressing go; it unfolds in time and solves the problem. He applies the same definition to the universe. There is nothing special about transistors; a Turing machine is an abstract computer, and the brain is the same kind of thing. Tell the internet that and people get angry. He files the argument under bike-shedding: a fight over definitions that dodges the work.

That view decides how the company is built. Humanity has spent huge amounts of time and money looking for drugs to restore vision, restore hearing, stop Parkinson's, or help paralyzed people move. Understanding biology in the molecular detail a drug requires has been very difficult. Humanity, he says, just is not that good at it. Small molecules are a random walk through nature. Even a highly engineered, patient-specific CAR-T can produce a giant immune overreaction. Devices are a different rhythm. Put electrodes in motor cortex M1 and a quadriplegic patient can be using a computer, even playing video games, in about an hour. Turn on a cochlear implant or a deep brain stimulator and the effect is immediate in a way almost no drug matches. Cochlear implants are among the biggest impacts in all of medicine; videos of a newborn turning one on for the first time are striking, and Prima is aiming at that class of result.

Do drug discovery for a decade, run a trial, turn over a card, and the answer might be no — then everybody goes home. With a device, each generation has a clear engineering direction, and the product can compound the way chips do. Working with the brain as a computer is how you get effect sizes medicine rarely sees. Science has raised mostly from tech investors, not biotech. There is an East Coast / West Coast split in that industry; they are more of a tech company that happens to be attacking historical biology problems. The only biotech VC Hodak sought out at Series A was Bob Nelson.

The trial's most important output was an existence proof. Patients filled in Sudoku and crossword puzzles. Patients read books. Hodak watched the videos, met a patient, talked to the surgeons. Nobody had previously restored form vision in a blind patient's mind's eye this way — a coherent face, a paragraph the eyes could scan. The current Prima is far from perfect. The field of view is small, like looking through a straw. It is only black and white. They need depth of grayscale. He thinks they can see a path to at least red and green; blue is trickier. The next device is going into animal studies now and, he hopes, humans next year. That is the compounding path: an engineering process that makes the product better and better.

The market is larger than a rare-disease story. AMD is age-related: about one in two people have some early-stage disease by 80, about one in ten at 85 actually have it. The current Prima is on the scale of hundreds of thousands of patients in the US and Europe; the next version should expand that to millions. Pricing is not public yet, and the team is cautious. Precedents are expensive. Second Sight, about a decade ago, had a different retinal prosthesis that never delivered form vision — flashes of light patients had to assemble into meaning — and still charged about $150,000 per patient in the mid-2010s, because there was almost nothing else. A gene therapy relevant to about 5 percent of patients in one narrow indication gets a 0.1-line improvement, mainly slows degeneration for some, and reimburses at almost half a million dollars per eye. Some of that price is the historical failure rate of development. Guo notes Hodak has said the field needs hundred-million-dollar run-rate companies. Becoming profitable, or at least able to do this forever, is a super-high priority. Restoring vision is a good business if you can actually do it: it is the dominant sense, losing it is debilitating, even minimal restoration is valuable, and almost everyone hits the problem with age.

BCI, Hodak says, is a category the way pharma is a category. When a VC says they have a BCI bet, he asks whether they would say they have a drug bet. Parkinson's is one hypothesis, neurodegeneration another, gene therapy and degraders still others. There is a definitional question inside that: is a retinal prosthesis a BCI? If yes, a lot of medicine comes into range that people were not thinking about. One end of the spectrum is silent-speech devices, some reading EEG, some even proposing radar through the face — basically hand substitutes. Hands already work well. Teams pitch thought-summoned Ubers; he would rather communicate unambiguously with the app than have it decide during a meeting that he was thinking about a ride. Extra hands are famously useful. That is still not where he spends time.

The other end is the one he works on. Get vision, hearing, balance, and a kilobit per second of motor control, and you are halfway to the Matrix. There is a distinction between communicating with a thing and redrawing the border around your brain. The two hemispheres operate as one bound system in a way talking to another person does not. All communication creates correlations: something lights up in his brain, serializes to language, and lights up a pre-shared concept space in yours. One mode is communicating over pre-shared structure, including with an AI model. The other is adding a new structural capability. He does not know exactly where the transition is. He thinks finding it is one of the most compelling research questions they have. A brain keyboard for walking-down-the-street monologues with an AI might be technically possible with some mix of EEG and MEG. He files it with AR glasses: attention was already fully occupied, and putting it on the face did not add a new increment. Speech-to-text and AI communication he will leave to other companies. Science does generating vision, generating hearing, and a larger word: substrate independence.

Hodak offers a crossover result that the AI community argues about under the name platonic representation hypothesis. Look inside large models at how they represent concepts, then look at the parts of the brain that represent concepts, and the geometry looks a lot alike. It is controversial; a faction does not want it to be true, for reasons he says are not totally clear. From where he sits, something real is happening. Science uses it constructively: they can get alignments between animal-brain neural recordings and AI models' internal representations. That was a clue that AI was on the right track, not a gimmick and not hitting a wall — that systems grabbing onto a true underlying data manifold is a fact about the universe. It feels, he says, like a law of physics: apply enough compute to matter and you get something that looks like intelligence.

Unknowns remain. Is the structure global or local? You can recover relational structure between ideas, but disconnected pieces might be placed in ways nobody understands yet. Those gaps create space for people to wonder whether the hint is as fundamental as it looks. He thinks it is. The ironic conclusion: if his beliefs are true, the most fertile way to study neuroscience today is probably to work on AI. He has neuroscience friends at OpenAI and Anthropic who joke that they left the field. Their line is no — it is just way easier to do neuroscience on the models.

A big missing piece is connectomics, mapping every connection in a nervous system. A human connectome is still relatively far. A mouse connectome is getting close, and that would be enormously useful. Even basic questions — what is the overall architecture of the brain — have answers that are not yet really detailed. He also holds a claim that will annoy people who want a neural shortcut around language: talking or writing is thinking. The sense that something is already fully formed in the head, waiting for a BCI to siphon it out faster, is misleading. Sit down to write and you find it was not formed. A famous cognitive bottleneck triangulates to about 10 bits per second — one demonstration is a person with perfect memory on a helicopter over Manhattan, asked to draw what they saw. Independent lines of evidence point at the same deeply evolved limit. That is why Science is not building a brain keyboard.

The thought experiment is unnecessary, Hodak says. That is what the skull is. The brain connects to the environment through a small number of wires, the cranial and spinal nerves. The optic nerve is cranial nerve two; the vestibulocochlear nerve that carries hearing and balance is nerve eight. Those little cables carry a person's entire interaction with the world, and that world is generated by the brain. So if you can get the visual, auditory, balance, motor, and somatosensory signals in and out, that is an end in itself. That is the central object. Organs get re-ranked under that view. The brain is the only organ you cannot even in principle transplant. Heart, pancreas, liver, lungs are support characters, there to keep brain activity interesting and going. Biology is alien nanotechnology we are completely dependent on and still understand poorly. He would rather hit the same goals with a toolbox humanity is much better at. "I'm going to be ultimately fairly disappointed if I'm murdered by my pancreas."

Science also studies consciousness directly. A conscious moment is parallel: seeing, hearing, feeling, smelling at once, as different elements of one experience. They want to know how the brain constructs each element, how it binds them together, and how it excludes everything else. You get your vision and your hearing. You never get his vision and your hearing. It sounds obvious because it is in your brain, not his. He wants a more fundamental explanation of how that partitioning happens. Guo asks the upload question: non-destructively scan someone into a computer, leave the person in hospice, let the software replica keep doing the venture job — does that help? Hodak does not think most people will feel that the replica is them. General anesthesia already produces a break; people are reticent, then they do it and realize it is fine. A copy you can talk to, then walk away from, is a different operator. Physics has creation and annihilation; minds can be created, destroyed, or changed while intact. The hard question is which changes still count as the same person.

He is in the continuity camp. People will accept significant drift in identity over time if experience is continuous. A software simulation that answers exactly as you would now, but is not phenomenally continuous, is less satisfying. Guo would take a dramatic morph with continuous experience; she is not sure she would take a significantly degraded IQ. Longevity investor Laura Deming has asked him the same thing. Life with provable characteristics is something humans have never seen. It might be transient: you would probably accept a few weeks of degraded IQ if it were backfilled later. Once you have substrate independence, you can take that almost anywhere. He is not that interested in where the AI race ends. He wants a better computer for his own brain, a richer one for understanding other people's experiences.

Success in twenty years, if Science works, is that the sense of jeopardy hanging over the human condition fades. People become much less fragile. They can upgrade and replace parts of themselves. Neurodegeneration still looks hard and still needs real investment. The two leading causes of death — cardiovascular disease and cancer that has not metastasized to the brain — he thinks become attackable through this kind of work. At the other extreme, if humans are serious about the stars, they will have to adapt themselves; they will not export Earth everywhere, and these bodies were designed for this planet. Preserving yourself and adapting are the same project: swappable parts plus substrate independence. Prima is the first existence proof that the track is right. Vessel, the bio-hybrid interface, and the rest of the line sit on that same vector.