B (3:02)

I'm taken by to what degree? Right now it's about sort of taking someone who has a condition or a disease and then bringing them, sort of restoring them to like sort of capability.

A (3:16)

Right.

B (3:17)

I think that's playing out in AI right now as well. Right. Like you had computers that had no ability to do like any sort of pure cognition or like, you know, and, you know, no neurons and then suddenly a bunch of neurons and then AGI is sort of like what a human can do. It's sort of like a restoration of capability. And then of course, there's like this other thing after that, which is, you know, ASI superintelligence. Do you ever think about what that might be down the road? You know, what is that for BCI?

A (3:48)

There are many types of BCIs. So it really is going to be a category like pharma. It's not one product. I don't think there's going to be like the BCI that people get. And there are different modalities that work for different things. So, for example, I don't work on ultrasound, but one of the things I think will be possible with ultrasound is like a digital Ambien or like a digital Adderall. So can you like stimulate part of the brain to cause focus or sleep and things like that would not surprise me if that was possible and that I could see as being more of a consumer application almost, and that won't require brain surgery. Hopefully right now that the high quality ultrasound stuff does require drilling through the skull, but I think that that will be overcome for the implantable BCIs. I mean, this is a very serious brain surgery. I think that's important to appreciate. So when you think about how do you actually get this into humans and who's going to use it, I mean, these are going to be very disabled patient populations. You always look at risk reward. You start at the most disabled patients, you get the most benefit for even relatively basic functional. Like I don't think that you or I would want to get one of the cortical motor decoders that you might have seen out there today. Because the reality is that like a keyboard and mouse is like great. It is a much higher performance. You can get like spoken word is like 40 bits per second. Many people can type in the like 20, 20ish bits. And so the 10 bit per second cortical motor decode is like not going to make your life better. I wouldn't get serious brain surgery for that. Now as it gets more powerful and as we are able to produce kind of access richer representations from more of the brain, especially bidirectionally, then you'll start to see the risk benefit change. Where my view on this is not that I think healthy 30 year olds are going to be getting these soon, but eventually many people become patients. Aging is the correlate of everything getting worse. And so there's some critical age where it crosses over, where it makes sense to have something that will restore some functionality that you had and then eventually that will kind of cross the origin and then you'll see people that have had something terrible happen to them who now have a capability that you're jealous of. And that will be kind of when you start to see it changing.

B (5:51)

Talk to me about how people who maybe never had sight, you know, why was the optic nerve not actually set up? Like, is that not something that you can do later? How does plasticity fit in? You know, do you have to get BCIs when you're incredibly young while the brain is still plastic? Like, how does all this come together?

A (6:10)

Neuroplasticity is really interesting and really misunderstood. There are genuine critical periods in early development that if you miss them, there are some things that will be very hard to wire up later. There actually are some cases of patients that were born blind, but it wasn't a loss of the optic nerve, it wasn't something in the brain. But they had congenital cataracts, so their vision was blurry from birth and they were never able to really form images. Who then had this fixed as adults and that did not work. Their brain could not make sense of the information. It was totally overwhelming. They would wear eye patches. Several of them committed suicide. And so there are clear critical periods, periods in early development where if you miss that, some things are not going to work. With that said, the brain stays way more plastic throughout life, in adulthood, than I think is widely appreciated.

B (6:58)

That's a relief.

A (6:59)

Yeah. If I put an electrode almost anywhere in your brain and then wake you up during surgery, and I show you a flashing light that flashes proportionally to how much that neuron is firing, at least almost anywhere in cortex, within a couple minutes, you can learn to control that neuron. And so the brain is very plastic under feedback. And this is partly how the cortical motor decoders work. Some of it is you're decoding what the brain was originally representing in terms of like a hand or an arm representation. But also just if you're getting these signals out of the brain and you're giving the patient feedback for what those signals are doing, then the brain also adapts to you. And so in the first experiments for this, they actually didn't fit anything at all. They just took a couple. They took two neurons or a handful of neurons and fixed the weights. So it said, when this neuron fires more, we're gonna go up the screen. When this neuron fire hours more, we're going to go down the screen and sideways. They fix the weights and let the brain figure it out. Let the brain learn. And again, the brain is very plastic under feedback and can do this.

B (7:54)

What a powerful moment. You have a learn. You know, we have two learning systems that can learn off of one another instead of sort of a fixed one with if statements on this side.

A (8:03)

Totally.

B (8:03)

Yeah.

A (8:03)

And the brain really, like, if you give the cortex information, it is really good at extracting the meaning. Now, in adulthood, I think one of the reasons that you don't see it as being so plastic is because it has already fit well to reality. And so there's like, if you think of it as this energy surface, and the state of brain states is this, like you've got these hills and valleys. So during normal development, typically for most people, there's this enormous basin in this energy surface. And so for most people during development, you descend into this basin and then you're down there and it's stable because you've fit to reality. And if I show you weird movies, it's not going to really push you out of that. You can. I think one of the theories of what psychedelics do is they kind of add, kind of anneal it, so it kind of shrinks the surface A little bit. So you kind of access these other states, but then when it wears off, you just immediately descend back down into the energy well that the brain had fit to. And so even though the brain is still plastic, it is in this stable part of the attractor system, so that you don't see the plasticity as much.

B (9:04)

But this was selected for.

A (9:05)

And this was absolutely selected for. Yeah. And so there's this tension between. There absolutely is ongoing plasticity. If there wasn't plasticity, you couldn't learn things. And so, like, your ability to learn new stuff is like. And have memory, like all memory is brain plasticity in many ways. And so we are constantly experiencing very dramatic plasticity. But there are also clear limits to it, especially in how, like, the modules of the different brain areas end up interconnected past these critical periods.

B (9:28)

I have like a million questions, honestly. I mean, one of the things that I'm super curious about is like, well, what is the qualia of the person who has prima? And what is, you know, I'd be curious, like, with the biohybrid approach, like, what does it feel like? And, you know, is it like having a second screen? Like, you know, is there an input or output? I'm very curious.

A (9:49)

Yeah. So for prima, actually, on the topic of plasticity, in the time that the patients are blind, the brain wants to see. Like, again, you. The thing you experience is this world model constructed by the brain, and that is this. Is this generative model that is conjuring your reality. And so when it's not getting input from the. From the optic nerve, it is still trying to see things. So it kind of turns up the noise. And so blind patients often report hallucinations and these internally generated percepts. When you first turn on the implant in these patients, you hit it with the laser, they'll say, oh, I see a flash. But then you can do a thing where you'll turn on the laser, they'll see a flash, and you'll play a tone. And you do this a couple times, and then you don't turn on the laser, but you play the tone. And they're like, I see the flash. And so for the first couple hours of rehab, they kind of just have to learn to dissociate the real percepts from the phantom percepts because the brain is so turned. Turned up the gain. Turned down the noise floor that just learning how to discriminate real information coming in from the optic nerve takes a little bit of rehab. The qualia of prima is normal sight. It's black and white, it's a small field of view, but it's vision. The deeper question is what is the qualia of a brain to brain of an ultra high bandwidth a bio hybrid neural interface? And that is just impossible to imagine. Like those devices will get built and we're going to find out. But there are some natural case studies. So there's a pair of conjoined twins in Canada that it's really like one head with four hemispheres. And what's really interesting is that the two hemispheres of each of the twins brains are connected normally, but they're not connected with each other except for this one cable connecting the thalami, like from the thalamus to thalamus. There's this big biological cable that you can see on an mri. And over this they can share meaningful elements of their conscious experience. And one of the open questions that hasn't really been studied in the depth that I would love to see it is when they can see to some degree through each other's eyes. But does this show up as new visual field? How do those get experienced directly? Most people have two image modes. You've got your eye open vision, but you also have imagination. Some people are aphantasic and they don't have internal imagery. Most people have two image modes. Do they have three image modes or four image modes or if they have internal monologue, they seem to each individually have internal monologue. But they also can clearly communicate over this channel because they've done tasks where they can coordinate without saying anything to do stuff. And they're conscious of it and they're conscious of it. And also they don't confuse it for each other. It's not with a schizophrenic where it's like, oh, I'm hearing voices and they're coming from internally generated me. It's misattributed monologue that doesn't happen to them. They can tell it apart, but they're experiencing it directly in some way. And so there's a question of is this like, when you look at that cable, are they sending the information in the classical way or is there an effective phenomenal binding happening over this cable where it's more like the two hemispheres of your brain that are bound together into one moment. And so there's these natural case studies that tell us that some really interesting things might be possible here. But it's kind of tough to imagine what it would feel like.

B (13:07)

Paint the picture for us. You're here, everything goes really, really well. Where are we in five to 10 years with this technology?

A (13:15)

I mean I do think that you can get to close to native acuity. So kind of like your normal 2020 vision. We're definitely not there yet, but I see a path to get there and be able to get color and fill in a lot of the field of view to be clear that it's not where we are right now, but in the next 10 years I think that that's possible. But beyond that, I'd say that our worldview or my worldview as a motivating idea behind the company is you can contrast this like there's like a drug discovery approach to medicine versus a neural engineering approach to medicine. I think this is much broader than the retinal prosthesis. We started with that because it's a huge unmet need and I think it's the most valuable BCI product on the horizon that I thought was doable. Now humanity just isn't very good at drug discovery. Every now and then you kind of find a thing. It's amazing. Like you find a GLP one or you find like there's every like there's a handful of drugs that are, that we were lucky to find. But it's much more common that you spend a decade going down this path and then at the end you run a study and the answer is no. And then it's like where do you go from there? There's been a huge amount of work that's gone into finding drugs to stop blindness getting worse or to reverse and restore vision to basically no effect. There's a million dollar per patient gene therapy that has really very marginal, like if any benefit to a very small percentage of patients in the first place. And with our retinal prosthesis that what we saw trial was we can take a patient who's been unable to see faces for a decade and allow them to read every letter on an eye chart. And so not only is the brain the only organ that really in some deep sense matters, we are also just empirically much better at engineering it. And so I think this allows a really fundamental reframing of medicine. And over the next decade, I think beyond you will need people see, hear, have balance, have a kilobit per second of motor control. That is like you and I think we have cochlear implants, we know how to do motor decoding. The thing we didn't know how to do is restore vision. We are making real progress on that. I think all of this adds up to something that speaks I think to the really foundations this paradigm shift in

B (15:14)

what's possible in healthcare, something like this. I remember reading about maybe 10, maybe even 20 years ago, they were able to stimulate the optic nerve with electricity directly, but it was very, very low resolution. And it was so invasive that it could probably only be done in a clinical setting or in a surgical setting.

A (15:31)

It's relatively easy to get flashes of light to cause a patient to kind of see these, these flashes. We call these phosphenes. There was a company a decade ago called Second Sight that had an electrical stimulator that was implanted in the eye. It was a four and a half hour surgery with a titanium box on the side of the eye. It stimulated a different layer of cells than we do. And they were able to get these flashes where like if a patient looked at it, they could say like, oh, there's some flashes here, there's some flashes here. It's connected, that's an A. And then we show the next letter and it's like, there's some here, there's some here, it's an H. But the brain doesn't assemble together these flashes of light into like a gestalt hole. That is an image in the mind's eye. Similarly, when you stimulate cortex, like the back of the head, where the visual cortical areas are, you can get these flashes of light and you can even in some cases get a lot of them. But again, the brain doesn't. It's kind of this more psychedelic effect. Like this doesn't get assembled together into form vision. And as far as I know, our clinical trial was the first time ever that formed vision had been created a coherent image in the mind's eye of a person.

B (16:32)

Is there something specific about macular degeneration that causes this to be possible for this set of patients?

A (16:39)

So there's a bunch of reasons why people lose rods and cones. There's macular degeneration, there's retinitis pigmentosa, there's some rare inherited diseases like Stargardt's disease, Diabetic retinopathy can do it, age related macular degeneration, and it's the most common. So this globally affects 200 million people. The severe form geographic atrophy is a million to a couple million. In that sense, it's a big need. One of the nice things about our device is that we're somewhat agnostic to the reason that you lost the photoreceptors. And so we think it'll also work for retinitis pigmentosa, for Stargardt's for these other indications. We're actually just about to start a new clinical trial on uninherited retinal disease, which affects much younger people. And again, this goes back to the drug discovery versus neural engineering view of the world. If you want to make a drug, then you care a lot about exactly what molecularly went wrong in the rod or cone. And that is different by disease. Then even if you figure this out, it's really hard to understand what to do about it. Here we don't really care why the rods are cone. We just care that we can get the visual signal back into the computer.

B (17:45)

I guess I'm just very fascinated by, you know, obviously as a computer scientist, we spend a lot of time thinking about inputs and signals. And then what I'm hearing is that like some of that thinking does actually translate into, from software, into wetware.

A (18:00)

Well, I mean the brain is a computer and it's going to. Saying that is going to get me yelled at by some corner of the field. But I think like, I think that you can take that like almost literally. It's a, it's a very different architecture than like a, like a von Neumann architecture electrical computer, but it processes information. It gets information down 1 of 12 cranial nerves, or 31 spinal. So all of the information that flows in or out of the brain goes through a small number of cables. The optic nerve we call cranial nerve 2, the vestibulocochlear nerve that carries hearing balance, cranial nerve 8. There's 31 spinal nerves that carry commands out to the muscles and sensory information into the brain. And you can think of that as like the API of the brain. And if you can get all the signals going down those, then the brain is not magically connected to the environment. Reality is whatever spikes are on the cranial and spinal nerves. And in that sense you've got this well defined interface to it. Then with the processing once it gets this information is enormously complicated. It constructs everything we experience. I think it's important to appreciate you experience yourself being in the world. You kind of see the walls and the room and the lights and everything. But that of course, you're not experiencing directly. You're experiencing a world model fabricated by your brain. But one of the interesting things that's come out of progress in artificial intelligence is we're seeing this big unification in neuroscience and AI. I think we're actually learning a lot from AI research, more than I think we thought we would learn from AI research. I can tell you 10 years ago, we thought it would go the other way and that the AI people would learn a lot from neuroscience. And it's really been the other way around.

B (19:28)

I'm always curious, I mean, you were mentioning second sight, sort of flashes of light, and yet here, how did you figure out the API? I mean, if I was trying to reverse engineer it, I guess I would try to measure the signals. Is it similar with biology?

A (19:45)

It's just, it's difficult to measure the signals. So brain computer interface research and development is limited by your ability to record and stimulate these signals. The neuroscience comparatively is actually pretty simple. As soon as you can record these signals, we've very quickly figured out what we talk about, neural representations, what they are. Second sight's instructive. So in the retina, there's three layers of cells that matter. There's 150 million rods and cones. This connects to 100 million bipolar cells, bipolar because they've got two ends. And that connects the rods and cones to 1.5 million optic nerve cells, called them retinal ganglion cells. Ganglion is like a fancy word for like, reaches a far distance and connects to somewhere. We stimulate the 100 million bipolar cells. Second site stimulated the 1.5 million ganglion cells. And so they were trying to get the signal into the brain past that 100x compression. And the retina was doing a lot of computation there. The Isaac camera light shines in from the front. It hits the rods and cones like that. The representation in the rods and cones is a bitmapped image. It's just like you take the image, you tile it across the rods and cones. That's what it is now in the 1.5 million optic nerve cells. It's not like that. If you just project an image onto them, you get a bunch of trash because at that point it's already compressed. Things like edges, relative motion, a bunch of other blobby shapes, color. And so if you stimulate a cell there, you're not going to get just like a pixel, you're going to get some edge direction, gradient thing. And when you excite that, you can't do that selectively because first of all, you just can't do it selectively enough. And we don't know the codec, we don't know how to pattern it appropriately. And so you end up getting these flashes of light. It was an empirical discovery of our study that if you excite the bipolar cells with an image, you get an image in the mind's eye because that is clearly the critical processing step in the retina that you wanted to preserve.

B (21:33)

Did you know that that would happen or did you have to try different parts?

A (21:36)

When we started the company, we. I think we're a little bit different than most medical device or biotech companies because they're often founded around like a specific asset, like a patent or some specific piece of IP that they're going to spin out of a university or maybe something that the founders have worked on. We weren't like that. We did, we had a couple ideas at the beginning. We had this neural engineering centric view of healthcare. We had a specific BCI probe idea and biohybrid. And we had a sense that the most valuable thing that we could build in the near term was a retinal prosthesis. And we thought the time was there, the technology was all there, that that would be possible circa 2021. And that was also further from stuff that I had worked on before. And so it felt like a good thing for us to kind of go explore. I think we took this very first principles approach. And you have to be careful with first principles in biology because first principles are not enough in biology. Like they'll get you very far in many other areas of engineering. But in biology you also have to understand what did evolution actually do. And there's a lot of other nuance there. But in this case we looked at the retina, there were kind of reasons, intuitions, to think that past that would be much harder. And so in the retina you've got this two by two matrix. You've got a choice of if you've lost the rods and cones, do you stimulate the bipolar cells or the optic nerve cells and do you do it electrically or with a technique called optogenetics? And we just went and explored all four quadrants of that. We very quickly figured out that stimulating the optic nerve cells is very difficult. For these reasons. You end up with this 1 million degree of freedom calibration that you have to do per patient that can't be done in practice. And so that led us to the bipolar cells, which was before this compression. And so then their question was, do you want to stimulate them electrically or using optogenetics? And we developed both. And so we have a state of the art optogenetic gene therapy in house. Published a paper last fall on the world's most sensitive optogenetics. Opt optogenic proteins. These are proteins that you can express in a neuron to make a neuron that is not normally light sensitive, responsive to light. But the drawback was that the Conventional optogenic proteins take a bright laser to activate them. And so what we were able to do were find optogenetic proteins that are so sensitive that they're sensitive to indoor office lighting. And so this you could use in very different ways. And then we could target them to the bipolar cells. But that still has like five to seven years of clinical translation away if it ends up working. And there's a bunch of of pitfalls it could run into along the way. And then we also just surveyed the world to see what was the state of the art for the best out there in electrical stimulation. And there was this technology that had been invented at Stanford about a decade ago that a small company in Europe had been kind of developing in the meantime. And we got convinced that that was the right way to go. And so we acquired them a few years ago. And this was kind of all from this bird's eye view of if you want to restore vision of the retina, kind of, how would you do that? What are the promising approaches? Narrow that down. And that brought us to here.

B (24:25)

That's insane. That's so cool. I wanted to jump to your start in tech broadly. I mean, did you start in bio and software and engineering? What was your sort of journey into what you're doing now, which is, I mean, giving people blindsight is the wildest thing. People watching might be asking themselves, like, well, you know, I hear a lot about B2B SaaS, but how do I actually become something more like you?

A (24:53)

I was certainly doing software, and my deepest hard skill is software. I have a degree in biomedical engineering, but I grew up programming, and so I was doing that well before I was doing any biotech stuff. My parents tell me a story about how I sat on the floor of a Barnes and Noble and cried until they bought me a Learn Visual Basic book. I was always interested in the brain. I was definitely inspired by science fiction. The Matrix had a big impact on me both because the idea of this world of bits was just so alluring for a bunch of fundamental reasons. When I look around at the world, it's hard to build things. Space is constrained, the Earth is small, the resources are intensely contested. Space is large, the speed of light is low. You don't have any of those constraints in the machine. And so if you could simulate a world, kind of anything was possible there. But then also, if you then kind of turn that inside out, if you realize that you can build this and that you couldn't tell the difference, then the corollary of that was must be like the thing that matters is the brain. And if you can engineer the brain and support the brain, then kind of all the rest of it is replaceable. And that just seemed like kind of a fairly deep insight that was not being borne out in the world in the way that it seemed like it should be.

B (26:10)

Some of it is if you can surround that consciousness with like the correct inputs.

A (26:16)

Yeah, I mean, this also gets into questions of like, what is consciousness like the how does the brain create our experience? There's this meme out there that BCI is an artificial intelligence adjacent story, and that the goal is to we have to merge unions and machines. And I do think that there's something to that. But I think the more immediate thing here is that ICPSI is really a longevity like healthcare adjacent story. If the end of the quest of artificial intelligence are super intelligent machines, then I think the end of the BCI quests are actually conscious machines. It might turn out that there's actually no measurement that we can take that will tell us if something is conscious or not or what it's like. And the only thing that you can actually know on that is your own. And so if that's the case, then to study consciousness, we will need to use brain computer interfaces to see it for ourselves. And once you've developed that, then I think that you kind of can understand the fundamental physics of what's happening there, whether that's new fundamental physics or it's emergent in some way. But if you can learn how to build kind of understand whatever the brain is taking advantage of that our universe supports, then eventually you get super intelligent conscious machines that we can be part of through these ultra high bandwidth connections. I think that's a very different narrative than how people usually think about BCI today.

B (27:27)

I mean, we're at the beginning of that, right?

A (27:28)

Oh yeah, we're at the very beginning of that.

B (27:30)

The current trial that you have, I mean, it's relatively low bandwidth, but it's going to get much higher bandwidth. And then, I mean, like anything you sort of bootstrap with the thing that works, which I think, you know, what you have is a clear breakthrough as it is. And then if you look at like the PC revolution, for instance, it's like, could you believe that all of this that we have today started with like a little blue box, like in Altair?

A (27:57)

It still takes some suspension of disbelief because I think biotech has just been so incremental. Like it's been so, like there's been big advances, but at the same time These time constants historically, I mean, you could easily spend 10 years on something that. And I think that one of the things that's so exciting about what's happening now is it no longer really feels so incremental to me. To me, it feels like we're firmly in the takeoff era now. Something new has happened on Earth. But I think it's also important to remember that this didn't start in 2019 or in 1999. This started in the late 1800s with the industrial Revolution, just a few years before the Industrial Revolution really kicked off. I mean, life was more or less unchanged in a fundamental sense for several thousand years. And they didn't really even have a concept of progress in many ways. And I don't think there's any way they could have imagined the way that their life would have changed over the course of the first 10, 15 years of the steam engine. And that is how I feel like looking at the next 15 years right now.

B (28:51)

Yeah, I mean, so we have an electrical stimulation right now, and then at the same time, you also do have a bio coupling. It's not purely just electrical. Would you call it a V2 or like sort of a next frontier?

A (29:04)

So this is a totally different area. I mean, you might be able to use it for vision. So one of the diseases that prima, our electrical stimulator doesn't treat is glaucoma, which is loss of the optic nerve itself. And so it's possible that you could use our biohybrid BCI technology for that, but that's not what we're doing right now. There are three elements to our pipeline at science. The first is our work in the retina in blindness, especially with the PRIMA implant. The second is our work in neural interfaces. And the third is our work in perfusion with our vessel program, the biohybrid neural interfaces. The idea here is if your brain is a bunch of neurons, how would nature solve this problem? We often look to nature for inspiration. Evolution is a way better engineer than we are, at least when dealing with biology. I think the intuition here started from your brain is composed of two hemispheres and they kind of process different halves of the world separately. But you don't experience two hemispheres or two hemifields. We would say you experience one integrated moment. And there's a cable that connects the two hemispheres of the brain called the corpus callosum. It's about 200 million fibers. And I was thinking like, if nature wanted to build a ultra high bandwidth brain to brain connection, like what Would how did. Or if you wanted to make a new cranial nerve. So instead of having an optic nerve or vestibulocochlear nerve, it wanted to have the Internet nerve. How would nature solve this problem is it would grow a new nerve. It would have a new fiber bundle with a USB port at the end. So the intuition here is if your brain is a bunch of neurons, what happens if I culture some neurons on your neurons? Do they when you do that in a lab, that neurons will typically grow together and wire up and form new biological connections. And so we have an approach to the device where we seed the implant with living neurons. These heavily engineered stem cell derived neurons that we've created.

B (30:54)

Are they related to your own neurons or no?

A (30:57)

So really interestingly, this is actually one of the deep areas of research. So it's one cell line and probably the single deepest area of IP on this is that we've hidden them from the immune system. So we're one of a really small number of companies that have, I think like pretty convincing what we call hypoimmunogenic stem cells. You don' manufacture it per patient, which would be really expensive and take much longer. We've got this hypoimmunagenic stem cell derived engineered neuron that we load into the device in a dish and then that kind of gets stuck there and then you engraft this onto the brain. So we don't place any wires into the brain. We also don't need to genetically modify your brain. Some of the other ideas out there, for example, using optogenetics or things like ultrasound, this requires using a gene therapy to genetically modify the neurons in your brain. Which first of all that's like a one way door and if it goes wrong, that can go really wrong. Whereas here, because we're adding, the only thing that has been edited are the graft cells that we add. And if those die off, then you're really not worse off than you were before for the most part. But it comes with the potential of growing throughout the brain, forming biological connections all over the place. And I mean that's what we've seen in the animal models. That's not in humans yet. But have you seen James Cameron's Avatar movies?

B (32:15)

Definitely.

A (32:15)

Like, you know, the ponytails that the aliens have. That's how I think about it. Basically. It's like it's a big new cranial nerve with a connector at the end. I think that's actually the avatar Q I think is like a pretty direct reference for how I think about Our biohybrid neural interfaces.

B (32:31)

So earlier you were saying sort of this, how do we find a USB port? I mean, obviously an avatar, that's, you know, one of the manifestations in the blue creatures. The optic nerve, in a way is like, like a port. And then jumping to Neuralink when you were co founding it, that sort of enters the brain and then there is not necessarily an obvious port. How do you think about that? Where do you attach and how does it work? And what did you learn from Neuralink that was useful here?

A (33:03)

Well, I mean, a lot of what I learned from Neuralink was just like, in many ways it was kind of the ultimate startup PhD. And so that was more about how do you execute a technically complex company that requires this type of multidisciplinary team and infrastructure.

B (33:17)

I'm very curious from those days. What was the V1? And then there's a hypothesis and then the outcome. And then here the outcome is very, very awesome. With science so far, not done.

A (33:28)

Obviously when you think about the brain, because I remember it being totally magical to me. How do you even understand what the brain's doing? What language is it speaking? How do we understand what's going on there? That seems impossibly complicated. The way that I would think about, about the brain from this information processing perspective is the brain is full of these things that we call representations. And so you can have a representation of hand activity, so there's a geometric object in the brain. If you record from some neurons, then when your finger is held open, a neuron will be firing. When it's closed, another neuron will be firing. There's neurons that kind of correspond to every possible state here and often in primary motor cortex, which is where many of the other BCI companies record from. Primary motor cortex is a couple synapses, often two synapses from the muscle. So it projects all the way from the top of the head down to the spine. And then there's another synapse from the spine out to the muscle. And so the representation that you get in primary motor cortex is kind of easy to understand because it looks like it directly corresponds to things that we can easily reason about, like hand stick and specifically, often joint torques. One of the things that I like to do sometimes with the LLMs is I'll pick a neuron to start from, for example, the retinal ganglion cell, and I'll be like, okay, go forward one synapse. What are all of the cells that we're connected to? I'll pick another one and Be like, okay, go forward one synapse. What are all the cells that we're connected to? Just kind of try to walk through the brain. And each generation of model, your ability to do this gets better. But one of the things that you see is that when you're close to an input or an output, like a muscle or a cochlear HA cell or a retinal ganglion, a rotter cone. In these cases, we think of the representations as being concrete because they correspond to things that are intuitive for us, like colors and, like, image intensities or frequencies of sound or muscle control. But as you go deeper into the brain, it very quickly kind of blows up into these very abstract things. And so, like, there's a part of the brain called infrotemporal cortex where the representation that it has is a map of objects or another area right next to it is a map of faces. You think about this map of object space, this neural representation of general objects. There's, like, one point. You can think of it as a long list of numbers, and there's some point in that that's like a vase. There's some point that's like the Eiffel Tower. There's some point that's a car. There's some point that's a person. There's some point that's like a zebra. And as you move around on this manifold, you get kind of the percept of any possible object. And there's millions of neurons there that are representing this space of possible objects that the brain could be identifying.

B (36:15)

Sounds like latent space.

A (36:17)

It is a latent space. Exactly. And so there's this huge unification going on between AI and neuroscience. And one of the most interesting things is that when you train AI models, like image models and even language models, the representations that you get inside them look a lot like the representations you see in the brain.

B (36:35)

Fascinating.

A (36:36)

And so this is like, a real hint that the AI people.

B (36:38)

I mean, that's really good.

A (36:39)

Are on the right track. Yeah, no, I mean, the whole idea that these things are like stochastic parrots or glorified autocompletes, like, these people just don't know what they're talking about. Many people in neuroscience have gone over to AI because they're basically still doing neuroscience, but it's just way easier to do it on the models.

B (36:52)

It sounds like it's very good news for you in that, like, there is actually some kind of latent space mapping. And then the job of science in terms of being sort of like The API to the brain.

A (37:04)

Totally. Exactly.

B (37:05)

It's like entirely possible.

A (37:07)

The neural activity that you. When you record neural activity from the brain, this is just another latent. And if you can translate this into another model, then you can do, we think, really cool stuff with that.

B (37:17)

So you have input now. And then you earlier were saying, I mean, a lot of the earlier BCI experiments involved figuring out, like, motor.

A (37:26)

Motor, yeah. So motor decoding is kind of this very classic task, and you can do it in any number of ways. But getting like cursor control or keyboard control in a human, that was first done in the late 90s. And so I think a lot of the BCI companies are doing that now just because we know it definitely works. There's some patient need. And it really is just like an electronics problem. Like, if you can shrink the electronics so that they're small enough and low power enough so they don't dissipate a lot of heat so that you can close the skin, then that is like a big advance. And that, I think, is really the first thing that neuralink has done. There were prior devices that could do that type of motor decoding, but they required a connector coming out through the scalp. And as long as the skin is open, there's a risk that, like, an infection will climb down that and then you're have a really bad day. So being able to close the skin is really important. But that was really difficult because it required really efficient electronics that were small enough to fully implant and also were power efficient enough that they wouldn't get hot. And so I think the thing that made this possible is what we call the smartphone dividend. Like, BCI couldn't have done this on its own, but Apple and Samsung and others have poured epic amounts of money onto making these types of electronics exist in the world so that people like us can use them.

B (38:33)

And then it feels like you have a really significant advantage around being a biohybrid. I mean, there are all these issues, famously, about sort of trying to electrically stimulate brain cells for a long period of time.

A (38:47)

Yeah, I mean, I think that there are different products here. I think that on the one hand, I mean, that's why I'm doing it. I think that's a good idea. On the other hand, I think some people look at, they're like, that is now you have a cell to deal with. You took a device and you added a bunch of biology to it. And I think we have a good handle on that. That's why we're doing it. But there's definitely a trade off there. And I think that BCI we're going to come to see is not a specific product in the way that pharma is not a product. I think there are going to be a bunch of BCI companies going after different applications where different types of probes will make sense. And I think biohybrid in particular is only really necessary for some of the very highest end things. And on the flip side, it will be harder to deploy for many other important medical needs and important applications along the way and will probably be a little backloaded relative to some other things in that scalable impact.

B (39:40)

So earlier you were referring to there's a third part of science which is vessel. Talk more about that because it feels like you're applying a lot of the first principles, thinkings that got you here to this thing that that is also pretty groundbreaking.

A (39:54)

So this is our smallest project. So there's this field of perfusion. You can think of it as they're kind of like heart and lung machines. And I was first clued into the need here about a decade ago when I read an article in a medical journal called the Lancet, which was this case study of the 17 year old living in Boston who was waiting for a lung transplant. And while he was was waiting for this lung transplant, he was being kept alive on an ECMO circuit. ECMO is sacrament for extracorporeal membrane oxygenation, this fancy word for heart lung machine. And in his case his heart was okay, but his lungs had failed. So this was keeping him alive. And after a while on the transplant list he was diagnosed with a complication that made him no longer a priority recipient for donor lungs. And so they took him off the transplant list. And so this article is kind of about the ethical dilemmas of like what do we do with him? But he's alive because he's, yeah, he's like playing video games, he's doing homework, hanging out with friends. If we turn off the circuit, he will immediately die.

B (40:50)

Well, don't do that.

A (40:51)

Then on the other hand, he's consuming a half a million dollar a month ICU suite. And so there are these quotes in this article from the doctors being like his family and friends derived benefits from his continued survival and how this raised fairness questions because if we support him for a longer period of time, then why don't we do this for everybody? And so I saw this, I'm like, those are great questions. I need answers to those questions. Because there seemed to be this big gap between what was technically possible and what was economic to deploy for some reason?

B (41:15)

I mean that's exactly what being a founder is about.

A (41:18)

Yeah, so I saw this and there's this database of medical literature called PubMed. And I realized that if I searched PubMed for the phrase ECMO ethical dilemma, there were multiple pages of results. So this was not like a one off. And when I looked at this literature, it was often a lot of it was talking about how ECMO shouldn't be used as a quote, bridge to nowhere care, and how many doctors were basically trying to discourage families from even pursuing it in these critical care cases because it would create this bridge of nowhere. And then what do we do? And it creates these dilemmas. And then I went and asked some doc, this was a long time ago, this was almost a decade ago now, like, oh well, why don't we consider it as a destination? The phrase is like a destination therapy versus a bridge therapy.

B (41:58)

What if the technology just isn't good enough yet and it needs to be improved?

A (42:01)

It needs to be improved, definitely. But that wasn't even the response that I got. The response that I got was just, just like shouting and throwing things. And so I was like, something feels wrong here. But I wasn't really in a position then to pursue it. But this was always a thing that was kind of, I saw that there was a really important unification here also. The same fundamental type of technology has really transformed organ transplantation. So there they call it NMP, Normothermic Machine Perfusion rather than ECMO, but it's the same idea. So 20 years ago, if you needed a kidney transplant or a liver, if the car crash happened at 3 in the morning, the surgery would happen at 4 or 5 in the morning. But now it gets scheduled for like the afternoon or the next day. And over 75% of liver transplants in the US use this type of perfusion technology now. But the systems that exist for this are like $500,000. They can only be moved by private jet. Like one of the big companies in the space. It turns out that their private jet logistics business is bigger than their medical device business. And it just like there was just clearly an engineering that could refine this. And so we looked at this and we thought like, well, what if you could refine this to the point where you could check a kidney luggage on a United flight to the east coast? Or what if you could make a thing that that 17 year old could have brought home as a backpack instead of just what they did in his case is they stopped changing the oxygenator filter and a week later it clotted and he died. That's what happened. There are other problems here, like being able to close the skin around the brain implant. You also need to make it so that the tubes that connect the blood supply to the circuit, the skin can heal to it. So that's not an infection risk. Otherwise you have to clean it very carefully. But just overall there's this huge gap between clearly where the scientific breakthroug were pointing and what was being done. I think that people don't appreciate is that in many cases you want to be a brain in a vat. This basically already exists. You can keep an end life patient alive in an ICU almost indefinitely. But this is very poor quality of life. And so patients ask for that to be withdrawn. Nobody wants to be basically a brain in a hospital bed connected to tubes. You need to be able to provide a high quality of life. And so you need something that people can live with, with. And I think to see this, if you can get vision, hearing, balance, motor control, the ability to be out in the world and doing things, I just saw this very fundamental way to reframe the problems of medicine here. And so that, like I said at Science, even though there's these several different projects, I really see them as one project over the next 10 years.

B (44:24)

So started as an engineer, first principles thinking, which often now is quite associated with Elon Musk. Ask how did Neuralink start? How did you get to know him and how did all of this sort of come together? Because I first met you when you were doing Y Combinator many years ago, my first stint at YC.

A (44:43)

So I got an email one night in early 2016 from Sam. Subject line, crazy question, be like, Elon starting a brain computer interface company, like who should run it? And I assumed they're talking to a lot of people. And my first reaction was actually I had some friends at MIT that I thought, I'm like, well these guys are really smart, you should talk to them. But then like an hour later I was like, wait a second. And so I emailed him back, I'm like, can I like? And Tam introduced me to Elon. And Elon was going around, he'd already had the idea like on his own that he wanted to start a company and he had the name Neuralink. I also think that he heard my name from enough people that he was talking to at the time time. And kind of over the second half of 2016, there was just this group of people that was kind of to some degree ever shifting that would meet once a week or so in the evening, and that snowballed into Neuralink. And of the initial group, a bunch of them were people that I knew from Duke. So Tim Hansen, the guy who had originally had the sewing machine idea, he was in the lab that I came from at Duke. He was a grad student, I was an undergrad working for him. And then the professor that he and one of our other friends had gone to at ucsf and then a collaborator there. So it was kind of a very small community.

B (46:00)

What was that like initially to talk about the idea of like, you know, connecting a computer to a human being's brain?

A (46:07)

Elon, I mean, he saw what was coming in AI like, very much more clearly than many other people much earlier. And I think the implications of, like, you gotta. This can't be a separate thing from humanity, and that needs to emerge somehow. I think that implication was just very clear to him. And so that was the genuine motivating factor of how do we make it so that this allows us to upgrade humanity rather than get left behind? I mean, if you look at the natural history of Earth, it's not like this is a totally speculative thing. Humanity has totally dominated the planet, and we keep our closest living relatives in glass boxes so they don't go extinct. And so there's a real history here of greater intelligence being very dangerous. Like, in the beginning, there wasn't like a specific technical idea necessarily, but there was that motivating force. And then the idea was we'd pull together the smartest group of people that he could find and enough resources to do whatever made sense and eventually got consensus around what you see now as the thin film polymer threads.

B (47:04)

You're one of the best examples of someone who came from a pure software world and then went into hard tech and now is actually doing real breakthrough type of research and work that is also commercializable. The people watching, they might be on a similar track. Knowing what you know now, like, what would you tell to the sort of 2016 version of yourself?

A (47:24)

So I think there's two things. The first is like, the thing that I did, and then there's the thing I didn't do. The thing that I did, I think, was I had. I had a clear sense of what I wanted. And then I was very high agency towards that. When I was in college, I knew that I wanted to work in brain computer interfaces. There was a great lab that was doing that work at Duke, where I went. And I was pretty persistent in figuring out how to, like, place Myself into that lab. Lab. It was in the medical center. They didn't usually take undergrads. They're like. It took me a little while to get in there. I eventually figured out that I could sneak in by taking an independent study in the chemistry department. That would be a backdoor into this primate neuroscience group. But then really, most of my education in college happened in that lab. So, yeah, I grew up programming in my deepest hard skill is software. But I've been doing primate brain computer interface, closed neural decoding stuff since 2008. And so that was just like you had to be pretty high agency and persistent in trying to follow through on that. But that only works if you have a sense of where you want to go. And so the first is figure out what you want. The thing I didn't do was. So after college, I started a company called Transcriptic. It was a robotic cloud laboratory. So the idea was, I mean, also in college had the experience of working in synthetic biology group. Group. Where I needed to go press a button on a device called a plate reader every three hours for three days to take a measurement that I wanted. And I was like, in software, we wouldn't do this. This just clearly doesn't make sense. We would automate it. This was also the time when AWS was just emerging and cloud computing was becoming a thing. And it seemed very obvious to me that instead of every researcher having their own lab and spending millions of dollars for all their equipment and then needing to press these buttons, what we should build is a central robotic cloud laboratory that expose APIs that scientists can use to run experiments over the intern Internet. I did that. Raised a bunch of money. When I stepped down as CEO in 2017 to join Neuralink, it had millions of dollars in revenue. I felt like we got it to kind of an early promising point. And then since then, over the last decade, that promise was not fulfilled. That was hard mode. That was a slog. That era, from 2012 to 2016, I strongly identified with Ben Horowitz's essay the Struggle. And I think the thing that I should have done earlier is go work for somebody like Elon Musk on because that just so dramatically leveled up my ability to do this and know how the game is played. And I think that often you'll see these really promising kids who are just like, I'm going to do it myself. I don't want to work for anybody else. I'm going to start my own company. I'm going to plow through it. And sometimes that works. Who am I to say. But I can tell you that very often running a startup is an oral tradition. There have been a couple nucleating times in history where a really remarkable group of people have figured it out from scratch. I think PayPal was like this, but almost always beyond that, it's like it's an oral tradition that you get passed down from one of this handful of Silicon Valley cultures that can make a huge difference on the trajectory of your career. To get that right when you're 20 versus when you're 26 or 28, well,

B (50:26)

science is the next and it sounds like you're assembling. You know, you've already assembled a really accomplished crew of people and then what we've learned from startups over the years is that once something works like, like more and more resources, more and more smart people sort of come together and then zooming out, that's what we really hope happens a whole lot more in exactly the spaces that you're in right now. So science sounds like one of those places to go to right now.

A (50:54)

It's pretty cool. Yeah. I mean I definitely feel pretty lucky that I get to do this because it's such an interdisciplinary problem and to innovate on it you need all these different areas and really great people in each of them. But the same time, it's just that the things that you can do today were unimaginable a few years ago. And yeah, I mean I think that, I think we have the best team in the, in the field.

B (51:16)

So I mean next 10, 20 years of, you know, science, BCI like I guess where do you see this going and you know, what are you most excited about?

A (51:26)

I have this like event horizon at 2035 now. Like when I was earlier in my life I always kind of prided myself on the ability to see the future and that is the next few years I think I have a sense of. But by 2035 it's just I can't see past it. I think it is very possible that the first people to live to a thousand are alive right now. And I think it might be many more people than you think. It's not going to be like one or two people on earth today. Earth as a whole is at a not unique moments in history have happened all the time before. But right now it's a time of exceptional change. This is going to be really, really influenced by the technological changes that are happening. And I think the twin plot lines of brain computer interfaces and artificial intelligence, people are beginning to get that artificial intelligence is real. It is still not priced in. People still don't appreciate it, I agree. But they really don't get what's coming in, what's possible with brain computer interfaces. And those are really parallel but very distinct stories. Intelligence is going to become widely available for those that have the agency to deploy it. And I am generally pretty optimistic about that. My PDOOM is pretty. It's not zero, but it's not 50%. It's well below that. Yeah, I don't know if we'll have cured all diseases. In fact, I definitely wouldn't use that term. I wouldn't say we'll have cured all diseases by 2035. But I think that there will be kind of new lateral options that totally reframe how we think about the human

B (52:49)

condition on that timescale and totally reconfiguring, basically that. That sort of interface between computers and humans.

A (52:56)

Yeah, and humans and each other. A brain computer interface is equivalent to a brain to brain interface in many cases. This takes you to totally new territory.

B (53:05)

Max, thank you so much for joining us. Thanks for building the future and we can't wait to see what you build next.

A (53:10)

Thanks, Gary.