The connectome is the wiring diagram of a brain, a big matrix that tells us what neurons talk to what other neurons. Understanding it is an important step to understanding how brains work, but a long way from the final answer. A big next step is understanding how neuronal circuits connect to and guide bodily behavior. Very recent work on mapping the fruit-fly connectome has brought us closer to that goal. I talk with neuroscientist Bing Brunton about the connectome, how we can study it to understand bodily motion in flies and other creatures, and where it's all taking us.
Chubbies is here to keep you comfy and looking good year-round. Get 20% off with code MINDSCAPE at chubbiesshorts.com/MINDSCAPE! #chubbiespod
Upgrade your denim game with Rag & Bone! Get 20% off sitewide with code MINDSCAPE at www.rag-bone.com. #ragandbonepod
Support Mindscape on Patreon.
Bing Wen Brunton received her Ph.D. in neuroscience from Princeton University.. She is currently a Professor of Biology and the Richard & Joan Komen University Chair at the University of Washington, with affiliations at the eScience Institute for Data Science, the Paul G. Allen School of Computer Science & Engineering, and the Department of Applied Mathematics.
Click to Show Episode Transcript
0:00:00.6 Sean Carroll: Hello, everyone. Welcome to the Mindscape podcast. I'm your host, Sean Carroll. As you are listening to this podcast, or listening to anything else, or looking at anything else, your brain is processing information. We can argue about how much information is in the podcast or anywhere else, but in some sense there are bytes of information being sensory inputted into your brain and then processed, and that affects what you do, how you behave. Now, as we've talked about in the podcast recently, there's other things going on in the brain and the nervous system and the body as well, it's not just information processing. There is absolutely information processing happening, but that's an abstraction, right? What there's actually happening are atoms, molecules, cells doing various physical things. And we find it very, very interesting and helpful to talk about those physical processes in terms of information being processed. And today we're not gonna worry about deep questions about whether or not that information processing is sufficient for consciousness or anything like that. We're gonna get our hands dirty a little bit and think about the connection between what goes on in our brains, our nervous systems, and our bodies. There's a constant interaction.
0:01:22.7 SC: In fact, it's even, of course, a little bit of a mistake to separate our brains from our bodies, because our brains are part of our bodies. So in reality, we're gonna be talking about interactions between two different parts of our bodies, how we move around in the world, and how our brains send signals back and forth, receiving signals and then transmitting them to the nervous system, which then does things. We've also talked recently on the podcast about the connectome, the idea that if you knew every neuron in a brain, or maybe some coarse-grained version of groups of neurons and how they connected to each other, you would have the wiring diagram of the brain. And so we have some wiring diagrams for simple organisms. Nowhere close to human beings yet, but we're working on that. What does that give us? Knowing the wiring diagram, knowing how that information flows around, how does that then go into controlling our bodies and what we do and our behavior? So that's what we're gonna be talking about today. Bing Brunton is a neuroscientist and biologist at the University of Washington, and she has been leading the charge in very recent days. We've mapped out the connectome of the fruit fly. You might know that we've mapped out the connectome of C. Elegans, the little worm that biologists like to study. It's only 300 neurons, right? The fruit fly has over 100,000 neurons, and now we've mapped out that. So that's a much more subtle system, a lot more intricate things going on, little subsystems doing different things. And so we're gonna be talking about how we can learn about the relationship between the fruit fly brain. Such as it is, there is a brain there. It's pretty impressive, actually. And how the fruit fly does things like walking around, flying, other kinds of things. This is absolutely new stuff, less than a year old and just the beginning of a forefront of really interesting research in biology and neuroscience. So let's go.
[music]
0:03:41.2 SC: Bing Brunton, welcome to the Mindscape podcast.
0:03:43.3 Bing Brunton: Thanks, Sean. Glad to be here.
0:03:45.3 SC: So I think that for this audience, it would be good to start pretty broadly, because the brain... The brain is kind of like time. I've written books about time, and what I've noticed when I wrote books about time is that everyone has an opinion about how time works, what it is, things like that. And I think that maybe the brain has a little bit of that. Right? We all have brains. [chuckle] People have their opinions about how it works.
0:04:09.9 BB: It's true. I'm rather attached to mine.
0:04:11.5 SC: Yes, exactly. But the connectome in particular is something we have talked about in the podcast before. But why don't you give us the high-level overview of what the connectome is, how the neurons work, all that fun stuff.
0:04:25.7 BB: Yeah, that's actually... Yeah. You went right for it. I think there's actually a little bit... Some of the confusion around connectomes is exactly what it is because people use that word in a different way. And I'm sure you know the terminology actually does matter here. So I think the rough definition... And my colleagues actually differ on this and so I'm gonna try to channel them a little bit. The rough idea is that we all know that the brain is composed of cells because it's an organ, like every organ in your body, so it has cells. And the cells work by electrical activity and they talk to each other through electricity. And so, unlike an anonymous net of cells that are just kind of passing messages forward and backwards, these cells actually have specific identities. Some of them have specific jobs. And they also have specific localization, some cells are found in different parts of the brain and nervous system, and some parts are not. And so there's essentially a wiring diagram, so to speak, of the brain. You can think about it in terms of if you're building a really big complicated building, you're building a skyscraper or something, you would have a wiring diagram, literally an engineering diagram of, "Okay, so this is where the transformers are, I'm gonna flip this switch and this thing's gonna turn on these lights over here," right? So you can sort of have a diagram of that. And that's sort of the connectome, roughly speaking, is that for all the cells and their connections and identities in the brain. Now, the difficulty comes in in terms of how do you actually define the units? Like, do you want a connectome that's necessarily at the scale of individual cells and how they're connected to other individual cells?
0:05:38.7 SC: Okay.
0:05:59.2 BB: Okay? So that's one way people have used that term. But there's sort of more what we call meso-scale connectomes that exist as well, in particular because there's certain animals that are so big, like humans, for example, or even smaller rodents, where we can't really get, technologically, we don't have the capability of getting the cell-by-cell connectome. We just can't do it. Some people think we should, some people think it's impossible, some people think even if we could have it, it's useless. But nevertheless, we have these, like... Like, if you hear about the human connectome, the human connectome is not the scale of cells and how they connect to each other. It's about, it's mostly like brain areas and how the brain areas connect to each other. So people use that term to mean like an area-by-area connectome as well.
0:06:49.4 SC: So there's some coarse graining involved.
0:06:51.4 BB: There's a lot of coarse graining. And so people don't agree on how they use that term. Right?
0:06:54.2 SC: Okay. Okay.
0:06:57.0 BB: So the whole omics thing in biology, so every word that ends in omics, like genomes, proteome, transcriptome, it's supposed to mean comprehensive map thereof. Okay? Now, people usually agree if I tell you, "Hey, Sean, I got a genome of a new, I don't know, spider that I found," you would expect that genome to be at the resolution of the base pairs, the A, C, Gs, and Ts, right?
0:07:20.5 SC: Right, right.
0:07:20.8 BB: Like, you have that expectation. If I gave you something else, you're like, "That's not a genome. I don't know what this is, but it's not a genome," right? So we don't have that in connectomes. Like, we don't quite agree on the scale of description of like, what is a... Like, do you need to have every single neuron in that spider for that to be the connectome of a spider? Right?
0:07:42.5 SC: But the human brain has like 85 billion neurons. We do have some maps of connectomes of more manageable creatures.
0:07:48.1 BB: We do. Some.
[laughter]
0:07:53.6 SC: We'll get there. I did notice you were kind of very careful there about talking about cells rather than talking about neurons.
0:08:02.1 BB: Oh, yeah, let's go there.
0:08:03.4 SC: I presume that's because there are other cells.
0:08:04.5 BB: There are other cells and they're clearly important. So the rough estimate, in my understanding, is that half of the cells in your brain are not neurons. Our word for not neurons is just glia, which doesn't mean anything except just the word for it.
0:08:20.5 SC: Yeah.
0:08:21.5 BB: And they're clearly important. People used to think that they are just there to... You know, kind of like custodial staff or something, but that's so trivializing. They do a lot more than that. They clearly are involved in all kinds of vital functions and have their own dynamics, but we don't understand. I think that's like a really, really exciting emerging field in neuroscience is understanding all of the other cells in your brain and what they do and what they do in concert with the neurons.
0:08:49.4 SC: Let me demonstrate how ignorant I am about biology. I mean, you said the body is made of cells, etcetera. Is it entirely made of cells? Like, is everything in our body cells? There's got to be like just some liquids and solids and things in there.
0:09:02.9 BB: Oh, for sure. Yeah, there's definitely stuff in the extracellular space. Yes. But I think all I meant was that all of life as we know it is made out of cells.
0:09:12.5 SC: Right.
0:09:13.3 BB: We can quibble about viruses later, but living organisms are composed of cells. Yeah.
0:09:18.9 SC: But I think one of the lessons that we're going to be bumping into over the course of the podcast is biology is messy...
0:09:25.1 BB: It's squishy, yeah..
0:09:26.0 SC: Things are more complicated than being... Squishy and interconnected and complex. And I mean, maybe one of the things to keep in mind is that a macroscopic organism is pretty much a matter of teamwork between different kinds of cells, but also cells and non-cell substances.
0:09:45.9 BB: Yep. Yeah. All the stuff, right? Like, for example, your bones, your skeleton, you probably know that it's made out of lots of inorganic compounds, like there's a lot of calcium in there, when you drink your milk, your mom tells you to drink your milk. But your bones, even though the skeletal elements of it, a lot of its material properties come from the calcium matrix and lots of other stuff that's going on that's kind of complicated, it's also this really intricate meshy structure that has blood vessels all inside it, right?
0:10:20.0 SC: Right.
0:10:21.1 BB: Because it needs to be vascularized, otherwise it's gonna die. It needs sugar to be fed, it needs oxygen to stay alive. And so even something that you think is structural, it's not like a stainless steel beam in a building. It's alive, right? And it's alive in a way that only cells can keep it alive. And so there's cells all inside it, and if you just zoom in, it's got very intricate structure.
0:10:42.9 SC: I do think this is not what we're talking about, but I do suspect that that's got to be the frontier of artificial organism building. Like when we build robots, we make steel beams, we don't make it out of cells, and that means it doesn't repair itself, etcetera.
0:10:57.5 BB: We think about that quite a bit. And so not only... I mean, this is relevant for our thinking of connectomes, but really it's just a really great fundamental question of biology is how organisms are able to recover from injury and repair ourselves, or sometimes not.
0:11:19.5 SC: Yes. Well, you're gonna... You and your friends are gonna figure out how to make all of my organs repair themselves and make it soon, okay?
0:11:25.4 BB: We're... I mean, it's... We're gonna try. It's gonna be fun.
0:11:28.2 SC: You're gonna try.
0:11:28.6 BB: Yeah.
0:11:28.8 SC: Okay, just the follow-up, the last little bit, very interesting that half of the cells in my brain are not neurons. They're the other things, the glial cells. So we are... Again, we have this cartoon picture in our brain of the neurons firing signals back and forth to each other. Is it that feature that distinguishes neurons from non-neurons?
0:11:41.5 BB: It is, yeah.
0:11:51.8 SC: And so the connectome is the... The fine-grained connectome, if you want to call it that, the level of cells.
0:12:02.1 BB: We can call it the cellular level one or the neuronal.
0:12:03.6 SC: Cellular level connectome.
0:12:05.2 BB: Neuronal.
0:12:06.8 SC: That would be just a big old matrix listing every single neuron and how it connects to every other neuron.
0:12:15.6 BB: Yep, exactly right. Yeah.
0:12:16.7 SC: Okay, good.
0:12:17.7 BB: From a computational perspective, because I am a computationalist, by the time it gets to me, it's that gigantic connectivity matrix. And it has structure, it's sparse, it's not at all random. It's all kinds of cool.
0:12:32.5 SC: Right. It's not symmetric either.
0:12:33.3 BB: Not at all.
0:12:33.9 SC: Neurons talk to one others, but they don't listen necessarily.
0:12:35.1 BB: That is correct.
0:12:37.8 SC: Okay, some asymmetry there.
0:12:38.7 BB: Yeah.
0:12:39.7 SC: And in the... But okay, so is technically the connectome just the wiring diagram, or is it that extra information about where information flows?
0:12:51.7 BB: So there's a lot of extra information in it. And so this is what the analogy I tried making earlier about the genome as well, like we don't even understand... We don't agree on the correct way of representing the information. So that giant connectivity matrix you talked about, Sean, is definitely a part of the information, but it's nowhere near all the information that we get out of this technology. So, for instance, it matters the identity of the cells, because the neurons are not... I mean, you probably heard about things like dopamine, serotonin, right? Like there's dopamine cells, there's serotonin cells, and if they both fire an action potential, they both say something, those messages are completely different, right? And so the identities of the cells matter.
0:13:13.8 SC: Yep.
0:13:36.0 BB: The other thing that really matters is how those messages are received, right?
0:13:42.0 SC: That's right.
0:13:43.3 BB: So in analogy with... It's very context-dependent language. Trying to think of an interesting social analogy. Like, if you say the same thing to two different people, depending on your relationship with them, they can hear very different messages.
0:13:48.1 SC: Right.
0:14:00.2 BB: Does that make sense?
0:14:01.1 SC: Yep, a hundred percent.
0:14:01.9 BB: So when cell A speaks to cell B and cell A says exactly the same thing to cell C, depending on the identities of B and C, they could hear very different messages and do very different things with it.
0:14:14.1 SC: Sure. If you say, "You're a bonehead," to your best friend, it's received differently than if you say that to your graduate students, right? [laughter]
0:14:21.3 BB: That is entirely correct. Right. So messages are received differently. Yeah. So that's why we care... Thank you for coming up with the analogy. Yeah. So the identities of the cells matter. There's lots of other really interesting but also very detailed biophysical properties of each cell that clearly do matter, but we don't know by how much. So the thing that I usually try to tell my graduate students when I'm first introducing them to this type of modeling that we do is... Like, say that I'm a civil engineer, I'm trying to build a building, and I need some materials to hold up the roof. And I say I need to know the properties of this beam so that I can hold a roof up. Now, the beam is made out of atoms, and I know that there's like down there somewhere, there's quantum mechanics.
0:14:49.8 SC: Right.
0:15:18.4 BB: But we are not solving Schrodinger's equations in order to design a roof. It's just way too much. So it's super interesting, and maybe you'd be interested on the side, but you don't need it for the task of building a roof. So that's sort of where we are right now. Like I know I don't need every single detail that is known about these biophysical parameters of these cells. They get really funky.
0:15:28.7 SC: Good.
0:15:39.1 BB: They're crazy nonlinear and they're super special, and they're almost impossible to measure. People will spend an entire PhD measuring one cell and characterizing a lot of detail. But do we need it for these very holistic models of the entire animal nervous system? Probably not.
0:15:56.8 SC: Okay.
0:15:58.0 BB: Where do we stop? It's hard to say right now.
0:16:01.5 SC: Right.
0:16:02.5 BB: Like I know I don't need every single detail, but I do not know which of them are actually crucial.
0:16:09.2 SC: And the individual neurons are different, not only sort of structurally or biologically, but even in terms of information processing, right?
0:16:17.0 BB: Yes.
0:16:17.8 SC: Like they have different, I don't know, I want to say algorithms for turning input into output. Is that fair?
0:16:24.4 BB: I think that's fair, yeah. So if you think of it computationally in terms of just maps, if you are able to define exactly what its inputs are and what its outputs are, then you can infer some kind of function that maps it from the inputs to the outputs, right?
0:16:40.0 SC: Right.
0:16:40.4 BB: I think that's a totally valid way of saying it. And I think that might be one of the clues, computational clues as well, in order to be able to run some of these simulations, is that you don't need every single detail of how that map is implemented to approximate its function.
0:16:54.9 SC: But is the specification of how each neuron maps inputs to outputs part of what we call the connectome, or is that a next step?
0:17:03.5 BB: It's not.
0:17:04.2 SC: It's not. Okay.
0:17:04.7 BB: I think it's hard... So, I mean, I don't know. It's hard to say, right? But I feel like this is partially why I, among some of my colleagues... I'll admit, the audience can't see, but I'm raising my hand right now as... I was skeptical. Okay? So this whole thing started when, I don't know, I feel like I was in grad school when I first heard about these really large efforts to produce more connectome datasets, and whatever, it was like, whatever, 15, 20 years ago. And I remember thinking that's like... Well, I won't tell you what I actually thought, [laughter] but I was skeptical.
0:17:40.8 SC: Skeptical
0:17:41.2 BB: I was skeptical. I was skeptical on a couple of different fronts. I was skeptical that it was even gonna work at all, right? Like, can we actually reconstruct one of these things at sufficient scale? Because it involves, I don't know, running a transmission electron microscope for six months straight, making zero mistakes. And so I was skeptical it was possible even to do it in the first place. And then I was further skeptical that if we could have it, like if somebody just handed it to you magically tomorrow, what would you do with that? Right?
0:18:10.8 SC: Right.
0:18:11.5 BB: Like, how could you even make sense of this giant spaghetti monster that somebody just handed you? And so I think some of our... I mean, it's only been pretty recently that some of the work that my lab has been doing with some collaborators has started to convince me that, hey, this this might actually... I think we might actually be able to do this. Now, the reason I was skeptical, and lots of other people were skeptical, so there are essays written, I don't know, ballpark 10, 15 years ago by lots of people in the field, including Eve Marder, Cory Bargman, is because they knew that there were so many other details that are not observable by the connectome. Like this information about all of the channels, the biophysical properties of some of these cells, we can't get them from the connectome. We know we can't. We never thought we could. Nobody thought that we could, right?
0:19:00.8 SC: Yeah.
0:19:01.4 BB: So the disagreement was whether or not the stuff that you can measure, effectively these connectivity matrices, is that sufficient? [0:19:08.0] ____ something.
0:19:08.8 SC: Is that good enough to do something? Yeah.
0:19:10.7 BB: Yeah. Versus, sort of the other logical extreme would be it's utterly useless because you actually need all of the other stuff, right? And so there's a giant continuum of opinions. And I was, you know, I was somewhere in the middle, but, you know, kind of on the little skeptical side. But I never actually worked in the connectome. I was simply fascinated by these efforts that some of my friends were undertaking. And my current opinion is swaying a little bit closer to the, "I think we can actually do something useful with this dataset."
0:19:41.8 SC: Having done useful things with them, I think that's a good opinion for you to have. So, what are the connectomes that we do know something about, even if the human cellular level connectome is far away?
0:19:55.2 BB: What do we know?
0:19:56.4 SC: What animals do we have the connectomes of?
0:19:58.9 BB: So the first one we got was actually like 30 years ago. We have a full a connectivity matrix of the C. Elegans nematode worm.
0:20:08.7 SC: Okay.
0:20:10.1 BB: It's not an earthworm, like the kind you see sometimes attempting to cross the sidewalk and perishing in the middle. It's not those. They're much smaller. They're flatworms or nematode flatworms. They're about a millimeter long, and they live in the soil. So if you scooped up any soil in your garden and looked at it under a microscope, you're very likely to be able to see them there. They're everywhere.
0:20:32.4 SC: Okay.
0:20:33.8 BB: And so they're millimeter long, and they have... This particular species has been studied a lot in molecular biology because they breed really quickly, and so we have tons of tools. They have about 1,000 cells and about 300 neurons. And so the connectivity matrix of those 300-ish neurons has been mapped out decades ago, many decades ago. And so if you talk to people in connectomes, one of the first things they always bring up is like, "But we've had the connectome of the C. Elegans worm for so long, and yet we still don't understand it. We do not understand it."
0:21:07.3 SC: Right.
0:21:08.8 BB: And there's actual good technical reasons why C. Elegans worms are actually really difficult from a connectomics perspective to understand. And so the one that has come out much more recently in the last year or two is a couple of efforts by lots of giant collaborative teams. I was not involved in any of these teams. I was simply cheering them on from the sidelines to map the full connectivity matrix of a Drosophila fruit fly.
0:21:25.3 SC: Fruit fly.
0:21:41.8 BB: Fruit fly, yeah. So this is a kind of fruit fly that... Every year at the end of the summer, my kitchen gets infested with fruit flies and I can't get rid of them. So you've seen them too in your kitchen. They buzz around anytime you have a little bit of rotten fruit or a pile of compost or something in your kitchen. That's where they live. So these little guys are more like 3 millimeters long and so they're like the size of a grain of rice. And their entirety of their nervous system is more like the size of a sesame seed.
0:22:01.1 SC: Okay.
0:22:20.5 BB: Okay. And so they're small enough that it has been possible to reconstruct the entirety of their brain and nervous system. So we have a brain in our heads, and we also have a spinal cord. So that constitutes our central nervous system. So the brain and spinal cord of humans and mammals, of vertebrates. They have an analogous structure. So they have also a central brain that's inside their head, it goes down their neck just like ours. And then the remainder of, instead of a spinal cord, insects and invertebrates have this thing called a ventral nerve cord. It's actually remarkably similar in terms of its structure and how it's organized to our spinal cord. But instead of being on their back, it's actually in their stomach side, so it's on their belly side. That's why it's called ventral nerve cord. Anyway, so that whole thing has been mapped out. And those, there's two of those data sets for one male and one female fruit fly. And that was only published in the last half a year or so.
0:23:22.0 SC: Wow. And how many neurons?
0:23:23.1 BB: So the brain has 150k, and then the ventral nerve cord has an additional 22k.
0:23:30.7 SC: Okay, good. So a much bigger matrix than our little C. Elegans.
0:23:35.1 BB: There's a much bigger matrix. And I think the important thing about the size of it, paradoxically, is that it's actually a little bit easier to understand from the connectivity matrix. Now, the reason that the C. Elegans connectivity matrix has been so hard to understand is, it took us a while to figure this out as a community, they do a lot of computation not using that connectivity matrix. There's a ton of chemical communication. They're constantly squirting out neurotransmitters and other chemicals at each other. There's a lot of mechanical computation. So it's a squishy thing that crawls around in a matrix, not a mathematical matrix, a soil matrix. And so there's a lot of mechanical stretching and reflexes that go on like that. You know the thing the doctor does when they like...
0:24:26.8 SC: Sure.
0:24:27.7 BB: Yeah. So they have those reflex loops that are mechanically coupled with their body, which is squishy, right? And so the physics of that is pretty complicated. So the short way of saying it is that the way that they function as an animal is taking advantage of lots of other computational properties. So they do chemical communication, they do mechanical computation in addition to neural computation.
0:24:53.7 BB: So the fact that we had the neural connectivity matrix was just not quite good enough to understand what they do. In contrast, it is some of our current understanding and perhaps hope that the connectivity matrix of the fruit fly, because of, that it is a little bit bigger, it has jointed limbs just like humans do, and it has enough cells that there're actual cell types. Like, not every single cell is just its own little snowflake. They actually have types of cells. All of those, we are hoping, makes it so that that connectivity matrix is more helpful, more directly helpful, helping us understand what the heck is actually going on.
0:25:36.0 SC: In other words, because individual neurons, etcetera, might be more specialized or something like that, rather than just every neuron pitches into every task.
0:25:44.1 BB: Right, yeah. So the C. Elegans neurons, some of them are... They're so not specialized that, you know how we have a visual system, so there's cells that detect photons, and we have an olfactory system, cells that detect smells, right? They have single cells that have multiple sensory modalities going into it because it's just so tiny, it's so compressed, right? They've had to multiplex in that way. And we don't see that as much in our understanding in the worm nervous system. And that's a feature, it's a computational feature of how their nervous system works that's in common with ours.
0:26:25.3 SC: And with the fly connectome, the fly neurons, I saw in one of your videos these images of these neurons. And I think that people, certainly I, have this image of a neuron as a little blob with a couple little spikes. But these are very spindly things. They're stretching across some non-trivial fraction of the size of the fly.
0:26:47.5 BB: Right. Do you know the longest cell in your body?
0:26:50.9 SC: I do not know the longest cell in my body.
0:26:53.8 BB: It's about as tall as you are.
0:26:57.0 SC: That's a little freaky. I don't want to think about that. What is...
0:26:58.7 BB: It's a little freaky. So you have these cells that actually the same cells we were talking about in the ventral... So in insects, in the ventral nerve cord. In your body, it's in your spinal cord, okay? You have these cells that are responsible for how... This is how you know you stubbed your toe, okay? So there's a cell that detects when you've stubbed your toe. So one end of it is at your big toe, okay?
0:27:24.7 SC: Okay.
0:27:26.1 BB: And the other end of it, it goes all the way up to your brain stem, so the very base of your skull. So talk about long and spindly.
0:27:33.6 SC: Why does it need one cell to do that? Can't a bunch of cells hand off the message? Isn't that the usual way?
0:27:38.2 BB: You can do that. No, this is the actual... This is the normal architecture.
0:27:43.5 SC: Okay.
0:27:44.5 BB: There are other cells involved and you can hand off the message. The advantage of having one cell do it is that you can do it really fast. Because as it happens, if you stub your toe, your brain really wants to know about it very quickly.
0:27:56.6 SC: Stupid brain. I don't think I want to know about it at all. [laughter] I just want to get on with it.
0:28:02.2 BB: You want to know something... Well, this is how you don't fall over. It's all the shit that your body does that, you don't think about, right?
0:28:06.5 SC: Fair enough. Yeah.
0:28:07.6 BB: You don't have to think about not falling over. If you're hiking and you kick a rock, you don't fall over. And you also don't want to waste your precious time thinking about how to not fall over. You simply want it done, right? You want to keep on having that conversation about number theory you're having with your buddy, right? You don't have to think about how to not fall over just because you kicked a rock.
0:28:25.5 SC: Which segues very nicely into the actual work you've been doing with the fruit fly connectome. So you have the connectome. That's good. And then there's this open question that you elucidated very nicely: Is it good enough to help us do anything? And you've been asking, what is the relationship between the connectome and walking in the fruit fly? Is that right?
0:28:38.8 BB: Yep. That's right.
0:28:52.1 SC: So I don't know, how do you even start with that? What do you do?
0:28:58.3 BB: Well, so the slightly longer story is that this is a longtime collaboration I've had with a friend and collaborator of mine, John Tuthill. And John is a fly experimentalist. His lab does neurophysiology and they study the ventral nerve cord and the sensors that come in as well as the motor control that goes out. That's what his lab does. And we've been collaborating for a decade now and have co-advised a series of graduate students and postdocs doing some combination of theory and modeling. And it's been super fun. And so John's also been really involved in some of these connectomics efforts. So a lot of what I said... The stuff that I know that is not wrong is because I learned it from John. [laughter] Stuff that is wrong, I made that up. I take responsibility. And so I remember a couple of years ago, John and I were taking a walk and we had a brand new PhD student who was thinking about joining our labs. And we're like, "Oh, what do we have them do? We got to think of something," right? And John and I were talking and he's like, well, we have... "We almost have a ventral nerve cord connectome."
0:30:09.3 BB: It's like it's almost ready because they were in the process of cleaning it up, curating it, trying to write it up, right? He's like, "What if we just simulated it?" And I said, "That's never going to work. Let me tell you all the ways this is not going to work." So I told him all the ways it was not going to work, some of which I summarized earlier, you know, the biophysics, all the parameters we don't know, blah, blah, blah. There's tons of stuff. There's lots and lots of reasons that wouldn't work. But by the end of this talk, we had come to, "Well, you know what, let's try it anyway." We don't lose anything. Let's just give it a good old grad school try.
0:30:40.6 SC: I do, by the way, think that half of the secret to succeeding in graduate school is listening to your advisor tell you that won't work and distinguishing when they're right from when they're wrong.
0:30:51.1 BB: Absolutely. So our student, Sarah Puglisi, listened to us and she said, "Okay." She went off and wrote some code. So of course, long story short, it took a couple of years, but we kept at it, partially because some of the preliminary stuff was actually kind of interesting. There were some hints, right? And what ended up happening is that we went after a question that biologists and neuroscientists have been asking for over 100 years, which is this question of how does the nervous system generate rhythms from non-rhythms? How does this happen? And to give you a context a little bit about why this is such an important question, all animal movements are rhythmic. Actually, not just animals, even bacteria move by spinning their flagellum, right?
0:31:46.1 SC: Right, right.
0:31:46.9 BB: So basically all biological movements are cyclic in some way. So you can be walking, running, swimming, slithering, crawling, basically all locomotion is rhythmic, right? And so the fact that your nervous system needs some way of generating the instructions for your muscles to move in a circle, that's fundamental.
0:32:11.6 SC: Okay.
0:32:12.7 BB: This is like one of the... And so we've been... Ever since the 1910s, some of the first experiments demonstrated that the generation of these rhythms is not by reflex only, is that your central nervous system, somewhere in your brain and spinal cord.
0:32:28.6 BB: And spinal cord, was capable of generating these cycles. But we didn't know exactly where, we didn't know which cells did it, we didn't know how they did it.
0:32:39.7 SC: And this... So just by the way, the idea of some system of mechanical things, cells or anything else, vibrating in periodic ways, that's one that appears all over the place.
0:32:53.4 BB: All over the place.
0:32:54.4 SC: We understand that. Yeah.
0:32:55.2 BB: We understand this in general. And if we have time, I'll come back to... Like, I love dynamical systems.
0:33:01.8 SC: We will. Okay, good.
0:33:02.9 BB: We can nerd out about the dynamical systems of oscillator equations a little bit later, and it actually has connections to our work in the connectome as well. But yes, absolutely. Yeah. And so in the intervening 100 years or so, lots of people have studied the idea of these circuits. And so the ability of your nervous system to generate rhythms is not only important for locomotion, it's also important for things like breathing, right? Because you have inhale, exhale, inhale, exhale. You can control it, but if you don't think about it, it just happens. And so that's generated by what we call a central pattern generator, a CPG circuit, as well. Digestion is cyclic, right?
0:33:25.3 SC: Yep. Well...
0:33:47.7 BB: So you have to churn the stuff in your digestive system. So there's a sequence of muscle contractions that gets the food down into your stomach.
0:33:58.8 SC: I see, yeah.
0:33:59.6 BB: And in your stomach, especially the stomach... So the most studied CPG circuit, central pattern generator circuit, is actually in the crab digestive system. There's a couple of these adorable little neurons that are responsible for churning what's in the crab stomach. It goes... [0:34:14.0] ____ And it makes that rhythm.
0:34:17.8 SC: And you know which neurons are in charge.
0:34:21.0 BB: This is the work of Eve Marder. She is known for having studied this for decades. And that system is so extraordinarily well understood, we actually understand. It's probably... Sometimes people are a little snarky and we say the crab digestive circuit is the only neural circuit we actually understand in all of neurobiology. It's a bit of an exaggeration, but it's not untrue either. We actually understand that circuit.
0:34:46.3 SC: And the thing we're looking at... So the idea of central pattern generators, these are little sub-circuits within the connectome?
0:34:50.5 BB: Yes.
0:34:54.7 SC: That are responsible for... Is it always cyclic rhythm motions, or is there a more general definition?
0:35:00.7 BB: That's probably the plainest definition of it. And then... So the CPG is... I mean, roboticists love the CPG. So a lot of modern robotics is built on these oscillator equations. So they don't even... I've talked to roboticists who actually have no idea about the neurobiology of central pattern generators because for them, they don't care. They just write an equation. We've done the same. So a lot of these computational models of locomotion in animals and robotics is just based on... You just write an oscillator equation. It just goes around in a circle. There's lots of them you can write. It doesn't really matter how it's implemented by cells. You just care that there exists a thing that goes in a circle, right? But we didn't know what actually were the cells and their connections in an actual nervous system that generated these rhythms for any animal that walks.
0:35:52.8 SC: Okay.
0:35:53.7 BB: So that's where we were a couple of years ago, is like, nobody had ever actually found what are the cells, what are their names, how do they work?
0:36:00.9 SC: So in other words, you knew from prior experience with digestive systems and breathing that there had to be these CPGs, central pattern generators, that would do these rhythmic motions. You also know that walking is kind of a paradigmatic rhythmic motion, but we hadn't quite identified and these are important.
0:36:17.6 BB: We hadn't quite identified the actual cells. And so to be fair, people have studied lots and lots of walking systems. There's tons of just whole bookshelves in the library about spinal circuits of walking. Invertebrates have these ventral nerve cords. How do they generate their wing flapping? How do they walk? People have tried, and there's tons of information, but we didn't know precisely which cells they were and how they worked.
0:36:48.9 SC: All right, so what are you gonna do?
0:36:51.1 BB: So we had an opportunity. It's not like we were smarter than all of these other people who have worked on it. It's just that we had an opportunity of having the complete connectivity map of the ventral nerve cord of a fruit fly. And we figured, whatever it is, it's got to be in there somewhere, right? Like, we don't know... Instead of starting from building it up from individual components that I can actually do experiments on, we took the reductionist approach. We're like, it's in here somewhere. We got it down to a network of 4,000 cells or so. We're like, it's got to be in here somewhere.
0:37:23.2 SC: So wait, when you say you got it down, you're basically saying, okay, we have 150 or 170,000 neurons, and you eliminate... You say, if I didn't have this one, it could still walk fine.
0:37:35.9 BB: Precisely. So we simulated... Okay, so first thing we did is to make it a little more manageable, we focused on only the two front legs. So insects have six legs, so we just got rid of the other four. We're like, okay, let's just have two legs, the two front legs. We got rid of all of the parts of the nervous system that don't control the two front legs. So we just have two front legs. That's how we got to 4,000-ish.
0:37:48.0 SC: Okay.
0:38:00.1 BB: Okay? So then we simulate that, and we were able to demonstrate that those 4,000 neurons were able to generate a cycle. They can generate motor rhythms and actuate the muscles that would have to move the leg. Again, you can't see me, but I'm moving my arm forwards and backwards, right?
0:38:17.6 SC: Yeah. Those are moving.
0:38:20.1 BB: It actuates these muscles right here on your shoulder, like the ones that move your shoulder forwards and the ones that move your shoulder back. Okay, so just by simulation, again, we were doing no reinforcement learning, there's no machine learning here, there's actually no deep learning going on at all. We're just doing brute force numerical simulations of this giant connectivity matrix.
0:38:36.4 SC: Shallow learning.
0:38:36.5 BB: [laughter] Regular numerical simulations.
0:38:40.9 SC: Right, okay.
0:38:41.9 BB: The stuff that we've been doing for a long time. You write a lot of code and you run it a million times. We can get these rhythms to come out, right? So then we asked, now that we have these rhythms, now that it's actually in here somewhere, now let's try to reduce it. Now let's cut away one at a time. We basically just started getting rid of cells. We're like, do I need this one? No. Do I need this one? No. And you just keep going until you've thrown away as many cells as you possibly could without losing the rhythm. And what you have left over is the minimal circuit. Does that logic make sense?
0:39:10.6 SC: I think it does. And so this is just one leg, or I guess it's symmetric...
0:39:15.0 BB: It's symmetric.
0:39:16.5 SC: The front two legs are doing the same thing. By the way, let's just take an aside to explain the fascinating question, which is the wings.
0:39:25.5 BB: Yeah. Yeah, I know. It's wild.
0:39:29.2 SC: So I would have thought from my mammalian-centric point of view that wings are just like arms that have grown wing-like, but flies are very different.
0:39:39.3 BB: Not so. Not so. So this is something my friend and collaborator Michael Dickinson is very fond of saying. The insect wings are actually novel limbs. I'll explain what that means. So for every other animal that flies, like bird wings are modified arms, bat wings are modified arms, right? Other animals that fly have wings that used to be not wings. Not so of insect wings. They're not modified legs. There's theories about exactly how they evolved, but they're actually novel structures. It's not like they took a pair of legs... It's not like they used to have eight legs and two of them became wings. These are just actual new things.
[overlapping conversation]
0:40:23.7 SC: And this is reflected in the nervous system.
0:40:27.6 BB: Say it again, please.
0:40:28.4 SC: This is reflected in the nervous system.
0:40:30.0 BB: It's very much reflected in the nervous system. So just like there's these little parts of your spine that correspond to... Like, you have parts of your spine that's like, this goes to the left leg, this goes to the right leg, this goes to your trunk, right? Same thing. They have parts of their ventral nerve cord that go to each of the six legs, and you can actually see them. They're like little balls that kind of stick out. They're a little bit bigger because they have more cells. And then they have the same thing, like, there's little clumps of cells that correspond to the wings.
0:40:55.7 SC: Cool. Okay. So there's a whole separate future research project understanding how flies fly. You're trying to understand how they walk.
0:41:03.6 BB: Yes, just walking for now.
0:41:05.3 SC: And how did that go?
0:41:05.9 BB: It worked great. [laughter] So the pruning study that I briefly described earlier, where we took a functioning system that was able to generate these CPG-like rhythms, and then we started just pruning it. We started cutting away everything computationally that didn't seem necessary for it to be there. I remember I was sitting in actually this office with Sarah and with John the day we figured out, "Okay, let's give it a try, let's do this pruning study." So remember we started out with 4,000 cells. And I remember telling Sarah, "Sarah, just go give it a try. If you get it down to a few dozen cells left over, like, if that's the minimum circuit, you need a few dozen cells to do this, I would be ecstatic. Like, that would be a really cool result." She went off and did it. The answer was three.
0:41:31.9 SC: Three cells.
0:41:57.8 BB: Three cells. That's the minimum you need. And they have names. We know who they are in the fly nervous system.
0:42:05.1 SC: Tell us their names. That'd be fun.
0:42:07.0 BB: That is a great question because I actually have no idea what their actual names are.
0:42:08.4 SC: Look it up. Okay. Names are known.
0:42:12.5 BB: They have... Their names are known and their lineages are known. So we sort of know where they came from. The names are... I can't handle this. The names are a series of letters and numbers and I can't remember what they are.
0:42:24.7 SC: You're the one who said we know their names. That's the only reason I asked.
0:42:27.7 BB: Okay, we, the royal we.
0:42:29.1 SC: The royal we.
0:42:30.5 BB: John knows what their names are. I have no idea what their names are. We gave them pet names, though.
0:42:33.3 SC: Of course.
0:42:36.1 BB: Of course we had to give them pet names. And they're not too cute. So there's three cells. And remember I told you earlier, the cells have identities. It kind of matters what type of cell they are. So two of the cells are excitatory. They make other cells more excited. And one of the cells is inhibitory. It makes other cells less excited. And so they're called E1 and E2 because there's two excitatory cells. And the last one is I1 because it's an inhibitory cell. And they are connected in a particular, very understandable architecture motif that explains why this tiny little circuit is capable of generating cycles.
0:43:12.4 SC: Well, maybe this is the place then to get into dynamical systems theory a little bit. I mean, because my next question was, how do three tiny neurons manage to tell the leg how to walk?
0:43:27.2 BB: So, okay, so I will be a slightly more precise and say that, we believe the three neurons are sufficient to generate the rhythm.
0:43:32.2 SC: Okay, the rhythm.
0:43:37.2 BB: They can generate the rhythm. They're not sufficient to actually control... They have, I don't know, dozens of individual muscles that need to be coordinated in their legs to be able to walk.
0:43:49.7 SC: Okay.
0:43:50.8 BB: We have many more, but you can get the idea, right? There are many more muscles than there are degrees of freedom in a limb. So actually controlling them to do something coordinated and not super clumsy is a little more complicated. But we believe these three neurons, our hypothesis is at least three neurons generates the basic rhythm. And then there's other cells involved to make it actually walk. Does that make sense?
0:44:10.5 SC: Okay, good. And that's just the lesson we're learning over and over again. There's a lot of teamwork in biology, a lot of responsibility shared among different subcommittees.
0:44:22.4 BB: I certainly don't feel like the nervous system is wasting cells.
0:44:26.6 SC: Right.
0:44:27.3 BB: We have all these cells. They're doing something, right? I just don't... People have all these ideas about low dimensional structures and neural manifolds and I don't know, there's words thrown around if you talk to some other neuroscientists. I don't think biology is wasteful in that way. There's redundancy and there's a good reason for the nervous system to be redundant in case it gets injured, etcetera, right? But I don't think there's waste. I don't think we have cells for no reason. If it's there, there's probably a pretty good reason it's there or it wouldn't be there.
0:45:00.6 SC: Well, it's possible... Well, what do I know, but I can imagine that it used to be useful and then the evolutionary use of it sort of went away, but the cell lingered for a while.
0:45:10.8 BB: Because the cells are so expensive to maintain. Neurons are some of the most expensive cells to maintain in your body. I think my hypothesis would be that if a cell is actually not necessary, the body would find a way for it not to be there over a longer time frame.
0:45:29.6 SC: I see. So it's actually more plausible to have vestigial organs in the body than vestigial neurons.
0:45:37.0 BB: If you're thinking of the vestigial organs that I'm thinking about, there's actually just, like... I mean, we can go off on a super long tangent if we wanted to. That's a different cycle we can go on about why those vestigial organs didn't go away. And there's usually a good reason because they got stuck, basically. Not that we had a use for them, but just because the way that evolution works, they got stuck.
0:46:00.2 SC: Okay, let's go back to our three neurons, E1, E2, I1.
0:46:05.4 BB: Right.
0:46:06.0 SC: And so there's a really oversimplified spherical cow version of this where it's literally a circuit.
0:46:08.2 BB: It is.
0:46:16.2 SC: And it is constructing a rhythm. And then there's the slightly more complicated version where there's external inputs and outputs and other influences going on. And how do you learn about all those?
0:46:30.0 BB: Yeah, so to learn about all of the other stuff, I think what my lab, our vision, and there's tons of collaborators who are involved in all of this because this is kind of a giant team effort, is to then actually embody the nervous system, the connectome in all of its glory, actually put it inside a body where it belonged all along.
0:46:55.1 SC: Like a mechanical body or...
0:46:57.4 BB: A simulated body. More like a video game body.
0:46:59.9 SC: Yeah. Video game body. Okay.
0:47:01.3 BB: Yeah. So my son's been playing Red Dead Redemption. He rides a little horse around in this virtual little environment. It's a clomp, clomp, clomp, clomp, clomp. That's just an animation. It doesn't really matter if it is biomechanically realistic, physically realistic, biologically interpretable, whatever. It's just a video. So we want to do that, but actually have it be biologically interpretable and then also physically realistic, so far as we can. But it would be a physics engine. Models F equals ma.
0:47:25.4 SC: Right. Good. So, sorry, is that going on? Does that exist? Did that help? Did that teach you anything?
0:47:30.9 BB: It's in progress. I think it's in progress. I'm really excited about it. I mean, this is a bit superlative, but I feel like I've rarely in my career felt so much conviction that something is the right thing to do. I just.. I feel... It's so obvious to me that the brain does not live in a jar.
0:47:42.5 SC: Right.
0:47:54.1 BB: It always controlled a body, and it always controlled a specific body with these limbs and these muscles and these joints and these sensors in order to move around the world and eat and collect information and do all the things that animals do. And so it's just so obvious to me that we need to be understanding the brain and nervous system in the context of the body that it interacts with to produce the behaviors that the animal actually does. And so that's the grand overall vision of what we're doing. I'd love to be able to. I mean, we are, we're early days, but it's just... This is... I'm really excited about it.
0:48:32.4 SC: Is there any usefulness in imagining doing it in good old-fashioned physical reality as well as virtual reality?
0:48:40.0 BB: Like robotics.
0:48:41.8 SC: Yeah, either a robot or can you hijack the nervous system of an actual fly?
0:48:46.1 BB: For sure. Super easy to hijack the nervous system of a fly. It's part of the reasons that we're working in the fly is because it was kind of the genetic organism of choice for a really long time. And so our ability to hijack every aspect of its nervous system, do gene engineering to put proteins in it, to shine lasers at it, all of that stuff already exists. And that is the reason we're working in flies is because the wealth of knowledge that has accumulated over the many decades of people working on the fly, we just know so much more about their everything than a spider, for example.
0:49:21.4 SC: Right.
0:49:22.2 BB: So, yeah, we can hijack it. So a lot of the things that... I mean, we're neuroscientists, we love lasers. So there's a lot of lasers going on. We shine lasers at them and we can make them do things. We can shine lasers at them when they're walking, flying, trying to sniff, stuff like that.
0:49:35.6 SC: I don't think that the sentence "We're neuroscientists, we love lasers" is that obvious to the outside world. I didn't realize.
0:49:41.7 BB: Well, it is.
0:49:43.3 SC: Lasers are the thing.
0:49:45.0 BB: Yeah, love lasers. I don't know if we... I think we might love lasers slightly more than physicists do because we just play with them. [laughter]
0:49:58.8 SC: Yeah, well, physicists are gonna join you there. That's okay. And it sounds like, maybe I didn't quite get it right, but it sounds like rather than learning about these three neurons by experimenting on the neurons, you almost guessed or you almost sort of figured out they have to be doing this in order to make it work.
0:50:15.9 BB: It is a guess at the moment. We do need to do the validation experiments. We need to also corroborate our predictions and our hypotheses by doing experiments on these actual neurons. For technical reasons, that stuff is ongoing. We haven't done it yet.
0:50:31.4 SC: Sure.
0:50:32.1 BB: So that's why this is still, I would say, a very strong hypothesis in my mind.
0:50:36.6 SC: Right.
0:50:37.2 BB: It's our... We have good reason to make this guess, but it's still a guess at this point until we can confirm it biologically. But I think one of the things that's kind of cool about this result is that as a computational modeling person, I've spent the majority of my career fitting data. Somebody has an observation, something they already know, and we're like, "Oh, sure, I can write some equations in code and we can recapitulate it. We can make a model that does the same thing that you already know." This is one of the few instances where I feel like the model actually came before the experiments. We were agnostic going in. We had this giant dataset. We're like, "Let's just simulate it." And then we made a prediction of things we didn't know before.
0:51:20.8 BB: And so part of this result I haven't talked about is that we haven't quite gotten to the three cells that we predicted to be the core CPG circuit, but there's other parts of the nervous system that we did predict. Like, we made some predictions of... So there's this one pathway that comes down from the neck, and in our model, it was a cell that has a name. And doesn't really.. I do actually know the name of this. It has a few letters and numbers. I know what it is. But this neuron that comes down from the central brain, and our model said, "Okay, if you zap it with a laser, it should make the leg tap. It should go back and forth." Okay?
0:51:59.9 SC: Okay.
0:52:00.7 BB: And nobody has ever even studied this neuron before because there's a lot of them. But somebody did actually make a cell line. There was a fly we could order that somebody had already made that had the correct proteins in it so that we can shine a laser at it and activate that cell. So we ordered it, we grew it, we cut its head off, and we glued it to a stick, and we shined a laser at it.
0:52:26.7 SC: Yeah.
0:52:27.7 BB: And I tapped this link.
0:52:28.5 SC: Oh, there you go.
0:52:30.3 BB: It's adorable. It's like, it was an actual model-driven prediction.
0:52:32.5 SC: Right.
0:52:34.8 BB: We had no idea, nobody had any idea what this neuron did. It was just in the nervous system. Most cells in the nervous system are like that. We're just like uh-uh.
0:52:43.8 SC: We don't know the names.
0:52:44.4 BB: We give it a name because we kind of know where it came from. We have a nomenclature, we have systematics. We don't know what it does.
0:52:50.0 SC: Well, I guess that was an obvious next question. If you have three neurons per leg controlling the rhythm, and there are six legs, that's 18 neurons. That leaves 150,000 minus 18 neurons to figure out what they do.
0:53:04.0 BB: So many things
0:53:04.7 SC: Is there an obvious roadmap to what we're able to do?
0:53:08.0 BB: I sure hope so. So part of it is this idea of the embodied brain that I talked about. And it goes by a couple of different names. So drawing analogies between what we're doing with these virtual models of animals with the nervous system and the biomechanics of the body, we and some other people have been calling them digital twins, which is a word that we're borrowing from industry and from industrial engineering. So the digital twins that exist in industry are digital twins of things like airplanes and cities. Like, there's a digital twin of the city of Singapore, right?
0:53:28.1 SC: Okay.
0:53:48.2 BB: And it's a simulation. It doesn't have every single light bulb... [chuckle]
0:53:52.1 SC: Sure.
0:53:53.3 BB: But it has many of the important parts of the city of Singapore, including its morphology, its connectivity, and it's hooked up to real-life sensors in the city so that they can sort of update the status of the city. And the city planners kind of use it to do things like predict disaster response, right, or to in real time shift, if they have to shift traffic patterns or whatever to relieve some congestion because there's an emergency in one place, stuff like that. Okay. So that's what people in industry have used these digital twins for. And in close analogy of that, the thing that we're thinking about building, I think, would be considered a digital twin of an animal, a behaving animal. So it would have a simulation of the nervous system and the interfaces between the nervous system and the body so that we know how information goes in and how information comes out. And it would be situated in a virtual reality environment that's capable of interacting with things, right, like surfaces that are not flat, right? You can walk on bumpy surfaces.
0:54:55.9 SC: Physics.
0:54:56.8 BB: Physics, yeah, just physics, right? They can also interact with other agents. So this would be an agent-based model. And so you can have two animals interacting with each other. They can even touch each other, for example, stuff like that. And so in that way, if we have a set of simulations that are developed in very close collaboration with our experimental collaborators, we should be able to come to a set of models that can predict what's going to happen in parts of these circuits that are hard to predict otherwise. 'Cause the thing is, the whole thing has just mad, mad feedback and recurrence, right?
0:55:29.5 SC: Yeah.
0:55:30.2 SC: And if it's one thing that I've learned about humans and our ability to reason through rational thought, it's humans are really terrible at reasoning through what happens with feedback circuits and recurrence. We can go forward. We can follow a path like A to B to C to D.
0:55:45.2 SC: That we can do.
0:55:46.8 BB: That we can do. As soon as there's recurrence, when D goes back to B and then C goes back to A, our intuition for what's going to happen is really poor.
0:55:54.8 SC: Okay.
0:55:55.4 BB: And that's one of the arguments that I make in motivating why we need these complicated computational models. We can't do it, but we have computers.
0:56:07.1 SC: Well, I guess an obvious issue that floats to mind is when you are simulating the biology on the computer, you have to make some choices about what to include, what not to include, what to model, what not to model. Is there any danger you'll get sort of get the right answer for the wrong reason?
0:56:28.7 BB: Yes. [laughter] So many. Probably more than not. I think we need to be really careful. So this is a... I mean, this is something that I think maybe we talked about briefly in person at some point is this idea of the Digital Sphinx paper that we wrote a couple of weeks ago. And the brief intro to that is that there's a lot... I was starting to see a lot of work and conversation in the field, including by my lab and our collaborators, where because the whole thing is so overwhelming and there's so many details and we know we can't possibly measure them all, it's literally impossible, we know we have to make a lot of assumptions, right? And so a lot of people, again, including us, we've been doing the same thing. The thing that... One thing that we can measure with a lot of fidelity and relatively easily is just the behavior output of the animal. We can get cameras and we can track what it's doing. We can see how it's moving its legs around. We can see where it's pointing its head, right? That we can do. Anything external with cameras we can do because we have cameras and we have really good computer vision. And so a lot of people are basically saying that, okay, this is the grounding, right? Like, if we can get a model that looks like it's behaving like the animal in that it matches what the animal was observed to do with a camera, then we've surely we've gotten something right.
0:58:02.3 SC: Sure.
0:58:06.8 BB: I know, I know. That's... I'm glossing over lots of details. Of course, lots of people are doing this in a really careful way. But what I was a little afraid of was that people were starting to do this in a not careful way.
0:58:14.5 SC: Okay.
0:58:22.2 BB: And in particular, there was some stuff coming out on social media by some startup companies, trying to fundraise, putting out work that I looked at it and lots of our friends looked at it and was like, "That's not... You're overselling this," right? "You're not doing what you said you did." And what they said they did was that they had uploaded a fly brain. That was the headline. Brain upload.
0:58:51.1 SC: Wow.
0:58:52.1 BB: They've uploaded a brain. They claim they did it. And of course, you have to read them, you have to read the details. You looked at it and was like, okay, this is not... They didn't actually do that. But that's what they said they did, right?
0:59:01.6 SC: Yeah, okay.
0:59:03.0 BB: But this thing got a lot of... It kind of went viral and it got a lot of attention, and not just among non-scientists. It actually, I think a lot of... I got a couple of friends who are not neuroscientists, they're chemists, where it was like, I heard this thing, they uploaded a fly brain, right? Like, that sounds really cool because they didn't know exactly the...
0:59:24.5 SC: Sure.
0:59:25.9 BB: Yeah, exactly. And so as an exercise, just as an exercise, I was sitting around with one of my postdocs, Elliot Abbe, and I was like, "Elliot, this is just this is nuts. Like, this is bananas, right? Like, this is not the right way of doing it." But to explain why it was not the right way of doing it took a lot of technical words. You have to understand reinforcement learning, you have to kind of understand the architecture of the nervous system, like, there's a lot of stuff to explain and it just takes a long time. And so Elliot and I are sitting around, we're like, "Well, what is the thing that we can do to point out how ridiculous this is? What is the logical extreme of what they're effectively doing?" We're like, well, they're not actually even using the fly brain connectome. This could be anything. It could be a random matrix.
0:59:36.5 SC: Ah.
1:00:10.1 BB: In fact, it might as well be a worm matrix. It might as well be a C. Elegans worm matrix. And so we looked at each other...
1:00:15.9 SC: Wait, sorry, what did they upload?
1:00:19.2 BB: They uploaded a portion of... They simulated a portion of the fly brain.
1:00:22.0 SC: Okay.
1:00:24.7 BB: Yeah. And crucially, since we spent so much time earlier talking about the ventral nerve cord and how that controls leg movements, they did not simulate the ventral nerve cord.
1:00:30.1 SC: Okay. All right.
1:00:35.6 BB: But even me saying that, right, like, that took conversation about explaining what the ventral nerve cord was and how that actually controls your limbs, right? Anyway, that was one of the things that they did not have. They did not have a ventral nerve cord, even though their animation definitely had little legs that were moving around, right? That was the animation part. So Elliot and I were like, okay, well, what if we upload a worm brain and and train it to control the fly body? We're like, we could totally do this. And so he and another graduate student in the lab, they just they did it. They downloaded the C. Elegans worm connectome, all 300 cells in its glory, and we popped it into a reinforcement learning algorithm that we've been using for lots of other things to control a biomechanically realistic fruit fly body walking around in a physics engine to imitate 3D kinematics of flies. Like, and it works. Like, it runs around just like we know the flies do.
1:01:39.8 SC: So the worm connectome in the fly body wriggles around the right way.
1:01:45.4 BB: Yeah. I mean, like, if you use enough deep learning and you can train it with good enough data, it is perfectly possible to get a worm brain to control a fly body.
1:01:56.2 SC: Okay.
1:01:56.8 BB: And so what we're learning from all of this is this is just... I mean, it's silliness, right? If you basically use that much deep learning in there and you are allowing all of these parameters to change in ways that are not obvious, the fact that you have a connectome and the fact that you have a hyperrealistic biomechanical body doesn't mean anything. This is not biologically meaningful.
1:02:08.1 SC: Right.
1:02:19.3 BB: You can get behavior fidelity without any biological fidelity.
1:02:23.3 SC: Well, and especially because you said that we're not even talking about the identities of the individual neurons or their maps from inputs to outputs. So how in the world can you expect to get something believable, realistic, I don't know what you want to call it.
1:02:42.0 BB: Yeah, so we actually even tried a little bit. So the worm connectome also has motor neurons. It has the neurons that would be controlling their muscles. And we used that population of motor neurons to wire it up to the fly body actuators just for fun, basically.
1:03:00.6 SC: Yeah, okay.
1:03:01.7 BB: But that's where the deep learning comes in, right? If you have an artificial neural network that maps, that does that mapping between the motor neurons and how it produces torque in the body, that's where the neural network was, and we trained that.
1:03:17.1 SC: But this is kind of just showing that I can run the same software on a Mac or a PC.
1:03:26.8 BB: Yeah, if you have the right interfaces, yes. I think that's a great analogy.
1:03:30.0 SC: I can emulate an engine on a different computer.
1:03:32.5 BB: It's an emulator, right.
1:03:33.4 SC: It's an emulator.
1:03:34.1 BB: It's totally an emulator. It's exactly what it is. It's like if you don't care about meaningful interfaces, you can get lots of things to plug together.
1:03:39.2 SC: Right. Okay. Good. So that's like...
1:03:45.9 BB: And just like HDMI matters, right? And the real trick to get something meaningful out of it is to actually engineer those interfaces.
1:03:54.7 SC: Right. Yeah. Okay, so that's a good cautionary tale. We should all read the popular science literature with a little bit of caution. But since we're near the end of the podcast, let's think big a little bit about the implications of everything that you're telling us. One of the messages I'm getting is the embodied nature of all of these neurons, all the brain and so forth. These days, forgetting about wild claims from startup companies, but there's still a lot of interest in AI and consciousness. I mean consciousness, both artificial and real. We've had a couple of podcasts recently that talked about whether biology was intrinsically important to the idea of consciousness.
1:04:22.7 BB: Okay.
1:04:50.8 SC: And not because of anti-physicalism or mystical woo stuff, but because maybe all of the little processes going on underneath the hood of a biological organism, the respiration, the metabolism, and the signals going forth, maybe all of those matter in some way over and above just the algorithm that is being run on the hardware. Do you take any lessons from your work for these kinds of ideas?
1:05:18.6 BB: Yeah, I think so. I'll maybe state what I'm about to say a little more strongly than I actually believe for the sake of conversation.
1:05:24.3 SC: Sure.
1:05:30.1 BB: We have no examples that we all agree on of agents that are intelligent and conscious except the ones that are embodied. We don't agree, right? The other ones may or may not be intelligent and conscious, but we don't agree yet. So the only ones we actually agree on are embodied agents. Furthermore, the nervous system, all nervous systems evolved starting about 500 million years ago to control a body, to sense from the environment and respond to those senses in order to move around in the world and seek food and mates and etcetera, etcetera. That's what animals do, right? And everything that we think of as reasoning, as consciousness, I mean, there's all of these words, I'm not going to list them all, right? All of that machinery, all of the capabilities for doing so evolved on top of the neural computations required for sensorimotor control, for sensing from the environment and moving the body around. And so in my guess, like I'm going to guess that understanding the platform, the substrate on which all of those other capabilities were built would be important for understanding the stuff above, as well as a strong constraint in how it could possibly have gotten there.
1:06:54.4 SC: I think that for claiming to say things stronger than you actually believe, that was incredibly reasonable.
[overlapping conversation]
1:07:04.1 SC: You fit all the caveats in there.
1:07:05.6 BB: All right, all right. I am a scientist.
1:07:10.9 SC: But to turn that around a little bit, if we are interested in artificial approaches to thinking, consciousness, whatever, there's a lot we still have to learn from the biological reality of it.
1:07:28.4 BB: That I actually don't feel that strongly about. I don't think it's actually important to understand the details of how biology implements something to build an artificial system that takes advantage of some of those insights. I mean, I feel like the field of bioinspired engineering is full of these examples where the concept that biology might do something was sufficient to inspire a perfectly good solution without understanding the details. I mean, bird flight is the only... First one everyone always thinks of, right? We knew that birds can fly, therefore we were inspired to fly. It turns out imitating the way that birds fly was an utter failure. We had to throw it all out the window and start over. Fixed-wing aircraft, that's where it's at, right? And so the details of how a bird actually flies is fascinating and we continue to study it, but understanding it was not necessary for us to fly.
1:07:55.0 SC: Fair enough.
1:08:21.7 BB: Right? So it's just that we can do it. I think that's a lot of times where the bio-inspiration comes from. The inspiration is sometimes just more like, isn't that cool? The central pattern generator circuits that I talked about earlier, same example, right? The observation that there must be something inside your spinal cord that generates rhythms, that's all the roboticists took.
1:08:43.0 SC: Yeah.
1:08:43.9 BB: They didn't need exactly how it worked, exactly what the cells are and how it works in an actual biological system. They're like, "Oh, I can implement this chromato oscillator on my onboard computing chip." Great, perfect. This works great. That's all they needed. So I'm not sure we need to know the details of biological intelligence to get to artificial intelligence. I don't think that's necessary. However, the concept that it may be necessary for that agent, for that intelligent system to be actually embedded in something that's interactive, that has a physics, so to speak. It doesn't have to be our physics, but I think it should have maybe some rules and constraints, some notion of energy, some notion of conservation laws, like not just limitless everything. Right. Maybe that's important. I don't know. This part I'm speculating, but I don't think it's necessary to understand how biology works to get artificial intelligence.
1:09:43.2 SC: I guess I would completely agree with you there. The difference... The failure of the analogy a little bit is we all know what flying is. We don't really know what consciousness is.
1:09:55.9 BB: Very true.
1:09:56.5 SC: And maybe there's a little bit more to be learned than just inspiration from the biological side of things.
1:10:01.8 BB: I think that's absolutely true. Yeah. I don't know. I think the biology of consciousness is difficult. There's the psychology and the philosophy of consciousness. Getting at the biological basis of consciousness is pretty difficult, as far as I understand.
1:10:19.1 SC: Of course, famously it has been labeled the easy problem of consciousness. [laughter] But to be fair, David Chalmers always said the easy problem is hard.
1:10:30.5 BB: Yeah, well, there's.. I mean, it is... I don't know. In biology, there are no easy problems. [laughter]
1:10:35.1 SC: There are no easy problems. That's perfectly fair.
1:10:37.5 BB: There are no easy problems. Every time you come up with that dichotomy, "Oh, surely it must be A versus B," 30 years later, it's both.
1:10:43.4 SC: It's both or C. Yeah. I guess, okay, then the last question will be, are there any potential therapeutic aspects of this? Are we learning enough about how connectomes work that we can help figure out ways to fix them when they're broken?
1:10:58.9 BB: I sure hope so. I think one of the most interesting applications that I actually think about a lot in building our embodied models of the, like I said, brain's not in a jar, it actually connected to a body, is exactly that interface between the nervous system and the musculoskeletal system, of which there are tons of pathologies and dysfunctions that are pretty terrible when they happen to you. We can think about, for example, spinal injury. So that affects both your nervous system as well as your neuromuscular control. Right?
1:11:31.5 SC: Yeah.
1:11:32.9 BB: You can have an injury to... Like if I have a bum ankle on one side, I start limping. That's a different neural strategy. And then over time, I might adopt a different gait and walk differently. And then a little bit longer, maybe the legs on one side of my body get stronger. So it's adapting at a different timescale as well. So understanding those interactions between how your nervous system controls movement in compensation to injury and attempting to repair it, or if you have an amputation, you can't repair the amputation, but you can repair the function. So all of those interactions are very important and poorly understood because they are holistic.
1:12:12.9 SC: Right.
1:12:13.5 BB: Our ability to understand those kinds of holistic, longer-term adaptations is very poor because we just haven't had the tools to be able to do so. And so that's one of the things we hope to be able to do in building these embodied models is to understand not just does it walk? That's kind of, hopefully that works, right? But after it walks, what happens when it breaks? What happens when we break it in different ways? How does it compensate? What is the role of plasticity? What is the role of growing new muscles versus growing new tendons versus growing new neural pathways? And if we understand that, perhaps we'll have some new clues about how to design therapeutics to help that process work better, right? Like your PT does a lot of crazy stuff, but not all PTs agree on how to rehabilitate after some kind of injury. Is there any guidance that we might be able to come up with by understanding the interactions between the nervous system and the musculoskeletal system a little bit better?
1:13:06.7 SC: This all sounds very complicated.
1:13:09.5 BB: Biology is complicated.
[laughter]
1:13:12.4 SC: That's why it's the science of the 21st century, I guess.
1:13:17.0 BB: It's squishy and it's fun. And I just feel so... I feel so privileged to be able to do it, to spend my time hanging out with my friends talking about brains and biomechanics. Sometimes I wake up and I can't believe this is a real job.
1:13:36.1 SC: I think that is the perfect place to end because that's an inspiration for everyone. So, Bing Brunton, thanks so much for being on the Mindscape podcast.
1:13:43.6 BB: Thank you, Sean. This was super fun.
[music]
Bing makes the very important point here that she believes that consciousness evolved in and with embodied biological organisms. She points out that every example of consciousness that science agrees on has been found in embodied biological organisms. No examples of consciousness have ever been found in any inanimate object. Consciousness is a huge evolutionary advantage for animals and helps them to survive and actualize their own biological, emotional, and group survival, reproductive and social and other interests. In fact, consciousness is such a huge evolutionary advantage that it is virtually impossible to conceive of an animal that lacks it. No one has ever identified an unconscious animal, and (notwithstanding Chalmers’ speculative zombies) unconscious animals are actually inconceivable, as they would have no mechanism for navigating their environment and getting around to finding food and mates.
Connectomes seem partly the key to species specific differences in the evolutionary ladder.
Assuming that chimpanzees and humans share 98% of their dna- humans write poetry, do maths, science and design rockets unlike chimpanzees. However both humans and chimpanzees share characteristics of social bonding, rituals etc. Chimpanzees and humans could share connectomes, but the capability differences between them could be due to the flexibility of the connectomes or neural plasticity. Humans share 99.9% of their dna with each other yet they differ so much in their talents/abilities. Differing connectomes possibly shaped partly by the environment during development could play a role here.
Reproductive isolation between species follows geographic isolation. Despite similar dna sequences differing connectomes shaped by differing environments could play a role in this.
Since most of the cells in our body and brain have undergone change or have been replaced by other cells over the course of our lives how do we retain the feeling that we are the same person now that we have always been?
Think of it like this:
o A river is always made of new water, but it still keeps its shape.
o A song played on different speakers is still the same song.
o A computer can replace its hardware parts, but the software and data preserve identity.
What actually stays the same?
Your brain’s connectivity map (the “connectome”) neurons form stable networks that encode:
o memories
o habits
o personality traits
o emotional tendencies
o your sense of “I”
These networks don’t get replaced wholesale. They evolve, but they don’t reset.
Philosophers have a name for this: the continuity of identity.
You are the same person because your brain preserves the same ‘information pattern’, even if the material changes.
This is the same logic behind why:
o a ship with replaced planks is still the same ship (Ship of Theseus)
o a computer with new hardware is still “your computer” if the data is intact
A non-obvious insight
You don’t feel like the same person because your body stays the same.
You feel like the same person because ‘your brain is very good at telling a coherent story about the self’.
The self is a narrative engine
Even if the underlying biology refreshes, the story continues.
Ref: Microsoft Copilot
After listening to this episode, I am now worried that if that long, long cell dies, I will not be able to know if I stubbed my toe.
So fascinating this conversation and how you go about coarse graining areas and circuits in the human brain. Also intriguing the point about human reasoning and poor intuition when feedback circuits and recurrence. Is there a reference article? Also wouldn’t learning work with feedback of sorts though? But I also thought that it is true at a certain physical level – it’s easier to run point to point, or in a loop-circuit – but a D>B, C>A would be a rather strange movement?