The History And Future Of Brain Implants - EP 31 Sumner Norman

It's kind of cool, right? I got it. I can just do it myself. Usually there's like 4 people that do all this stuff.

Let's pretend that there is though. Just be like, hey, John, Bob, you guys ready? Guys, we're rolling.

Be quiet. Be quiet.

Keep the cat down.

Be quiet. Um, if you're ready, are you, you want to chill or you, you good?

I think I'm, I don't know what I would do in the chill time.

We are just talking. So it's not, it's not, it's not overly complicated. Otherwise, okay, well then I will introduce you. Okay. Sumner Norman, you're a co-founder and CEO of Forest Neuro. More than that, you are, you're someone I met, I feel like, I don't know, 3, 4 years ago now.

It's either 6 months or 6 years, but either way, time dilation is real.

One of those. But you're, you, we got introduced, I suspect, kind of through Adam Marblestone and, and, and, um, I think that's James Fickle or some, some brain network. And then you're someone I just ended up You've been immensely helpful in so many things I do in the BCI, brain-computer interface, space.

That's all any of us hope to be is helpful. I think I just like nerding out with you. We did some good nerding out.

We've done good nerding out, but you're— I always feel like you're— there's so many things going on in this brain implant, brain-computer interface world. I always feel like you have a balanced take on things. And so it's been helpful to me to calibrate. What is going on. And then, so we'll get into Forrest Nero in a second. And, um, but I always like to let people get to know folks a little bit, you know. And, and, and so I was gonna dive in a little, little biographical journey for a second. Um, one thing, when I was— I never— I hadn't like properly researched you in a while. I'm sure I did the first time I was coming down to LA Um, to interview you, but then, yeah, there's something, I don't know, I just had not filed this away properly, but you were, you're like a mechanical engineer by training as opposed to neuroscience. And then you seem to know like everything about neuroscience as well.

Um, yeah, I, I don't know everything about neuroscience and luckily for us, actually the neuroscientists also don't know everything about neuroscience. That's how I get away with it. But yeah, I was, I was, uh, actually trained as a mechanical and aerospace engineer and some background in physics as well.

And you got like a all the way up to a PhD in mechanical engineering. Yep.

Yep. So yeah, undergrad in physics and mechanical specifically. And then my master's and PhD were in mechanical and aerospace by division, like the actual department of the school.

Yeah.

But my focus is in control theory and brain-computer interfaces. And I— that— where that really marries is the control of exoskeleton robotics with brain-computer interfaces for therapeutic use.

When you were either high school years or the beginning of college years, did you know you were going to brain stuff or you thought you would be making rockets then?

I thought I'd be making rockets.

Okay.

Um, yeah, I think when, so just like I'll back way up.

Yeah. Yeah. Let's do it.

I grew up in a very, very small town in Utah, something like 5,000 people. It was like Mormons, cows, and then this like one nerdy brown kid that didn't fit in. Um, And so I, I was the first person in my family to go to college. I was trying to, like, find my bearings. And one thing I knew I loved was math. So I was just a huge math nerd. I knew I wanted to be a mechanical engineer because we went on a family trip to the Boeing factory when I was 11 years old. My dad was a pilot, and so I met a mechanical engineer and I remember this and I was like, that guy's like got powers because there's something on his computer screen and on his desk is the thing he designed.

Okay.

These days it's like green lines on black screen, you know? So I had my mom write it down, mechanical engineer. And I was like, that's what I'm going to do.

Was your dad a professional pilot or a hobbyist pilot?

Um, actually it went in both directions. So he was professional, um, but like really, really small aircraft. Um, because in Utah, like farm boys used to fly airplanes for delivery and crops and stuff like that.

Yeah.

And then, um, when he got older and that became like less common, then he, he flew, uh, private just like himself for a while. But then eventually even that became kind of costed out for many people, including us.

So he's from Utah and grew up in the same sort of setting.

Yeah. So yeah, he grew up in Spanish Fork, Utah, which is a really small town.

Okay. Okay. And so he was, he was just like flying out in the, in the country.

Uh, yeah. Out in the country.

Okay. Yeah.

Out in the country.

And did your mom work or stay at home or?

Uh, my mom was a bank teller.

Okay.

Yeah. Which is, so like my dad was a small businessman and my mom was a bank teller and the one thing I definitely got was math.

Okay.

And I still love math. Um, Yeah.

No, it sounds like your dad was, if he's taking you to the Boeing factory and stuff, I mean, I know he's in the aviation, but it must have been inclined that way a little bit.

Yeah. Yeah. So I wanted to go that direction. I was like, I want to design stuff that goes fast. I'm a huge car nerd. Um, love old cars. And I thought that was the direction I wanted to go. It wasn't, it didn't take very long for that to, that illusion to be broken.

Okay.

Because as I actually got into the math of it in undergrad, I was like, it was very much like, well, you know, this guy came up with this equation and this woman came up with that equation. And now, but you know, we'll teach you all that theory, but in the end you're just gonna go to the lookup table in the back of the textbook because that's been solved. You don't need to redo all that work. Like NASA did that in the '70s. So just go look up their airfoil designs. They're categorized very well. Yeah. And I was just like, man, where all the creativity has been like sucked out of this. Like there's of course now like a resurgence, but that didn't, didn't feel like it was there for me. So I was, so I, and I don't think I did this in a logical way. Just intuitively started pushing into like, well, where are the corners of this space that there is a lot of creativity? And robotics was the big one. And control theory was kind of the area of that, that just like, for me, my math brain just like hooked onto that immediately. Like, how do you control these machines? And why is it so hard to get a robot that is, you know, more powerful in many ways than a human and more stronger and greater, whatever. But we just can't get these things to act naturalistically or interact with humans in an interesting way. That was like, really grabbed my brain. So I joined a haptics lab actually.

Well, I did haptics in undergrad.

In undergrad. Okay.

Yeah. Okay. And you did undergrad in, was it LA or University of Utah?

University of Utah. Yeah. But I actually, my first research experience, I didn't know even how PhDs worked, um, to be honest with you, or like how labs work. But, you know, credit to them, there was a paid research experience for undergraduate, uh, thing at Brigham Young University in Provo. I had to go through the like Mormon screening process as a non-Mormon, which is a whole thing. Um, but that was the first lab I joined, uh, and that was a theoretical physics lab. So I was doing X-ray crystallography, and that was really where I learned to program. And then I was like, that's just too theoretical. And And then the, like, mechanical stuff was too solved. And then I found my middle ground, and that was a robotics haptics lab at the University of Utah.

Okay. Not to, like, totally side-drag us, but what are they screening for in the non-Mormon screening program?

Um, they just, like, want to make sure that you're not a menace. Um, you know, you can't just be, like, doing drugs and fornicating with all the students that they've, like, very carefully curated to be, like, model Mormons.

Okay. Were your parents Mormon?

Uh, no, huh?

No. Oh, so you just— you guys were just out there.

Yeah, we're just out there doing our thing, just trying to get along. Okay.

All right, what do we do at Core Memory? We cover innovative, fast-moving, forward-thinking companies, which is why Core Memory is sponsored by Brex, because Brex is the intelligent finance platform for many of these companies. 30,000 companies, from startups to the world's largest corporations, rely on Brex's technology for their finances. They've got smart corporate cards, high-yield business banking and expense automation tools that are fantastic. I hate doing my expenses, and Brex's AIs and software run right through those expenses, figure out where we're spending money, and take care of so much stuff for you so you don't have to waste your time on it yourself. Go to brex.com/corememory to learn more and just, you know, get with the program. Let's get going. Let's get out of this Archaic finance software and move toward the future, core memory and Brax. As you're going through this journey, when, when you start gravitating towards brains and neuroscience and adding that in.

Yeah, that, that really kind of came towards the end of undergrad where I was working again on these kind of haptic robotic interfaces. So I was thinking about teleoperated surgical robotics and we had come up with these, uh, the lab there had come up with these ways of like stretching your skin around. It turns out that your, your fingertips, especially your index finger and thumb, are so sensitive. And so one of the papers I wrote was about how you could actually discriminate 16 different types of tactile stretch cues, which is like pulling your finger around. Well, it's actually really cool for surgical robotics where you don't want to have force feedback because you can imagine like a, a doctor is working on a patient carefully cutting with the scalpel and then they like slip off the robot a little bit and the force feedback kicks and now the scalpel is flying. Okay. You can't have force feedback. So we were developing this way of actually like pull it, like giving you tactile sensation but with forces that are minuscule.

And you— wait, on the stretching the skin part, you would be— you would be feeling that level of detail? Okay.

And so yeah, imagine like, you know those little red trackpad things that are in the middle of the IBM laptops?

Yeah, yeah, yeah.

It's like that in reverse. So instead of you like pushing the red thing around, it's pushing you around. But these tiny little forces can tell you a great deal about what you're doing.

Oh, I see what you mean. But you could feel it in your fingertips for the— the surgeon? Okay, okay. So There's so much sensitivity there that it gives you this finer control. Okay. Okay. So you, you start, so you're starting at this from this robotics lens and digging in and then, and then you have to start at some point learning some neuroscience as part of this or, or—

So, so what happened was just totally serendipity that, um, I built this experimental device and it was really just to, it was a psychophysics experiment to understand how people used the sense of touch in robotic environments.. And someone mentioned to me from another lab, like, hey, you know, that device in and of itself could actually be really useful for stroke rehabilitation because it's providing like really good feedback that's real time. It requires proprioception and sense tactile sensation, and you can teach people to move again in like high repetition movements. It's a low cost device. Like maybe you should write a paper for this special IEEE Transactions journal that on rehab. And so I said, okay, I just thought it was a great path to publication. And then I like got my mind blown as I'm writing this thing because now I'm reading all these publications about stroke rehab and I'm going, hey, there's a control system of a robotic actuator of some type. There's a control system of the human being. And these interfacing in and of itself is really interesting. But then when you damage one, but it has plasticity, right? The brain can change and learn and adapt. And now you can use the robot to assist and teach and protect that other control system, like, it just created this, like, a more intimate intersection of human and machine interaction that just— I, that's like really what drives me deep down. So that was where I really hooked on to it. And so as I considered my options and thought about grad school, I knew I wanted to do something where it was like a— the most intimate form of human-machine communication. And that's how I found my way to brain-computer interfaces eventually.

And then what do you— you have to— like, I would feel if I was doing this, like, I was so far behind on the neuroscience stuff. And I mean, to your point earlier, I've interviewed enough neuroscientists to know, like, there is this degree to which we still don't know that much. But you, you would, you would have to go through the whole corpus of, of, like, you know, key texts, I assume. I mean, did you feel like you had to catch up and then start?

Oh yeah, for sure. I felt very behind in the beginning. Um, But I think there were a couple things that helped. One is, um, I just had an excellent PhD advisor that was patient and thoughtful and pointed me to all the right resources and was like, if you read these, you'll start to understand. Um, and I do think that like a PhD, really what you actually learn in a PhD is how to teach yourself. And so he was very good at teaching people to teach themselves, um, hands-on without being overbearing by any stretch of the imagination. And, um, it was just a lot of reading papers. There's no, there's no secret to it. It's just like lots and lots and lots of exposure, a few classes, but honestly not that many. A lot of it was really just exposure to the literature itself.

All right, this may be an unfair question, but if I, if as like a journalist or, uh, somebody who just wants to start getting into this stuff, I mean, are there 2 or 3 key texts that get you the farthest?

Well, I mean, there's the Bible, uh, Eric Kandel's Principles of Neuroscience, which I admittedly have not read, but would— I've read parts of it. Yeah. Um, and I think that like, that's going to tell you about neurobiology all the way from kind of the atomic level up to like what systems neuroscience is. Um, I don't have like a magic bullet here. Yeah. Um, I'll think about it. If I, if I can, we like put it in the pod notes. If I've got like a neuroscience for dummies, maybe that's it. Neuroscience for dummies.

We will do this. I'm sure that probably exists.

It probably does.

Right. Um, no, I just, you know, I did, I think you answered the question even just there's, there's always one. It feels like there's always one seminal text that seems to get people pretty far.

So yeah, I think I got lucky. I had a great mentor. And then, you know, when I joined Caltech later on, the group that I joined was kind of led by a legendary neuroscientist. I was actually just scared of that. He was like, oh, don't worry, actually. And I asked during my job talk, like, hey, is this something I should be like brushing up on or learning more? He's like, don't worry. I generally find that hiring engineers and teaching them neuroscience works really well.

Okay.

He said, I have a very hard time hiring neuroscientists and trying to teach them engineering, and you need both for neurotech. And so that was, that was my entry into that.

Was that, was that Mikhail Shapiro, or, uh, that was Richard Anderson.

Richard Anderson.

Okay. There's many, many, many different approaches to brain-computer interfaces at the moment. I would kind of categorize, I think, Neuralink as maybe the most aggressive. They're doing— they have an implant with tons of electrodes on it. They're going— they're doing a full-on surgery, going into the brain. There's other people that want to sit right above, like, sort of the first layers of the brain. There's even people that I just interviewed this week where you kind of dig a little groove in the skull and you're just sitting there. And then there's people going in blood vessels like Synchron. And so there's all these different approaches, some more invasive than others. And then, and then a lot of them are doing this electrodes into the motor cortex and things like that. When we get to Forest, you know, you guys are doing, um, this different approach with ultrasound with an implant. But before we got there, I just wanted to give people some context on like where we're heading. And you're kind of looking to treat, um, to understand and treat some different conditions than a company like Neuralink is with movement through the motor cortex. But one of the things that I've always enjoyed talking with you about is like the history of how we got to this point, because it can feel just because there's been so much startup capital and everything that's gone in this field, all of a sudden it sort of feels new. A lot of the work that's being done is piggybacking off this Utah array, which is kind of like 20, 25 years in the making. But we go— it actually, you know, people messing with the brain goes back even farther.

Oh yeah.

Yeah. I mean, what were— what, like, when did people start poking at brains with electrodes?

Well, I mean, yeah, with electrodes, I mean, it's well over 100 years. I know the EEG just had its 100-year birthday. So electroencephalography, or non-invasive recording of electrical activity, which people often talk about as brainwaves, right? It was Hans Berger in 1924. So we're just over 100 years there. And I think that measuring extracellular electrical activity potentially Oh geez, that was the large squid axon, or giant squid axon, and I forget the exact year, but it wasn't too long after that, '40s, '50s, somewhere in there.

Um, so like, what happens that people just realize the brain, like the body, has electricity running through it and they start to get interested?

And that goes all the way back to Luigi Galvani, so that's like the 1700s. Um, it was when I think that he stimulated some dead frog legs and they twitch, which now sounds obvious, but at the time we didn't understand how movement was generated in the body. And so now you have, okay, these muscles are twitching, what's going on here? What is it about electricity and musculature that's, that's connected? And so that's where we start to understand the peripheral nervous system. And then over the past 100 years, we start exploring electrical activity in the nervous system in a real way— or sorry, in the central nervous system and in the brain.

Okay.

Um, and so that's kind of where our, our foray into brains starts, really neurotech proper.

And in the 1920s, they're doing what?

They're doing Yeah, so there they're just measuring the activity and saying like, you know, look, this person's asleep, this person's awake. It's very crude. And honestly, EEG hasn't really progressed that far beyond that these days.

They're just putting some sensor on the outside of the skull and just, they're just seeing that there's stuff going on.

Yep. There's, there's a voltage. So it's called a montage. You have these electrodes that you place over the skull. You can put them in different arrangements and then a voltage has to be across some distance. So you have a, like, you know, one electrode's going to be a reference, maybe you use it over here in the mastoid or something. And then you have another one on top of the head that you're measuring a signal. And then you might have 16, 32, 64 channels. And then you see these voltage shifts. So for example, when someone falls asleep, you see these alpha waves, which are roughly at like 5 Hz and lower. You see these kind of like slow moving, relatively slow moving waves, and they increase in magnitude. And then when someone's awake, those kind of go away. And then there are some other effects, like when someone's preparing to move, you see this beta. So this is like this 12 to 15, uh, hertz signal actually disappears right before someone moves, and it's a movement preparation signal.

Okay.

We actually don't know that much about it, but we think there's a kind of constant, um, inhibition of movement that's happening between your cortex, the kind of surface of the brain, and the thalamus. And this cortical thalamic loop is just constantly going, inhibiting you from moving. Moment, which is probably an evolutionary thing that it's better to do that. And then it actually gets interrupted, and then you move. Um, so you get these kind of correlates, but they're really kind of crude, um, and not that specific. But you can get something out of EEG. Uh, and then as time goes on, extracellular electrophysiology, which is kind of what you're describing— you put an electrode down into the tissue itself, get really close to the neuron— now instead of just getting the summed voltage of millions or billions of neurons, you're actually getting the activity of one or or maybe a couple nearby neurons. And that tells you something really interesting because instead of these waves, what you see or hear are neuron spikes. And so you hear like neurons firing or spiking or depolarization. That's what we're talking about. So you see as a neuron becomes active, it does a little pop. It has this big voltage jump and then recovery. And this happens over just a couple of milliseconds. And what's interesting is that some neurons are active for certain types of activity and then otherwise completely silent. Movement. And then sometimes they're active for many types of activities but not others. And so there's this kind of what's called mixed selectivity. They actually pick up certain aspects of different behavior. So to make that concrete, let's imagine that I told you to move your finger or thumb— let's do thumb— to the right. When you move your thumb to the right, there's a set of neurons in your motor cortex that are connected through all these fibers into your thumb that are doing that and that are preparing that movement and executing that movement. Movement. And so if I'm lucky and I put my electrode in the right spot in the motor cortex, I can actually find the neuron that pops every time you— or depolarizes or fires every time you move your thumb to the right. And now I have a little, like, thumb right neuron. And then if I find another one that's a thumb left neuron, and up and down and so on, now if I have a couple hundred electrodes recording from a bunch of different neurons, I can start to mix those together and say, not only can I know when he's going to move to the right, I see that one's— the right one was firing and the up one was firing, he's probably going at a 45-degree angle. You can use simple machine learning methods, or decoders we call them, to decode that thumb movement. This is precisely what groups like BlackRock and Neuralink and the BrainGate studies that go all the way back into the early 2000s have done, is record from the motor cortex of human beings who are paralyzed, unable to move, and then reconstruct the intention or the willed intent of movement that they just can't do because their spinal cord is not allowing signal to pass through.

So the person is still thinking, I want to move my thumb to the right, and, and, and we can pick the— pick up that intention.

Yep, exactly. Yep. So we're just correlating intended behavior. So we have to train these decoders, right? So we tell the person, hey, try to move to the right. They can't. We assume that they do the best they can, and we record the neural activity and we look for neurons that are firing. And then after a while, our decoder learns what the person is trying to do. And then, yeah, exactly, we can recreate through an external device. Could be a cursor on a screen, could be a robot, doesn't really matter. It's easy to turn it into whatever once you have the control signal.

And when it— when we first started going into the brain, I assume this was like in animals first, or were people just diving right into humans?

Animals. Um, but the translation path wasn't that long though. Like, what we— I think so, we see Ebb Fetz, um, at the University of Washington. I think he was a postdoc at the time, I think. Um, but I think that paper was 1969. He taught a monkey to control the firing rate of a single neuron. This actually taught us a couple things. So this was before we could do multicellular recordings. We're just listening to one neuron, and then you have a little meter that's counting how many neuron spikes it sees over time. And he teaches the monkey to— if he can like tip this needle up in the right way, he'll get a juice reward. And so this teaches us a couple things. One, one, that we can actually detect these signals, and two, that they're not only controllable, but in this case, it's sort of abstracted away from any kind of specific behavior. The monkey just learns to control this through like, hey, I want to make that needle move, and over time, the monkey connects that. So 1969 is when we see the first kind of like BCI in the motor neuroprosthetic way we think of BCI.

And it's like a wire that's in the brain? Okay, just with—

exactly.

I'm just picturing electronics at that stage, but I guess it's, it's You could still make thin wires with a little electrode at the end.

Yeah, yeah. Thin wires have been around a long time. I mean, I think where we actually see the biggest jumps in groups like Neuralink and others today is in the commercializability of these things. So they need to last 10 years in the human body. That's an incredibly hard problem. Like making a wire that you stick in the brain for a day and making a wire that you stick in the brain for 10 years and it's, you know, it's going to work and not shock somebody and not break down or be toxic. That's an infinitely harder problem, right? And so, like, you, you mentioned earlier that, like, what Neuralink is doing is aggressive. And what I would actually say is that it is very technically aggressive and it's medically aggressive, but it's scientifically actually very conservative because the science goes back 50-plus years in some ways. And, but it's really the technical aspect of making these things work.

Well, and people ding them for that sometimes. They're like, oh, you're just replicating stuff that's gone on for years and years and years. And, and yeah, I always come down, I think, where you are, where it's— I don't want to put words in your mouth. I, I, I've— it's a blend, right? Because yeah, okay, you've proven some science out, but the machines you did that on were enormous, and the person had to be in a hospital, and there had to be like 8 people managing this process. And then Neuralink, like you said, I mean, it's got this tiny coin that's essentially a consumer electronics device, and I've been with the patients, they're at their house using this thing on their own. And, and, you know, the Um, like sometimes I get it because Neuralink has a big press conference and they're making a big song and dance about what they've done. And, and so people in the field are like, oh, we did that a long time ago. But you know, then I— so I understand their point of view, but then I also feel like they're so into their field that they're discounting all the work that goes into making this thing work in these particular circumstances.

I think wanting wanting groups like Neuralink to push the frontier of science is really fair. I'm with all of those people that like, yeah, let's do more. Let's see where the edge of this really is. And I'm 100% certain that Neuralink feels the same way. I think it's just a completely wrong argument. It's like saying like, you know, a 747 flies overhead or F-18 and people go, wow, holy shit, did you see that? And someone next to them goes, the Wright brothers did that. That they came up— they did came up with that concept years and years ago. It's like, okay, true, technically the concept's been around a long time, but turning that into something like you said, where a person go— can go home, they don't need a team of scientists hiding behind the camera, which is what the academic studies really are.

Yeah.

Um, and racks full of computers and equipment that cost millions of dollars. Like, to shrink that down into something that fits inside the skull is an insanely hard problem.

Yeah. Okay, if, if we Am I fast-forwarding too far if we go from some of the experiments you were talking about to the Utah Array?

Not really. Um, yeah, so if we go from like Ed Fetz, like 1969, and then you start to see, um, so you see tetrodes and, you know, multi-electrode arrays coming on the scene. Um, I think it was 1992, '97, somewhere in there that the Utah Array— I forget exactly— Richard Norman at the University of Utah— no affiliation, by the way, um, two N's in his name, uh, that creates this little— for anyone that hasn't seen it, it looks like a tiny little bed of spikes, but it is, I mean, really tiny. It's 4 by 4 millimeters like that. And that has a 10 by 10 array of spikes that stick out of it. And then you basically stick that into the brain and it's just right in kind of the surface of the cortex. And now you have 100 channels of multi-unit electrophysiology. And so recording single units, single neurons, and you can start to mix together these signals. So if you put it in motor cortex, for example, you can start to decode these kind of fine motor controls. They didn't go straight into humans. They went into monkeys first. So we see groups like Richard Anderson, my former mentor at Caltech, and Andy Schwartz and John Donoghue, these kind of like early big names, Miguel Nicolelis, all around the same time in the '90s and early 2000s start to do these multi-unit recordings in monkeys. Monkeys and start to show monkeys controlling robotic arms, feeding themselves with a robotic arm, or controlling a cursor on a screen. And not too dissimilar from what we eventually see Neuralink doing. And in fact, then the BrainGate group pops up, and this is John Donoghue and a few others. And this is the effort to translate this from monkeys into humans for the first time.

Okay, and John Donoghue was at Brown?

Yep.

Okay, so, and, and that's, that's the beginning of like human human trials.

Yeah, I think that's, that, that's where I think people would go, oh, I recognize that as a BCI.

Okay.

Because it's like, there's a human doing the thing that Neuralink is doing today, not nearly the same level of performance. Like, it's, you know, a lot jankier, a lot more expensive. Um, it's a long ways away from commercialization, but that's where it's like, I recognize that.

All right, I will bastardize this and you can help fix it, but that, that small— when you talk about the Utah Array, I I mean, can we roughly say like fingernail-sized, maybe smaller than a fingernail with spikes on the end and it gets kind of like pneumatically hammered into the, into the brain with some taps, right? In, in, in.

Yeah. Um, although I won't say where or how I heard this, but in many cases the pneumatic applicator just doesn't work that well.

And the surgeon goes, just gives it a shot of a final, a final poke. Okay. Okay. And then, but then to read the signals out, I mean, that thing is small, but then you've got a hole in the brain. And then on top of that, there's like, I always compare it to like a fist. I mean, maybe it's a little, it depends on the size of your fist, I guess. But I mean, there's like this hardware element that I guess they call it a pedestal that's sitting on top of your head. And that is quite clunky. And then that has traditionally they had cables that stretch into a computer. And so, so you've got a person who's maybe paralyzed or has ALS in quite dire circumstances sitting in a hospital room with this thing coming out of their head wired into like a rack of computers. I mean, and that was probably sort of 20, 15 years of, of that.

Yeah, at least. Um, yeah, at least. Yeah. So I think the first, uh, paper was in 2006, but I think the patient was implanted before that, 2003, 2004, 2005, somewhere in there. Um, so Yeah, we're, you know, 20, over 20 years of, we're still doing this. There are still many patients implanted with Utah Race.

Right. And then, and then like you said, for those 20 years, we see these things where people who are paralyzed are controlling robotic arms. They're, they're sending out thoughts, words of what they would like to see happen. Right. And translating some of this. And then when we have, if I remember right, Neuralink comes around, around like 2016. Um, in the— at least from where I was sitting, the push forward is, okay, we've had this thing in this research setting for 20 years, it's been incredibly useful. People haven't really advanced the underlying microelectronics behind this. Um, there wasn't really a business use case to do this, so there wasn't the same kind of money pouring in. And then all of a sudden Brian Johnson gets interested in brains, Elon gets interested in brains, then everybody's interested in brains. And so money floods in, and now we've seen these different attempts to either miniaturize, tweak, different takes on what amounts to Utah, right? In different flavors, sort of.

I think that's sort of true. I would say that like, during that entire time, there was a ton of effort going into making better, thinner electrodes. And so, you know, some of the tech that went into Neuralink came out of Lawrence Livermore National Labs, the electrodes, Vanessa Tolosa's group there. Flip Sabes and Tim Hanson are working on a sewing machine to actually get these things into the brain in a way. So they're literally like sewing these thin wires in. People are working on longevity to make them last a long time. So all that's happening.

So people were moving away from this kind of rigid— these rigid, um, rigid nodes, I guess, and, and the, the Utah Array itself being kind of like this, um, they were a chip. Yeah, they're like a fix to this chip. So people are— they were trying to move away from that.

And I think like right back to very beginning, even in the like late '90s, early 2000s, people knew that that was the way forward. It's a hard engineering problem, they were working on it. But the fact is, like, academia is set up to solve $3 to $5 million problems. There's like chunked in that way. And at some point, these just become very expensive engineering problems, and they're not really as interesting as like fundamental research. And so, and even if they are, like, there's only so much national funding to support it along the way. And translating that into human use is a very expensive, difficult, long proposition. So like, once the Utah Array makes it into humans, that really— actually, a lot of that early success came from DARPA funding. And so you see this big injection that was kind of one big catalyst to like, okay, now we've got electrodes in humans and like we've moved the field forward. And then you see that kind of like fade a bit, and the NIH is driving things forward, but there's really not a lump sum sitting around anywhere that people can say, I'm going to go take those really fine threads that are much more biocompatible, and I'm going to take that delivery method, I'm going to take that processing and that integrated circuit. That didn't exist. There's no way to do that. Neuralink came along and said, hey, look, all the pieces of the puzzle exist. The science is far enough along that we really kind of know it's going to work. That's what I meant by like, it's not aggressive on the science front. Like, they knew it would work from day one. With a large enough investment, we could actually put these together and get a like killer BCI. AI out there into, into humans.

Okay, since day one, the Core Memory Podcast has been supported by the fine people at E1 Ventures. They are a young and ambitious VC firm in Silicon Valley investing in young and ambitious companies and people. Thank you so much to E1 Ventures for all your support. Actually, it's not that dissimilar from SpaceX in a lot of ways. Like, we knew how rockets worked, um, we've been doing them for a long time, there'd been this consumer electronics revolution that the aerospace industry hadn't really—

Control systems, man. Yeah, you know, inverted pendulum.

Yeah. And they just hadn't— they— there was a— it was like stuff got a little bit fixed in time where, where you had space-grade radios. It's like, we can't trust regular radios. And, and software had obviously modernized. I mean, I've always thought of SpaceX as you know, the physics of rockets are just well understood and there's limitations, but they assembled all these modern parts and started with this clean slate and kind of integrated everything in a way the other folks hadn't and modernized it that way.

Yeah, I think that's right. I think Neuralink takes some of those same pages out of the playbook, which is like, let's take the pieces, the engineering problem that we know, if we put it together, we can solve it with enough people, talent, time, time and capital. Yeah, we don't need to revolutionize the neuroscience, but maybe once we have the tool, we can start to push the forefront of neuroscience. And I would argue in some ways they have. I mean, the fastest bitrate decoding that we have with a BCI, a motor BCI today, came from that group. And it's built on science that came from speech decoding groups. So like Sergey Stavisky's group at UC Davis is really pushing on this and making incredible progress as an academic group. And before that, Krishna Shenoy's group at Stanford, just down the road, was doing similar work. So yes, it's building on science that's coming out of the lab, but it's translating it faster and faster, and it's driving it further and further to the edge.

Okay. All right. I want you to explain what Forest does because I have a ton of BCI questions about all these various players. And so I want to, I want to place you guys in where all this fits. I, I'll, I can do my journalism journalist stab at this for a second.

Yeah, yeah, let's do the journalist stuff.

Okay, okay. I mean, it is, it is an implant in the vein of, of there's a surgery required and you're, you're putting— it's like not horribly dissimilar in shape and size a bit from Neuralink. I mean, the circular device going into the skull and, um, and then, but you're— you have a chip on there that is able to to pulse ultrasound through the brain. And so whereas like a Neuralink implant sitting in the motor cortex and it's just a couple millimeters in there and it's just able to see that really specific group of neurons, with the ultrasound you can see, I think it's like 80 to 90% of coverage across the brain and the activity.

We'll get there. We're not quite there yet. Okay. But yeah, we're closer to like 20, 25% now.

20, 25.

But with a single device.

Okay. So, but the hope is that you're not just seeing this tiny clump of neurons. You're seeing— broader-based brain activity. And then instead of focusing in the near term on things like movement and having the motor cortex generate those kinds of outputs, you're looking more at like, how does our brain operate if I'm, if I'm depressed? What does depression look like? And hopefully, again, a lot of these studies end up being okay, somebody's depressed, we have them in this very clinical machine in the hospital. I'm going to study how your, your mental state while we have all these people watching you in this very clinical setting. Instead, you could have an implant that you're wearing for a month, for 2 months. You could be at home. And now hopefully we're getting like a more accurate baseline of what your brain looks like when you're— am I doing okay?

Yes. And you can use it to then stimulate stimulate. So you can actually try to correct these brain states into more normative, healthy ones.

Okay.

And then you have a closed-loop monitor, stimulate— did that have the effect I want, yes or no? Back to the control systems, right? Control and system theory here. But can we drive the brain into more normative, healthy brain states through closed-loop feedback? Which—

okay, and this is gonna open up a whole other can of questions, but, but we've had things like deep brain stimulation, you've had like electroshock therapy. I mean, that's— that —way back, decades. So people didn't understand why electroshock therapy worked basically like at all, from my understanding. But we knew. Yep. So somebody did some experiment somewhere where they started shocking people and something decent happened.

Yeah. Yeah. I mean, I think deep brain stimulators in particular are really interesting because that's like really— it looks like neurotech, right? There's a needle that's going down into the brain. It's stimulating a part of the brain. It's alleviating, alleviating complex systems despite being really pretty like crude as a device just delivering voltage somewhere in the brain. Um, and that goes all the way back into like '50s and '60s. Um, Robert Heath at Tulane University has like this crazy round of experiments. This is like pre-FDA, so like a lot of this was like borderline unethical, just to be clear. Um, or he's just like testing this in like every— you could do that back then. Yeah, you kind of could. Um, you could test it just like every psychological, um, psychiatric indication you could possibly imagine— depression, OCD, bipolar disorder, all of them— and he's finding ways that this is going to work. And so in some ways, like, he is the, the control system, right? He's like, I'm going to stick this thing in and I'm just going to zap the hell out of the brain in this, like, particular person that I think of as a hopeless case. And if it works, it works. If it doesn't, oh well, like, we're just going to try somewhere else and I'll poke it in again somewhere else. And it was very, you know, fly by the seat of your pants compared to how we do things day, um, but made an incredible amount of progress. And deep brain stimulation may have never come to fully exist the way it does. I think we've— there's 200,000-ish people currently implanted with deep brain stimulators for depression, um, primarily for Parkinsonian tremor. Okay, so it's for motor dyskinesia, um, so people that have Parkinson's, you kind of see this sort of like—

yeah, you see those videos where they flick it on and—

yeah, and they'll literally be going like this and they just stop. Stop. Yeah, it's, it's like magic. It's super cool.

And like epilepsy, I think. Yeah.

So, uh, NeuroPace really kind of innovated a lot there where they were doing closed loop for epilepsy. So they're actually like looking for a seizure and when they see it, they try to interrupt it. Um, and in fact, Frank Fisher's, uh, on our impact committee at Fora. So he was the, one of the co-founders and CEO of, uh, NeuroPace, which I think is still the only company to go all the way from a concept of a neurotech device all the way to full commercialization in humans, uh, without getting killed or acquired. Okay. So, so a badge of honor.

It is. And then, but, and then, but in severe mental health cases as well, right?

It's being investigated in deep brain stimulators for depression. It's already approved for OCD. It's being investigated for depression. People are interested in using it for substance use disorder and other disorders of compulsion. So I think that there's still like quite a lot of alpha to be found in deep brain stimulation. And many neurotech companies are aware of this and interested in going after it. After it. And in some ways, you could even say that FOREST is a deep brain stimulator. It's just not the traditional one. We can deliver doses of energy deep into the brain by focusing the ultrasound beam, but we don't— we're not invasive in the sense that we're not actually going into the brain parenchyma, into the brain tissue itself. We're sitting completely outside the brain and also completely outside the dura mater, which is this protective membrane that goes around your brain and really separates the outside world from the central nervous central nervous system kind of lives on its own immune system. You really don't want to mess with that. That's where the risk from brain surgery comes from. Okay. We sit outside that, which means we're almost more like a plastic surgery than a brain surgery.

For people outside of brain land, I mean, I hear this come up all the time. It's like you're either on top of the dura or underneath. And if you're underneath, you're in a whole different world of opportunity and problems. And if you're outside, considered much, much safer. So you're, you're, you're drilling a hole in the skull, the implant's going in, and then you're essentially like sitting on top of these protective layers before you have to start dealing with as much scarring, doing damage to brain tissue, all of that stuff. And then when you're— so when you're talking about your form of stimulation, it is— it's just you're sending in ultrasound pulses. That's right. And you can— you aim them with— with what, like, like software on this chip? That's software on the chip.

Okay. So this is what's called a phaser array. So in the current instantiation of the chip, we have about 9,000— 8,960 elements. Each one of those can vibrate at ultrasonic frequencies. And so you can take this big 2D array, and I can energize one side and the other side at the same time, and then kind of stair-step my way into the middle, and that will create a focus beam. If I do it the opposite, I can create a diverging wave or a plane wave or any shape that I want, and I can holographically construct deep into the brain what, uh, whatever pattern I really to.

And so the idea is like if you're depressed, you have maybe some neural pathways that are overactive or dormant, and, and this is help— helping, um, break up that, that sort of fixation of, of these neurons. Is— I don't know, I just— I'm sure I did a terrible job.

No, no, that's fine. Um, yeah, so actually maybe just like backing up a little bit too, I think one of the things that that really stood out to myself and Tyson, my co-founder, and the entire Forest team, and people are kind of aware of this, is like, these electrodes are amazing. You put them in, the person can control a robotic arm, but there's this really tricky problem where a patient will come to you and say, you know, hey, I really love coming in and working with you guys and controlling the robotic arm, it's fun, but I'm also just depressed, man. Sad. And like, that's really common among people with spinal cord injury. Um, and he's like, can, can the implant do anything for me? And it's like, well, no, you'd either need to have a whole other surgery with a totally different type of investigative device, probably a deep brain simulator. And, you know, that's an added risk, totally different thing. Um, and so there's just not a transferability or scalability of those types of implants. And So kind of going back to the drawing board, when I was finishing up my PhD, I was sort of disillusioned. I'd used everything under the sun. I'd worked with patients. And when you work with patients, unless you're like sociopathic, you're like, man, like you become really motivated to like, I want something that works. And I became disillusioned with EEG, MEG, SEEG, MRI, electrode arrays. And I was just like, what would work? And ultrasound physics are just kind of cool. They go through soft tissue— ultrasound waves go through soft tissue really well, and this is why we can use it to look at an unborn baby, right? Um, and it also turns out that their resolution is diffraction limited by the size of the wave to about hundreds of microns, which is— 100 microns is about the width of a human hair. It's incredibly precise, and it really just hadn't been used in the brain because it doesn't go through bone. Bone. But if you're willing to remove the bone, which the neurotech industry is, then you have access to this whole space. And all of these things that are big societal problems, like we mentioned— depression, OCD, bipolar disorder— um, it's something like 21 million people that have severe and treatment-resistant, meaning they failed 3 or more drugs, forms of those. So these are debilitating, just like, people out of the workforce, they're depressed, they're like borderline suicidal, or so compulsive that they just can't live a normal life. And like, we're just like failing them every day we don't create something like this. And ultrasound felt like such a good match, and there was just a lack of, of research in that area. Um, so the idea is if you can bring ultrasound in, we know that it can monitor, and that was really the work I did at Caltech, was showing that it had enough sensitivity to monitor function of the brain. And then going back to like '50s and '60s, we know that you can then focus that beam to actually upregulate or downregulate neural activity. And it's, it's more of like a bump rather than like a precise thing today. And Forrest wants to take that further. Um, and so like you said, going back to depression, your original question, if you see a biomarker that is suggestive of or correlated with that person's depressive state, can you then stimulate parts of the brain, much of which we're learning from deep brain stimulators, uh, the electrode kind, by the way, as a good starting point? Can we do that in a less invasive way and a way where, like, we can retarget every day? So then if that person that was implanted originally for epilepsy is like, hey, you know, I'm also depressed, we go, oh, we can solve that for you. Yeah, it's a change in software. We don't need any change to the hardware device itself. So it's really a paradigm shift where one device can do many, many things. And that's like actually the change that Forest cares about, like ultrasound, whatever. Those are all just tools. What we want is the one device that can do many things. I see. Okay.

And as far as like learning about this and proving it out, um, you and I talk about things outside of this podcast, so I don't want to, I don't want to overstep on some things that we've talked about in the past. What can we say about about what stage you guys are at, um, with, with, with trials and, and where— what you've learned.

Yeah, so, um, we have two lines of hardware stood up, um, actually much of which was created in this building. Yeah, we should—

we'll give Playground Global a shout out. Where is it? So my podcast studio is in Playground Global's offices, and they're, they've been a backer of the, the company.

Um, no, no, so they're, they are also a venture capital firm, um, but we are actually a nonprofit, so we only accept donations. That being said, one of our anchor donors, Eric Schmidt, is also kind of part of the Playgroundiverse, let's call it. Okay. And so they've been very, very helpful and kind and really helped us out a lot in getting lab space set up and treated us almost like a portfolio company. So shout out. Okay. Okay. So yes.

So some of the Playground, some of the, the Forest team sits not, not too far away from me. Yeah. Yeah. Okay. So there's, so 2 devices.

Yep. 2 devices. And I think we have 5 approved non-significant risk IRB. So these are early feasibility studies in human beings. Um, so I'm not sure how many hours we're up to, but, you know, several hundred, probably a couple hundred hours worth of, of data recording now. Um, and yeah, I can kind of jump into whatever level you want.

Well, I mean, what's like, what's some of them? What have, what have we learned?

What have we seen?

Yeah, is it working? It does work.

Yeah, I, um, um, I won't name names, but actually someone at the company who they're like right at the end of the pipeline, so that means they're the ones that see the data first. Um, I actually wasn't there, which I'm really sad, but apparently she was just like crying at her desk one day, and then people are like, what's going on? It was like, it was the very first data we'd ever collected from humans, and like it just came out, and it was like, boom, it works. And just to give you like a sense there, when we started this, um, company, which again is a nonprofit research org— we can talk about that— um, I would say the majority of people just said like, hey, you know, it's a cool concept, but there's no way you're ever going to get this miniaturized ultrasound on chip to get that level of sensitivity. Because up until that time, like, the stuff we used at Caltech, it's like 2 server racks full of equipment, you know, a million-dollar scanner, and you get out of that 128 channels. But it's really sensitive and very programmable. And everyone's like, yeah, that's great, but if you try to put that in this tiny little thing, it's really hard to make stuff small. Right?

Yeah, because you have to put ultrasound in this tiny device.

Yeah, yeah. And we're like, we're going to do a 9,000-channel array, it's going to be the size of your thumb, and it's going to be more sensitive than any research scanner you've ever seen. People were like, okay bro, I'll believe it when I see it. Um, and so for like some of the very first images we got to just work, and they were by no means optimal, we've improved immensely since then. Um, but yeah, we've detected functional signal in the brain, we've detected— we talked earlier about cortical thalamic loops— we've detected activity deep in the brain and on the surface of the brain in concert. Um, and then we've had clinically meaningful stuff as well. So, um, I'll maybe just like tease this a little bit, but one of the things that we'll want to share with the, the world pretty soon was actually looking at the brain states of people who are in a coma. And we were able to detect this, um, this phenomenon called covert consciousness, where a person is functionally unable to move, disconnected from the outside world. They're in a coma, so their family's asking the kind of usual questions like, hey, are Are they going to wake up? Can they hear me? Should I sit here and talk to them? Is it worth my time? And we know from fMRI, so credit where credit's due, like fMRI literature tells us that in some of those people, you can actually detect brain states that indicate that they can actually hear you and they're interpreting what you're saying. And in some cases can even respond through simple brain activity. So you say like, hey, the first study was actually, you'll like this, was like, if your name is Susan, in, then please imagine playing tennis. Like, try to hit a tennis ball as hard as you can, because she was a tennis player. Yeah. And they would see this big spike in activity at the part of your brain that would be active if you were trying to move your right arm, or whatever her dominant arm was. Um, and so you can create rudimentary communication with someone who's literally in a coma. How long has that been going on? I want to say that— I don't remember exactly— it's like 2004, 2014. Okay.

I feel like this is like one of these big questions. This is like everybody's nightmare, right? Is that you're in a coma— It is. —and you can't communicate out.

And I think the reason that you don't hear as much about this is because it's, like, amazing, but it's not super actionable. And the reason why is because a person who's in a coma might be in a neuro ICU. Taking them off life support or wheeling their life support somehow into an MRI scanner is so difficult. And running that study, it's functional MRI. This isn't like your normal, you know, neighborhood doctor is going to do it. This is like you need a specialized team that's going to do it, analyze the data. It's really— it's a science project. Um, and so although there's proofs of concept, it's not— it's super far away from being kind of standardized clinical care. But here now, for us, for— we have this little tiny device that we can just stick on the head bedside. It's a wearable, but again, only a wearable if you're missing your skull, right? But for many people who are in a coma, if they've had a traumatic brain injury that's put them in the coma in the first place, they'll have a decompressive craniectomy where they actually remove a part of your skull to allow for swelling to happen safely. And during that time, there's an unobstructed view into the brain that's just through soft, you know, skin, scalp, and right into the brain. So we can actually monitor the brain that way. And so we were actually able to detect in, I think, one of the very first patients we had not only the kind of audio listening parts of her brain, but the parts that actually interpret speech active whenever we would talk to her. And it was not us talking, it was a recording. And then interestingly, if you take that same recording and play it backward, so now it's kind of like it's the same volume, same waveform, so mathematically it's kind of the same thing, but it's unintelligible because you can't really understand when something's being played backwards, then that we'd see that signal disappear. So we could actually watch like these speech interpretation areas ramp up every time she was hearing spoken language.

And yeah, okay. And that means like, so she was an active participant on some level then? Yeah, yeah. I mean, it's not like I, I didn't ask that question very well. I mean, if you're, um, If you're playing it backwards and it's sort of nonsense and your brain's not responding, I mean, I would still hear something. Wouldn't I be trying to like decode it?

Or— You might, yeah. But you're— it depends on what area of the brain you're in. So your kind of auditory cortex is just gonna be going crazy because it's still hearing sound and doing its thing. But your language interpretation areas are kind of like, they're downstream of that and they don't really have anything to interpret. So, you know, somewhere in this processing pipeline, things are getting cut down.

So I think that's what I was trying to fish out, if it was like passive or active that—

I think that might turn into a philosophical question. Somewhere in between those. Yeah. Yeah. I mean, I think what it suggests is that on some level, like functionally speaking, she was hearing and interpreting the people around her. And therefore, like, actionably,, for her family, it might make more sense to spend time chatting with her. And not only that, but people who— we know this from these great fMRI studies, and there was a great one in New England Journal of Medicine just last year that kind of concatenated all of them and said, oh, across these 400 or so people that have been in a study like this, um, those that had cognitive motor dissociation, this kind of COVID consciousness effect, um, were— I forget the exact number— but much more likely to wake up and their post-wake-up recovery was much more substantial. So these are— it tells you something actionable about the course of action. I mean, the scary part about this, like you said, the nightmarish part about this is the idea that someone's going to pull the plug while you're sitting there listening to them talk about it, right? And that's the number one cause of death in these folks, because about 25% of people in a coma do have cognitive motor dissociation.

We think, okay, that's what I was going to ask, is, is if you haven't sustained some kind of tremendous brain damage, um, like, do we think that most people in a coma then are, are, are alert to all this?

Yeah, yeah, not most, but 25%. Yeah, good chunk. You're talking thousands of people every year that, um, are at risk of getting a plug pulled that shouldn't.

Like, the positive seems to be that you could flag this to the family, you could have people then spending more time talking to you, reading to you. Obviously, the horrible downside is that you know this and you still can't kind of get things out. Yeah, we—

maybe I'm oversharing, but whatever. We've been internally calling this Project Orpheus because Orpheus in the Greek lore, like in Greek mythology, he goes down into the underworld to rescue his wife who had been, um, uh, had died. And so he's communicating with someone in the underworld. And so that's kind of like where that's coming from. And so we want to eventually get to the point that we can actually allow them to communicate back to us from their coma state. And then later down the road, which is, you know, we're not— haven't really started on this yet, but I maybe jokingly, maybe not, call it Project Lazarus, which is like, can you actually stimulate the brain in a way that would bring that person back? And because now you have this closed-loop feedback, can you listen to the brain, monitor these levels of consciousness on some of our biomarkers of consciousness, and then try to stimulate them to come higher and higher and higher? And there's come some early evidence to suggest that that may be possible. Oh my God.

Well, yeah, do you remember the name of that podcast I sent you about all that?

Yeah, yeah, I listened to— I'm like halfway through.

Okay. Wait, vamp for a second. I'm gonna, I'm gonna look this up. Um, so you're doing, you're doing these studies and, um, now that I feel on the spot, I have nothing to say.

You gotta carry me here. Um, yeah. So I mean, like, this is really just blink, blink, blink.

That's it. Yeah. Okay. So it's just fully recommended. I don't think it's like, as I, when I first recommended it to to— now that I've listened to the whole thing, I'm not sure it's like fully apples to apples because this guy, he basically has this— I'm not going to divulge too much of the story because it's crazy and it has, it has a juicy ending. But, um, he has this terrible reaction to smoking heroin and so it's like this, this like chemical-induced coma. Um, but basically he's in this coma for a really long time and then he comes out and he was like, yeah, I was awake and I saw like everything you 'What are you guys doing to me?' And I just heard people fighting all around me. And, and, um, and so I feel like his is a little bit different because of the circumstances. But yeah, I mean, I was just like, oh my God, if we could, um, if you made progress on this front, it seems like such a fascinating scientific avenue.

Yeah. And I think, you know, just to highlight too, um, because we're this focused research organization, a nonprofit group, And I mentioned earlier, like the real thrust here is one device, many applications. It's not that we're fundamentally only interested in this, but it was pretty exciting that one of the first indications that we were interested in, like boom, right off the bat just showed something so promising. And so I hope that ultrasound-based neurotechnology can help in that space. And at the same time, we're learning, you know, we just partnered with Aria and a group in the UK where we're kind of starting to go down the route for depression to understand and eventually and treat depression. So that's a neurostimulation study. And we're interested in a bunch of other indications as well. I mentioned we have 5 of these IRBs. Each one's a little different in how we use this with really the goal to show like, yeah, not only can it do this thing in coma, but it can do this thing in depression. It can do this other thing.

Okay. Okay. I'm going to ask you what's like maybe an unfair question because I don't think it has a black and white answer and I sort of wanted to, but okay. We've got Neuralink doing surgery into the brain. We've got companies like Precision Precision Neuro doing this layer of electrodes. I think they are under the dura, but they're not going like into the tissue. They're kind of like sitting on the brain. Um, Synchron goes through a blood vessel in your neck, almost like a stent, um, into your brain and kind of hangs out in the blood vessel. Downside is the resolution's not great. Like, like, like everything has this trade-off. Everything, um, We're also in this spot where almost all these devices right now are being used for people in quite dire straits who might have ALS or they might be paralyzed. You know, we're seeing quite far from like the consumer, we're merging with the machine as a general populace era. And, and so I get bombarded with all these different approaches and everybody makes a pretty strong case to me on some level about why why, like, it feels like we're heading to this world where we're treating people who are sick in very specific ways, where you have— you've got this one implant that it's like, okay, well, for your condition, you kind of want the less invasive one, you know. And then for this one, here's a little more serious, we're going to go a little bit deeper. But that also feels like we have almost like too many options, or— and I know it's early days, and so this is all shaking out. But I'm just so curious, like, how you think this plays out. I mean, do you end up with one device that does everything, or do we end up with like 6 different implants for, for 20 different conditions?

Yeah, I, I think we end up with one device that can do everything. Okay. Um, but it's not tomorrow, and that's why, like, I think all of these devices you just mentioned, every one of them has a place right now. So there's this— you mentioned also that there is a trade-off between invasiveness and performance. The fact is, the closer you get to the neuron today with an electrode, when you're talking about electricity, the closer you get to the neuron, the more precise the signal is, because you're getting closer to the origin of the signal itself. The further you get away, so you move out of the deep brain, you move into the cortex, eventually move on top of the cortex like Precision, or you move epidural, or you move into the vasculature like Synchron is doing, you get kind of further and further away. So performance degrades., but you get this invasiveness trade-off or less invasive trade-off. Um, and I think that if you are beholden to that electrophysiology phenomenon, that trade-off is pretty hard to break. You're not going to jump off of that line very easily. Um, so I've been calling, uh, these like generation 1 or level 1 interfaces. These are the first generation of commercialized neurotech that I think are really going to make an impact on people's people's lives. But at the same time, what you're seeing is, if you look a little further down the line, entirely new types of physics and approaches being used— ultrasound being one of them— um, that start to break that mold a little bit and go, we're actually going to put aside the biophysics of electricity for a second and let's look at other mechanisms by which we can detect neural function. And now we can do it with remote sensing. And I think that whether it's ultrasound or it's light or or its mag— magnetic— magnetism. Remote sensing is kind of the way forward. There's this great concept by Marcus Meister called the butcher number, which is kind of a brutal way to say it, but it's the number of neurons that you have to kill. Oh, right. To record from one. And so the Utah Array, which we talked a lot about, has a butcher number of something like 200. So you kill 200 neurons putting this spike into the brain and you record from one of the bottom. So it's fundamentally not scalable. You're going to end up damaging a lot of brain to put a lot of these in.

Definitely not elective or something like that.

Yeah. Um, and then, you know, Neuralink brings that down and Neuropixels, which is another more kind of R&D technology, brings that down to probably single digits. So it's like 3, 4, something like that. Neuropixels is— I don't—

Neuralink's never published theirs, but yeah, I think they're a little secretive on that. Um, the— I don't want to butcher this because I'll get in trouble, um, if I get it wrong, but, you know, I mean, I hear them cite like 99.999% are, are untouched, you know, of the neurons as you're going through. But I don't know how that translates into the butcher number.

Yeah, I mean, yeah, I also will get in trouble if I— so we'll just say that it's good. Yeah, but it's non-zero. And fundamentally, if you need something to scale to whole brain structures, then you need a butcher number of zero. And, um, you know, there are ways of, like you said, like sliding something on top of the brain but kind of like, um, underneath the dura, but there's complications and risk of infection and everything else. And so for me, the long-term, this kind of generation 2 interfaces are departure from traditional electrophysiology. Okay, like, because when you're saying remote sensing, you mean what?

You mean—

I mean something that sits on top of the brain or outside of the brain tissue itself, so you're not you're actually touching the brain tissue and you're sending energy into the brain tissue. And that energy can be light, magnetism, sound, et cetera. Okay.

So you think if we use these different approaches, we will get enough of the information we need to not be in— not need to sit right next to a neuron to do amazing things.

That's right. Yeah. And so I mentioned earlier, there's these 21 million people with psychiatric and cognitive disorders, and that's just your treatment-resistant resistant severe cases. That's like the worst of the worst. Well, 21 million is a lot bigger than the roughly 1.5 million people that are living with severe forms of paralysis. So like, just the opportunity size to help people grows tremendously as you move beyond motor neuroprosthetics. And so, but you need whole brain coverage— or not whole, but a lot of the brain— so we'll do this devices. So either you have a ton of Neuralinks all over the place— and actually, I'm not even sure that would work because you have to deep, um, although there are efforts working on that too. Um, I'll be careful again. Go on. Um, and but, you know, you're going to turn the skull into Swiss cheese trying to do that. I think the Neuralink— we back of the napkin, you know, right now, despite being a technological tour de force, again, nothing to take away from them, but covers about a sixth of 1% of surface cortex. That's to say nothing of all these sulcal folds or deep brain structures that we know are so intimately involved involved in all of these psychiatric disorders. So if you want to get to whole brain, you need something that is stable over long periods of time, covers huge amounts of the brain, and the physics will tell you that electricity just doesn't get you there unless you're literally, you know, just implanting all over the place.

Okay, so is— and this looks like a helmet or a— it eventually is just multiple devices devices?

Well, with remote sensing, that's what's cool, right? So with ultrasound, we can put one device on, you know, one part of the skull here, but we can send out a beam just about anywhere in the skull. And so, and then we can electronically steer this thing and sweep it around. And I mentioned earlier, with one device, we're getting today something like 20% of the human brain by volume. So if you wanted to cover, you know, let's call it 75%, covering 100% is always going to be hard because they're just kind of edge cases and funny geometry. But something like 4 devices is going to get awfully close to everything you could possibly need. Okay.

So kind of, I mean, I don't think it fully worked the way intended, but like Kernel, Kernel kind of like, I mean, conceptually in the ballpark of where some of that might be heading.

Yeah. So, so Kernel, um, they're, so for those that don't know, the Kernel headset is non-invasive, sits on top of the head entirely and uses functional near-infrared spectroscopy. Neuroscopy. So what does that mean? You're sending light down into the brain. And so again, remote sensing, yeah, is kind of going down through the skull and scalp and through the dura and into the brain, and it's scattering all over the place. But near-infrared scatters less than other wavelengths. And so you have what's called an optode. So you have a source that's sending that laser light down in, and then you have, um, a detector. And so most of those photons scatter all and kind of disappear, but some of them happen to scatter back to the detector in this kind of probabilistic, almost banana-shaped curve. And so you can say that summed activity of everything in that curve will tell me something about the neural activity, because the laser light transversing through the brain is being absorbed by hemoglobin, which carries oxygen in your bloodstream, right? And it— you send in two wavelengths, and then you use this concept called the Beer-Lambert law, where one wavelength absorbs hemoglobin more or less than the other one. And so you kind of get the differential between these and you can go, oh, there's more or less oxygenated blood in this banana-shaped curve over time, and you can watch this signal change. Now, why do we care about blood? As neurons are active, they use metabolic resources like oxygen, and that has to be resupplied in the bloodstream. So you get this neural correlate that's sort of temporally blurred, but in this, this case, spatially blurred, and that will recover some correlate of, of human brain function. But it's when you're doing it non-invasively, it's pretty noisy. Good enough.

Yeah. Okay. And okay. And in this future that you're talking about, we can use these techniques possibly to help millions and millions of people in different conditions. If we're talking about a future where we are heading towards a more consumer thing, this is an elective surgery. We're trying to merge with the machines. We're trying to keep up with the AI. Same thing, or that needs to be improved.

Implanted in the brain? It's going to need to be implanted first. Now, there's like a leap of faith that I'm happy to take and say like, maybe, like, you know, every naysayer that says, oh, that will never happen is just proved wrong by history. So I'm not going to be that guy. But there is not an obvious path to a non-invasive high-bandwidth brain-computer interface today. And the physics that we have available to us right now now just suggest that we're very, very far away from it. I'm very confident that the first high-bandwidth interface will be an invasive one. That's not to say it won't translate eventually, but you're going to see that high performance first in an invasive one. Okay.

Um, if you would say never, and then we could like replay this on YouTube a bunch when you're wrong, I would get— that would be good for—

that would be good for me. Not gonna do it. Not gonna do it. Um, and I think that there are like certain tricks that can get you closer, but ultimately like the highest performance interface is going to be the that's, that's there, invasive and closest to the neurons. Okay. So, so I kind of talked about Type 1 electrophysiology, like in there touching the brain. Type 2 interfaces are these remote sensing that buy you whole brain coverage, but they, they make a sacrifice, right? They're not touching the neuron anymore. And so they lose some performance. So in Forrest's case, we also use hemodynamics. So we lose the temporal precision. So we're not a very good motor neuroprosthetic. We're great at, you know, neuropsychiatric and like detecting these things like disorders of consciousness. But, and so we kind of serve that 21 million population, but we don't really serve that 1.5. And then in the future, I'll just tease this a little bit, like what I'm most excited about is where you actually start to combine the hard tech, this remote sensing, new modalities of physics with biologics. And so you start to actually modify the neurons themselves or grow new neurons into the brain. And we are seeing these sorts of things start to come back. And proliferate, I mean, like, efforts in the space that are doing this. But once you combine biologics and hard tech, you can take the precision of Type 1 interfaces and the coverage of Type 2 interfaces, and you can put them together, and now you have like the ultimate high-bandwidth brain-computer interface. And that's actually, to me, the like, you know, decades away, decade away, uh, solution that actually gets you to like fully Ray Kurzweilian.

So, okay, so Science Corporation, Max Hodak, I mean, they're working on, on growing these neurons that are sitting outside of the brain, and then you kind of fuse them into your— or they, you know, they grow axons and dendrites into your tissue and give you this extra horsepower. That, that's, that's one. That would be—

that would, in my mind, so that biohybrid interface is to me one of the first real, like, big attempts at creating a Gen 3, or Type 3 interface, as I've been calling them. Okay, I keep thinking there's— it's old, you know that, right?

No, when you said people have been doing this before, I was—

okay, so the very, like, first, uh, well, is it the very first? It might be the first human implanted BCI was a biohybrid interface. Do you know the story of Phil Kennedy? No, dude, you gotta, you gotta, you gotta find Phil Kennedy. Like, I'm dying to meet this guy. Okay, so Phil Kennedy— find Phil. Uh, you gotta find— hashtag find Phil. Um, I actually think I know where he is. So you, you take, in this case, a glass cone, and you have a gold strip running down into it, and then you put, um, these— you actually grow neurons in the cone itself, and the neurons neurons actually, these grow out into the brain tissue itself. And then you basically have a link, an organic link between the normal brain, these kind of lab-grown neurons, and then it's touching these electrode contacts. So the electrodes never actually go into the brain. You never penetrate the brain. So butcher number zero, probably. Um, and so Phil Kennedy actually came up with this, and I don't know if he was actually the inventor, but he was one one of the pioneers in this space, and this is back in, I think, the '90s. And, um, then he wants to test it on humans, running into multiple kind of roadblocks and hurdles, and eventually decides to actually implant himself. So it goes like full rogue scientist, flies to Mexico, gets this thing implanted, and then actually comes home and gets like 10 days of recording or something like that before it gets infected. So the U.S. healthcare system— no doctor in the U.S. was willing to put the thing in. That's why you have to go to Mexico. Which is why I had to Mexico. But then, uh, they were like, well, it's life-threatening now, so we'll take it out. And so, like, I think he got away with, like, minimal neurological damage. But, like, just the craziest story of, like, rogue scientist implanting himself with, like, a biohybrid interface. He hasn't really talked about this since? Uh, he has, um, but, like, he's not an active voice. Okay.

How are they making synthetic neurons? They're lab-grown. Were they from an animal?

Or I, in, in his case, um, I, they might have been, uh, stem cell derived, but I think they were because it was like pre-IPSC, isn't it?

Yeah, exactly. Yeah. Um, okay. Okay. Now my, my mind is racing. Um, so where, where are you on the, you think human, okay, I guess two questions. So it sounds like you do Do you think at some point in the future humans will not only have the option to have these kinds of implants, but will, will elect to do them?

You do? Yeah, I— it just seems actually like really obvious that the Overton window on this will move when the benefit is there. Today it seems totally crazy because it is totally crazy. Like, why would you have a hole drilled in your skull so you can move a cursor on a screen? Like, I can do that with my mouse and pretty good at it. Yeah. Um, so yeah, of course it seems crazy, but I mean, I don't remember exactly what the numbers are, but the number of people that have elective plastic surgery for what is ultimately a psychological benefit, right? And in most cases it's like a temporary bump and then you kind of level back off to where you started and hedonic treadmill kicks in. Um, but plastic surgery is a huge industry, single-payer driven. So people are paying out of pocket for surgeries that give them a minor psychological psychological bump. So if you have a device that can be implanted one time and not only provide you a minor psychological bump but a permanent knob on tunable mood, better sleep, improved depressive or anxious symptoms, um, name it, right? All of a sudden that seems to actually be like totally reasonable. Um, same thing happened with pacemakers. I love the pacemaker story. In the 1950s, like the very first pacemaker that was put in a human was done sort of fly-by-night in, I think, Sweden. And it was in a— they literally cast it in a, in a, like, a shoe polish tin, just to give you a sense of, like, how advanced this was. And this is a device that beats a person's heart. If it stops working, that person dies. Yeah. And so, and then eventually, over 20 years or so, these become nuclear powered. So you're talking about in the '50s and '60s, it's like, hey, I'm gonna put in a nuclear device in your chest that's going to beat your heart for you. That must have also sounded crazy. Yeah, we don't even think about it now.

It's just like, well, of course. Yeah, I mean, there's a lot that comes once we start messing with our mind, right? I mean, I mean, well, I'm not going to break new ground here on where this science fiction future could go, but, um, you know, just, just us sitting around pleasuring ourselves all day, sitting in a chair with, with happiness flowing over us, and/or, uh, this is actually one of the things that makes me kind of lab.

I mentioned earlier that like I really do believe that the high-performance interfaces start as invasive, right? Which means they do, unless you're like the one-off rogue Phil Kennedy of the world, need to be prescribed by a surgeon. It's a, you know, needs to go through the FDA's most rigorous pre-market authorization approvals. It's a Class 3 device. So there's already actually regulation and ethical boundaries set up around this through the medical system and regulatory system that mean that we'll actually have a little bit of a moratorium, even if we do end up creating non-invasive consumer-type devices, where we'll learn what these devices are capable of and we'll learn where the downsides and the risks are with, like, regulation in place today. So I think when people get, like, really nervous about, like, where's the regulation on this? Like, dude, there's a lot.

I don't know, man. I think Elon— they're gonna pop one. Elon's just gonna chuck that in his head and, uh, it'll be like, you thought I could do a lot of stuff before all of a sudden he's going to be like 65 and just super sharp.

I think he honestly would just play more Polytopia.

I mean, you do. I have noticed like just about everybody who gets one of these Neuralink implants, they end up playing just video games.

It's like, that's because it's what it's good at. Yeah. Right. It's a cursor on a screen, a click and off you go.

Yeah. And then, and where are you on the humans need to do this anyway to compete with AI and, and, um, and, and have a place in this universe.

I'm not like a full doomer in the Silicon Valley or SF scene. Um, but what I would say is that like AI is going to be transformative. Like we're at the elbow on another, um, you know, transformational curve in human history. The agricultural revolution, industrial revolution, computer revolution, now the intelligence revolution. Revolution. So it's going to have a lot of churn and turmoil, and jobs will be changed. I'm not going to say lost, right? It's like, it's always a temporary— like, well, I don't do this anymore, now I do this. Um, and I think that it's like, we're in for a very fascinating, not boring decade.

Well, yeah, because it's so bizarre because we're advancing these artificial minds at this moment in time when we're learning far more about the human mind than we've ever known because all of a sudden we have tons of computing going into brains and all this money. Um, so like which curves these two things go on is super fascinating.

Um, yeah, unfortunately I think that BCIs, because we build these things in atoms, right, these are— it's hardware and it's slow and it's regulated and, um, is not on the same like runaway curve that AI is. So I— when people say like, oh, this is going to be the solution to alignment, that— I think that, that will— that just raises a lot of questions for me. What do you mean by that? How does that actually align AI? And I hosted workshops on this in the past and thought a lot about it. And I do think to stay relevant as a species in a world where intelligence is no longer only human, BCIs have a place, but they're not your one-stop solution.

Do you think that there's all— oh, this is a very hard thing to predict. One thing I, I, this is like some intuitive thing just I have from having covered companies for many, many years in different markets is I look at the amount of money all these BCI startups have raised and the market that they have to play in for the foreseeable future is this medical market. You pointed out that in some cases the number of patients is actually relatively small. I wonder, I feel like there's going to be, there's going to be a natural thing where a lot of these startups go away because they can't, they can't advance their technology and find, find a big enough market for it. I even think though, there's going to be, you know, that even for the most successful companies, that could be a protracted period of time where you're trying to make money off of people who are sick and it has cost you an awful lot to get to this spot. And I, I guess maybe this technology will always be interesting enough at this point that somebody will invest in the next thing. I just wonder if we're going to have like, um, a little bit of a valley of despair when it takes longer than people think to find a multi, like billion dollar gold mine at the end of this. Yeah.

Yeah. This is increasingly the interesting space though, right? Is deep tech investments, which is what we're, you know, VCs would call this and others would call it is they're fundamentally interesting because they're hard. They have long timelines, big capital requirements. But if, if you can make it to the end or even some like modicum of the end, you're interacting with every facet of the human experience, what it literally means to be human, to interact with the brain. So even if you don't capture the whole thing, like, it's like, yeah, we shot for like the full like augmented human merge with AI. And at the end, all we got was like, oh, we solved depression. Right. That's actually not so bad. Yeah. That's a great outcome. Like I had a professor that used to talk about failure modes. So in engineering, we think a lot about failure modes. You know, does a plane falling out of the sky is a bad failure mode, which is why we spend a lot of time building redundancy into the system. The flip side of that is like an escalator. Escalators have the best failure mode. They just become stairs. And like, that's actually kind of the, the world that I see for BCIs where like, yes, big capital investments, but like, if you never try, you'll never know. If we had never scaled large language models or transformers, we would have possibly never figured out that the AI revolution that we're experiencing now. You have to try it to know. And the failure mode is we solve some of the most debilitating diseases on Earth. That to me is fundamentally good for humanity.

That seems okay. Um, all right, last question: downloading brains. Do we download our brains? Do we, do we live eternally in a hard drive in the sky?

Oh, I'm gonna get myself in so much trouble. Why did this have to be the last question? Um, I really love the show Pantheon, by the way. Yes. Pantheon's amazing. It's awesome.

I watched it with my kids and like takes place in Palo Alto, like pretty much like in this office is the headquarters of the— oh yeah. Like, like if you look at where it probably is, it's right here.

Credit where credit's due. Max Hodak told me to watch that, um, a while back before it made it onto Netflix. And I was like dying to find it. And I was watching some like bootleg sketchy copy. And then in the middle of watching it, it came out on Netflix. And the resolution is much better.

However, have you done season 2? Yeah, you have. Okay. So most people I know, I have not. And it's like season 1 is amazing. Then for some reason, because of these weird rights issues, everyone sort of ditched season 2 and now it's on Netflix. And anyway, okay, that's good. All right. Season 2 is good.

It's, it's not as good as 1 until the end. Okay. And then it gets really good.

Okay. But in, in relation to this question, there's some brain downloading.

All right. So you're going to make me answer the question. Yes.

Yes, I will.

Um, all right, I have a contrarian to all my friends take. Um, I fundamentally find brain uploading to be problematic because a brain in a vat is banal. It's not interesting. Like, a brain, it's, it's information content is created by the entropy of the world around it. It is its experience. It is creating connections. It is learning. That is kind of what makes you a human. And so, like, just uploading the weights of a brain onto a hard drive, I think, is missing something. It needs a couple of things. One, I do believe it needs embodiment. Now, that can be silicon-based embodiment, which means that you take that— it's not just the brain, it's the whole central nervous system that you have to emulate, and you have to then connect that to some kind of input/output state. It may not be like a humanoid-type virtual body, it could be just about anything, but it needs to be wired up to something in a way, and that thing needs to live in a physical construct that has enough consistency that it can be modeled. So you have to create a world model. So you need the brain, the body, and the world model. And also, by the way, we're kind of social beings, so we should probably put other agents in there, whether artificial or biologically uploaded intelligences. And now you've basically just reconstructed the world atom for atom. And for me, that is like, sure, but that, um, so many of the time I just feel like nuance is lost here. So, so many of the timelines you hear is like, well, we uploaded a fly brain, we uploaded a worm brain, we uploaded whatever— or uploaded— many of these aren't even functional connectomics, right? But let's say that's mapped it. Yeah, so we've mapped it, and then we're going to get to mapping, um, the, the mouse brain in the next few years, I think, really, with some of the efforts in the space, many of which I know you're aware of. And then human brains are like tangibly there, but far away still, probably, you know, tens of billions of dollars in a decade might get you there.

Just to map a brain in a snapshot of time.

Exactly. And on— and that doesn't take into account the functional aspect. Now you need to actually simulate the connections between those, those neurons. And that's— and then there's kind of something missing there too, where maybe you actually need more resolution than that. You need to know what's happening at every synapse. You need to know the synaptic cleft in every connection. And that's where plasticity happens. So there's probably memory formation that's happening at the molecular level. Do you need to simulate at the molecular level, the atomic level? Like, how deep does the rabbit hole go? We don't know. But let's say that you don't and you just— you have the snapshot, and then you need to make the snapshot functional, and then you need the central nervous system, and then you need the multi-agent and the physics simulation. I, I think it's just like too big. It's, it's big. It's a really, really big problem that we should definitely keep pushing forward on, and I hope I'm wrong. Um, but, um, maybe it's just because I'm— it's not my primary focus. But AI, bro. AI, man. Something something AI, something something profit.

Um, okay, well, this was fascinating. My brain like actually physically hurts. Um, you, you— that was awesome. Thank you so much. Um, thank you so much.

Yeah, it was great to Thanks, Ashlee.

Um, and well, I'm sure we'll have you on again if you will, if you'll—

I'll come back anytime. I've got endless brain thoughts.

I want you to be our resident brain expert. That sounds fun.

Can I get a title though? I need something for my LinkedIn.

We can— yeah, I mean, what do you want? Brain in residence. Core memories, like, uh, like memory, uh, chief memory officer or something. Oh, I just got upgraded. Now we're getting somewhere. Sumner, thank you so much. We will put links to all your endeavors and then, yeah, I think we should, I'll try to go extra on this one and put some links to some of this history in so people can deep dive on some of these studies and different things. Awesome, love that. And don't you, do you have like a, don't you have a YouTube presentation? There's a few out there.

Yeah, yeah, yeah. Foresight Institute hosted me for a few different talks.

Talks and Protocol Labs and where you walk through some of this, right? Yeah. Yeah.

I go through some of the history stuff. I think I mentioned bits and pieces of that today, but yeah, I find it can be very easy to get lost in the swirl of all the things happening when you're trying to build a company or organization that's looking 30, 50 years in the future. And it helps to just look back 30 or 50 years. Yeah. And so I, it's my hobby.

No, I think it's awesome. All right. Thank you so much, man. I appreciate it, and we'll speak again soon.

Sounds good. Later.

The Core Memory Podcast is hosted by me, Ashlee Vance. It is produced by David Nicholson and me. Our theme song is by James Mercer and John Sortland, and the show is edited by the John Sortland. Thanks as always to Brex and A1 Ventures for making this possible. Please visit our Substack, YouTube, and podcast channels to get more of what Core Memory makes. Thanks, y'all.