从恢复视力到重新构想大脑：Max Hodak 对谈

The brain very literally, very clearly, plainly is a computer. You can solve computational problems by arranging matter in a certain way and then taking your hands off and pressing go. We talk about being a brain in a vat. That’s what the skull is. The brain is connected to the environment through a small number of wires: the cranial and spinal nerves.

These are little cables that carry your interaction with the world. If you can get the visual signal, auditory signal, balance, and motor signals in and out of the brain, that is an end in itself. That is the central object. The retinal prosthesis right now, I think, is a great proof of concept that we’re on the right track. Nobody had previously ever been able to restore a form-vision image in the mind’s eye of a blind patient in this way.

We need to add depth of grayscale. We think we can see a path to get at least red and green. There are ways that we can compound upon this path through an engineering process to make a product that’s better and better.

Hi, listeners. Welcome back to No Priors. Today, I'm here with Max Hodak, the founder and CEO of Science, formerly of Neuralink. We talk about Prima, the implant that helps people who have gone blind see again, which just got regulatory approval in Europe, their quest to sustain the human experience, and substrate independence for the brain. We also talk about alignable representations between AI models and the future of neuroscience. Welcome, Max. Max, thanks so much for doing this.

Thanks for having me.

For anyone who is not familiar with Science, can you describe a little bit about why you started the company, leaving Neuralink, and what the mission is?

1. Science Builds Vision Restorers

Fundamentally, we’re a medical device company. But I think the mission of lowercase-s science is to use a differentiated understanding of the universe to improve the human condition. That’s the mission of uppercase-S Science. That’s what we do.

We use, specifically, an understanding of how to work with the brain to get big effect sizes that you don’t get in medicine often. Our main product is a retinal prosthesis. You can think of it like a cochlear implant for the eye. Cochlear implants are some of the biggest impacts in all of medicine. If you’ve ever seen a video of a newborn turning it on for the first time, it’s striking. Our goal is to build things like that, including our Prima retinal prosthesis.

And for people who are not familiar with that, it’s a chip—

Yeah, so—

—that’s inserted with glasses.

It’s a tiny chip that’s implanted under the retina in the back of the eye for patients who have gone blind due to loss of the light-sensitive cells in the eye. Specifically, these are diseases like macular degeneration, which our clinical trial was done in. We’re about to do studies in retinitis pigmentosa, Stargardt disease, and a couple of other diseases.

It’s a chip that sits under the retina and then converts— The patient wears glasses that have a laser projector that projects an image onto the implant. That then stimulates the retina to bypass the dead rods and cones and stimulate the retina directly to get a visual signal back into the brain.

How did you go from, “We should have, like, a chip invasively in the brain,” to this particular form factor as the first premise?

2. Retina Becomes The First Bet

When we started the company, we had a couple of ideas. One of the ideas was the biohybrid neural interface direction, where instead of placing metal wires into the brain or genetically modifying the brain, what we do is graft in living neurons that grow in and form new biological connections. That’s a big research project. It’s very exciting research, but also needed to be paired with another nearer-term business.

We asked ourselves what the most valuable thing we could do was, and we thought that we could restore vision to the blind with the resources available to us and where the state of the field was in early 2021. If you want to do that, we had to start from this understanding of how the brain gets vision. What is vision in the brain?

You could look at the retina, which is obviously how vision gets into the brain—the first place it’s created. The first stop of the optic nerve into the brain is a structure called the lateral geniculate nucleus in the thalamus, so you could think, “Oh, we’ll stimulate the LGN.” The connection from there is the visual cortex. It’s like half a billion cells up at the back of the brain.

If you want to restore vision, you can think: I can go in through the retina, I can go in through the thalamus, or I can go in through V1. There are a bunch of scientific and technical reasons that lead you to think that if you have an optic nerve, you want to be in the retina. From there, you have a choice of what to stimulate. There are 2 types of cells, and there are a couple of different ways you could stimulate them. We explored all variants of that early on.

We developed an in-house gene therapy that affected the retina in one direction. We did a survey of electrical stimulators. We looked at ultrasound. What we ended up doing is developing, indigenously, a state-of-the-art gene therapy, which is probably going into humans next year, as well as finding the state of the art out there in the world of people electrically stimulating the retina.

There was a company in France called Pixium that, back in late 2022, had by far the state-of-the-art work. It was originally developed by an inventor at Stanford and then licensed to this small French company, and they were in the middle of clinical trials. We got to know them over the course of a couple of years and then were in a position to acquire them when we saw something that I think nobody else really saw at the time. That deal has turned out to be great.

Can you talk about the recent CE designation and regulatory approval you got?

3. Prima Enters The Market

It took us about 2 years after the acquisition to get it to the place where this was possible, but we just, in July, got marketing approval in Europe for Prima to start commercially selling it there. That’s a major milestone. That means it’s really commercially available. The first sales will happen in the coming weeks.

That’s amazing. I think most people think of anything in the BCI field as a moonshot project that may or may not pan out 10 years from now.

People forget that the moonshot worked. We went and left boot prints on the moon. This comparison has gotten used in Silicon Valley to mean things that have extremely long odds and are unlikely to work, and therefore we can vaporize a bunch of investor money just fine. When we went to the moon, we did it. Historically, the success rate of moonshots is higher than I think people give them credit for.

One of the most important things is having a real business here, and this is the start of that.

Can I ask how, when you were exploring different signaling pathways and form factors and just conditions to go after, you thought about the scope of the timeline, engineering cost, and risk? Were you just looking for something big enough to be useful and feasible in some period of time, or how did you think about funding the project and how long it could take?

4. Science Builds A Three Part Pipeline

There are 3 elements to our pipeline. The first is our work in vision, the second is our biohybrid neural interfaces, and the third is our work in a different area of medicine: perfusion, a program called Vessel.

These 3 things together form the minimum set of things that I think, if they’re successful in the timescale of 10 to 15 years, could really drive a significant revolution in medicine broadly. People have spent huge amounts of time and money looking for drugs to restore vision or hearing, to stop Parkinson’s, or to help paralyzed people move again.

Understanding the biology and the molecular detail required to make a drug has been very difficult. Humanity just isn’t that good at that, to be totally honest. On the other hand, the brain is a computer, and when you deal with the brain as a computer, you get these things to work. You don’t see demonstrations in medicine like a cochlear implant being turned on or a deep brain stimulator being turned on.

You can implant a quadriplegic patient’s motor cortex and have them playing video games in an hour. You just don’t really see things like this in medicine, in most drugs.

In small-molecule random walk, sifting in nature.

Small molecules especially are super hard. You can do some super-highly-engineered, patient-specific CAR-T, and instead you get an autoimmune-like giant immune overreaction. It’s like, if I put electrodes in M1, you’ll probably be using a computer in an hour.

It’s easier. It’s more amenable to biology in many ways. You can do drug discovery for a decade, run a clinical trial, and you’re going to turn over a card. The answer might be no, and then what? Everybody goes home, whereas here we have a clear sense of how to make the thing better.

The retinal prosthesis right now, I think, is a great proof of concept that we’re on the right track.

Nobody had previously been able to restore a form-vision image in the mind’s eye of a blind patient in this way. But at the same time, it’s a small field of view. It’s like looking through a straw. It’s only black and white. We need to add depth and grayscale. We think we see a path to get at least red and green. Blue is a little bit trickier. And so there are ways that we can compound upon this path through an engineering process to make a product that’s better and better.

Can you talk a little bit about what you saw in the clinical trial in terms of variation between patients, or what the ceiling was so far?

Yeah. In the clinical trial, the main thing was just the existence proof of, like, that success—

People can say—

—that was a possible outcome, right? Like, we had patients filling in Sudoku puzzles or crossword puzzles. There were patients who were reading books. I saw some of these videos, met with one of the patients, and talked to the surgeons. This is one of those things that seems too good to be true. It’s—

How do clinicians react to all of this? Would the people that you work with say at the beginning, “Yes, Max is right. The brain is a computer. This should definitely work. It should work at a higher likelihood and better rate of progression than our random walk and biological understanding”?

5. The Brain Is A Computer

Well, if you want to make people angry, you should tell the internet that the brain is a computer.

Okay.

Yeah.

We’ll start by doing that.

Yeah.

All right. Yeah.

Yeah. That kind of starts you off in a defensive place.

Why don’t people like that?

I don’t know. This is one of those things that just feels like bike-shedding to me. I don’t mean that metaphorically. The brain very literally, very clearly, plainly is a computer in my understanding of the world. I also view the universe generally as a computer. We can solve computational problems by arranging matter in a certain way and then taking your hands off and pressing go. And so the fact that it unfolds in time to solve some computational problem, I think of that as a computer. The brain’s the same thing. And I don’t think there’s necessarily a deeper—

That’s a broader definition of computer than I had before.

—a deeper point there.

Yeah.

Yeah, there’s nothing special about transistors. We understand computers in this idealized way as a Turing machine. That’s an abstract computer. It’s just that you’re going from state to state in ways that are subject to laws that mean that the transformations are interesting and meaningful.

But no, I think this was fairly contrarian, both in the sense that BCI has this broader interpretation than motor decoding, as well as the question, “Is a retinal prosthesis a BCI?” That’s also kind of a minor definitional question, but if you think that it is, then that opens up this interpretation of a lot of areas of medicine that could be accessible to it that people weren’t really thinking about.

Clearly, there was interest in looking into this. It wasn’t that contrarian. It’s a different approach, and I think we come from a different culture than a lot of the conventional biotech industry. There’s always been kind of an East Coast–West Coast divide in biotech, especially. And we are more of a tech company than a conventional biotech company.

Mm-hmm.

Our device view of a lot of historical biology problems makes us even more of a tech company by biotech standards. So we mostly raised from tech investors, not that much from biotech investors. In fact, there’s only 1 VC that I sought out at all at the Series A that I went to pitch, which was Bob Nelson, who’s a biotech investor.

When you describe different types of BCI products and missions, I think you have a really good way of explaining it that is on a spectrum. Can you talk about just the landscape of what devices and approaches people are working on in BCI today?

Yeah, I think BCI is a category, kind of like how pharma is a category. Sometimes I’ll talk to VCs, and they’re like, “Oh, we have a BCI bet.” I’m like, “Do you have a drug bet? You made 1 bet out of drugs.” That’s how you think about the category.

Mm-hmm.

Versus thinking about it in neurodegeneration and Parkinson’s, or a specific condition.

Yeah, or maybe even—

Yeah.

—different bets within neurodegenerative.

Mm-hmm.

You’ve got a degrader, and maybe you’ve got a gene therapy, and maybe you’ve got something else, because they’re—

Different hypotheses.

Yeah.

Yeah.

And similarly, on 1 end of the spectrum, you’ve got silent-speech devices that may be BCI to a greater or lesser degree. Maybe they’re recording a neural signal, like EEG. Maybe they’re using something just like radar through the face, which I know is an idea out there. But these are all basically hand substitutes.

On the 1 hand, hands are great. 2 hands are great. They work really well. Or I’ll talk to teams that say, “It’d be really nice if, to go to your next thing, you didn’t have to open the Uber app and call an Uber. You just thought of it, and it was there.” I’m like, “You probably want to communicate really unambiguously with the Uber app. It’d be pretty annoying if you started spontaneously getting notifications during meetings that said, ‘Oh, it thought that you were thinking about an Uber; therefore, it decided to summon 2 for you.’”

And so you’ll probably want these to be pretty explicit. And to the degree that that is a volitional intent, you already don’t need to do a lot to get your hands to do things. Now, could you have extra hands? Extra hands: famously useful. And so having some easier way to communicate might be useful. That is kind of outside the scope of things I spend a lot of time thinking about.

Because if you get vision, hearing, balance, and 1 kilobit per second of motor control, you’re halfway to The Matrix. And this takes you into some really trippy reinterpretations of medicine, and that’s the stuff that we work on. I think other people will do things like speech-to-text and AI communication.

There is a distinction. Let me come back to your broader question in a second. But there’s some point where you go from communicating with a thing to redrawing the border around your brain. And we don’t have a great sense of exactly where that transition is yet, but there’s a sense that there is one.

The way that you use the 2 hemispheres of your brain as 1 integrated, bound thing is different than the way that you talk to another person. And it’s not just that there’s correlation, because all communication is about creating correlations between brains when we speak. There are big correlations that are being driven between our brains because all communication is premised on that.

If we didn’t pre-share a language or some common education, some sense of math, then we wouldn’t be able to communicate those concepts. Because there’s something that’s lit up in my brain. I can serialize that to language and send that to you. That lights up the same pre-shared concept spaces.

And so there’s 1 mode where you’ve pre-shared some structure between the 2 brains, whether this is an AI model or a biological brain, and then you’re communicating over that channel. The other is that you’ve added some new structural capability. I think figuring out where that transition happens is a really compelling area of research for us.

What are you most personally interested in exploring in terms of that boundary yourself?

Yeah. Well, that is a—what is “you”? That’s a really central question here. If the end of the artificial intelligence quest is—

I don’t really care. What if I just want my brain to be a better computer or a richer one—

Well, I think—

—in terms of understanding other people’s experiences?

There is an important question here. If I just scanned your brain into a computer, and there was a software simulation of you, does that count as you? Would that make you feel better about dying of cancer? If you were diagnosed with lung cancer and said, “Okay, well, we’ll scan you into a computer.” So imagine that we do it non-destructively. You’ll still be there.

Mm-hmm.

But then you’re talking to the software replica of you.

Mm-hmm.

And then you’re like, “Okay, I’m going to go to hospice, but this thing will keep doing my venture-investing job.”

Mm-hmm.

Does that make you feel better, that much?

Well, I think on this question—have you ever been under general anesthesia?

Yeah.

Yeah. Well, I—

And that produces a break there, and that's the type of thing that you have to explain: why does that feel different?

Mm-hmm.

Because I think it does feel different. I think that people are reticent to undergo general anesthesia, but they do it and they survive, and they realize it's fine. And then, if I could make a copy of you and you can talk to that copy, and you're like, "Okay, I will go away now," I just don't think that many people are going to be like, "This is it." And so you have to answer why it's different.

There's an asymmetry in the... Some of the operators that actually change things in physics are a creation or an annihilation operator, and we get these in life, right?

Mm-hmm.

You can create a new life or a new mind or a new soul. And then there are times when they can be annihilated. They can get destroyed. And then there are ways that they kind of change while intact.

Do you study consciousness at Science in a sequential way, or directly and explicitly today, when you talk about the operators that are part of it, let's say?

Your conscious moment is—you’re experiencing a bunch of things in parallel. You're seeing things, and you're hearing things, and you're feeling things, and you're smelling things, and these things happen just simultaneously together. But they're kind of different elements of the experience, and we want to understand how the brain constructs each of those, and how it causes them to be perceived together to the exclusion of other things.

You have your vision and your hearing. You never get my vision and your hearing. You might think, "That sounds really obvious. It's in my brain, it's not in your brain." But we need some more fundamental explanation for how that partitioning really happens.

Okay, so you think that's a foundational component?

Yeah. I'm in the camp that continuity is greatly important. People will accept significant drift in their identity over time as long as they have continuity.

But if you preserve the sense of identity—you have a software simulation that answers exactly like you would now, but it's not phenomenally continuous—that is less satisfying.

Yeah, that's an interesting trade. I think I would take dramatic morph, but continuous experience.

Yeah.

I don't know if I'd take significantly degraded IQ. Laura Deming asked me this.

Life with provable characteristics is a thing we've never seen before, and it might be transient. You'd probably accept degraded IQ for some period of time if it then got backfilled some number of weeks later, and then you got some... I mean, at that point, where you achieve substrate independence, you can really take that almost anywhere you want.

Which is why that's really, really interesting. One of the big missing pieces here is connectomics. That is getting pretty close, I think, to the point where the project could be done. We're still relatively far from a human connectome, but I think we're not that far from a mouse connectome. That would be enormously useful for facilitating this research and understanding how all this works.

We need to understand even really basic questions like, "What is the overall architecture of the brain?" We have some answer for that, but I don't know that it's a really, really detailed one at this point.

You are of the view that it makes sense that there's this increased interest, this surge of investor interest and founder and engineer interest in BCI as a field, given the progress of AI model research, because the representations actually should be shared—or they empirically seem to be.

Yeah, I mean, this is the idea called the Platonic Representation Hypothesis, and this is really interesting. It is controversial in the community, but from where I sit, there's clearly something real happening here.

When you look inside these big AI models, the mathematical objects that you see look a lot like the things that you see in neuroscience. If you look at how these AI models represent concepts, then you look at the parts of the brain that represent concepts, you see very similar geometry.

That, to me, was one of the first clues. When I really saw that, we used that practically. We use that constructively at Science. We know that that is true because we can get alignments between animal-brain neural recordings and AI model internal representations.

That was a big clue to me that AI was on the right track, and this was not a gimmick and not hitting a wall. There was something deeper that's true here. There's some fact about the universe where, as these things are learning, they're grabbing onto some true underlying data manifold. It feels like a law of physics. If you apply enough compute to matter, you get this thing that looks like intelligence.

Why do you think that's controversial, or why is it controversial in the field?

There's some faction of people that don't want this to be true for reasons that are not totally clear to me. It is also not clear—we don't fully understand whatever phenomenon is happening here. It's unclear if the structure is global or if it's local in some sense.

You can recover relational structures between ideas, but it might be that this works locally. Where disconnected things might be placed might... This gets fairly technical quickly, but there's a bunch of stuff that we just don't know, and I think this creates space for people to wonder, "Is this giving us as fundamental of a hint as it might seem?" I think that it is.

What do you think are the most fertile ways to study neuroscience today if your set of beliefs is true?

Yeah. Well, ironically, it's probably working on AI.

Yeah.

I have a couple of neuroscience friends at OpenAI and Anthropic. We would joke, "Oh, you left neuroscience." Like, no, no, no. It is just way easier to do neuroscience on the models. But the degree to which it is neuroscience is fascinating.

I want to talk a little bit about the future, and maybe the very short and then the medium timescale for Science, short of changing the boundary of who we are. What does it look like to commercialize the first program for you? You said that there need to be $100 million run-rate businesses in this field. How do you get there?

6. Science Turns Vision Into A Business

Yeah, well, becoming profitable, or at least having the ability to do this forever, is a super-high priority. Restoring vision to the blind is pretty good business if you can actually do that.

Especially since almost everyone has the problem as an age-related problem.

Yeah, so AMD—it's like 1 in 2 have some early stage by 80. It's like 1 in 10 at 85 that actually have it. It is definitely a major issue, and not just vision, but these topics in general affect everybody.

We don't have firm pricing yet. This is a thing that we are being a little cautious about how we talk about publicly because we aren't totally sure yet. But the precedents for vision are all—I mean, I think we could say they're expensive.

Second Sight, 10 years ago... There was a company about a decade ago that had a retinal prosthesis that worked differently than ours does. It did not get the type of performance that Prima does, but was briefly approved because, again, there's really nothing for these patients. There's always been a lot of enthusiasm for anything that could possibly help them.

They didn't get what we call form vision. They didn't get a coherent face or paragraph that your eyes could scan over. They got these flashes of light that patients could look at and kind of think about assembling into what they meant. They got paid about $150,000 per patient in the mid-2010s.

There's a gene therapy that works, that is only relevant in the first place for about 5% of patients in one narrow indication, and it really doesn't work that well. It gets an 0.1 line to improvement. It kind of slows the rate of degeneration for some patients. That reimburses at almost half a million dollars per eye.

Some of this is a function of just how expensive it is to develop these therapies and how high the failure rate has been historically. And then some of it is that vision's a very dominant sense for us. If you lose that, that's totally debilitating, and restoring it is very important, even just minimal vision.

So the TAM will grow over time. For this first version, it's on the scale of hundreds of thousands of patients in the US and Europe. For the current version of Freedom, it's probably hundreds of thousands of patients, and then the next version, which is going into animal studies now, will be in humans hopefully next year, should expand that to millions.

You said something that surprised me—an intermediate point between vision and fully understanding consciousness. How does oncology—or how do other indications—fit into the picture in terms of what you might work on?

I mean, the thing that makes you you—the only organ that you can’t even in principle transplant is the brain. The heart, the pancreas, the liver, and the lungs, as far as I’m concerned, are really support characters. They’re there to keep the brain activity interesting and going.

I think we’re gonna get to a point where, because the biology is so difficult—you’ve got this alien nanotechnology that is around us, completely surrounding us, that we are completely dependent on, that we understand still very poorly—instead of needing to solve that, there are ways where we can accomplish the same fundamental goals using a toolbox that humanity is much more advanced in. I’m gonna be ultimately fairly disappointed if I’m murdered by my pancreas, and I think that’s the worldview: the thing that matters is the brain.

The brain is the computer that gives us this. You could talk about being a brain in a vat or these upload thought experiments, but you already are. That’s what the skull is. The brain is connected to the environment through a small number of wires: the cranial and spinal nerves. The optic nerve is cranial nerve II. The vestibulocochlear nerve, which carries hearing and balance, is cranial nerve VIII. You’ve got these little cables that carry your interaction with the world.

That world is generated by the brain. If you can get visual signals, auditory signals, balance, somatosensory and motor signals in and out of the brain, that is an end in itself. That is the central object.

Okay.

Through a mix of the BCIs that allow you to change the thing that it’s interacting with and our perfusion medicine program, we think that there are ways to significantly improve not just lifespan but healthspan and create a better quality of life for many patients in ways that I think will feel like a lateral move rather than just solving many of the things that people have seen on the horizon.

For people who are interested in working at or investing in science, if you are successful, what will the change to human experience be 20 years from now, besides you not worrying about your pancreas as much?

Yeah, I mean, that’s it. There’s a fragility that we all live with. There’s this jeopardy that we all live under as part of the human condition, and I think that if we’re successful, what will happen is that sense of jeopardy will fade. We will just become much less fragile.

We will have the ability to upgrade and replace parts of ourselves. Neurodegeneration, we don’t know about that one. That one still seems difficult. That still needs real investment.

The 2 leading causes of death, though, are cardiovascular disease and cancer that metastasizes to the brain, and I think both of those are gonna be really attackable through this type of work. At the other extreme, if we are serious about exploring the universe and going to the stars, we’re gonna have to adapt ourselves to that environment. We’re not gonna export Earth with us everywhere we go.

These bodies are great, but they’re designed for this planet. Adapting ourselves to the hard vacuum of space is definitely going to be the thing that we wanna do in the long run. Ultimately, those are the same project.

Being able to preserve yourself and being able to adapt yourself—

Swappable parts—

Or update—

And substrate independence, yeah.

Yeah. Substrate independence, I’ll use that phrase.

The simplest premise for a company in the BCI domain today is that you can, in some way, invasively or noninvasively, talk to an AI model in a high-bandwidth way. That is not your focus of interest. Why?

Yeah. Well, first of all, I think that talking or writing is thinking. I think this idea that there’s stuff that’s just preformed in your brain, that if you could access it through a BCI, it would be faster, is probably not the case.

You don’t think there’s some special latent state that’s not language?

No, I think it’ll feel like it’s fully formed, but until you really sit down and try to write it out, it isn’t really, and I think that feeling is misleading.

There’s this famous 10-bits-per-second cognitive bottleneck. There’s this observation that the brain seems to process information. There are a bunch of ways you can triangulate this. You can put somebody with a perfect memory on a helicopter ride over Manhattan and ask them to draw what they saw, and then you look at all the details. It works to about 10 bits per second over the course of an hour or two. There are a bunch of different independent lines of evidence for this.

There’s this deeply evolved cognitive bottleneck at about that, and I think that this kind of rolls up through language. But even so, even if you take that, it probably would be nice to be able to walk down the street with a cap on and ask questions to my AI through monologue. That might be possible. There’s probably some combination of EEG and MEG that might be capable of this.

That is still just a different type of product. That is not the thing we are trying to do. Brain keyboard might be valuable. It might turn out to be like AR glasses, where our attention was already fully 100% occupied, and putting it on the face didn’t really change that. We were already consuming all of the time.

But at the other end of that spectrum are things like generating vision, generating hearing, or achieving substrate independence. Those are the things that we are focused on, not a brain keyboard. Both of these are potentially BCI problems or products, but very different types of companies will build them, just as I think you have a huge range of drug companies.

Find us on Twitter at nopriorspod. Subscribe to our YouTube channel if you wanna see our faces. Follow the show on Apple Podcasts, Spotify, or wherever you listen. That way you get a new episode every week. And sign up for emails or find transcripts for every episode at no-priors.com.