AI transcript
0:00:06 [MUSIC]
0:00:07 Pushkin.
0:00:34 My guest today is a brain surgeon who also has a P_H_D_ in electrical engineering from M_I_T_, which is to say he is extremely well prepared to figure out how to implant electronic devices in people’s brains, which is what he’s doing. And in fact, as it happens, he’s actually been preparing to do this kind of his whole life.
0:01:03 You know, I I sort of was born into the business. My dad is a neurologist uh who started down his career as an electrical engineer. You know, electrophysiology, clinical neuroscience and um you know, neurology and neurosurgery have been a part of my life forever as far as I can remember. And um you know, brain computer interfaces, the way we talk about them today, didn’t exist in the nineteen eighties. But the fundamentals were there and uh so it that’s been percolating in in some way forever.
0:01:34 I’m Jacob Goldstein and this is What’s Your Problem, the show where I talk to people who are trying to make technological progress. My guest today is Ben Rappaport. He’s the co-founder and chief science officer at Precision Neuroscience. Ben’s problem is this. Can you build a device that allows someone who is paralyzed to use a computer with only their thoughts? And can you do it without sticking needles into their brain?
0:01:55 Before he started Precision, Ben was a co-founder of Neuralink. Neuralink is probably the best known brain computer interface company, and it was founded in two thousand sixteen, right around the moment when modern A_I_ was just emerging. And Ben told me the A_I_ revolution was really what inspired the foundation of Neuralink.
0:02:21 The kind of founding principles of Neuralink were, you know, here’s a here’s a point in time when we’re thinking broadly about how the human brain is gonna interact with artificial intelligence. And if if breakthroughs in artificial intelligence are scaling at an exponential rate, you know, how’s m how’s the human brain gonna keep up with that? How are we gonna keep communicating with uh artificial intelligence in a way that is feasible and productive? [SPEAKER_TURN]
0:02:33 Really different that’s not how can we help people who are paralyzed. That’s a much more sort of cognitive centric. It’s about like the nature of human thought in the context of A_I_ Is. that right? [SPEAKER_TURN]
0:02:54 Yeah Interesting. [SPEAKER_TURN]
0:03:07 Mm-hmm.
0:03:11 Mm-hmm.
0:03:21 Mm-hmm.
0:03:42 Uh in my view required making a few different design decisions uh than what we’d made it at Neuralink, you know, so those were the founding principles of precision. [SPEAKER_TURN]
0:03:54 You leave Neuralink to found precision. Tell me tell me about what you w you know what you’re setting out to create a precision when you’re launching a company. Like what is it that you wanna do and how is it different than what everybody else is doing. [SPEAKER_TURN]
0:04:14 Uh the the the goal then and is today uh to build a safe, scalable, brain computer interface that can become the standard of care in um the treatment of patients, people with a variety of diseases of the brain and nervous system that today are untreatable. That that includes various forms of paralysis and uh inability to communicate. [SPEAKER_TURN]
0:04:22 And tell me about the tell me about the technology. Like w tell me about the thing you’re building and how it’s different from what other people are building. [SPEAKER_TURN]
0:04:43 Uh our philosophy has been that in order for a brain computer interface to really work in the w in the real world and to unlock the potential of this technology for many millions of people, first of course it has to be incredibly safe. We s we see because the view is the term minimally invasive a lot, but really in in my view it has to not damage the brain. [SPEAKER_TURN]
0:04:45 So wh what does that mean in practice? [SPEAKER_TURN]
0:05:15 Yeah, the tissue interface with the electrode involves um kind of like little needles. Well the electrodes are little needles and they penetrate into the brain. And there’s been a lot of innovation in doing it trying to do that very safely, but in my view the the h the most safe version of that is a version that just kind of caresses the brain but doesn’t penetrate it. And it was at first thought you know certainly when we found a precision many people thought that it was not possible to extract high quality signals from the brain without penetrating and we and others have [SPEAKER_TURN]
0:05:38 that in fact it’s not only possible to do but has many advantages. So not that it’s the only way or necessarily better or worse, but from the standpoint of people who have untreatable diseases and already have a very low threshold for damage to the brain not, doing any incremental damage to the brain for, us is very very important. So that that was sort of part one uh of precision. [SPEAKER_TURN]
0:05:46 trade-off? [SPEAKER_TURN]
0:06:16 So there the we always get this question, you know and uh yeah right, no it’s, a absolutely it’s a good question right, and so there’s this false dichotomy I think that um more penetration into the brain equals higher quality signal and if you don’t do that then you somehow s s sacrifice signal quality but, it’s it’s really not a one dimensional as one dimensional as that. If you’re a neuroscientist then there’s a trade-off. If you care about recording from one neuron at a time and you’re studying the behaviour of individual [SPEAKER_TURN]
0:06:46 neurons and you care about that, then you you want intracort what we call intracortical penetrating microelectroids, the ones that can come up to up close to an individual neuron and listen to those individual action potentials. And that that’s something that neuroscientists care about. Um so you you you don’t wanna use the same electrodes that we use for uh precision. But if you but if what you care about is um is treating paralysis uh or disorders of communication, what you care about is stable [SPEAKER_TURN]
0:07:16 high quality signals uh over a long period of time. And in in that area uh arguably just based on the data you know the cortical surface electrodes that we use at precision are at least as good if not better. And I think you know time will tell because there’s a few of these different systems that are now out there in the real world. What’s really exciting is that this has come out of the laboratory, out of animal experiment territory, into human pilot clinical trials that we and Neuralink and Synchron and uh others are engaged in
0:07:18 And that that’s really where it’s at. [SPEAKER_TURN]
0:07:26 So tell me where you are now. I I know you’ve done some amount of experimental work in PEPEL, right? What is the frontier of your work right now? [SPEAKER_TURN]
0:07:56 Yeah, we’ve now um implanted our electrode arrays in almost thirty patients over the last two years. These are pilot studies across four major medical centres in in the U_S_ that are partnering with us. And all of those studies are really uh they’re temporary placements of the electrodes. So they’re they’re studies that are run in patients who have volunteered to have the electrodes placed alongside clinical standard electrodes as part of a uh a neurosurgical procedure that they’re already undergoing. And we’ve been using [SPEAKER_TURN]
0:08:13 those opportunities to basically validate the quality of the electrode activity that we can record on those electrodes, and to demonstrate that uh our algorithms can in fact basically decode intention and thought as intended by uh health essentially healthy volunteers. [SPEAKER_TURN]
0:08:49 So the brain lives in the s in the skull, so it’s a it is a soft tissue that’s uh kind of jelly-like in consistency, and so the best way to gently interface with it is with something also that is soft and flexible. And the service of the brain, as many of us have seen in pictures, is curved or undulating. And so the our electrode array is a uh a thin polymer that’s many times thinner even than a human hair, so [SPEAKER_TURN]
0:09:19 a film, and embedded in that film are tiny little dots of platinum, each one connected to a very very very thin platinum wire. And so the that that film with the tiny little dots of platinum inside can be placed over the brain surface, and it conforms to that uh curved surface. Uh so each of those little platinum electrodes touches the surface of the brain at a very discrete point, and so it can record the [SPEAKER_TURN]
0:09:23 activity from the area of the brain just under that it’s touching basically. [SPEAKER_TURN]
0:09:32 Okay, so so in these trials you put this implant on a patient’s brain uh and then what? [SPEAKER_TURN]
0:10:02 So uh let me describe maybe one of the paradigms that we use at at one of our partner sites. So Iain Cajigas is the neurosurgeon at Penn, who’s our partner, and he is a surgeon who specializes in the treatment of Parkinson’s disease. One of the ways of treating Parkinson’s disease is uh a procedure called deep brain stimulation in which electrodes are placed deep within the brain to stimulate those areas that are responsible for modulating the tremor. Uh Dr. Cajigas among many others performs these
0:10:32 at least a part of them awake in order to make sure effectively that the exact right place is being targeted. And the the brain doesn’t feel pain, and so um it’s it’s po not only possible but beneficial to do these procedures, at least partially awake. So in those procedures we take a p you know basically a fifteen minute window and uh and Dr. Ahegas places the precision electrode directly over the motor cortex uh portion of the motor cortex that controls hand movement. And this has provided, you know, for for us and for the [SPEAKER_TURN]
0:10:39 Mm-hmm. [SPEAKER_TURN]
0:10:41 Uh-huh uh-huh uh-huh. [SPEAKER_TURN]
0:10:50 Of a postage stamp, okay. [SPEAKER_TURN]
0:11:05 Mm-hmm. [SPEAKER_TURN]
0:11:13 So all that critical computation and activity is happening very very close to the surface. And so um good for us, okay. [SPEAKER_TURN]
0:11:32 Yeah. [SPEAKER_TURN]
0:11:33 Okay. [SPEAKER_TURN]
0:11:43 Yeah. [SPEAKER_TURN]
0:12:20 the A_I_ portion of it because this is this is a s the so-called training data. So this this is a calibration phase in which uh our algorithms learn what the brain’s signals to the hand look like in a given patient. So there’s a there’s a characteristic signature electrical signature that happens in the moments before an action is done. And it’s a little bit different in in each person. And learning learning that signature for that person allows us to recognise when the brain is telling the hand to make a particular gesture. [SPEAKER_TURN]
0:12:45 And the fingers are supposed to move in a particular way. When the hand opens it closes. And after about three to five minutes of training we then have a trained algorithm that can recognise um not just movement, but the intention to move. And so we then use the balance of the time that we have with those patients to ask the patient to move and validate that we’re predicting the correct movement and then to to imagine movement without moving. [SPEAKER_TURN]
0:12:58 Right. [SPEAKER_TURN]
0:12:58 Right. [SPEAKER_TURN]
0:13:24 You mentioned that each person’s uh is different in terms of the patterns of of uh neuron activity for each hand motion in this context How. different Is. it like kind of like a southern accent versus a New York accent? Is it like an entirely different language if that kind of metaphor works? [SPEAKER_TURN]
0:13:48 It’s a characteristic metaphor. Uh and and it’s it’s kind of like that. So you know you if you’re trying to learn a new language or a dialect, you know that there are words uh and and you know that they’re spoken in a particular frequency range. So you kind of know what to listen for and you kind of know the cadence. So when there’s a word, you know that’s a word. But you might not know what it means until you listen in to conversation and you’ve seen the context. [SPEAKER_TURN]
0:13:57 Algorithm and put it on my brain because, [SPEAKER_TURN]
0:14:23 Uh it doesn’t work to make a generic algorithm, but that’s an area where there’s been a lot of just fascinating development Um. and so a good example of this is you know Siri works out of the box for most people pretty well right, Right. It it it works right, It. works pretty well and then you need to train it to make it better. And then it listens to you in the background and gets even better. And so that that’s a good that’s a good analogy. [SPEAKER_TURN]
0:14:51 It is possible for us to build you know a translation algorithm that works somewhat out of the box. But we build into a a calibration phase that knows something about the structure of brain signals and how they interact with and relate to movement or speech. And uh that’s what basically allows us to use only relatively small amounts of calibration data. I mean we you know we can do a lot with a small amount of calibration data. [SPEAKER_TURN]
0:15:23 So um I wanna be careful about what I say uh before it happens, but we do anticipate being able to in the very near future extend what are now you know short duration pilot studies that last only the span of time that we have access to the brain within a standard neuro-surgical procedure which, is uh relatively short. We anticipate having ways of extending that with regulatory [SPEAKER_TURN]
0:15:23 Okay.
0:15:39 possible to hopefully many days and weeks uh within the calendar year. And then of course this is all in the service of permanent implants that wirelessly communicate with the outside world. And that will be the basis of our pivotal clinical trial a couple of years hence. [SPEAKER_TURN]
0:15:52 Still to come on the show, Ben and I discussed the possibility of using brain computer interfaces in healthy people. Also the meaning of consciousness.
0:16:07 Just before the break Ben, mentioned that pivotal clinical trial that they’re building up to. And so I asked him what exactly they’re gonna be doing in that trial. [SPEAKER_TURN]
0:16:37 So the the first clinical application is gonna be for the treatment of severe paralysis. Uh and the device will be um an implant that has the electrodes on the brain and an implant within the chest wall that provides power and data transfer to the outside world to communicate with the you know external devices like a computer. And that system will allow for example a p a person with a spinal cord injury really to hold the desk job, that will allow them to operate effectively a word processing program, email, surf the internet, [SPEAKER_TURN]
0:17:07 a zoom conversation, operate an expel a cell spreadsheet, use PowerPoint, have the ability to re-enter the workforce with a level of personal and economic self-sufficiency that allows them to you know cert certain freedoms that they don’t have and that are core to being a part of modern society. That is for us a major goal number one. I’m quite sure that as the technology becomes provenly safe and effective, that other disorders and conditions that are perhaps less [SPEAKER_TURN]
0:17:37 you know will benefit from this and uh and other forms of technology. And part three is there’s a lot that I’m sure that we’re not even imagining right now. You know the the brain computer interface at the end or at least the precision system is really in some ways a platform technology because it’s it it translates the wet and difficult to access delicate you know biological signals you know of the brain into robust digital bit streams and allows us to compute on them in a scalable way. The [SPEAKER_TURN]
0:18:07 computer interface is not a substitute for a keyboard and a mouse. It’s not a substitute for a gestural interface or a uh a voice interface. It’s a another kind of interface with the brain. Just like it was would have been impossible to predict based on the keyboard alone or the you know graphical user interface alone, all of the different applications that uh have emerged. I think as long as we build a safe reliable interface uh and make that responsibly available, kind of the sky’s the limit and I I I can’t even hazard a [SPEAKER_TURN]
0:18:18 asset some of the things that that will come next. So I think there’s a there’s a whole generation of discovery and innovation waiting to happen after we get this across the line into patients to you know to become standard of care. [SPEAKER_TURN]
0:18:26 Could you imagine it being used in in healthy people for you know the computer and the brain application? [SPEAKER_TURN]
0:18:29 Yeah I, could eventually. In a sense I would love that to be the case. [SPEAKER_TURN]
0:18:34 I think I’m ambivalent about that one. T uh tell me tell me why you’d love that to be the case. [SPEAKER_TURN]
0:18:36 Well because it will have meant that we’ve um [SPEAKER_TURN]
0:18:42 Well yes, it’ll mean your thing works really well, it is wildly safe. Yes, that’s true. Yeah. [SPEAKER_TURN]
0:18:56 Mm-hmm. [SPEAKER_TURN]
0:19:03 And that’s that’s the kind of system that we’re trying to engineer. [SPEAKER_TURN]
0:19:12 Yes, f from that point of view it makes perfect sense then, i if if that is true then you then you have built a wildly safe and effective device. [SPEAKER_TURN]
0:19:25 Exactly. So if you and I were having this conversation and you said to me gosh I would love to ha right I mean that would mean that all those doubts had been erased and uh and in order to erase those doubts we have to prove certain things to the world and that’s that’s really our our job. [SPEAKER_TURN]
0:19:33 you would you want if uh if you were healthy would, you want to have your device in your brain if it were safe and effective? [SPEAKER_TURN]
0:19:52 Uh it ha it wouldn’t have to do certain things that uh that that the device can’t do yet. But uh but I wouldn’t definitely wouldn’t rule it out when we get there and I mean it’s like sometimes with technology it’s it’s hard to wrap your mind around what’s gonna happen in a generation right, of two little kids and uh we’re always talking about like should the kids actually get to use an iPhone. [SPEAKER_TURN]
0:20:01 out for as long as you can. So right ’cause it so it’s not exactly a choice right, that’s the that’s the thing. You think like oh an iPhone, great. [SPEAKER_TURN]
0:20:19 By the way I’m, very I’m very very permissive uh and uh yeah. Yeah. Yeah. So the but the reason I bring that up is that you know like our parents could not even have conceived of even that question right, [SPEAKER_TURN]
0:20:40 But I mean the other way to think about that is like you know I’m pro-progress and pro-technology, but like having kids makes me wish iPhones didn’t exist right, Makes me wish f like sure give, ’em a flip phone so they can text their friend and call me if something goes wrong. But I don’t know. But on the other hand I’d make podcasts for a living which, is great. Yeah. [SPEAKER_TURN]
0:21:14 born now knowing how to swipe and navigate the phone interface, right. So my my point is that uh in twenty years it’s gonna be a different conversation. There’s a lot of kids of uh people in the company and they know what we’re doing. You know, my girls know what we’re doing. And their view on the technology is is different. They see it as something that exists. And when you’re born into it, you have kind of a different sense of what’s okay and what’s normal. And that’s the generation that’s that’s growing up today is gonna grow up with great computer interfaces
0:21:15 being a normal thing. [SPEAKER_TURN]
0:21:21 Yeah, maybe your grandkids will feel about brain computer interface as the way your kids feel about iPhones. [SPEAKER_TURN]
0:21:22 It’s gonna have fantastic impact.
0:21:29 We’ll be back in a minute with the lightning round.
0:21:44 Tell me about the metabolic factors limiting performance in marathon runners. [SPEAKER_TURN]
0:21:46 Okay, right. So um
0:22:01 That was a paper that I wrote now more than a decade ago. So I’m a I’m a dedicated marathon runner. I’ve run forty-something uh marathons over twenty plus years. It’s a longer story which we don’t have time for now as to why I wrote that paper. [SPEAKER_TURN]
0:22:04 Give me wh yeah, what’s the short version of that story of why you wrote the paper? [SPEAKER_TURN]
0:22:14 The short version is it shouldn’t be metabolically possible to run a marathon because everybody everybody thinks the paradox is that, you know, you can’t eat enough pasta to get through twenty six miles. [SPEAKER_TURN]
0:22:19 If you do the math, there’s not enough energy stored in the body? [SPEAKER_TURN]
0:22:49 if you do the simple math, there seems to be a paradox that you can’t you can’t eat enough pasta to run the marathon, right? Th everybody thinks you gotta run eat pasta before you run the marathon. It turns out that you can’t really eat enough pasta to run a marathon. So how is it even possible? And the reason it’s possible is that you’re burning some fat as you go. And then everybody knows that there’s this phenomenon of hitting the wall where you you know many runners collapse or uh have a a major impact at some point, you know, along the way usually about two thirds of the way through the race where they just can’t keep going or can’t keep going at the same [SPEAKER_TURN]
0:22:55 Mm-hmm. Uh-huh.
0:22:58 Uh-huh, uh-huh.
0:23:19 Uh-huh.
0:23:23 That’s one of the core rate-limiting metabolic factors in in marathon writing. [SPEAKER_TURN]
0:23:29 And so the I mean so what was it that you figured out that got published in whatever it was P_L_O_S_? [SPEAKER_TURN]
0:23:36 Uh yes, I figured that out. Uh-huh And. I think I think I figured out how to how to model that mathematically. [SPEAKER_TURN]
0:23:57 Well I I I I I learned how to face myself in a more quantitative way. And uh and I l I learned how to how to structure my pre-race diet and my training diet in a way that was much better than I had in in the years before that. [SPEAKER_TURN]
0:23:59 Did you get faster? [SPEAKER_TURN]
0:24:21 I got ma I got significantly faster. Yeah, I run a bunch of sub three hour marathons around the time I figured that all out. And for a period of time, I don’t know if it’s still the case, but maybe embarrassingly, that was my it still is I think my only single author paper. And for a period of time, it was I h most cited paper So. that’s so hard. [SPEAKER_TURN]
0:24:35 And basically says this proves that quantum is not a complete description of reality ’cause there’s no way it could be true. Uh and he was wrong, right? Um. [SPEAKER_TURN]
0:24:50 What’s one tip that comes out of that? Like do I is there like a model I could plug in? I ran my first marathon this year. I did not know about your paper. Um is there something you can tell me just qualitatively from it that I’m doing wrong? [SPEAKER_TURN]
0:24:57 Yeah, take a look. Uh there’s a there’s a little formula there basically that allows the average person to estimate their optimal marathon pace. [SPEAKER_TURN]
0:25:01 Um Boston marathon or New York marathon? What do you like better? [SPEAKER_TURN]
0:25:29 Well you know I’m a native I’ve run both many times uh I’ve run Boston for the last twenty four years consecutively and I I’ve run New York I think I forget now how many times but more than ten and uh I love them both and I’m not gonna go I’m not gonna say in public which one I love more but they’re very different uh they’re very different and uh yeah, that’s all that’s all that’s all I’ll say but they’re wonderful races and uh lot of special things about both. [SPEAKER_TURN]
0:25:36 What is one thing we don’t understand about the brain that you wish we understood? [SPEAKER_TURN]
0:26:08 So the question of what is consciousness I think is a been a big one in philosophy and neuroscience for a long long time, right. You know I I think that the tools of brain computer interfaces are probably have already given but certainly will be giving us in the next couple of years um ways to answer that in a really rigorous and quantitative way. And not just that, but I think to have an impact in disorders of consciousness. And uh so I think that’s an area where brain computer interfaces are gonna have a perhaps
0:26:10 and surprisingly major impact.
0:26:16 What’s a disorder of consciousness I? don’t think I know that phrase what, like what does that mean? [SPEAKER_TURN]
0:26:28 Well uh you know I think many people are familiar with the coma right, so there’ll be people who are alive but not commencementist in the in the ways that you and I are when we’re talking. Th that’s just a dramatic example of that. [SPEAKER_TURN]
0:26:39 Has the work you’ve done I mean either as a as a brain surgeon or as in in developing brain computer interfaces, how has that changed the way you think about consciousness? If it has. [SPEAKER_TURN]
0:26:47 Uh I’m not sure it has yet, but uh at least not in a ways that I wanna talk about it in public, but uh I mean watch this space carefully. [SPEAKER_TURN]
0:26:56 Say one more thing about that. That’s th it’s very intriguing to me. I I feel like there’s something you’re thinking that you’re not saying. [SPEAKER_TURN]
0:27:26 I think so so a lot of it is public uh I think in a really really interesting way. So uh I’d highlight some recent work or recently published work by um you know Nico Schiff and others demonstrating that some people who seem to be in a minimally conscious state actually have the ability to communicate if you give them the tools to do so. And that just has profound implications uh for the diagnosis of certain types of severe brain injury, for prognosticating you know the subsequent course
0:27:56 of people who have such injuries and all kinds of philosophical, ethical and really just most importantly practical aspects of um how do we take care of people with that kind of severe brain injury. Many of whom pose tremendously difficult questions um to family and caregivers uh who can’t predict what’s gonna happen next and and can’t communicate with their loved ones and there’s always this question in such situations you, know, is that person the person we knew still there alright, and will that person come
0:28:25 back, so to speak, uh or not. And uh answering that question is this one aspect of getting at what is consciousness and how does it fluctuate and how do we quantify it and how do we create or restore it when it’s lost or damaged. So you know, that has been the realm of philosophy for most of human history and um I think it is very exciting for me now that that’s that’s changed in the last several years and I do think that [SPEAKER_TURN]
0:28:33 technology of brain computer interfaces is gonna have an impact in making some of the discoveries that have come to light actionable. [SPEAKER_TURN]
0:29:01 Ben Rappaport is the co-founder and chief science officer at Precision Neuroscience. Today’s show was produced by Gabriel Hunter-Cheng. It was edited by Lydia Jean Cott and engineered by Sarah Boughere. You can email us at problem@pushkin.fm I’m Jacob Goldstein and we’ll be back next week with another episode of What’s Your Problem?
0:00:07 Pushkin.
0:00:34 My guest today is a brain surgeon who also has a P_H_D_ in electrical engineering from M_I_T_, which is to say he is extremely well prepared to figure out how to implant electronic devices in people’s brains, which is what he’s doing. And in fact, as it happens, he’s actually been preparing to do this kind of his whole life.
0:01:03 You know, I I sort of was born into the business. My dad is a neurologist uh who started down his career as an electrical engineer. You know, electrophysiology, clinical neuroscience and um you know, neurology and neurosurgery have been a part of my life forever as far as I can remember. And um you know, brain computer interfaces, the way we talk about them today, didn’t exist in the nineteen eighties. But the fundamentals were there and uh so it that’s been percolating in in some way forever.
0:01:34 I’m Jacob Goldstein and this is What’s Your Problem, the show where I talk to people who are trying to make technological progress. My guest today is Ben Rappaport. He’s the co-founder and chief science officer at Precision Neuroscience. Ben’s problem is this. Can you build a device that allows someone who is paralyzed to use a computer with only their thoughts? And can you do it without sticking needles into their brain?
0:01:55 Before he started Precision, Ben was a co-founder of Neuralink. Neuralink is probably the best known brain computer interface company, and it was founded in two thousand sixteen, right around the moment when modern A_I_ was just emerging. And Ben told me the A_I_ revolution was really what inspired the foundation of Neuralink.
0:02:21 The kind of founding principles of Neuralink were, you know, here’s a here’s a point in time when we’re thinking broadly about how the human brain is gonna interact with artificial intelligence. And if if breakthroughs in artificial intelligence are scaling at an exponential rate, you know, how’s m how’s the human brain gonna keep up with that? How are we gonna keep communicating with uh artificial intelligence in a way that is feasible and productive? [SPEAKER_TURN]
0:02:33 Really different that’s not how can we help people who are paralyzed. That’s a much more sort of cognitive centric. It’s about like the nature of human thought in the context of A_I_ Is. that right? [SPEAKER_TURN]
0:02:54 Yeah Interesting. [SPEAKER_TURN]
0:03:07 Mm-hmm.
0:03:11 Mm-hmm.
0:03:21 Mm-hmm.
0:03:42 Uh in my view required making a few different design decisions uh than what we’d made it at Neuralink, you know, so those were the founding principles of precision. [SPEAKER_TURN]
0:03:54 You leave Neuralink to found precision. Tell me tell me about what you w you know what you’re setting out to create a precision when you’re launching a company. Like what is it that you wanna do and how is it different than what everybody else is doing. [SPEAKER_TURN]
0:04:14 Uh the the the goal then and is today uh to build a safe, scalable, brain computer interface that can become the standard of care in um the treatment of patients, people with a variety of diseases of the brain and nervous system that today are untreatable. That that includes various forms of paralysis and uh inability to communicate. [SPEAKER_TURN]
0:04:22 And tell me about the tell me about the technology. Like w tell me about the thing you’re building and how it’s different from what other people are building. [SPEAKER_TURN]
0:04:43 Uh our philosophy has been that in order for a brain computer interface to really work in the w in the real world and to unlock the potential of this technology for many millions of people, first of course it has to be incredibly safe. We s we see because the view is the term minimally invasive a lot, but really in in my view it has to not damage the brain. [SPEAKER_TURN]
0:04:45 So wh what does that mean in practice? [SPEAKER_TURN]
0:05:15 Yeah, the tissue interface with the electrode involves um kind of like little needles. Well the electrodes are little needles and they penetrate into the brain. And there’s been a lot of innovation in doing it trying to do that very safely, but in my view the the h the most safe version of that is a version that just kind of caresses the brain but doesn’t penetrate it. And it was at first thought you know certainly when we found a precision many people thought that it was not possible to extract high quality signals from the brain without penetrating and we and others have [SPEAKER_TURN]
0:05:38 that in fact it’s not only possible to do but has many advantages. So not that it’s the only way or necessarily better or worse, but from the standpoint of people who have untreatable diseases and already have a very low threshold for damage to the brain not, doing any incremental damage to the brain for, us is very very important. So that that was sort of part one uh of precision. [SPEAKER_TURN]
0:05:46 trade-off? [SPEAKER_TURN]
0:06:16 So there the we always get this question, you know and uh yeah right, no it’s, a absolutely it’s a good question right, and so there’s this false dichotomy I think that um more penetration into the brain equals higher quality signal and if you don’t do that then you somehow s s sacrifice signal quality but, it’s it’s really not a one dimensional as one dimensional as that. If you’re a neuroscientist then there’s a trade-off. If you care about recording from one neuron at a time and you’re studying the behaviour of individual [SPEAKER_TURN]
0:06:46 neurons and you care about that, then you you want intracort what we call intracortical penetrating microelectroids, the ones that can come up to up close to an individual neuron and listen to those individual action potentials. And that that’s something that neuroscientists care about. Um so you you you don’t wanna use the same electrodes that we use for uh precision. But if you but if what you care about is um is treating paralysis uh or disorders of communication, what you care about is stable [SPEAKER_TURN]
0:07:16 high quality signals uh over a long period of time. And in in that area uh arguably just based on the data you know the cortical surface electrodes that we use at precision are at least as good if not better. And I think you know time will tell because there’s a few of these different systems that are now out there in the real world. What’s really exciting is that this has come out of the laboratory, out of animal experiment territory, into human pilot clinical trials that we and Neuralink and Synchron and uh others are engaged in
0:07:18 And that that’s really where it’s at. [SPEAKER_TURN]
0:07:26 So tell me where you are now. I I know you’ve done some amount of experimental work in PEPEL, right? What is the frontier of your work right now? [SPEAKER_TURN]
0:07:56 Yeah, we’ve now um implanted our electrode arrays in almost thirty patients over the last two years. These are pilot studies across four major medical centres in in the U_S_ that are partnering with us. And all of those studies are really uh they’re temporary placements of the electrodes. So they’re they’re studies that are run in patients who have volunteered to have the electrodes placed alongside clinical standard electrodes as part of a uh a neurosurgical procedure that they’re already undergoing. And we’ve been using [SPEAKER_TURN]
0:08:13 those opportunities to basically validate the quality of the electrode activity that we can record on those electrodes, and to demonstrate that uh our algorithms can in fact basically decode intention and thought as intended by uh health essentially healthy volunteers. [SPEAKER_TURN]
0:08:49 So the brain lives in the s in the skull, so it’s a it is a soft tissue that’s uh kind of jelly-like in consistency, and so the best way to gently interface with it is with something also that is soft and flexible. And the service of the brain, as many of us have seen in pictures, is curved or undulating. And so the our electrode array is a uh a thin polymer that’s many times thinner even than a human hair, so [SPEAKER_TURN]
0:09:19 a film, and embedded in that film are tiny little dots of platinum, each one connected to a very very very thin platinum wire. And so the that that film with the tiny little dots of platinum inside can be placed over the brain surface, and it conforms to that uh curved surface. Uh so each of those little platinum electrodes touches the surface of the brain at a very discrete point, and so it can record the [SPEAKER_TURN]
0:09:23 activity from the area of the brain just under that it’s touching basically. [SPEAKER_TURN]
0:09:32 Okay, so so in these trials you put this implant on a patient’s brain uh and then what? [SPEAKER_TURN]
0:10:02 So uh let me describe maybe one of the paradigms that we use at at one of our partner sites. So Iain Cajigas is the neurosurgeon at Penn, who’s our partner, and he is a surgeon who specializes in the treatment of Parkinson’s disease. One of the ways of treating Parkinson’s disease is uh a procedure called deep brain stimulation in which electrodes are placed deep within the brain to stimulate those areas that are responsible for modulating the tremor. Uh Dr. Cajigas among many others performs these
0:10:32 at least a part of them awake in order to make sure effectively that the exact right place is being targeted. And the the brain doesn’t feel pain, and so um it’s it’s po not only possible but beneficial to do these procedures, at least partially awake. So in those procedures we take a p you know basically a fifteen minute window and uh and Dr. Ahegas places the precision electrode directly over the motor cortex uh portion of the motor cortex that controls hand movement. And this has provided, you know, for for us and for the [SPEAKER_TURN]
0:10:39 Mm-hmm. [SPEAKER_TURN]
0:10:41 Uh-huh uh-huh uh-huh. [SPEAKER_TURN]
0:10:50 Of a postage stamp, okay. [SPEAKER_TURN]
0:11:05 Mm-hmm. [SPEAKER_TURN]
0:11:13 So all that critical computation and activity is happening very very close to the surface. And so um good for us, okay. [SPEAKER_TURN]
0:11:32 Yeah. [SPEAKER_TURN]
0:11:33 Okay. [SPEAKER_TURN]
0:11:43 Yeah. [SPEAKER_TURN]
0:12:20 the A_I_ portion of it because this is this is a s the so-called training data. So this this is a calibration phase in which uh our algorithms learn what the brain’s signals to the hand look like in a given patient. So there’s a there’s a characteristic signature electrical signature that happens in the moments before an action is done. And it’s a little bit different in in each person. And learning learning that signature for that person allows us to recognise when the brain is telling the hand to make a particular gesture. [SPEAKER_TURN]
0:12:45 And the fingers are supposed to move in a particular way. When the hand opens it closes. And after about three to five minutes of training we then have a trained algorithm that can recognise um not just movement, but the intention to move. And so we then use the balance of the time that we have with those patients to ask the patient to move and validate that we’re predicting the correct movement and then to to imagine movement without moving. [SPEAKER_TURN]
0:12:58 Right. [SPEAKER_TURN]
0:12:58 Right. [SPEAKER_TURN]
0:13:24 You mentioned that each person’s uh is different in terms of the patterns of of uh neuron activity for each hand motion in this context How. different Is. it like kind of like a southern accent versus a New York accent? Is it like an entirely different language if that kind of metaphor works? [SPEAKER_TURN]
0:13:48 It’s a characteristic metaphor. Uh and and it’s it’s kind of like that. So you know you if you’re trying to learn a new language or a dialect, you know that there are words uh and and you know that they’re spoken in a particular frequency range. So you kind of know what to listen for and you kind of know the cadence. So when there’s a word, you know that’s a word. But you might not know what it means until you listen in to conversation and you’ve seen the context. [SPEAKER_TURN]
0:13:57 Algorithm and put it on my brain because, [SPEAKER_TURN]
0:14:23 Uh it doesn’t work to make a generic algorithm, but that’s an area where there’s been a lot of just fascinating development Um. and so a good example of this is you know Siri works out of the box for most people pretty well right, Right. It it it works right, It. works pretty well and then you need to train it to make it better. And then it listens to you in the background and gets even better. And so that that’s a good that’s a good analogy. [SPEAKER_TURN]
0:14:51 It is possible for us to build you know a translation algorithm that works somewhat out of the box. But we build into a a calibration phase that knows something about the structure of brain signals and how they interact with and relate to movement or speech. And uh that’s what basically allows us to use only relatively small amounts of calibration data. I mean we you know we can do a lot with a small amount of calibration data. [SPEAKER_TURN]
0:15:23 So um I wanna be careful about what I say uh before it happens, but we do anticipate being able to in the very near future extend what are now you know short duration pilot studies that last only the span of time that we have access to the brain within a standard neuro-surgical procedure which, is uh relatively short. We anticipate having ways of extending that with regulatory [SPEAKER_TURN]
0:15:23 Okay.
0:15:39 possible to hopefully many days and weeks uh within the calendar year. And then of course this is all in the service of permanent implants that wirelessly communicate with the outside world. And that will be the basis of our pivotal clinical trial a couple of years hence. [SPEAKER_TURN]
0:15:52 Still to come on the show, Ben and I discussed the possibility of using brain computer interfaces in healthy people. Also the meaning of consciousness.
0:16:07 Just before the break Ben, mentioned that pivotal clinical trial that they’re building up to. And so I asked him what exactly they’re gonna be doing in that trial. [SPEAKER_TURN]
0:16:37 So the the first clinical application is gonna be for the treatment of severe paralysis. Uh and the device will be um an implant that has the electrodes on the brain and an implant within the chest wall that provides power and data transfer to the outside world to communicate with the you know external devices like a computer. And that system will allow for example a p a person with a spinal cord injury really to hold the desk job, that will allow them to operate effectively a word processing program, email, surf the internet, [SPEAKER_TURN]
0:17:07 a zoom conversation, operate an expel a cell spreadsheet, use PowerPoint, have the ability to re-enter the workforce with a level of personal and economic self-sufficiency that allows them to you know cert certain freedoms that they don’t have and that are core to being a part of modern society. That is for us a major goal number one. I’m quite sure that as the technology becomes provenly safe and effective, that other disorders and conditions that are perhaps less [SPEAKER_TURN]
0:17:37 you know will benefit from this and uh and other forms of technology. And part three is there’s a lot that I’m sure that we’re not even imagining right now. You know the the brain computer interface at the end or at least the precision system is really in some ways a platform technology because it’s it it translates the wet and difficult to access delicate you know biological signals you know of the brain into robust digital bit streams and allows us to compute on them in a scalable way. The [SPEAKER_TURN]
0:18:07 computer interface is not a substitute for a keyboard and a mouse. It’s not a substitute for a gestural interface or a uh a voice interface. It’s a another kind of interface with the brain. Just like it was would have been impossible to predict based on the keyboard alone or the you know graphical user interface alone, all of the different applications that uh have emerged. I think as long as we build a safe reliable interface uh and make that responsibly available, kind of the sky’s the limit and I I I can’t even hazard a [SPEAKER_TURN]
0:18:18 asset some of the things that that will come next. So I think there’s a there’s a whole generation of discovery and innovation waiting to happen after we get this across the line into patients to you know to become standard of care. [SPEAKER_TURN]
0:18:26 Could you imagine it being used in in healthy people for you know the computer and the brain application? [SPEAKER_TURN]
0:18:29 Yeah I, could eventually. In a sense I would love that to be the case. [SPEAKER_TURN]
0:18:34 I think I’m ambivalent about that one. T uh tell me tell me why you’d love that to be the case. [SPEAKER_TURN]
0:18:36 Well because it will have meant that we’ve um [SPEAKER_TURN]
0:18:42 Well yes, it’ll mean your thing works really well, it is wildly safe. Yes, that’s true. Yeah. [SPEAKER_TURN]
0:18:56 Mm-hmm. [SPEAKER_TURN]
0:19:03 And that’s that’s the kind of system that we’re trying to engineer. [SPEAKER_TURN]
0:19:12 Yes, f from that point of view it makes perfect sense then, i if if that is true then you then you have built a wildly safe and effective device. [SPEAKER_TURN]
0:19:25 Exactly. So if you and I were having this conversation and you said to me gosh I would love to ha right I mean that would mean that all those doubts had been erased and uh and in order to erase those doubts we have to prove certain things to the world and that’s that’s really our our job. [SPEAKER_TURN]
0:19:33 you would you want if uh if you were healthy would, you want to have your device in your brain if it were safe and effective? [SPEAKER_TURN]
0:19:52 Uh it ha it wouldn’t have to do certain things that uh that that the device can’t do yet. But uh but I wouldn’t definitely wouldn’t rule it out when we get there and I mean it’s like sometimes with technology it’s it’s hard to wrap your mind around what’s gonna happen in a generation right, of two little kids and uh we’re always talking about like should the kids actually get to use an iPhone. [SPEAKER_TURN]
0:20:01 out for as long as you can. So right ’cause it so it’s not exactly a choice right, that’s the that’s the thing. You think like oh an iPhone, great. [SPEAKER_TURN]
0:20:19 By the way I’m, very I’m very very permissive uh and uh yeah. Yeah. Yeah. So the but the reason I bring that up is that you know like our parents could not even have conceived of even that question right, [SPEAKER_TURN]
0:20:40 But I mean the other way to think about that is like you know I’m pro-progress and pro-technology, but like having kids makes me wish iPhones didn’t exist right, Makes me wish f like sure give, ’em a flip phone so they can text their friend and call me if something goes wrong. But I don’t know. But on the other hand I’d make podcasts for a living which, is great. Yeah. [SPEAKER_TURN]
0:21:14 born now knowing how to swipe and navigate the phone interface, right. So my my point is that uh in twenty years it’s gonna be a different conversation. There’s a lot of kids of uh people in the company and they know what we’re doing. You know, my girls know what we’re doing. And their view on the technology is is different. They see it as something that exists. And when you’re born into it, you have kind of a different sense of what’s okay and what’s normal. And that’s the generation that’s that’s growing up today is gonna grow up with great computer interfaces
0:21:15 being a normal thing. [SPEAKER_TURN]
0:21:21 Yeah, maybe your grandkids will feel about brain computer interface as the way your kids feel about iPhones. [SPEAKER_TURN]
0:21:22 It’s gonna have fantastic impact.
0:21:29 We’ll be back in a minute with the lightning round.
0:21:44 Tell me about the metabolic factors limiting performance in marathon runners. [SPEAKER_TURN]
0:21:46 Okay, right. So um
0:22:01 That was a paper that I wrote now more than a decade ago. So I’m a I’m a dedicated marathon runner. I’ve run forty-something uh marathons over twenty plus years. It’s a longer story which we don’t have time for now as to why I wrote that paper. [SPEAKER_TURN]
0:22:04 Give me wh yeah, what’s the short version of that story of why you wrote the paper? [SPEAKER_TURN]
0:22:14 The short version is it shouldn’t be metabolically possible to run a marathon because everybody everybody thinks the paradox is that, you know, you can’t eat enough pasta to get through twenty six miles. [SPEAKER_TURN]
0:22:19 If you do the math, there’s not enough energy stored in the body? [SPEAKER_TURN]
0:22:49 if you do the simple math, there seems to be a paradox that you can’t you can’t eat enough pasta to run the marathon, right? Th everybody thinks you gotta run eat pasta before you run the marathon. It turns out that you can’t really eat enough pasta to run a marathon. So how is it even possible? And the reason it’s possible is that you’re burning some fat as you go. And then everybody knows that there’s this phenomenon of hitting the wall where you you know many runners collapse or uh have a a major impact at some point, you know, along the way usually about two thirds of the way through the race where they just can’t keep going or can’t keep going at the same [SPEAKER_TURN]
0:22:55 Mm-hmm. Uh-huh.
0:22:58 Uh-huh, uh-huh.
0:23:19 Uh-huh.
0:23:23 That’s one of the core rate-limiting metabolic factors in in marathon writing. [SPEAKER_TURN]
0:23:29 And so the I mean so what was it that you figured out that got published in whatever it was P_L_O_S_? [SPEAKER_TURN]
0:23:36 Uh yes, I figured that out. Uh-huh And. I think I think I figured out how to how to model that mathematically. [SPEAKER_TURN]
0:23:57 Well I I I I I learned how to face myself in a more quantitative way. And uh and I l I learned how to how to structure my pre-race diet and my training diet in a way that was much better than I had in in the years before that. [SPEAKER_TURN]
0:23:59 Did you get faster? [SPEAKER_TURN]
0:24:21 I got ma I got significantly faster. Yeah, I run a bunch of sub three hour marathons around the time I figured that all out. And for a period of time, I don’t know if it’s still the case, but maybe embarrassingly, that was my it still is I think my only single author paper. And for a period of time, it was I h most cited paper So. that’s so hard. [SPEAKER_TURN]
0:24:35 And basically says this proves that quantum is not a complete description of reality ’cause there’s no way it could be true. Uh and he was wrong, right? Um. [SPEAKER_TURN]
0:24:50 What’s one tip that comes out of that? Like do I is there like a model I could plug in? I ran my first marathon this year. I did not know about your paper. Um is there something you can tell me just qualitatively from it that I’m doing wrong? [SPEAKER_TURN]
0:24:57 Yeah, take a look. Uh there’s a there’s a little formula there basically that allows the average person to estimate their optimal marathon pace. [SPEAKER_TURN]
0:25:01 Um Boston marathon or New York marathon? What do you like better? [SPEAKER_TURN]
0:25:29 Well you know I’m a native I’ve run both many times uh I’ve run Boston for the last twenty four years consecutively and I I’ve run New York I think I forget now how many times but more than ten and uh I love them both and I’m not gonna go I’m not gonna say in public which one I love more but they’re very different uh they’re very different and uh yeah, that’s all that’s all that’s all I’ll say but they’re wonderful races and uh lot of special things about both. [SPEAKER_TURN]
0:25:36 What is one thing we don’t understand about the brain that you wish we understood? [SPEAKER_TURN]
0:26:08 So the question of what is consciousness I think is a been a big one in philosophy and neuroscience for a long long time, right. You know I I think that the tools of brain computer interfaces are probably have already given but certainly will be giving us in the next couple of years um ways to answer that in a really rigorous and quantitative way. And not just that, but I think to have an impact in disorders of consciousness. And uh so I think that’s an area where brain computer interfaces are gonna have a perhaps
0:26:10 and surprisingly major impact.
0:26:16 What’s a disorder of consciousness I? don’t think I know that phrase what, like what does that mean? [SPEAKER_TURN]
0:26:28 Well uh you know I think many people are familiar with the coma right, so there’ll be people who are alive but not commencementist in the in the ways that you and I are when we’re talking. Th that’s just a dramatic example of that. [SPEAKER_TURN]
0:26:39 Has the work you’ve done I mean either as a as a brain surgeon or as in in developing brain computer interfaces, how has that changed the way you think about consciousness? If it has. [SPEAKER_TURN]
0:26:47 Uh I’m not sure it has yet, but uh at least not in a ways that I wanna talk about it in public, but uh I mean watch this space carefully. [SPEAKER_TURN]
0:26:56 Say one more thing about that. That’s th it’s very intriguing to me. I I feel like there’s something you’re thinking that you’re not saying. [SPEAKER_TURN]
0:27:26 I think so so a lot of it is public uh I think in a really really interesting way. So uh I’d highlight some recent work or recently published work by um you know Nico Schiff and others demonstrating that some people who seem to be in a minimally conscious state actually have the ability to communicate if you give them the tools to do so. And that just has profound implications uh for the diagnosis of certain types of severe brain injury, for prognosticating you know the subsequent course
0:27:56 of people who have such injuries and all kinds of philosophical, ethical and really just most importantly practical aspects of um how do we take care of people with that kind of severe brain injury. Many of whom pose tremendously difficult questions um to family and caregivers uh who can’t predict what’s gonna happen next and and can’t communicate with their loved ones and there’s always this question in such situations you, know, is that person the person we knew still there alright, and will that person come
0:28:25 back, so to speak, uh or not. And uh answering that question is this one aspect of getting at what is consciousness and how does it fluctuate and how do we quantify it and how do we create or restore it when it’s lost or damaged. So you know, that has been the realm of philosophy for most of human history and um I think it is very exciting for me now that that’s that’s changed in the last several years and I do think that [SPEAKER_TURN]
0:28:33 technology of brain computer interfaces is gonna have an impact in making some of the discoveries that have come to light actionable. [SPEAKER_TURN]
0:29:01 Ben Rappaport is the co-founder and chief science officer at Precision Neuroscience. Today’s show was produced by Gabriel Hunter-Cheng. It was edited by Lydia Jean Cott and engineered by Sarah Boughere. You can email us at problem@pushkin.fm I’m Jacob Goldstein and we’ll be back next week with another episode of What’s Your Problem?
Ben Rapoport is the co-founder and CSO of Precision Neuroscience. Ben’s problem is this: Can you build a device that allows a paralyzed person to use a computer with only their thoughts – without damaging their brain?
See omnystudio.com/listener for privacy information.