Enhance Your Learning Speed & Health Using Neuroscience Based Protocols | Dr. Poppy Crum

Andrew Huberman: Welcome to the Huberman Lab Podcast, where we discuss science and science-based tools for everyday life. I'm Andrew Huberman, and I'm a professor of neurobiology and ophthalmology at Stanford School of Medicine.

Andrew Huberman: My guest today is Dr. Poppy Crum. Dr. Poppy Crum is a neuroscientist, a professor at Stanford, and the former chief scientist at Dolby Laboratories. Her work focuses on how technology can accelerate neuroplasticity in learning and generally enrich our life experience.

Andrew Huberman: You've no doubt heard about and perhaps use wearables and sleep technologies that can monitor your sleep, tell you how much slow wave sleep you're getting, how much REM sleep, and technologies that can control the temperature of your sleep environment and your room environment. Well, you can soon expect wearables and hearable technologies to be part of your life. Hearable technologies are, as the name suggests, technologies that can hear your voice and the voice of other people and deduce what is going to be best for your immediate health and your states of mind.

Andrew Huberman: Believe it or not, these technologies will understand your brain states, your goals, and it will make changes to your home and working and other environments so that you can focus better, relax more thoroughly, and connect with other people on a deeper level. As Poppy explains, all of this might seem kind of space age and maybe even a little aversive or scary now, but she explains how it will vastly improve life for both kids and adults and indeed increase human-human empathy.

Andrew Huberman: During today's episode, you'll realize that Poppy is a true out-of-the-box thinker and scientist. She has a really unique story. She discovered she has perfect pitch at a young age. She explains what that is and how that shaped her worldview and her work. Poppy also graciously built a zero cost step-by-step protocol for all of you. It allows you to build a custom AI tool to improve at any skill you want and to build better health protocols and routines. I should point out that you don't need to know how to program in order to use this tool that she's built. Anyone can use it. And as you'll see, it's extremely useful. We provide a link to it in the show note captions.

Andrew Huberman: Today's conversation is unlike any that we previously had on the podcast. It's a true glimpse into the future, and it also points you to new tools that you can use now to improve your life.

Andrew Huberman: Before we begin, I'd like to emphasize that this podcast is separate from my teaching and research roles at Stanford. It is, however, part of my desire and effort to bring zero-cost-to-consumer information about science and science-related tools to the general public. In keeping with that theme, today's episode does include sponsors.

Andrew Huberman: And now for my conversation with Dr. Poppy Crum. Dr. Poppy Crum, welcome.

Poppy Crum: Thanks, Andy. It's great to be here.

Andrew Huberman: Great to see you again. We should let people know now we were graduate students together, but that's not why you're here. You're here because you do incredibly original work. You worked in so many different domains of technology, neuroscience, et cetera. Today I want to talk about a lot of things, but I want to start off by talking about neuroplasticity, this incredible ability of our nervous systems to change in response to experience. I know how I think about neuroplasticity, but I want to know how you think about neuroplasticity. In particular, I want to know, do you think our brains are much more plastic than most of us believe? Like, can we change much more than we think and we just haven't accessed the ways to do that? Or do you think that our brains are pretty fixed and in order to make progress as a species, we're going to have to, I don't know, create robots or something to do the work that we're not able to do because our brains are fixed? Let's start off by just getting your take on what neuroplasticity is and what you think the limits on it are.

Poppy Crum: I do think we're much more plastic than we talk about or we realize in our daily lives. And just to your point about creating robots, the more we create robots, there's neuroplasticity that comes with using robots as humans when we use them in partnerships or as tools to accelerate our capabilities. So neuroplasticity, where I resonate with it a lot, is trying to understand, and this is what I've done a lot of in my career, is thinking about building and developing technologies, but with an understanding of how they shape our brain. Everything we engage with in our daily lives, whether it's the statistics of our environments and our context or the technologies we use on a daily basis, are shaping our brains in ways through neuroplasticity. Some more than others. Some we know as we age are very dependent on how attentive and engaged we are as opposed to passively just consuming and changing. But we are in a place where everyone, I believe, needs to be thinking more about how the technologies they're using, especially in the age of AI and immersive technologies, how they are shaping or architecting our brains as we move forward.

Andrew Huberman: You go to any Neuroscience 101 medical school textbook and there's something you'll see a few pages on something called the homunculus. Now, what is the homunculus? It's a data representation, but it's this sort of funny creature when you see it. But that picture of this sort of distorted human that you're looking at is really just a data representation of how many cells in your brain are helping, are coding and representing information for your sense of touch. Right? And that image, though, and this is where things get kind of funny. That image comes from Wilder Penfield back in the 40s. He recorded somatosensory cells of patients just before they were to have surgery for epilepsy and such. And since we don't have pain receptors in our cortex, he could have this awake human and be able to touch different parts of their brain and ask them to report what sensation they felt on their bodies. And so he mapped that part of their cortex. And then that's how we ended up with the homunculus. And you'll see it'll have bigger lips. It'll have, you know, smaller parts of your back and the areas where you just don't have the same sensitivities.

Andrew Huberman: Well, fast forward to today. When you look at that homunculus, one of the things I always will ask people to think about is, you know, what's wrong with this image? This is an image from 1940 that is still in every textbook. And any Stanford student will look at it and they'll immediately say, well, the thumb should be bigger because we do this all day long. And I've got more sensitivity in my fingers because I'm always typing on my mobile device, which is absolutely true. Or maybe they'll say something like, well, the ankles are the same size and we drive cars now a lot more than we did in the 40s. Or maybe if I live in a different part of the world, I drive on one side versus the other. And in a few years, we probably won't be driving and those resources get optimized elsewhere.

Andrew Huberman: So what the homunculus is, is it's a representation of how our brain has allocated resources to help us be successful. And those resources are the limited cells we have that support whatever we need to flourish in our world. And the beauty of that is when you develop expertise, you develop more support, more resources go to helping you do that thing. But they also get more specific. They develop more specificity so that, you know, I might have suddenly a lot more cells in my brain devoted to helping me. You know, I'm a violinist and my left hand, my right hemisphere, my somatosensory cortex, I'm going to have a lot more cells that are helping me feel my fingers and the tips of everything so that I can be fluid and more virtuosic. But that means I have more cells, but they're more specified. They're giving me more sensitivity. They're giving me more data that's differentiated. And that's what my brain needs, and that's what my brain's responding to.

Andrew Huberman: And so when we think about that, my practice as a musician versus my practice playing video games, all of these things influence our brain and influence our plasticity. Now, where things get kind of interesting to me and sort of my obsession on that side is every time we engage with a technology, it's going to shape our brain, right? And it's both, you know, our environments, but our environments are changing. Those are shaping who we are. I think you can look at people's hearing thresholds and predict what city they live in.

Andrew Huberman: Really?

Poppy Crum: Absolutely, yes.

Andrew Huberman: Can you just briefly explain hearing thresholds and why that would be? I mean, I was visiting the city of Chicago a couple years ago. Beautiful city. Amazing food. Love the people. Very loud city. Wide downtown streets. Not a ton of trees compared to what I'm used to. And I was like, wow, it's really loud here. And I grew up in the suburbs, got out as quickly as I could. I don't like the suburbs. Sorry, suburb dwellers, not for me. I like the wilderness and I like cities. But you're telling me that you can actually predict people's hearing thresholds for loudness simply based on where they were raised or where they currently live?

Poppy Crum: In part, it can be both, right? Because cities have sonic interference types of noise, things that are very loud cities, but also what's creating that noise, right? That's often unique, the inputs, the types of vehicles, the types of density of people, and even the construction in those environments, it is changing what noise exists. That's shaping people's hearing thresholds. At the lowest level, it's also shaping their sensitivities. If you're used to hearing certain animals in your environment and they come with a heightened response, you're going to develop increased sensitivity to that, right? Whereas if it's really abnormal, you know, to hear chickens... I have a neighbor who has chickens in the city.

Andrew Huberman: Roosters too?

Poppy Crum: Yes. Yes.

Andrew Huberman: I grew up near a rooster. I can still hear that rooster. Those sounds are embedded deeply in my mind.

Poppy Crum: There's the semantic context and then just the sort of spectrum, right? And the intensity of that spectrum. When I say spectrum, I mean the different frequency amplitudes and what that shaping is like.

Andrew Huberman: High pitch, low pitch.

Poppy Crum: Yeah, yeah. And that affects how your neural system is changing even at the lowest level of what your ear is, your brain, your cochlea is getting exposed to. But then also where, you know, so that would be the lower level, what sort of noise damage might exist, what exposures, but then also there's the amplification coming from your higher level areas that are helping you know that these frequencies are more important in your context, in your environment.

Poppy Crum: There was a funny film called, I think it's The Sound of Silence, and it started, I love Peter Sarsgaard, he was one of the actors in it, and it was sort of meant to be a bit fantastical. But in fact, to me, the filmmakers had talked to me a lot to inform this sort of main character and the way he behaved, because I have absolute pitch. And there were certain things that they were trying to emulate in this film. He ends up being this person who tunes people's lives. He'll walk into their environments and be like, oh, you know, things are going badly at work or your relationships because you've got this tritone, or your water heater is making this pitch and your teapot is at this.

Andrew Huberman: Oh, my God, this would go over so well in L.A. People would pay millions of dollars in Los Angeles. It's totally funny. Do you do this for people?

Poppy Crum: No.

Andrew Huberman: Okay, okay. I will tell you, I will walk into hotel rooms and immediately if I hear something, I've moved. And so that is—

Poppy Crum: Because you have perfect pitch.

Andrew Huberman: Could you define perfect pitch? Does that mean that you can always hit a note perfectly with your voice?

Poppy Crum: There is no such thing as perfect pitch. There's absolute pitch. And so I think only because the idea of, so like ah that would be A equal 440 Hertz, right? But that's a standard that we use in modern time. And what A is has actually changed throughout our lives with aesthetic, with what people like, with the tools we use to create music. And in the Baroque era, A was 415 hertz.

Andrew Huberman: Can you hit that?

Poppy Crum: Awesome. Awesome. And in any case, so that's why it's absolute. Because guess what? But as my basilar membrane gets more rigid as I might age or my temporal processing slows down, my brain's going to still think I'm singing 440 hertz, but it might not be.

Andrew Huberman: Basilar membrane is a portion of the internal ear that converts sound waves into electrical signals, right?

Poppy Crum: Yeah.

Andrew Huberman: Okay, fair enough. I'm talking to an auditory physiologist here. Mechanically, it moves the hair cells. I teach auditory physiology, but I want to just make sure because I'm sitting across from an expert.

[AD BREAK]

Andrew Huberman: Okay, so our brains are customized to our experience.

Poppy Crum: Yeah.

Andrew Huberman: Especially our childhood experience, but also our adult experience.

Poppy Crum: Yes.

Andrew Huberman: You mentioned the homunculus. This representation of the body surface. And you said something that I just have to pick up on and ask some questions about, which is that this hypothetical Stanford student, could be any student anywhere, says, wait, nowadays we spend a lot of time writing with our thumbs and thinking as we write with our thumbs and emoting, right? I mean, when we text with our thumbs, we're sometimes involved in an emotional exchange.

Poppy Crum: Yeah.

Andrew Huberman: My question is this. The last 15 years or so have represented an unprecedented time of new technology integration, right? I mean the smartphone, texting. And when I text, I realized that I'm hearing a voice in my head as I text, which is my voice because if I'm texting outward, I'm sending a text. But then I'm also internalizing the voice of the person writing to me if I know them. But it's coming through filtered by my brain, right? So it's like I'm not trying to micro-dissect something here for the sake of micro-dissection. But the conversation that we have by text, it's all happening in our own head. But there are two or more players. Group text was too complicated to even consider right now. But what is that transformation really about?

Andrew Huberman: Previously, I would write you a letter, I would send you a letter, right? Write you an email, I'd send you an email. And so the process was really slowed. Now you can be in a conversation with somebody that's really fast back and forth, right? Some people can type fast, you can email fast, but nothing like what you can do with text. I can even know when you're thinking because it's dot, dot, dot, or you're writing. And so is it possible that we've now allocated an entire region of the homunculus or of some other region of cortex brain to conversation that prior to 2010 or so, the brain just was not involved in conversations of any sort.

Andrew Huberman: In other words, we now have the integration of writing with thumbs. That's new. Hearing our own voice, hearing the hypothetical voice of the other person at the other end and doing that all at rapid speed. Are we talking about like a new brain area or are we talking about using old brain areas and just trying to find and push the overlap in the Venn diagram? Because I remember all of this happening very quickly and very seamlessly. I remember like texting showed up and it was like, all right, well, it's a little slow, a little clunky. Pretty soon it was autofill. Pretty soon it was learning us. Now we can do voice recognition. And it's, you know, people pick this up very fast. So the question is, are we taking old brain areas and combining them in new ways? Or is it possible that we're actually changing the way that our brain works fundamentally in order to be able to carry out something as what seems to be nowadays trivial but as basic to everyday life as texting? What's going on in our brain?

Poppy Crum: We aren't developing new resources. We've got the same cells that are, or I mean, there's neurogenesis, of course, but it's how those are getting allocated. And just one quick comment from what we said before when we talk about the homunculus. The homunculus is an example of a map in the brain, a cortical map. And maps are important in the brain because they allow cells that need to interact to give us specificity, to make us fast, to have tight reaction times and things, because you've got shorter distance and things that belong together. Also, there's a lot of motility in terms of what those cells respond to, potentially dependent on our input. So the homunculus might be one map, but there are maps all over our brain. And those maps still have a lot of cross input.

Poppy Crum: So what you're talking about is, are you having areas where we didn't used to allocate and differentiate in the specificity of what those cells were doing that are now quite related to the different ways my brain is having to interpret a text message? And the subtlety and the nuance of that, that actually now I get faster at. I have faster reaction times. I also have faster interpretations. So am I allocating cells that used to do something else to allow me to have that? Probably. But I'm also building, where like think about me as a multisensory object that has to integrate information across sight, sound, smell to form a holistic object experience. That same sort of integration and pattern is happening now when we communicate in ways that it didn't used to. So what does that mean? It means there's a lot more repeatability, a lot faster pattern matching, a lot more integration that is allowing us to go faster.

Andrew Huberman: I completely agree. I feel like there's an entire generation of people who grew up with smartphones for which it's just part of life. I think one of the most impactful statements I ever heard in this general domain was I gave a talk down at Santa Clara University one evening to some students. And I made a comment about putting the phone away and how much easier it is to focus when you put the phone away and how much better life is when you take space from your smartphone and all of this kind of thing. And afterwards, this young guy came up to me, he's probably in his early 20s, and he said, listen, you don't get it at all. I said, what do you mean? He said, you adopted this technology into your life and after your brain had developed. He said, when my phone runs out of charge, I feel the life drain out of my body. And it is unbearable. We're nearly unbearable until that phone pops back on, and then I feel life return to my body and it's because I can communicate with my friends again. I don't feel alone. I don't feel cut off from the rest of the world.

Andrew Huberman: And I was thinking to myself, wow, his statements really stuck with me because I realized that his brain, as he was pointing out, is indeed fundamentally different than mine in terms of social context, communication, feelings of safety, and on and on. And I don't think he's alone. I think for some people it might not be quite as extreme, but for many of us to see that dot, dot, dot in the midst of a conversation where we really want the answer to something or it's an emotionally charged conversation can be a very intense human experience.

Poppy Crum: That's interesting.

Andrew Huberman: So we've sped up the rate that we transfer information between one another. But even about trivial things, it doesn't have to be an argument or like is it stage four cancer or is it benign, right? Like those are extreme conditions, right? Are they alive? Are they dead? You know, did they find him or her or did they not? Those are extreme cases, but there's the everyday life of, and I noticed this, like if I go up the coast sometimes or I'll go to Big Sur and I will intentionally have time away from my phone. It takes about an hour or two or maybe even a half day to really drop into the local environment where you're not looking for stimulation coming in through the smartphone. And I don't think I'm unusual in that regard either. So I guess the question is, do you think that the technology is good, bad, neutral, or are you agnostic as to how the technologies are shaping our brain?

Poppy Crum: It goes in lots of different directions. One thing I did want to say, though, with smartphones specifically and sort of everything, in audio, our ability to carry our lifetime of music and content with us has been because of huge advances in the last 25, 30 years and maybe slightly more around compression algorithms that have enabled us to have really effective what we call perceptual compression, lossy perceptual algorithms and things like MP3 and, you know, my past work with companies like Philby. But whenever you're talking about what's the goal of content compression algorithms, it's to translate the entirety of the experience, the entirety of a signal with a lot of the information removed, right? But in intelligent ways. When you look at the way someone is communicating with acronyms and the shorthand that the next generations use to communicate, it is such a rich communication, even though they might just say LOL. I mean, it's like, or they might, you know, it's actually a lossy compression that's triggering a huge cognitive experience right?

Andrew Huberman: Can you explain lossy for people who might not be familiar with it?

Poppy Crum: Lossy means that in your encoding and decoding of that information, there is actually information that is lost when you decode it. But hopefully that information is not impacting the perceptual experience. Imagine I have a song and I want to represent that song. I could take out, to make my file smaller, I could take out every other 500 milliseconds of that and it would sound really horrible, right? Or I could be a lot more intelligent. And instead, basically, if you look at early models like MP3, they're kind of like computational models of the brain. They might stop at like the auditory nerve, but they're trying to put a model of how our brain would deal with sound, what we would hear, what we wouldn't. If this sound is present and it's present at the same time as this sound, then this sound wouldn't be heard, but this sound would be. So we don't need to spend any of our bits coding this sound. Instead, we just need to code this one. And so it becomes an intelligent way for the model and the algorithm of deciding what information needs to be represented and what doesn't to create the same, the best perceptual experience, perceptual meaning what we get to take home.

Poppy Crum: I think one of the things that's important then, why I think whenever I used to have to teach some of what it means to represent a rich experience with minimal data. You think with minimal information, some of the acronyms that exist in mobile texting, they've taken on a very rich life in internal life. LOL, ONG.

Andrew Huberman: Yeah, well, those are simplistic ones.

Poppy Crum: But I think people can have communication now that we can't understand entirely.

Andrew Huberman: Is it because you have a 10-year-old daughter? Does she have communication by acronym that to you is cryptic?

Poppy Crum: Sometimes, but I have to figure it out then. But yes. But the point is, that is an example of a lossy compression algorithm that actually has a much richer perceptual experience, right? And it often needs context, but it's still, you know, you're using few bits of information to try to represent a much richer feeling in a much richer state. And if you look at different people, they're going to have a bigger physiological experience dependent on how they've grown up with that kind of context.

Andrew Huberman: It sounds to me, I don't want to project here, but it sounds to me like you see the great opportunity of data compression. Let's just stay with the use of acronyms and texting. That's a vast data compression compared to the kind of speech and direct exchange that people engaged in 30 years ago. So there's less data being exchanged, but the experience is just as rich, if not more rich, is what you're saying, which implies to me that you look at it as generally neutral to benevolent. It's good. It's just different. I'm coming up on 50 in a couple months. And as opposed to somebody saying, well, you know, when I was younger, we'd write our boyfriend or girlfriend a letter. You know, I would actually write out a birthday card. I would go – you'd have a face-to-face conversation and you've got this younger generation that are saying, yeah, whatever. You know, this is like what we heard about I used to trudge to school in the snow kind of thing. It's like, well, we have heated school buses now and we've got driverless cars. So I think this is important and useful for people of all ages to hear that the richness of an experience can be maintained even though there are data or some elements of the exchange are being completely removed.

Poppy Crum: Absolutely. But it's maintained because of the neural connections that are built in those individuals, right, and that generation.

Andrew Huberman: I always think of, okay, and the nervous system likes to code along a continuum, but like yum, yuck, or meh. Like, do you think that a technology is kind of neutral? Like, yeah, you lose some things, you gain some things. Or do you think like this is bad? These days we hear a lot of AI fear. We'll talk about that. Or you hear also people who are super excited about what AI can do, what smartphones can do. I mean, some people like my sister and her daughter love smartphones because they can communicate. It gives a feeling of safety at a distance. Like quick communications are easier. It's hard to sit down and write a letter. She's going off to college soon. So the question is like, how often will you be in touch?

Andrew Huberman: It raises expectations about frequency, but it reduces expectations of depth. Because you can do a quick "hey, I was thinking about you this morning," and that can feel like a lot. But a letter—if I sent a letter home during college saying, "Hey, I was thinking about you this morning. Love, Andrew"—they'd be like, "Okay, that didn't take long." So I think there's a seesaw. You get more frequency, and it comes with different levels of expectation.

Poppy Crum: My daughter's at camp right now, and we're only allowed to write letters for two weeks. Handwritten letters.

Andrew Huberman: Handwritten letters. How did that go over?

Poppy Crum: It's happening. I lost my home in a flood years ago, and one of the only things I saved was this communication between one of my ancestors during the Civil War. They were courting, and all those letters were saved—back and forth between the women. It's from 1865.

Andrew Huberman: Do you have those letters?

Poppy Crum: I do. I had them in my computer backup until I flew up here. They were on parchment, and even though they went through a flood, they didn't run. It's a very different era of communication, and it's wonderful to have that preserved because that history doesn't translate otherwise.

In any case, I am a huge advocate for integration of technology. For me, the world is data. I look at the way my daughter behaves and think, "Okay, what data is coming in? Why did she respond that way?" We were talking about neuroplasticity. We are creatures of three things: our sensory systems and how they've evolved—be it by intrinsic noise degrading our sensory receptors or external strength. My brain is going to have access to about the same amount of information as someone with hearing loss if I'm in a very noisy environment. So suddenly you've compromised the data I have access to. Then there are our experientially established priors.

Andrew Huberman: Which is Bayesian. We should explain for people. Deterministic would be input A leads to output B.

Poppy Crum: Bayesian is it depends on the statistics of what's happening externally and internally. These are probabilistic models. There's a likelihood of A becoming B, but also a probability that A will drive C, D, or F.

Andrew Huberman: Absolutely.

Poppy Crum: Some of the things that make us most effective in our environments is how fast and effective we are at dealing with those probabilistic situations. Probabilistic inference is a great indicator of success—be it in a work environment or just walking down the street. It's our situational intelligence in the world. We can break that down into many ways.

We are the products of our sensory systems, our priors—the statistics and data our brain uses to weight how it will behave and the decisions it makes—and also our expectations and context. There's a funny story. My daughter, when she was two and a half, we were in the planetarium at the Smithsonian. We started in LA, zoomed out on our way to the sun, and passed that quintessential NASA image of the Earth. It was totally dark and silent. My daughter, as loud as she possibly could, yelled, "Minions!" I was like, "What?" Then I realized—her experientially established prior of that image was from the Universal logo.

Andrew Huberman: That says Universal. I love it.

Poppy Crum: It was totally valid. But it was an honest and true part of what it is to be human. Each of us experiences very different things from the same physical information. It's driven by our exposures, our priors, and our sensory systems—that trifecta—and our expectations of the moment. Once you unpack that, you really start to understand the influence of technology.

I am a huge advocate for technology improving us as humans and improving the data we have to make better decisions. At the same time, I think sometimes we're penny wise, pound foolish with how we use technology. The quick things that make us faster can also make us dumber and take away our cognitive capabilities. People who use technologies to write papers all the time might be putting themselves in a place where they are compromised trying to do anything without that technology. You end up with bigger differentiations in cognitive capabilities based on how you use a tool.

One of my things is teaching at Stanford—we have that in common.

Andrew Huberman: I need to sit in on one of your lectures.

Poppy Crum: My class is called Neuroplasticity and Video Gaming. I'm a neurophysiologist, but I'm really a technologist. Video games are powerful because they create a closed-loop environment: you get feedback, data on your performance, and you control what you randomize and how you build. Our aim is to build technology and games with an understanding of the neural circuits you're impacting and what you want to train. I have students who are musicians, computer scientists, and some of Stanford's top athletes. It's always focused on dissecting some aspect of human performance and amplifying sensitivity or access to that type of learning in a closed-loop way.

For anyone unfamiliar with the role of gaming in neuroscience, there have been great papers. Take a gamer versus a non-gamer. A typical gamer has what we would call more sensitive contrast sensitivity functions—the ability to see edges and differentiation in a visual landscape. They can see faster and are more sensitive to that differentiation.

Andrew Huberman: So someone who says they're not a video game player has trained it. Like a first-person shooter game—I've played occasionally in an arcade. A lot of it is based on contrast sensitivity: is that a friend or foe? You have to make decisions very fast, right on the threshold of reflexive, with rapid iteration. Then the rules switch—suddenly you're supposed to turn other things into targets.

Poppy Crum: You're spot on. Take a self-identified non-gamer, make them play 40 hours of Call of Duty, and now their contrast sensitivity looks like a video game player. It persists—measure them a year later. Forty hours of playing Call of Duty and I see the world differently, not just in the game. I have foundational shifts in how I experience the world that give me greater sensitivity to my situational awareness. In real life.

Andrew Huberman: Because that's a low-level processing capability. I love intersecting those.

Poppy Crum: What's even more interesting—and this was a great study by Alex Pouget and Daphne Bavelier—is going to that next level of probabilistic decisions where things aren't deterministic. A video game player will make the same decisions as a non-player in those probabilistic inferential situations, but they'll do it a lot faster. That edge is phenomenal. When you can tap into that, it becomes very powerful.

Then I like to take it further. I had a couple of Stanford's top soccer players in my course this year. Our focus was: what data do you not have, and how can we build a closed-loop environment so you're gaining better neurological access to your performance? We used data like acceleration and velocity—not at the end of a two-hour practice, but in real time—and gave auditory feedback to tap into more neural training. We had sensors on their calves measuring acceleration and velocity, giving feedback in real time during a gamified training environment. It fed them different sonic cues so they were building resolution. As a novice, I can't tell if I've accelerated successfully or not. But if you give me more gradation in the feedback from that closed-loop behavior, my neural representation starts differentiating more. That's where auditory feedback comes in.

I'd love for you to share the story about your daughter improving her swimming stroke.

Andrew Huberman: She's not a D1 athlete yet, but she's a swimmer. In the past, if you wanted to get better at swimming, you needed a swimming coach. If you wanted to get really good, you'd need a really good coach and work with them repeatedly. You took a slightly different direction that points to how beneficial and inexpensive this technology can be.

Poppy Crum: First, I'll say this: having good swimming coaches is important. I'm not trying to do away with them. Parents who are data-centric and like building technologies can sometimes be red herring distractions, but hopefully not.

Andrew Huberman: Let's keep the swimming coaches happy.

Poppy Crum: For example, if you go to swimming camps, they use cameras to record and assess your strokes. But you can use tools like Perplexity Labs or Replit—online resources—to quickly build a computer vision app that gives you data analytics on your strokes in real time.

Andrew Huberman: How does that work? You take the phone underwater?

Poppy Crum: In this case, I'm using a mobile phone above water. You film your daughter doing freestyle, breaststroke, or butterfly. There are core things you might care about—for backstroke and freestyle, things like the roll, how high they're coming above the water, velocity on entrance, how far in front of their head their arm is coming in, how consistent their strokes and cadence are across the pool. You suddenly have access to better resolution and more analytics that give you insight.

The important thing is, my 10-year-old isn't going to change her velocity based on my telling her. But it gives me information to understand how consistent she is on things coaches have told her to do. What I love is that this isn't just for elite athletes. You don't have to code anymore to build these apps. You should leverage AI for development. You tell AI to write code to analyze trajectory jumping into the pool, how it could be improved if the goal is to swim faster. You use AI to build an app that gives you access to whatever data you want.

You could do the same for running, gait, or in a work environment to understand vulnerabilities and weaknesses. There are two places where I see AI acceleration and tool building having major impact: democratizing data analytics and information normally reserved for the elite to everyone who is engaged. That has a huge impact on improving performance because that data is useful for understanding learning. It also has applications in work environments for understanding success in a process or skill, giving you new data to think about with regard to what I call a digital twin.

When I use the term "digital twin," the goal is not to digitize and represent a physical system in its entirety. It's to use different interoperable data sets—from different sources—to gain insights. Digitized data of a physical system, environment, or world—be it a hospital, airplanes, my body, or my fish tank—to give me insights that are continuous and in real time that I otherwise wouldn't have access to.

[Sponsor break for AGZ by AG1]

[Sponsor break for Rora]

Andrew Huberman: As nerds speak, but domain specific. The lowest-level example I can think of, which would be very useful to me, would be a digital twin of my refrigerator that places an order for the things I need, not the things I don't. It would eliminate the need for a shopping list. It would just keep track: "You usually run out of strawberries on this day." It would keep track in the background, and the stuff would just arrive. It would eliminate the need to go to the store.

This morning I walked to the corner store and bought some produce. I had the time to do that—eight minutes. But really, I'd like the fridge to be stocked with the things I like and need. I could hire someone, but that's expensive. This could be done trivially and probably will be soon. I don't even need to build an app into my phone. I like to think in terms of low-level but highly useful and easily available technologies.

When it comes to students learning information, we've heard that AI is a bad thing—like they'll use it to write essays. But there's a use of AI for learning. I know this because I'm still learning. I teach and learn all the time for the podcast. I've been using AI to take large volumes of text from papers—verbatim, not hallucinating. I've read those papers, printed them out, taken notes. Then I've been using AI to design tests for me on what's in those papers. About eight months ago, when researching a podcast on how to study and learn best, the data all point to the fact that when we self-test—especially away from the material, like thinking, "What is the cascade of hormones driving the cortisol negative feedback loop?" on a walk instead of looking it up—that self-testing is most impactful for memory. Most of memory is anti-forgetting.

So I've been having AI build tests for me. It asks me questions like, "What is the signal between the pituitary and the adrenals that drives the release of cortisol? What layer of the adrenals does cortisol come from?" I love that. The information it's drawing from is accurate to the best of science's knowledge. It's testing me, and it's learning where I'm weak and where I'm strong. I ask it, "Where am I weak and where am I strong?" It says, "Naming and third-order conceptual links need work." I say, "Test me on it," and it does. It's amazing. I'm blown away that the technology can do this. I'm not building apps—I'm just using it to learn better.

Poppy Crum: Whether you're building apps or tools, you're using it as a tool to optimize your cognition, find weaknesses, and get feedback on your performance to accelerate your learning.

Andrew Huberman: That's the goal. But you're still putting in the effort to learn.

Poppy Crum: Even the ways I'm using it—computer vision with mobile devices—AI is a huge opportunity. Using cameras and collected data for more sophisticated input is huge. In both cases, you're shaping cognition. You're using data to enrich what you can know. AI is incredibly powerful in those spaces.

The place where I think it separates—and I sort of divide it into two categories, maybe too simplistic—is: am I using the tool to make me smarter, give me more information, make me more effective cognitively, and gain different insights? Or am I using it to replace a cognitive skill I've done before to be faster? It doesn't mean you don't want to do those things. GPS in our car is a perfect example of replacing a cognitive tool to be faster and more effective. Take away your GPS in a city you drive around, and we're not very good.

Andrew Huberman: I remember paper maps. I remember the early studies of the hippocampus based on London taxi drivers who had mental maps of the city.

Poppy Crum: Absolutely. With all due respect to London taxi drivers, up until GPS, those mental maps were necessary. They had more gray matter in their hippocampus. Today, they don't have to have that because the people in their back seats have more data, more information, eyes from the sky. Satellite data is huge in our success. It can anticipate things locally you can't. It's been replaced. But it still means when you lose that data, you don't expect yourself to have the same spatial navigation without it.

Andrew Huberman: I love your two buckets. You're either using it to make you cognitively better, or you're using it to speed you up. But you have to be careful.

Poppy Crum: Cognitively or physically. You're still trying to gain insight and data that makes you a more effective human.

Andrew Huberman: I think where people are concerned, including myself, is when we use these technologies that eliminate steps and make things faster, but we fill the additional time or mental space with things that are neutral to detrimental. It's like saying, "I can get all the nutrients I need from an eight-ounce drink." That's not true. Then the question is, how do I make up the rest of my calories? Am I making up with nutritious food that keeps me at neutral health? Or am I eating stuff that brings in a bunch of bad stuff with those calories? In the mental version, things are sped up, but people are filling the space with things that are making them dumber in some cases. There was a recent paper from MIT that is very much what I spend a lot of my time talking about.

Poppy Crum: Could you describe that study?

Poppy Crum: The upshot of the paper first was that people have a lot less mental process or cognitive process going on when they use LLMs to write papers. They don't have the same transfer, and they don't really learn the information. Surprise, surprise.

So just to briefly describe the study, even though it got a lot of popular press, MIT students were writing papers using AI versus writing papers the old-fashioned way where you think and write. There were three different categories. People had to write the papers using their brain only, which was case one. Case two was they got to use search engines, a sort of middle ground. And case three was they used LLMs to write their paper.

They were looking at what kind of transfer happened and what kind of neural response occurred. They used EEG to look at neural patterns across the brain to understand how much neural engagement happened during the writing process and what they could do with that information down the road. It's a really nice paper, so I don't want to diminish it by summarizing it.

What I think is a really important upshot of that paper, and how we talk about it, is that I talk a lot about cognitive load. You can measure cognitive load in the diameter of your pupil, body posture, and how people are thinking. It's really about how hard my brain is working right now to solve a problem. There are a lot of different cues we give off as humans that tell us when we're under states of different load, whether we are aware of it or not.

There's something called cognitive load theory that breaks down what happens when our brains are under states of load. That load can come from three different places. The intrinsic load is from the difficulty of the material I'm trying to understand. Some things are easy to learn, some things are a lot harder. Extraneous load comes from how the information is presented. Is it poorly taught? Is it poorly organized? Or even the environment. If I'm trying to learn something auditorily and it's noisy, that's introducing extraneous cognitive load. It's not the information itself, but everything else happening with that data.

The third is germane cognitive load. That's the load used in my brain to build mental schemas, to organize that information, to really develop a representation of what that information is that I'm taking in. That germane cognitive load is the work. If you don't have germane cognitive load, you don't really have learning.

What they found is that the germane cognitive load is what gets impacted most by using LLMs. It's a very obvious thing. Using LLMs, you're not engaging the mental effort to build cognitive schema, to build neural schemas and the mental representation of the information so you can interact with it later and have access to it later. This is really important because without that, you won't be as intelligent on that topic down the road.

Let me give two examples. I have a doctor and a lawyer, and both of them use LLMs extensively for searches or for building information. In one case, it's for patient aggregation of patient data. In another case, it's for history of case files. That is the GPS that's happening in those spaces because those are the tools that are quickly adopted.

Where you have someone from a different world who has learned that information and worked with data in a different way, their representation of that information is going to be better at extrapolation. It's going to be better at generalization. It's going to be better at seeing patterns that would exist. The brain that has done everything through LLMs is going to be in a place where they will get the answer for that relevant task using the tools they have. But you're not at the same level of richness and depth of information or generalization or extrapolation for those topics as someone that has learned in a different way.

There's a generational difference in understanding, not because they don't have the same information, but there is an acknowledgment that there's a gap, even though we're getting to the same place as fast. That's because of the learning that's happened. That's your germane cognitive load. You've got to do the work. Your brain has to do it.

Andrew Huberman: What was beautiful about your descriptions is when you were talking about how you were using it to test yourself and find your vulnerabilities. In the MIT paper, which I think are things that are somewhat obvious but we just have to talk about them more, people with higher competency on the topic use the tools in ways that still engage more germane cognitive load but help to accelerate their learning.

Poppy Crum: The biggest vulnerability and gap is when you're trying to learn a new domain fast or you're under pressure, and you're not putting in the germane effort. You're not using the tools that AI can enable to amplify your cognitive gain, but instead to deliver something faster, more rapid, and then walking away from it.

Andrew Huberman: I'm going to try and present two parallel scenarios in order to go further into this question of how to use AI to our best advantage to enrich our brains as opposed to diminish our brains. I could imagine a world, because we already live in it, where there's this notion of slow food. You cook your food, you get great ingredients from the farmer's market, a peach that really tastes like a peach. You make your own food, you cook it, and it's delicious.

I can also imagine a world where you order a peach pie online, it shows up, and you take a slice and you eat it. You could take two different generations of people, people that are currently 50 or older and people that are 15 or younger. The older generation would say, "Isn't the peach pie that you made so much better? These peaches are amazing." I could imagine a real scenario where the younger person, 15 to 30, would say, "I don't know, I actually really like the other pie. I like it just as well." The older generation is like, "What are you talking about? This is how it's done."

What's different? From a neuroscience standpoint, it very well could be that it tastes equally good to the two of them, just differs based on their experience. The person isn't lying. It's not that this kid isn't as fine-tuned to taste. It's that their neurons acclimated to what sweetness is, what contrast between sweet and saltiness is, and what a peach should taste like because they had peach gummies and that tastes like a peach. We can be disparaging of the lower level or diminished sensory input, but it depends a lot on what those neural circuits were weaned on.

Poppy Crum: I love the peach pie example. Making bread is another example. In the 90s, everyone I knew when they graduated from high school got a bread maker that was shaped like a box and created this loaf of bread with a giant rod through it. It was the graduation gift for many years. You don't see those anymore. If you look at what happened with the millennial generation in the last five years, especially during the pandemic, suddenly bread making and sourdough became a thing.

What's the difference? You've got bread with the bread maker, it's fresh, and it is not at all desired relative to bread that takes a long period of time and is tactile. The process of making it is clearly much more onerous. I think the key part is in the appreciation of the bread. The process is part of it. That process is the development of the germane knowledge and the commitment and connection to that humanness of development, but also the tactile commitment, the work that went into it is really appreciated.

It's the same way that that peach pie for one person comes with that whole time series of data that wasn't just about my taste, but was also smell, physical, visual, and saw the process evolve and build a different prior going into that experience. That is part of the richness of human experience. Will it be part of the richness of how humans interact with AI? Absolutely. Or interact with robots? Absolutely. The relationships we're building and how integrated these tools are in our existence will shape us in different ways.

What I am particularly bullish on and excited for is the robot that optimizes my health, my comfort, my intent in my environment, be it in the cabin of a car or in my rooms and spaces.

Andrew Huberman: What would that look like? Could you give me the lowest level example? Would it be an assistant that helps you travel today when you head back to the Bay Area? What is this non-physical robot?

Poppy Crum: I think we already have some of these. It's the point where HVAC systems actually get sexy. Not sexy in that sense, but they're actually really interesting because they are the heart of it.

Andrew Huberman: HVAC systems. Heating, ventilation, air conditioning.

Poppy Crum: You think about a thermostat. A thermostat right now is optimizing for an AI thermostat, optimizing for my behavior. But it's trying to save me resources, trying to save me money. It doesn't know if I'm hot or cold. It doesn't know my intent, what I'm trying to do at that moment. It doesn't know what my optimal state is for my goal in that moment in time. But it can very easily talk to me. It can also know what my state of my body is right now. If it's 1 a.m. and I really need to work on a paper, my house should not get cold.

Andrew Huberman: My Eight Sleep mattress, which I love, knows what temperature adjustments need to be made across the course of the night. I put in what I think is best, but it's updating all the time now because it has dynamically updating sensors. I'm getting close to two hours of REM sleep a night, which is outrageously good for me. Much more deep sleep. That's a little micro environment. You're talking about integrating that into an entire home environment.

Poppy Crum: Home, vehicle, yes. It needs to treat me as a dynamic time series. It needs to understand the context of everything that's driving my state internally, everything that's driving my state in my local environment, meaning my home or my car, and then what's driving my state externally from my external environment. We're in a place where those things are rarely treated as interacting together for the optimization and the dynamic interactions that happen. But we can know these things. We can know so much about the human state from non-contact sensors.

Andrew Huberman: We're right at the point where the sensors can start to feed information to AI to be able to deliver. A lower level example would be like the dynamically cooling mattress or dynamically heating mattress. I discovered through the AI that my mattress was applying heat toward the end of the night, which increases your REM sleep dramatically, whereas cooling it at the beginning of the night increases your deep sleep. It's been immensely beneficial for me to be able to shorten my total sleep need, which is awesome because I like sleep a lot, but I don't want to need to sleep so much in order to feel great.

Poppy Crum: You want to have your own choice about how you sleep. Given the day, it's helping you have that.

Andrew Huberman: Sometimes I have six hours, sometimes I have eight hours. Here's where I get stuck. I've been wanting to have a conversation about this with a neuroscientist who's interested in building technologies for a very long time. I'm hoping you can solve this for all of us, Poppy.

We're talking about sleep. We know a lot about sleep. You've got slow wave sleep, deep sleep, growth hormone release at the beginning of the night. You have rapid eye movement sleep, which consolidates learning from the previous day and removes the emotional load of previous day experiences. We can make temperature adjustments, avoid caffeine too late in the day. AI and technology is doing a really great job, as is pharmacology, to try and enhance sleep. Sleep's getting better despite more forces potentially disrupting it.

Here's the big problem in my mind. We have very little understanding or even names for different awake states. We have names for the goal like, "I want to be able to work." What's work? What kind of work? "I want to write a chapter of a book." What kind of book? Based on what? We talk about alpha, beta waves, theta waves. But I feel like as neuroscientists we have done a pretty poor job as a field of defining different states of wakefulness. The technology and AI don't know what to shoot for. They don't know what to help us optimize for.

Whereas with slow wave sleep and REM sleep, we've got it. I ask questions of myself all the time. Is my brain and what it requires in the first three hours of the day anything like what my brain requires in the last three hours of the day if I want to work in each one of those three-hour compartments? I don't think we really understand what to try and adjust to.

Here's my question. Do you think AI could help us understand the different states that our brain and body go through during the daytime, give us some understanding of what those are in terms of body temperature, focus ability, and then help us optimize for those the same way that we optimize for sleep? Whether it's a conversation with your therapist, a podcast, playing with your kids, or Netflix and chill, the goal is what people have spent so much time, energy, and money on. Whether they're drinking alcohol, caffeine, taking Ritalin or Adderall, or running, humans have spent their entire existence trying to build technologies to get better at doing the things that they need to do. Yet we still don't really understand waking states. Can AI teach it to us? Can AI teach us a goal that we don't even know we have?

Poppy Crum: I would say AI is part of the story. But before we get AI, we need better, more data. Imagine I am very focused right now. I need to know the context of my environment that's driving that. What's in that environment? Is it internal focus that's gotten me there? What is my external environment? Understanding my awake state for me is very dependent on the data and interactions that happen from these different environments.

Let me give an example. If I'm in my home or in a vehicle and you are measuring information about me, and you know I'm under stress or experiencing joy or heightened attention, those are different states you may want my home or my system to react to and mitigate. If you get sleepy in a self-driving, smart vehicle, it will make adjustments. Potentially, it will make adjustments, but not necessarily right for you. That's an important part: optimizing for personalization and how a system responds.

An HVAC system or the internal state of the vehicle is going to adjust sound, background music, haptic feedback, temperature, lighting, the position of your chair, dynamics of what's in your space. All of these different systems can react. The important thing is how you react is going to shift me. The goal is not to measure me, but to actually intersect with my state and move it in some direction.

Andrew Huberman: I always think of devices as good at measurement or modification. Measurement or modification.

Poppy Crum: Measurement is critical. But measurement not just of me, but also of my environment and understanding of the external environment. This is where things like Earth observation come in. We're getting to a place where we're getting really good image quality data from satellites going in the sky at much lower distances, so you now have faster reaction times between technologies and the information they have.

Andrew Huberman: Can you give me an example where that impacts everyday life? Are we talking about weather analysis?

Poppy Crum: Sure. Weather predictions, car environment, things happening.

Andrew Huberman: What about traffic? Why haven't they solved traffic yet, given all the knowledge of object flow and how to optimize for object flow? We've got satellites that can look at traffic and open up roads dynamically, like change the number of lanes. Why isn't that happening?

Poppy Crum: The traffic problem gets resolved when you have autonomous vehicles in ways that don't have the human side of things. That gets resolved. Autonomous vehicles would solve traffic. You probably wouldn't have traffic in the ways that you do with human behavior. That's reason alone to shift to autonomous vehicles. It is that injection from the human system that is interrupting all the models.

I think the world right now, we think about wearables a lot. Wearables track us. You have smart mattresses which are wonderful for understanding. There's so much you learn from a smart mattress and ways of both measuring and intervening to optimize your sleep, which is the beauty. It's this nice, incredible period of time where you can measure so many things.

In our home, I used the example of a thermostat. It's pretty dumb about what my goals are or what I'm trying to do at that moment in time. But it doesn't have to be. There's a company called Passive Logic. I love them. They have some of the smartest digital twin HVAC systems. Their sensors measure things like sound. They measure carbon dioxide, your CO2 levels. When we breathe, we give off CO2.

Imagine a dynamic mixture of acetone, isoprene, and carbon dioxide that's constantly exchanging when I get stressed or when I'm feeling happiness or suspense. That dynamic cocktail mixture in my breath is both an indicator of my state, but it's also something that the spaces around me have more information to contribute about how I'm feeling. It can also be part of that solution in ways that I don't have to have things on my body.

I have sensors now that can measure CO2. You can watch my TED Talk. I have given examples. We brought people in when I was at Dolby and had them watching *Free Solo*, the Alex Honnold movie where they're climbing El Cap. Stressful. Carbon dioxide is heavier than air. We could measure carbon dioxide from just tubes on the ground. You could get the real-time differential of CO2 in there.

Andrew Huberman: Were they scared throughout?

Poppy Crum: I like to say we broadcast how we're feeling. We do that wherever we are. In this, you could look at the time series of carbon dioxide levels and be able to know what was happening in the film without actually having it annotated. You could tell where he summited, where he had to abandon his climb, where he hurt his ankle.

There's another study where they had different audiences watching *Hunger Games*. Different days, different people. You can tell exactly where Katniss's dress catches on fire. We really are sort of digital exhaust of how we're feeling. Our thermals, we radiate the things we're feeling. I'm very bullish on the power of our eye.

Andrew Huberman: Our eye?

Poppy Crum: Our eye, yes. The diameter of our pupil.

Andrew Huberman: Our pupil size. Back when I was a physiologist, I worked with a lot of species on understanding information processing internally in cells, but also I would very often use pupillometry as an indicator of perceptual engagement and experience. Bigger pupils mean more arousal, higher levels of alertness.

Poppy Crum: More arousal, cognitive load, or obviously lighting changes. The thing that's changing from 20 years ago, 15 years ago, it was very expensive to track the kind of resolution and data to leverage all of those autonomic nervous system deterministic responses. Those ones are deterministic, they're not probabilistic. Those are the ones where the body's reacting even if the brain doesn't say anything about it.

Andrew Huberman: Below conscious detection.

Poppy Crum: Today we can do that with open source software on our laptops or our mobile devices. Every pair of smart glasses will be tracking this information when we wear them. It becomes a channel of data. It may be an ambiguous signature in the sense that there's changes in lighting. Am I aroused or am I...

Andrew Huberman: Those can be adjusted for. If you wear eyeglasses that are measuring pupil size, the eyeglasses could have a sensor that detects levels of illumination in the room at the level of my eyes. It could measure how dynamic that is. We just make that the denominator in a fraction. Then we just look at changes in pupil size as the numerator in that fraction. More or less.

Poppy Crum: You just have to have other sensors. All you need to do is cancel. As you walk from a shadowed area to a brighter area, the pupil size changes, but then you can adjust for that change. You normalize for that. You end up with an index of arousal.

Andrew Huberman: Which is amazing. You could also use the index of illumination as a useful measure of whether you're getting compared to your vitamin D levels, or maybe you need more illumination in order to get more arousal. It could tell you all of this. It could literally say, "Hey, take a five-minute walk outside to the left after work and you will get your photon requirement for the day." Not just measuring steps. All this stuff is possible now. I just don't know why it's not being integrated into single devices more quickly.

Poppy Crum: You'd love to also know that person's blood sugar instead of drawing their blood and taking it down to the floor. You think about the resident that's been up for 13 hours because that's the standard in the field, and they're making mistakes on a chart. I think at some point we're just going to go, "I can't believe we used to do it that way." It's crazy.

Andrew Huberman: A lot of the consumer devices and computation we can do from cameras, exhalant, or other data in our environments tell us about our physical state in some of these situations. Why isn't it happening?

Andrew Huberman: A lot of the reasons are simply the regulatory process is antiquated and not up to keeping up with the acceleration of innovation that's happening. Getting things through the FDA, even if they're deemed in the same ballpark and supposed to move fast, the regulatory costs and processes is really high. And you end up many years down the road from when the capability and the data and technology actually should have arisen to be used in a hospital or to be used in a place where you actually have that kind of appreciation for the data and use.

The consumer-grade devices for tracking of data of our biological processes are on par and in many cases surpassed the medical-grade devices. And that's because they just have. But then they will have to bill what they do and what they tracking in some way that is consumer, you know, is not making the medical claims to allow them to be able to continue to move forward in those spaces. But there's no question that that's a big part of what holds back the availability of a lot of these devices and capabilities.

I'd like to take a quick break and acknowledge one of our sponsors, Function. Last year, I became a Function member after searching for the most comprehensive approach to lab testing. Function provides over 100 advanced lab tests that give you a key snapshot of your entire bodily health. This snapshot offers you with insights on your heart health, hormone health, immune functioning, nutrient levels, and much more. They've also recently added tests for toxins, such as BPA exposure from harmful plastics, and tests for PFASs, or forever chemicals.

Function not only provides testing of over 100 biomarkers key to your physical and mental health, but it also analyzes these results and provides insights from top doctors who are expert in the relevant areas. For example, in one of my first tests with Function, I learned that I had elevated levels of mercury in my blood. Function not only helped me detect that, but offered insights into how best to reduce my mercury levels, which included limiting my tuna consumption. I'd been eating a lot of tuna while also making an effort to eat more leafy greens and supplementing with NAC and acetylcysteine, both of which can support glutathione production and detoxification. And I should say, by taking a second Function test, that approach worked.

Comprehensive blood testing is vitally important. There's so many things related to your mental and physical health that can only be detected in a blood test. The problem is blood testing has always been very expensive and complicated. In contrast, I've been super impressed by Function's simplicity and at the level of cost. It is very affordable. As a consequence, I decided to join their scientific advisory board, and I'm thrilled that they're sponsoring the podcast. If you'd like to try Function, you can go to functionhealth.com slash Huberman. Function currently has a wait list of over 250,000 people, but they're offering early access to Huberman podcast listeners. Again, that's functionhealth.com slash Huberman to get early access to Function.

Okay, so I agree that we need more data and that there are a lot of different sensors out there that can measure blood glucose and sleep and temperature and breathing and all sorts of things. Which raises the question of, are we going to need tons of sensors? I mean, are we going to be just wrapped in sensors as clothing? Are we going to be wearing 12 watches? What is this going to look like? I am an advocate for fewer things on you, not having all this stuff on our bodies.

Poppy Crum: There is so much we can get out of the computer vision side, from the cameras in our spaces and how they're supporting us in our rooms. The sensors in our HVAC systems, I brought that up earlier. So now you've got effectively a digital twin and sensors that are tracking my metabolic rates just in my space. They're tracking carbon dioxide. They're tracking sound. You're getting context because of that. You're getting intelligence, and now you're able to start having more information from what's happening in my environment. The same is true in my vehicle. You can tell whether I'm stressed or how I'm feeling just by the posture I have sitting in my car.

And you need AI. This is AI interpretation of data. But what's driving that posture might be coming from an understanding of what else is happening in that environment. So it's suddenly this, with contextual intelligence, AI-driven understanding of what's happening in that space that's driving the state of me. How do I keep leaning to the side because I'm thinking about the way I move and sit? It's a proxy for what's actually happening inside me. And then you've also got data around me coming from my environment. What's happening if I'm driving a car or what's happening in my home, in the weather, in threats that might be outside, in noise that's happening not inside the space, but things that give context to have more intelligence with the systems we have.

So I'm a huge believer that we aren't anywhere until we have integration of those systems between the body, the local environment, and the external environment. And we're finally at a place where AI can help us start integrating that data. In terms of wearables, though, some of the big companies, we've got the watch we have on our hand that has a lot of information that is very relevant to our bodies. The devices we put in our ears, you may not realize, but a dime-sized patch in your concha, we can know heart rate, blood oxygen level. Because of the electrical signature that your eye produces when it moves back and forth, we can know what you're looking at just from measuring a signature, measuring your electrooculogram in your ear. We can measure EEG, electroencephalograms. You can also get eye movements out of electroencephalograms, but you can get attention. You can know what people are attending to based on signatures in their ear. So our earbuds become a window to our state. And you've got a number of companies working on that right now.

So do we need to wear lots of different sensors? No. Do we need to have the sensors, the data we have, whether it's on our bodies or off our bodies, be able to work together and not be proprietary to just one company but to be able to integrate with other companies? That becomes really important. You need integrative systems so that the data they have can interact with the systems that surround you or the mattress I'm sleeping on. Because you've had a lot of specialty of design come from different developers. And that's partly been a product of the FDA and the regulatory pathways because of the cost of development. It tends to move companies towards specialization unless they're very large. But where we're at today, we're getting to a point where you're going to start seeing a lot of this data get integrated, I think. And by all means, hopefully we're not going to be wearing a lot of things on our bodies. I sure as heck won't. The more we put on our bodies, it affects our gait. It has ramifications in so many different ways.

When I got here, I was talking to some of the people that work with you and they asked, "What wearables do you wear?" And I actually don't wear many at all. I have worn rings. I've worn watches at different times. But for me, the importance is the point at which I get insights. I am a big believer in as little on my body as possible when it comes to wearables.

One interesting company that I think is worth mentioning is Pison. They have a form factor like a Timex watch, or they partnered with Timex, but they are measuring psychomotor vigilance. It's like an ENG, electro neural modulation, and they're trying to understand fatigue and neural attentiveness in a way that is continuous and useful for high-risk operations or training, whether in sport. What I like about it is it's actually trying to get at a higher level cognitive state from the biometrics that you're measuring. That, to me, is an exciting direction: when you're actually doing something that you could make a decision about how I engage in my work or how I engage in my training or my life based on that data about my cognitive state and how effective I'm going to be. And then I can start associating that data with other data to make better decisions, better insights at a certain point in time. And that becomes your digital twin.

Andrew Huberman: It's interesting. Earlier you said you don't like the word gamification, but one thing that I think has really been effective in the sleep space has been this notion of a sleep score where people aspire to get a high sleep score. And if they don't, they don't see that as a disparagement of them, but rather that they need to adjust their behavior. So it's not like, "Oh, I'm a terrible sleeper and I'll never be a good sleeper." It gives them something to aspire to on a night-by-night basis. I feel like that's been pretty effective. When I say gamification, I don't necessarily mean competitive with others, but I mean encouraging of oneself.

So I could imagine this showing up in other domains too for wakeful states. Like I spent—I had very few highly distracted work bouts or something like that. I'd love to know at the end of my day I had three really solid work bouts of an hour each at least. That would feel good. It was like a day well spent even if I didn't accomplish what I wanted to in its entirety. Like I put in some really good solid work. Right now it's all very subjective.

We know the gamification of steps was very effective as a public messaging. 10,000 steps a day. We now know you want to get somewhere exceeding 7,000 as a threshold. But if you think about it, we could have just as easily said, "Hey, you want to walk at a reasonable pace for you for 30 minutes per day." But somehow the counting steps thing was more effective because people I know who are not fanatic about exercise at all will tell me, "I make sure I get my 11,000 steps per day." People tell me this. I'm like, "Oh, okay." So apparently it's a meaningful thing for people. So I think quantification of performance creates this aspirational state. I think that can be very useful.

Poppy Crum: Data and understanding the quantification that you're working towards is really important. Those are summary statistics effectively. Maybe they're good on some level to aim for if it means that people move more. All for it. And it's something that if I didn't move as much before and this is making me do it, that's great. But it's also great when now through a computer vision app, I can understand it's not just 10,000 steps, but maybe there's a small battery of things I'm trying to perform against that are helping shape me neurally with the feedback and the targets that I'm getting so that there's more nuance towards achieving the goal I'm aiming for, which is what I'm all about from a neuroplasticity perspective.

So I just don't like the word gamification. I believe everything should be fun. Everything. Training can be fun and gamified in some ways. My life has been predominantly in industry, but I've always loved teaching and I've always been at Stanford. I try to use technology and merge it with the human system in a way that does help optimize learning and training from a neural circuit first perspective. How do we think about the neural system and use this more enjoyable, understandable target to engage with it?

One of my favorite examples though, there was a period from 2018 to 2020 into the pandemic where the students I noticed had a lot of projects. Their final project, they could build whatever they wanted. They had to do projects where they build neural brain computer interfaces. They had to build projects in VR. They had to build AR projects. They had to build projects that use any sort of input device. They have to use different sensor-driven input devices, and that's all part of what they develop. Around 2018, 2020, I started to see almost every project had a wellness component to it, which I loved. I thought that was a very notable shift in the student body, and maybe you've seen that too.

But I still got this—one of my favorite games today was this VR game where I'm in a morgue, I wake up, I've got to solve an escape room, I've got zombies that are coming out of me and they're climbing out of the morgue and they're getting closer and there's people breathing on my neck. And it's a wellness app. Go figure. It was their idea of, "Look, this is what I feel like. I've got to—because I'm also measuring my breath and heart rate and I've got to keep those biological signatures. Everything about how the zombies in solving my escape room problems, they're going to get closer to me if my breath rate goes up, if my heart rate goes up. I've got to keep—"

Andrew Huberman: So it was about stress control, basically.

Poppy Crum: Exactly. Yes. But it was in that environment and it was realized for them how they felt. You can do it in much simpler ways, but I'm a huge fan of how do we use the right quantification to develop the right habits, the right skills, the right acuity or resolution in a domain we might not be able to break into the pieces we need, but it's going to help us get there because my brain actually needs to learn to understand that sophistication.

Andrew Huberman: It's clear to me that in the health space, giving people information that scares them is great for getting them to not do things, but it's very difficult to scare people into doing the right things. You need to incentivize people to do the right things by making it engaging and fun and quantifiable. I like the example of the zombie game.

Okay, so fortunately we won't have to wear dozens of sensors. They'll be more integrated over time. I'm happy to walk through a cheat sheet later for building out a computer vision app for quantifying some of these more personalized domain-related things that people might want to do.

Poppy Crum: That would be awesome.

Andrew Huberman: Yeah. And then we can post a link to it in the show note captions, because I think that the example you gave of creating an app that can analyze swimming performance, running gait, focused work bouts—I think that's really intriguing to a lot of people. But I think there's a gap there between hearing about it, thinking it's really cool, and how to implement. So I'd certainly appreciate it. I know the audience would too.

Poppy Crum: And that's very generous of you. Thank you.

Andrew Huberman: Yes, absolutely. And we're in an era where all you hear about is AI and AI tools. And there are tools that absolutely accelerate our capabilities as humans. But we gave the examples of talking about some of the LLMs. I was at a film premiere and I was sitting next to a few students who happened to be from Berkeley. They were computer science students and double engineering. And one of them, when he knew what I talk about or care about, he said, "I'm really worried my peer group can't start a paper without ChatGPT." It was a truth, but it was also a concern. So they understand the implications of what's happening.

We're in an era of agents everywhere. Reid has said, and a number of people have said, we'll be using AI agents for everything at work in the next five years. And some of those things we need to use. Agents will accelerate capability. They will accelerate short-term revenue. But they also will diminish workforce cognitive skill. As a user of agents in any environment, as an owner of companies employing agents, you have to think hard about what the near-term and long-term ramifications are. It doesn't mean you don't use your agents in places where you need to, but without the germane cognitive load, there is a different dependence that you have to have down the road. You also have to think about how you engage with the right competence to keep your humans that are engaged with developing their cognitive skills and their mental schemas to be able to support your systems down the road.

Let's talk more about digital twins. I don't think this concept has really landed squarely in people's minds as a specific thing. I think people hear AI, they know what AI is more or less. They hear about a smartphone, they obviously know what a smartphone is. Everyone uses one, it seems. But what is a digital twin? I think when people hear the word twin, they think it's a twin of us. Earlier, you pointed out that's not necessarily the case. It can be a useful tool for some area of our life, but it's not a replica of us, correct?

Poppy Crum: Not at all in the ways that I think are most relevant. Maybe there are some side cases that think about that. Two things to think about. When I talk about digital twins to companies, I like to frame it on how it's being used, the immediacy of the data from the digital twin. So let's go back 50 years. An example of a digital twin that we still use: air traffic controllers. When an air traffic controller sits down and looks at a screen, they're not looking at a spreadsheet. They're looking at a digitization of information about physical objects that is meant to give them fast reaction times, make them understand the landscape as effectively as possible. We would call that situational awareness. I've got to take in data about the environment around me and I've got to be able to action on it as rapidly as possible to make the right decisions that mitigate any potential problems or risks. That's what you're trying to engage a human system. The visualization of that data is important, or doesn't have to be visualization, the interpretation of it. It's not the raw data. It's how is that data represented? You want the key information in a way that the salient most important information about planes is able to be acted on by that human or even autonomous system.

Andrew Huberman: Could you give me an example in a more typical home environment? We both are into reefing. I built an aquacultured reef in my kitchen, partly because I have a child and I wanted her to understand the fragility of the ecosystems that happen in the ocean and things we need to worry about and care about. Initially when I started, when you build a reef tank and do saltwater fish, you're doing chemical measurements by hand, usually weekly, biweekly. There's 10 different chemicals that you're measuring. I would have my daughter doing that so that she would do the science part of it. You know the ranges, the tolerances you have, and you're also observing this ecosystem and looking for problems. By the time you see a problem, you're reacting to that problem. I can tell you, it was very unsuccessful. There's lots of error and noise in human measurements. You don't have the right resolution of measurements. Every few days is not enough to track a problem. You also have the issue of being reactive instead of proactive. The point at which it's visible to you, it's probably too late to do anything about it.

Poppy Crum: If you look at my reef tank right now, I have a number of digital sensors in it. I have dashboards. I can track a huge chemical assay that is tracked in real time so that I can go back and look at the data. I can see, "Oh, there was a water change there." I can tell what's happening by looking at the data. The spectrum of my lights is on a cycle of effect that's representative of the environment that the corals you're aquaculturing would, their deterministic systems are looking for. So you've built this ecosystem. When I look at my dashboards, I have a digital twin of that system. My tank is very stable. My tank knows what's wrong. I can look at the data and understand that an event happened somewhere that could have been mitigated, or I can understand that something is wrong quickly before it even shows up. It's amazing.

Andrew Huberman: I think for people who aren't into reefing, it might seem—I know people that are soon to have kids. Most everybody nowadays has a camera on the sleeping environment of their kids so that if their kid wakes up in the middle of the night, they can see it, they can hear it. Camera and microphone. Do you think we either have now or soon will have AI tools that will help us better understand the health status of infants? Parents learn intuitively over time based on diaper changes, cries, frequency of illnesses, how well their kids are doing before kids can communicate that. Do you think AI can help parents be better parents by giving real-time feedback on the health information of their kids, not just if they're awake or asleep or if they're in some sort of trouble, but really help us adjust our care of our young? What's more important for our species than supporting the growth of our next generation?

Poppy Crum: No, absolutely. But I'd even go more on the biological side. Think about digital twins. I'll get to babies in a moment. If you've ever bought a plane ticket, that's a very sophisticated digital twin. Not the air traffic controllers looking at planes, but the pricing models for what data is going into driving that price in real time. You might be trying to buy a ticket and you go back an hour later and it's double or it's gone up. That's because it's using constant data from environments, from things happening in the world, from geopolitical issues, from things happening in the market that's driving that price. That is very much an AI-driven digital twin that's driving the value of that ticket. There are places where we use digital twins that affect our lives but we don't think about them as a digital twin.

Then you think about a different example where you've got a whole sandbox model. The NFL might have a digital twin of every player that's in the NFL. They know the data. They're tracking that information. They know how people are going to perform many times.

Andrew Huberman: What do they care about?

Poppy Crum: They want to anticipate someone might be high risk for an injury so that they can mitigate it.

Andrew Huberman: They're using those kinds of data?

Poppy Crum: Absolutely.

Andrew Huberman: Interesting. I think the word twin is the misleading part. I feel like soon that nomenclature needs to be replaced because people hear twin, they think a duplicate of yourself.

Poppy Crum: Yes.

Andrew Huberman: I feel like these are—

Poppy Crum: Well, it's a duplicate of relevant data and information about yourself, but not just trying to emulate myself. It's to emulate key—imagine me as a physical system. I'm going to digitize some of that data. Whatever data I have, I interact with it to make intelligent insights and feedback loops in the digital environment about how that physical system is going to behave.

Andrew Huberman: So it's a digital representative.

Poppy Crum: Yes.

Andrew Huberman: More than a digital twin.

Poppy Crum: Yes. I'm not trying to split it. There are many digital twins. Even you've got data, you live with lots of what the world would call a digital twin. But I like a digital representative, and it's informing some aspect of decision making and it's many feedback. So I'm digitizing different things. In that situational awareness model, can I give a quick example? Imagine I digitize an environment. I can digitize the space we're in right now. Would that be a digital twin? In situational awareness, there's the state of the sensor limitations, the acuity of the data I've actually brought in. That's like perception. Same with our sensory systems. Then there's comprehension. Comprehension would be like, "Okay, that's a table. That's a chair. That's a person." Now I'm in those semantic units of relevance that the digitization takes. Then there's the insight. What's happening in that environment? What do I do with that? That's where things get interesting. That's where a lot of the future of AI products is, because then the feedback loops of what's happening with that input and that data become interesting and important when you start having multiple layers of relevant data that are interacting that can give you the right insights about what's happening, what to anticipate, in that space. But that's all about our situational awareness and intelligence in that environment.

Andrew Huberman: Yeah, I can see where these technologies could take us. I think for the general public right now, AI is super scary because we hear most about AI developing its own forms of intelligence that turn on us. I think people are gradually getting on board the idea that AI can be very useful. We have digital representatives already out there for us in these different domains.

Poppy Crum: Yes, absolutely.

Andrew Huberman: I think being able to customize them for our unique challenges and our unique goals is really what's most exciting to me.

Poppy Crum: I love that because I think what I was trying to say is exactly what you said. They are out there and these are effectively digital twins. Every company that you're interacting with on social media has an effectively a digital twin of you in some place. It's not to emulate your body, but it's to emulate your behaviors. In those spaces, you're using tools that have digital twins for things you do in your daily life. So the question is, how do we harness that for our success, for individual success, for understanding an agency of what that can mean for you? If the NFL is using it for a player, you can use it as an athlete at any level. It's that digitization of information that can feed you. For my baby, you can better understand a great deal about how they're successful or what isn't successful about them. Not your baby's always successful, I don't want to say, but what maybe isn't working well for them. The exciting places about digital twins come in once you start integrating the data from different places that tell us about the success of our systems, and those are anchored with actual successes.

Poppy Crum: I think you used an example of your mattress and sleep, or even like you, one I liked was: "I had three good, very focused work sessions." You may have used different words, Andy. But the idea is, okay, you had those, but when you can correlate it with other systems and other outputs, that becomes powerful. The way a digital representative or a digital twin becomes more useful is thinking about not the resolution of the data, but the data source—where the data is coming from. Meaning, is it biometric data? Is it environmental data? Is it the context of the state of what else was happening during those work sessions? And how is that something that I don't have to think about, but AI can help me understand where I'm successful and what else drove that success or what drove that state? Because it's not just my success. It's intelligence. I like to call it situational intelligence. It's the overarching goal that we want to have. And that involves my body and systems having situational awareness, but it's really a lot of integration of data that AI is very powerful for. Thinking about how it optimizes and gives us the insights. It doesn't have to just have systems behave, but it can give us the insights of how effectively we can act in those environments.

Andrew Huberman: Yeah, I think the AI is being able to see what we can't see.

Poppy Crum: Yes.

Andrew Huberman: So, for instance, if I had some sort of AI representative that paid attention to my work environment and to my ability to focus as I'm trying to do focused work, and it turned out—obviously I'm making this up—but it turned out that every time my air conditioner clicked over to silent or back to on, that it would break my focus for the next 10 minutes.

Poppy Crum: Yes.

Andrew Huberman: And I wasn't aware of that. And by the way, for people listening, this is entirely plausible because so many of our states of mind are triggered by cues that we're just fundamentally unaware of. Or that it's always at the 35-minute mark that my eyes start to have to reread words or lines because somehow my attention is drifting, or that it's paragraphs of longer than a certain length. It's a near infinite space for us to explore on our own, but for AI to explore it, it's straightforward.

Poppy Crum: Right.

Andrew Huberman: And so it can see through our literal, our cognitive blind spots and our functional blind spots. And I think of where people pay a lot of money right now to get information to get around their blind spots: things like when you have a pain and you don't know what it is, you go to this thing called a doctor. Or when you have a problem and you don't know how to sort it out, you might talk to a therapist. People pay a lot of money for that. I'm not saying AI should replace all of that, but I do think AI can see things that we can't see.

Poppy Crum: Two examples to your point, which I love. The reading—potentially there's a point at which you're experiencing fatigue. Ideally, much like the fish tank, you want to be not reactive, you want to be proactive. You want to mitigate it. You could have your devices integrate that data and respond to give you feedback when your mental acuity, your vigilance, or your effectiveness has waned. But also on the level of health, we know AI is huge for identifying a lot of different pathologies out of data that as humans, we're just not that good at discerning. Our voice. In the last 10 years, we've become much more aware of the different pathologies that can be discerned from AI assessments of our speech—not what we say, but how we say it.

Andrew Huberman: There's a lab up at the University of Washington. I think it's Sam Golden's lab, who's working on some really impressive algorithms to analyze speech patterns as a way to predict suicidality, and to great success, where people don't realize that they're drifting in that direction.

Poppy Crum: Yeah.

Andrew Huberman: And phones can potentially warn people, warn themselves, that they're drifting in a particular direction. People who have cycles of depression or mania can know whether or not they're drifting into that. That can be extremely useful. They can discern who else gets that information. I think it's all based on tonality at different times of day, stuff that even in a close, close relationship with a therapist over many years, they might not be able to detect if the person becomes reclusive or something of that sort.

Poppy Crum: Absolutely. I mean, neural degeneration shows up in short assessments of how people speak. They've definitely been able to show potential likelihood of psychosis, and that's with syntactic completion and how people read paragraphs. Neural degeneration, though—things like Alzheimer's—show up in speech because of the linguistic cues. Sometimes 10 years before a typical clinical symptom would show up that would be identified. And what I think is important for people to realize is it's not someone saying, "I don't remember." It's nothing like that. It's not those cues that you think are actually relevant. It's more like an individual says something like that—what I just did, which was I purposely stuttered. I started a word again. It's what we might call a stutter in how we're speaking, sometimes duration of spaces between starting one sentence to the next. These are things that as humans we've adapted to not pick up on because it makes us ineffective in communication, but an algorithm can do so very well. Diabetes, heart disease, both show up in voice. Diabetes shows up because you can pick up on dehydration in the voice. Again, I'm a sound person in my heart and my past. If you look at the spectrum of sound, you're going to see changes that show up. There are very consistent things in a voice that show up with dehydration in the spectral salience. As well as with heart disease, you get a sort of flutter that shows up. It's a proxy for things happening inside your body with cardiovascular issues. But you're going to see them as certain modulatory fluctuations in certain frequency bands. Again, we don't walk around as a partner or a spouse or a child caretaking our parents and listening for the four-kilohertz modulation, but an algorithm can. All of these are places where you can identify something that you can potentially mitigate proactively before there's a problem. And especially with neural degeneration, we're really just getting to a place where there are pharmacological opportunities to slow something down. You want to find that as quick as possible. So where do you want to have that input so that you can do something about it? You asked me about the babies. The type of coughs we have tell us a lot about different pathologies. For a baby, their cry—if I were thinking about a digital twin, where would I be most interested in using that information if I had children? From the lowest touch, most opportunity, it's to identify potential pathologies or issues early based on the natural sounds and the natural utterances that are happening, to understand if there is something that could be helped. You could proactively make something much better.

Andrew Huberman: Let's talk about you.

Poppy Crum: Oh, boy.

Andrew Huberman: And how you got into all of this stuff, because you're highly unusual in the neuroscience space. I recall when we were graduate students, when you were working on auditory perception and physiology, and then years later, now you're involved in AI, neuroplasticity, you were at Dolby. What is, to you, the most interesting question that's driving all of this? What guides your choices about what to work on?

Poppy Crum: Human technology intersection and perception is my core. I say perception, but the world is data. How our brains take in the data that we consume to optimize how we experience the world is what I care about across all of what I've spent my time doing. For me, technology is such a huge part of that. I like to innovate. I like to build things. But I also like to think about how do we improve human performance. Core to improving human performance is understanding how we're different, not just how we're similar. The nuances of how our brains are shaped and how they're influenced, and thus why I've spent so much time in neuroplasticity. It is at the intersection of everything: how are we changing and how do we harness that? How do we make it something that we have agency over, whether it's from the technologies we build and we innovate to the point of "I want to feel better. I want to be successful. I don't want that to be something left to surprise me."

Andrew Huberman: Right.

Poppy Crum: So you asked me how I got there. One thing—I was a violinist back in the day. I'm still a violinist and music is a part of my life, but I was studying music and engineering when I was an undergrad. I think we alluded to the fact I have absolute pitch. For anyone that doesn't know, it's not anything that means I always sing in tune. What it means is I hear the world like people see color.

Andrew Huberman: Okay.

Poppy Crum: And I can't turn it off really. I can kind of push it back.

Andrew Huberman: Wait, sorry. Don't we all hear sound like we see? I mean, I hear sounds and I see colors. Could you clarify what you mean?

Poppy Crum: When you walk down the street, your brain is going, "Oh, that's red, that's black, that's blue, that's green." My brain is going, "That's an A, that's a B, that's a G, that's an F."

Andrew Huberman: I see. You're categorizing.

Poppy Crum: There's a categorical perception about it. And because of the nature of my exposure to sound in my life, I also know what frequency it is. I can say that's 350 hertz or that's 400 hertz or that's 442 hertz. It has different applications. I can transcribe a jazz solo when I listen to it. That's a great party trick. But it doesn't mean that it's necessarily a good thing for a musician. As well as you do, categorical perception is something we all have different forms of, usually for speech and language. Units of vowels or phonetic units—you can hear many different versions of an "E" and still hear it as an "E." That's categorical perception. My brain does the same thing for a set of frequencies to hear it as an "A." That can be good at times, but when you're actually a musician, there's a lot more subtlety that goes into how you play with other people and what key you're in. If you ask me to sing "Happy Birthday," I'm always going to sing it in the key of G if I am left to my own devices. And I will get you there somehow if we start somewhere else. So what happened to me when I was in music school, in conservatory and also engineering school, is I knew that I had to override my brain because it was not allowing me the subtlety I wanted to play my Shostakovich or play my chamber music. I was having to work too hard to override these categories of sounds I was hearing. So I started playing early music—Baroque music. For anyone, I think I said earlier, "A" is a social construct. Today, as a standard, "A" is 440 hertz. If you go back to the 1700s, "A" was 415 hertz in the Baroque era. And 415 hertz is effectively a G sharp. So it's the difference between "ah" and "ah." What would happen to me when I was trying to override this is I was playing in an early music ensemble. I would tune my violin up, and I would see "A" on the page, and I'd hear G sharp in my brain. It was completely—I was terrible. It was really hard for my brain to override. Brass and wind players do this all the time. It's transposition, and they modulate to the key that they're in, and their brains have evolved through their training in neuroplasticity to be able to not have the same sort of experience I had. Anyhow, long story long, I was also taking a neuroscience course. In this neuroscience course, we were reading papers about map making and neuroplasticity. I read this paper by a professor at Stanford named Eric Knudsen. Eric Knudsen did amazing, seminal work for how we understand the auditory pathways, as well as how we form multisensory objects and the way the brain integrates data across our modalities—sight and sound. In this paper, what he was doing was he had identified cells in the brain that optimally responded. Their receptive fields—that set of data that optimally causes that cell to respond. For these cells, they cared about a particular location in auditory and visual space, which for mammals, we don't have the same sort of cells because we can move our eyes back and forth in our sockets, unlike owls. He studied owls. Owls have a very hardwired map of auditory visual space. If I hear a click off to my right, I turn my head to the right.

Andrew Huberman: You turn your head. It triggers a different vestibular ocular response that moves all of that, yes.

Poppy Crum: But in this case, he had these beautiful hardwired maps of auditory visual space. Then he would rear and raise these owls with prism glasses that effectively shifted their visual system by 15 degrees. Then he would put them—key to developing neuroplasticity—in high-importance situations where they had to do something critical to their survival or their well-being. So they would hunt and they would feed with this 15-degree shift. Consequently, he saw the cells—the auditory neurons—he saw their dendrites realign to the now 15-degree visually shifted cells. It was this realization that they developed a secondary map that was now aligned with the 15-degree shift of the prism glasses, as well as their original map. It was super interesting for understanding how our brains integrate data and the feedback and neuroplasticity. So I go back to my Baroque violin where I'm always out of tune, tuning up my Baroque violin, and I realize I had developed absolute pitch at A415. So I developed a secondary absolute pitch map. Then I would go play Shostakovich right after at A440. And I had that map. I have nothing in between, but I could modulate between the two. That's the point at which I said, "I think my brain is a little weird. I just did something that I need to go better understand." So that's how I ended up here as a neuroscientist.

Andrew Huberman: I know Eric's work really well. Our labs were next door. Our offices were next door. He's retired now.

Poppy Crum: Yeah, I know. He knows. I told him the story. He's wonderful.

Andrew Huberman: I think one of my favorite things about those studies, I think people will find interesting, is that if an animal, human or owl, has a displacement in the world—something's different, something changes and you need to adjust to it. It could be new information coming to you that you need to learn in order to perform your sport correctly or to perform well in class, or an emotionally challenging situation that you need to adjust to. All of that can happen, but it happens much, much faster if your life depends on it. We kind of intuitively know this, but one of my favorite things about his work is where he said, "Okay, well, yeah, these owls can adjust to the prism shift. Their maps in the brain can change, but they sure as heck form much faster if you say, 'Hey, in order to eat—in other words, in order to survive—these maps have to change.'" I like that study so much because we hear all the time that it takes 29 days to form a new habit or it takes 50 days to form a new habit. Actually, you can form a new habit as quickly as is necessary to form that new habit. The limits on neuroplasticity are really set by how critical it is. Of course, if you put a gun to my head right now and said, "Remap your auditory world," I mean, there are limits at the other end too. I can't do that quickly. But I think it's a reminder to me, and thank you for bringing up Eric's work, that neuroplasticity is always in reach. If the incentives are high enough, we can do it. So I think with AI, it's going to be very interesting. With technology generally, our ability to form these new maps of experience, at least with smartphones, has been pretty gradual. I really see 2010 as the beginning of the smartphone. And then now by 2025, we're in a place where most everyone, young and old, has integrated this new technology. I think AI is coming at us very fast and it's unclear what form it's coming at us and where. And as you said, it's already here. I think we will adapt for sure. We'll form the necessary maps. I think being very conscious of which maps are changing is so key. I think we're still doing a lot of cleanup of the detrimental aspects of smartphones—short wavelength light late at night, being in contact with so many people all the time, maybe not so good. I think what scares people, certainly me, is the idea that we're going to be doing a lot of error correction over the next 30 years because we're going so fast with technology, because maps can change really, really fast.

Poppy Crum: They do change. Sam Altman had a—I saw him say this, and I actually thought it was a really good description. Gen X, or that group, is using AI as a tool, sort of novel and interesting. Then you have different—millennials are using it as a search algorithm. Maybe that even includes Gen X, but it's a little more deeply integrated. But then you go back to younger generations, and it's an operating system. And it already is. That has major changes in neural structure for not just maps, but also neural processes for how we deal with information, how we learn. The idea that we are very plastic under pressure, absolutely. And that's where it gets interesting to talk about different species too. We're talking about owls and that was under pressure, but what is successful human performance in training? It's to make those probabilistic situations more deterministic. When you are training as an athlete, you're really trying to not have to think and to have the fastest reaction time to very complex behaviors, given complex stimuli, complex situations and contexts. That situational awareness or physical behavior in those environments, you want that as fast as possible with as little cognitive load as possible. That execution is critical. You love looking across species, so do I. Looking for these ways where a brain is changing or you've got a species that can do something that is absolutely not what you would predict, or it's incredible. How it can evade a predator, how it can find a target, find a mate. It's doing things that are critical to it being able to survive, much as you said. If I make it something that is absolutely necessary for success, it's going to do it.

Andrew Huberman: One of my favorite examples is a particular moth that bats predate on—echolocating bats. Echolocating bats are nature's engineered amazing predatory species. Their brains, when you look at them, are just incredible. They have huge amounts of their brain just dedicated to what's called an FM, constant frequency FM sort of sweep. Some of the bats elicit a call that's sort of like "woo, woo," but really high. So we can't hear it.

Poppy Crum: Yes.

Andrew Huberman: What does that do for them? It's doing two things. That constant frequency portion is allowing them to track the Doppler in a moving object. They're changing subtly how they elicit the call so that it always comes back in the same frequency range, because that's where their heightened sensitivity is. They're modifying their vocal cords to make sure that the call comes back in the same range. And then they're tracking how much they've had to modify their call. Just so that people are on board—bats echolocate.

Poppy Crum: Yeah.

Andrew Huberman: They're sending out sound and they can measure distance and they can essentially see in their mind's eye. They can sense distance. They can sense speed of objects. They can sense shape of objects by virtue of sounds being sent out and coming back. And they're shaping the sounds going out differently so that they can look at multiple objects simultaneously. But also they're shaping the sounds they send out so that whatever comes back is in their optimal neural range. So that they don't have to go through more neural plasticity—they have circuits that are really dedicated to these certain frequency ranges. So they send it out. And then they're keeping track of the deltas. They're keeping track of how much they've had to change it. And that's what tells them the speed. So that constant frequency is a lot like the ambulance sound going by. That's the compression of sound waves that you hear as a "whoo" when things move past you at speed. That's the Doppler effect. And then there also, it has a really fast FM frequency modulated sweep. And that lets me take an imprint of—so one's telling me the speed of the object. Another one's telling me what the surface structure looks like. That FM sweep lets me get a sonic imprint of what's there so I can tell topography. I can tell if there's a moth on a hard surface. So what's beautiful about other species is you've got a little moth and you've got nature's predatory marvel. And 80% of the time, that moth gets away. How? Multiple things. I call it an acoustic arms race that's happening between the two. There's a lot of acoustic subterfuge from the moth. But there's also beautiful deterministic responses that they have. Deterministic behaviors—be it an athlete, be it effectiveness, being fast, quick in making good decisions that get you the right answer—are always important. Moths have just a few neurons. When that echolocating bat is flying, at a certain point, when those neurons start firing, they will start flying in more of a random pattern. You'll see the same thing with seals when there are great white sharks around. It's decreasing the probability that it's easy for them to continue to track you. So they'll fly in a random pattern, and then when their neurons saturate—when those calls get close enough—the moth will drop to the ground with the idea that in a natural world, the ground is wheat, grass. It's a difficult environment for an echolocating bat to locate you. So that is a deterministic behavior that will happen regardless. But then the interesting part is their body is reflecting meta-reflectors effectively, so that the bat may put out its call and it deflects the energy of the call away from its body. So you're deflecting it away from critical areas. The changes in the physical body are interesting, but then it's the behavioral differences that are really key. How fast does that moth react? If it had to question, or if it were cognitively responsive instead of being deterministic in its behavior, it wouldn't escape. But it gets away.

Andrew Huberman: I've never thought about bats and moths. I never got the insect bug—no pun intended. I never got the insect bug because I don't think of things in the auditory domain. I think of things in the visual domain. And some insects are very visual, but it's good for me to think about that. One of my favorite people, although I never met him, was Oliver Sacks, the neurologist and writer. He claimed to have spent a lot of time just sitting in a chair and trying to imagine what life would be like as a bat, as a way to enhance his clinical abilities with patients suffering from different neurologic disorders. So when he would interact with somebody with Parkinson's or with severe autism or with Locked-in syndrome or any number of different deficits of the nervous system, he felt that he could go into their mind a bit to understand what their experience was like. He could empathize with them, and that would make him more effective at treating them. He certainly was very effective at sorting out their experience in ways that brought about a lot of compassion and understanding. He never presented a neural condition in a way that made you feel sorry for the person. It was always the opposite. And I should point out, I'm not trying to be politically correct here, but when I say autistic, I meant the patients he worked with were severely autistic to the point of never being able to take care of themselves. We're not talking about along a spectrum. We're talking about the far end of the spectrum of needing assisted living their entire lives and being from a sensory standpoint extremely sensitive, couldn't go out in public, that kind of thing. So apparently thinking in the auditory domain was useful for him. So I should probably do that. So I have one final question for you, which is really two questions. First question: why did you sing to spiders? And second: what does that tell us about spider webs? Because I confess I know the answers to these questions, but I was absolutely blown away to learn what spider webs are actually for. And you singing to spiders reveals what they're for. So why did you sing to spiders?

Poppy Crum: Two things. You can watch me sing to a spider on a TED Talk I gave a few years ago. We'll put a link to it. Maybe this comes back to I have absolute pitch, so I know what frequencies I'm singing. But I also recognize by having absolute pitch, I know my brain is just a little different. Again, you asked me what threads drive me. It's always been that we do experience the world differently. And I believe that our success, everyone's success, and the success of our growth as humans is partly dependent on how we use technology to help improve and optimize each of us with the different variables we need.

Andrew Huberman: So different species and how they respond to sound is very interesting to me. And as much as you, I know Andy, you look at how different species respond to color and to information in the world, be it cuttlefish or such. I have jellyfish too. And I can see how their pulsing rates change with their photoreceptors with different light colors. It's very obvious that some clearly make that when they're under stress versus when they're in a more calming state. And so it's like understanding the stimuli in our world that shape us. Those changes are a huge part of being human in my perspective.

Poppy Crum: In this case, this happens to be an orb spider, the one I sing to. And when I hit about 880 hertz, you will see the spider kind of dances. But what this particular species, and not all spiders will do this, is predated on by echolocating bats and birds, which makes sense that then, it tunes its web effectively. And orb weavers are all over California. They show up a lot around Thanksgiving, if you are October, November, for anyone that's out here on the West Coast. They're not bad spiders. They are not spiders you need to get rid of. They're totally happy spiders. There are some that maybe you should worry about more.

Anyhow, they tune their webs to resonate like a violin. And when you'll see it, as I hit a certain frequency, it'll effectively tell me to go away. And it's a pretty interesting sort of deterministic response. Other insects do different things. The one kind of funny for that was when my daughter was, I think at the time she was about two and a half or three, and she kind of adopted asking me when we would see spiders if it was the kind we should sing to or the kind we shouldn't touch. Those are the two classes.

Andrew Huberman: So amazing. So if I understand correctly, these orb spiders use their web more or less as an instrument to detect certain sound frequencies in their environment.

Poppy Crum: Resonances, absolutely.

Andrew Huberman: So that they can respond appropriately.

Poppy Crum: Yeah. Either by raising their legs to protect themselves or to attack or whatever it is.

Andrew Huberman: The spider web is a functional thing, not just for catching prey. It's a detection device also. And we know that because when prey are caught in a spider web, they wiggle and then the spider goes over to it and wraps it and eats it. But the idea that it would be tuned to particular frequencies is really wild.

Poppy Crum: Yeah, not just by any vibration, right? Yeah, there's the idea that any vibration means I got food somewhere, I should go to that food source. But instead, it's something that if I experience a threat or something, I'm going to behave. And that is a more selective response that I've tuned it towards.

Andrew Huberman: It's so interesting because if I just transfer it to the visual domain, it's like, yeah, of course, like if an animal, including us, sees something, like a looming object coming at us closer to dark, our immediate response is to either freeze or flee. That's just what we do. The looming response is one of the most fundamental responses, but that's in the visual domain. So the fact that there would be auditory cues that would bring about deterministic responses seems very real. I feel like the wail of somebody in pain evokes a certain response.

Yesterday, there was a lot of noise outside my window at night, and there was a moment where I couldn't tell if these were shouts of glee or shouts of fear. And then I heard this kind of high pitch fluttering that came after the scream. And I realized these were kids playing in the alley outside my house. And I went and looked. I was like, oh, they're definitely playing. But I knew even before I went and looked, based on the kind of flutter of sound that came after the shriek.

Poppy Crum: I can't reproduce the sound at that high frequency.

Andrew Huberman: No, no, but that's super. So the idea that this would be true all the time is super interesting. We just don't tend to focus just on our hearing unless, of course, somebody's blind, in which case they have to rely on it much more.

Poppy Crum: So two interesting things to go with that. So like crickets, for example, crickets have bimodal neurons that have peaks in two different frequency ranges for the same neuron. And each frequency range will elicit a completely different behavior. So you've got a peak at 6K and you've got a peak at 40K. And this is the same neuron. And a cricket hears 40K from a speaker, runs over to it because that's got to be its mate. And you hear 40K and they run away. And it's very predictive behavior.

I spent a good period of time working with non-human primate species, marmosets. Marmosets are very interesting when you get to a more sophisticated neural system. Marmosets are very social. It's critical to their happiness. If you ever see a single marmoset in the zoo or something, that's a very unhappy animal. But they're native to the Amazon, New World monkeys native to Brazil and the Amazon, but they're arboreal. They live in trees and they're very social. So that can be in conflict with each other because you're in dense foliage, but yet you need to communicate.

So they've evolved very interesting systems to be able to achieve what they needed to. One, if you ever see a marmoset, they're very stoic, unlike macaque monkeys that often have a lot of visual expression of how they're feeling. Marmosets always look about the same. But their vocalizations are almost like birdsong. And they're very rich in the information that they're communicating. They also have a pheromonal system. Like, you can have a dominant female in the colony who may not be—because you have to have ways of communicating. When one sense is compromised, the other senses sort of rise up to help assure that the success of what that species or system needs is going to thrive.

And in the case of marmosets, you can have the dominant female effectively cause the ovulation of the biology to change of all the other females. And you can have a female that you put just in the same proximity but now as part of a different group and her biology will change. I mean it's very powerful, the pheromonal interactions that happen because those are things that can travel even when I can't see you.

One thing when I was working with them that I thought was, and I never, I like writing pads more than publishing papers, but these things are real because I was studying pupillometry. I understood the power of their saccades. I could know what they were hearing based on their eye movements, right? So if I play marmosets—they have some of their calls that are really antiphonal. They're to see, hey, are you out there? Am I alone? Who else is around? It's like texting for humans.

Andrew Huberman: Yeah.

Poppy Crum: Yeah. And sometimes it's light or sometimes it might be like, oh, be careful, there's somebody around that we got to watch out for. Maybe there's a leopard on the ground or something, right? And then sometimes it's like, you're in my face. Get out of here now, right? And those are three different things. And I can play that to you and I can tell you without hearing it. And I know exactly what's being heard.

In the case of the antiphonal, hey, are you out there? You see the eye will just start scanning back and forth, right? Because that's the right movement. I'm looking for where's this coming from.

Andrew Huberman: Yeah, they paired the right eye movement with the right sound.

Poppy Crum: Exactly. In the case of, look, there's something to be scared of, threatened of, you're going to see dilation and you're also going to see some scanning, but it's not as slow. It's a lot faster because there's a threat to me. My autonomic system and my cognitive system are reacting differently. And in the case of you're in my face, without seeing you, if I hear another sort of aggressive sound, I'm going to react. I'm going to be—I'm not scanning anywhere, but my dilation is going to be fast and I'm also going to be much more on top of things.

But we do this as humans too, right? And it's like you can walk into a business meeting, walk into a conference room, and these subtle cues are constant. We can't always suppress them. We show them whether we think we do or we don't. But when you look at species like that, it's very much like, OK, there's a lot of sophistication in how their bodies are helping them be successful, even in a world or an environment that has a lot of things that could maybe come after them.

Andrew Huberman: So interesting to think about that in terms of our own human behavior and what we're optimizing for, especially as all these technologies come on board and are sure to come on board even more quickly. Poppy, thank you so much for coming here today to educate us about what you've done, what's here now, what's to come. We covered a lot of different territories, and I'm glad we did because you have expertise in a lot of areas. And I love that you are constantly thinking about technology development.

I drew a little diagram for myself that I'll just describe for people because if I understood correctly, one of the reasons you got into neuroscience and research at all is about this interface between inputs and us. And what sits in between those two things is this incredible feature of our nervous systems, which is neuroplasticity, or what I sometimes like to refer to as self-directed plasticity, because unlike other species, we can decide what we want to change and make the effort to adopt a second map of the auditory world or visual world or take on a new set of learnings in any domain. And we can do it. If we put our mind to it, if the incentives are high enough, we can do it. And at the same time, neuroplasticity is always occurring based on the things we're bombarded with, new technology. So we have to be aware of how we are changing. And we need to intervene at times and leverage those things for our health.

So thank you so much for doing the work that you do. Thank you for coming here to educate us on them and keep us posted. We'll provide links to you singing to spiders and all the rest. My mind's blown. Thank you so much.

Poppy Crum: Thank you, Ed. Great to be here.

Andrew Huberman: Thank you for joining me for today's discussion with Dr. Poppy Crum. To learn more about her work and to find links to the various resources we discussed, please see the show note captions. If you're learning from and or enjoying this podcast, please subscribe to our YouTube channel. That's a terrific zero cost way to support us.

In addition, please follow the podcast by clicking the follow button on both Spotify and Apple. And on both Spotify and Apple, you can leave us up to a five-star review. And you can now leave us comments at both Spotify and Apple. Please also check out the sponsors mentioned at the beginning and throughout today's episode. That is the best way to support this podcast.

If you have questions for me or comments about the podcast or guests or topics that you'd like me to consider for the Huberman Lab podcast, please put those in the comment section on YouTube. I do read all the comments.

For those of you that haven't heard, I have a new book coming out. It's my very first book. It's entitled Protocols, an Operating Manual for the Human Body. This is a book that I've been working on for more than five years, and it's based on more than 30 years of research and experience. And it covers protocols for everything from sleep to exercise to stress control, protocols related to focus and motivation. And of course, I provide the scientific substantiation for the protocols that are included. The book is now available by presale at protocolsbook.com. There you can find links to various vendors. You can pick the one that you like best. Again, the book is called Protocols, an operating manual for the human body.

And if you're not already following me on social media, I am Huberman Lab on all social media platforms. So that's Instagram, X, Threads, Facebook, and LinkedIn. And on all those platforms, I discuss science and science-related tools, some of which overlaps with the content of the Huberman Lab podcast, but much of which is distinct from the information on the Huberman Lab podcast. Again, it's Huberman Lab on all social media platforms.

And if you haven't already subscribed to our Neural Network newsletter, the Neural Network newsletter is a zero-cost monthly newsletter that includes podcast summaries as well as what we call protocols in the form of one- to three-page PDFs that cover everything from how to optimize your sleep, how to optimize dopamine, deliberate cold exposure. We have a foundational fitness protocol that covers cardiovascular training and resistance training. All of that is available completely zero-cost. You simply go to hubermanlab.com, go to the menu tab in the top right corner, scroll down to newsletter and enter your email. And I should emphasize that we do not share your email with anybody.

Thank you once again for joining me for today's discussion with Dr. Poppy Crum. And last but certainly not least, thank you for your interest in science.