Inner Cosmos with David EaglemanEp83 "Why Do Your 30 Trillion Cells Feel Like a Self?" Part 2
Does our sense of self emerge from our brain's skill at lumping things into unchanging categories? What can we learn watching a caterpillar brain transition to a butterfly brain? Can we think of a memory as a pattern that stays alive and has its own life? Does an ant colony have a sense of self? Join Eagleman and biologist Michael Levin at Tufts – one of the most energetic and original thinkers in the field -- to dive into new territories of the self.
Why do you think of your thirty trillion cells as a self? Does an ant colony have a sense of self? And could you think of all those ants as a liquid brain? What does any of this have to do with how the brain of a caterpillar transitions to the brain of a butterfly, or how we might think of a memory as a pattern that stays alive and has its own life. Welcome to Inner Cosmos with me David Eagleman. I'm a neuroscientist and an author at Stanford and in these episodes we dive deeply into our three pound universe to uncover some of the most surprising aspects of our lives. In the last piso, I talked about who we are and how we change through time. The pieces and parts of every single cell in your body degrade and get replaced continuously, such that you are physically speaking, a totally new person every few years, and yet we experience the illusion that we are the same person we've always been. We have this illusion of constancy. So last week we explore this by considering the thought experiment of the Ship of Theseus.
The story here is that.
Each plank of a famous ship is replaced one by one over time, which raises the question, is it's still the same ship even after every plank has been replaced and nothing of the original remains. We consider this question, of course, because like the ship, we too exist with physical changes that never stop, and yet we perceive ourselves as constant through time. The planks of theseus's ship map onto the cells and molecules in our bodies. So what maintains your sense of self over time? Okay, so maybe the thing that links our different selves through time is our memory. But there's a problem here as well, which is that memory is notoriously unreliable. It's constantly being reshaped and revised. So this is all very strange. And there's an added difficulty, which is that we're constantly changing, and even when we recognize that we have changed, we always assume that we're not going to change very much into the future, and that's always incorrect. This is a cognitive bias called the end of history illusion. We tend to believe that our current preferences and personalities are fixed, even though we will in fact continue to evolve. So that's our situation. We constantly change, but we're reaching back into our not so good memory to try to understand who we are, or.
At least who we were.
All this encourages us to start looking for new frameworks when we think about the self. For example, what if we thought of the memories themselves like their own little creatures and they are competing to stay alive. So in this episode, I want to dive into new ways of thinking about all of this, and there is no one better to ring up for that than my colleague Michael Levins. He's one of the most energetic and original thinkers in the field. Michael is a developmental biologist and a synthetic biologist, meaning he makes new kinds of organisms. He's a Toughts University and what I love about Michael is his extremely broad interests like bioelectrical signals by which cells communicate, and what cognition looks like across totally different body plans, and how you get similar forms and functions across all kinds of different scales and biology. So I'm going to have Michael back to talk about some of these other topics on a future episode, but today I want to zoom in with him on his recent thinking about the self and memory. So Mike, tell us how you think about the self.
I work in diverse intelligence. I'm interested in all different kinds of implementations of minds, and I think a self is a useful way to think about selves is as a system that has goals of some particular size. It's what the dark hop Statu would call a strange loop. It's an observer of the outside world, but it's also an observer of itself. It's something that can loop back in and interpret what its own memories mean and what it is doing, and make decision and going forward. Those are the components of being a self. And so do selves change? Yeah? Part of the paradox of being a self is that you have to change in order to stay alive. You have to change in order to persist. Every time you learn something, every time you make a decision that then feeds back and alters the inputs that you receive and thus changes your own cognitive system and your own behavior going forward. You've now changed. I you know, we selves face this this paradox that in order to to to to persist, they must change and transform over time.
Why do you suppose we have the illusion of ourselves when we think about you know yourself and your life, you have the illusion that it's unchanging.
What do you suppose that's about. Yeah, I think that, Uh well, let's let's just think about very primitive, basic, fundamental, basic agents at the very beginning of of of of that of that spectrum. If if you're any kind of a system that's going to survive in the real world, what you can't afford to do is to be a kind of placium demon where where you're tracking the micro states of you know, every single stimulus, every molecular impact on your membrane or whatever you can. You can't afford to track all that. If you try to do that, you will run out of time, You'll run out of energy, You'll be eaten and dead in no time. And so I think that what we have in at least at least in biological evolution, is an incredible pressure to coarse grain, that is, to look out into the world and to group inputs and sensations and stimuli and observations that you make, to group them into categories, and to try to understand your world by reducing its complexity, and by trying to make models of the world that have these large things in it that are themselves agents. You know that do things, and that you can then make decisions about what they do and you don't have to pay attention to all the micro states. If you're going to do that, you fundamentally need a way to make models of persistent things. You have to have some way of saying that this is an object I need to stay away from. More conversely, this something I need to approach, or this is something I need to recognize meanwhile, of course, so let's just look at vision, you know. For example, So you have these these pixels and impacting onto your retina, these these visual stimuli, and depending on how you're looking at something, the details will be completely different, the lighting will be different, the motion, the way that it's angled. But your job as a successful cognitive system is to recognize as well as possible that this is always the same thing, right that you can, despite the fact that it's now just shifted, tinted green or whatever it is, that that hey, wait a minute, I recognize it. I know what this is. And so I think from the very beginning to be a successful being that can survive in a world war, time and energy are extremely precious. These are precious resources. You have to get good at noticing and in fact magnifying invariance. You know, things that stay the same. And then I think you turn that on yourself and you say, wait a minute, I am also an agent that does things. And I don't mean consciously, and we don't do this consciously. But every system, every living system, I think, has an internal self model, and part of that model is to recognize you as a persistent thing even though things change. And then and then you know, those those of us that have self consciousness then have a have a story that we tell ourselves about being a persistent thing as opposed to I think what we much more close to what we really are, which is a flowing process.
So so we all walk through life and we have our notion of self, and we observe the selves in others that we love, and we try to figure out what this is all about. You have a very cool approach, which is that you think about diverse intelligences. Tell us what that term means.
Yeah, diverse intelligence is the name of a of a of a feel that on the one hand is the emerging now and it's a very exciting kind of feel. On the other hand, this is a collection of ideas that have really been around for a very long time. And it goes back to the efforts of trying to understand what what do we actually mean by intelligence, by minds, by cognition, by all all these kinds of all these kinds of terms. And we have familiar examples of them in brainy animals and and and you know, in ourselves. But actually the key thing that that we need to understand is that all of us, at one point, both on an evolutionary scale and on a developmental scale, we were all a single sell once. You know, at one point we were a little blob of quies and cytoplasm and unfertilized noocyte. And we would look at that and we would say, well, there's there's a little a little blob of chemicals that obeys physics and chemistry. And then at some point we have something that we would say has an inner perspective, It has a mind, it has preferences, it has goals, it has memories, and and you know it and it has a full blown self conscious kind of meta cognition and so on. Now, how did we get there? Right? Because we started out you know, this this amazing journey from physics to mind. You know, you start off with with a little blob of chemistry, and then eventually here we are, so diverse. Intelligence is is I think, is predicated on the on the idea that we need to understand how it is that minds are embodied in the physical world, and that by understanding that we emerge slowly and gradually, this is we are not. You know, there's no there's no magical category called human that suddenly snaps into existence at some particular moment of embryonic development. But this is a slow, gradual process. It's it's an effort to understand what those processes are that give rise to collective intelligence. And by the way, all intelligence, I think is collective intelligence because all of us are made of parts, and we have to understand how the competencies of those parts give rise to whatever the larger scale system is capable of doing. And as soon as as soon as you frame the problem that way, to understand what is it that the components are doing to scale that cognition from the competencies of a single cell to that of an animal or a human, then then immediately you start to ask yourself, okay, so not only is there a history going backwards where we see increasingly simpler forms and ask what is their intelligence like? But then we can really try to shed some of the limitations that we're given to us by our evolutionary firmware that make it easy for us to recognize, you know, large intelligence, large large, medium sized objects moving at medium speeds through three dimensional space, and we so, okay, that's an intelligent animal doing whatever it's doing. But to learn to recognize intelligence and unfamiliar guises, you know, what other spaces can intelligence be operating and what what else can it be made of? What other processes give rise to the scale up of intelligence. That's diverse intelligence. It's the broad attempt to understand intelligence as it can be. What's an example of that?
For example, ants in a colony understanding that as an intelligence system where they're laying down memories in particular ways. It's very different than the way we think about the human brain. Is that an example?
Yes, I'll give you a few examples and get the sort of progressively progressively weirder with them. Traditional collective intelligence is like ants and beehives and termite colonies and blocks of birds and things like that. Those are typical collective intelligence. They are what our Carsole calls liquid brains because they're moving around relative to each other. Our neurons tend to stand still relative to each other. And yet we too are a collective intelligence, right, We are a pile of neurons and other cells that work together to give rise to an emergent being with memories and preferences and goals that don't belong to any of the individual cells. So, no matter what you are, whether you're a collection of neurons or your collection of birds or of termites or anything like that, what you need is a kind of cognitive glue. You need these policies that are implemented by the pieces that allow the collective to be more than the sum of its parts in relevant ways. And so if we define intelligence as problem solving competencies in some space, so the ability to and so this is kind of William James's definition of intelligence, same goal by different meats, right, the ability to navigate some space to reach your objective and do so despite novel despite perturbations, and and so on. Then we can get we can get progressively more inclusive with this. We can say, okay, so we have we have animals that solve problems. We have unfamiliar things like slime molds, right, so so so there are people including us, who study slime mold behavior, in slime mold learning and things like that. So slime molds are a very unusual organism. It's a single cell. It can be quite large, and yet it can it can solve all kinds of problems. People have studied memory and learning in in bacteria, in unicellular organisms and plants, and and then you can get then you can get you know, even even more sort of broad with this, and you can ask, well, what happens beyond three dimensional space? For example, the cells that make up our bodies and our and our tissues and our organs, they operate in physiological state space, so the space of all possible physiological conditions. They operate in gene expression space. Embryonic and regenerative cells operate in anatomical space. They have to navigate from whatever shape they have now to some kind of complex organ structure. And in all of these spaces you can see if you look for them using the techniques of behavioral science, you can find them solving problems. You can find them getting to their goal in very creative ways despite various problems So that's kind of the range of the diverse intelligences. And then beyond that you have embodied robotics, and you have maybe software intelligent agents, and maybe someday exobiological beings and so on.
Okay, so the challenge is that everything's changing all the time in our brains and all these other biological systems and colonies whatever, things are changing all the time. Cells are dying or aging or getting mutations. And yet somehow, when it comes to memory of oneself, we tend to have some sort of memory. It's not perfect, it only has a few details, things like that, but we retain this memory through time. Now, what is your framework for thinking about that?
Yeah, I'm not sure we retain memory through time as much as we reconstruct memory through time. And of course there's a lot of human neuroscience being done about this. But I want to take a step back and give you another example that kind of drives my thinking on this. Consider what happens from a butterfly to a caterpillar. So in the caterpillar, you have the soft bodied creature that lives in basically a two dimensional world where it crawls around and eats leaves. Now, one thing that it has to do is that has to turn into a butterfly. During that process of metamorphosis, its brain is massively rebuilt. Many of the cells die, all the connections are broken, it's completely refactored, and you get a new brain suitable for driving a now hard bodied kind of vehicle, completely different controller you need for that. And now it flies and it lives in the three dimensional world and so on. Now it turns out that if you train the caterpillar the let's say you train it to associate a specific color to a specific leaf that it wants to eat, what you will find is that the butterfly, despite having a totally refactored brain, will remember that information. And for years I thought the amazing thing here was how do you hold on to information when your information medium is being totally ripped apart and refactored. Right, we don't, you know, our computer technology doesn't doesn't like that. We don't have anything that has that that robustness to it. But if you think about it more, what turns out is that the amazing thing is not holding onto the memory because actually the memories, the exact memories of the caterpillar, are of no use to the butterfly whatsoever, because it doesn't move the same way. It doesn't want the leaves a drinks nectar, right, And so just holding onto that memory is of no use. What you need is to actually transform and remap that memory into a completely novel context where it's like not leaves but food for example. Right, So you have to generalize a little bit, and you have to now remap it on a different on a different controller. And you know, numerous people who have done experiments in memory transfer, most recently David Landsman and others show that that it's it's really wild how you can move either tissues or extra extracts let's say, RNA extracts or whatever, and introduce them into a new into a new creature and and have that creature like take on that that initial learning.
So let me interrupt for a second. So tell us about David Landsman's experiments in these sea slugs. Yeah, Well, specifically in David's work, what he was doing is training these c slugs to a particular to a particular task. I mean it was a simple, you know, simple reflex and extracting RNA from from the brain. Because the hypothesis was that memory is stored in some way in this in this medium of RNA, and then he was injecting it into a naive animal and showing that they now have you know, they show a recall of that information I mean to me, I mean, it's it's a fabulous body of work.
To me. One of the most interesting things about it is that when you've got this RNA, you don't have to go put it exactly in the right neuron where it was supposed to be, you know, and that you just kind of inject it somewhere in the brain of a second of a second sea slug exactly of the recipient right of the host seasug, and it somehow gets taken out. And that's that's a that's a theme that that is like central to all this because what I think this is, this is telling us is so now, so now let's let's walk into this from from the other end. Whatever you are, human or anything else, you don't have access to the past. What you have access to is the end grams, the memory traces that the past has left in your brain or body or wherever that were formed by past experience. And so what you now have to do is to at any given moment and it's you know, these moments. I don't know how many milliseconds they are, but but some some number of milliseconds. You have to reconstruct that memory to be meaningful to you now because you don't know what it used to mean. This is you have to do this on the on the fly. And I think there's a lot of good neuroscience showing how plastic these memories are and how there how you know, even even recall of the memories means they're getting changed. There's no non destructive read. Accessing these memories changes them and and and so so the way I think about this is as an architecture in the shape of a kind of bow tie. And this is for the computer science listeners. You might imagine like a like a an auto encoder architecture where there's a funnel on the left side which receives the primary experiences, the raw data that come in the sense impressions, and that the process of learning has to and and and this is fundamental to intelligence, is abstracting from individual instances of things you experience to a rule, to some kind of some kind of pattern, and it's the pattern that you remember it, so you compress all those experiences, you throw away all the irrelevant details and you form a memory. You store that in some sort of enngram. Uh. You know. Sometimes people think of this as synaptic modifications. Sometimes other people think it's in the RNA or in a cyber skeleton wherever. So so you store this but now but now here comes here comes the really interesting part. When you need to recall this. You have to know, here comes the right side of that, on the right side of that bow tie. You have to sort of reinflate that compressed memory. And because you've lost all kinds of information, this process, this, this recall process is creative because you don't have all the details that were there, and nor can you really be sure at a later time what the meaning was to your past self. In other words, I also think of memories as literally messages from your past self. So I tend to think of memory and recall as communication events. And it's just you know, it's a communication with a past version of you, but it's the same. They sent you a message. It was encrypted and compressed in these n grams, and then you try to reinflate it. And your goal at any given time is not to have any kind of allegiance to what the memory meant before. Your goal now is to reinterpret it the way the butterfly does in whatever is the most optimal adaptive way that makes sense. Now, so this really, you know, the first part is kind of algorithmic, which is the compression. But now here comes a creative process. It's not really deductive. It's a creative process where you take that prompt. It's more of a prompt than anything else, and you say, okay, what does this mean to me? Now? How can I incorporate this into my current constantly evolving model of the self, of the outside world and on. So it's very much And this goes on, goes on all the time. And this is consistent with the plasticity of memories. It's consistent with confabulation, which has been seen in all kinds of experiments with human subjects, you know, split brain and so on.
Wait, let's take a second to talk about confabulation. So this is where brains seem to make something up. You know, there are these experiments, for example, the cutaneous rabit illusion. If I tap you twice somewhere on your forearm, and then I tap you a third time. Let's say, in a different location, you will feel like you felt three taps, that we're all that we're moving in the direction. Even though the second tap was in the same spot, you feel like you felt it on the way to the third one. This is one example of lots of confabulation where the brain is retrospectively making things up. Now, the interesting part is we look at confabulation generally as something bad. For example, with large language models, we talk about hallucinations. But there is another way to look at this, which is as hypothesis, as generating new creative ideas. So tell us how that fits into the way you think about confabulation.
Yeah, I mean so, here's another another example of confabulation that is similar to what I'm talking about. Two examples. One is there was a patient that had an electrode that was placed in their brain for I think aplepsy was the idea, and it happened to land in a region of the brain that when you stimulate that electrode, it makes them laugh, makes the mouth laugh. So the person will be sitting there thinking about something serious. You can see and there's a video of this I saw somewhere. The scientist pushes the button. They start laughing, and then you ask them why are you laughing, And the answer is never, Gee, I don't know. I was sitting here thinking about something serious and then my mouth started laughing. That's never the answer you get. The answer you get is, well, I thought of something funny. I thought of a joke, And you get the same thing out of split brain patients. When the one hemisphere causes the other side of the body to do something that the language hemisphere doesn't understand, what they usually do is make up a story about it on the spot, and they don't know, you know, consciously, they won't report that they're making up a story. So I think what we mean by confabulation is really a fundamental skill and necessity of sense making of your world. You have a model of the outside world. You have a model of yourself and what you are and how you behave and sometimes that model gets updated, but sometimes you just incorporate other world events that the go on. You incorporate them into that model and you interpret them in a way that makes sense to you. Now, you know, and something similar to what you said about the tapping, you've seen the rubber hand illusion.
Yeah, in the rubber hand illusion, your hand Let's say your left hand is covered up. You're not seeing it, but you're seeing a rubber hand, and you see somebody stroking that rubber hand, and every time that rubber hand gets stroked, you feel a stroke on your hand too. The person is stroking them both at the same time, even though you can't see your own hand, and then they hit that hand with a hammer and you withdraw in terror because it feels like it's become your hand.
Yeah. So, to me, the amazing thing about that that is the plasticity. Look, we've been tetrapods for I think almost four hundred million years something like that, and so for millions of years we had a brain that knows exactly how many limbs you have, and within what seven minutes of new experience, you now have decided that you have this extra hand. It's the plasticity is crazy.
By the way, One thing that has been extraordinary on this is experiments in VR, where, for example, you give somebody a third arm that comes out of their chest, and you control your natural two arms with two controllers, and you can see your arms and you also see this third arm which you control by changing your wrist orientations, and that can control the third arm, and within a few minutes people can get very good at controlling this third arm. And you know they're doing a game where you pick up boxes of certain colors, and yes, so you can add limbs and subtract limbs readily. The homunculus, the little model of your body, is totally flexible in that way.
That right there, the ability to adapt to novel situations in this way that can fabulation to tell a story that makes sense, not that it's necessarily true relative to what your past was giving you, but to what makes sense for you now is fundamental to buyology. And I think biology doesn't preserve the fidelity of memories. It preserves the salience of memories. It tries to remap them in the way that it makes sense to you now, not necessarily to what it meant. And I think that this is fundamentally an intelligence ratchet for life. And here's what I mean by that. Let's look instead of the memories of a single of a single human or animal, Let's look on an evolutionary timescale. You come into the world as an embry Oh, You've been given all of this genetic and cytoplasmic and other kinds of information that are the accumulated really the memories of your of your lineage agent, right, the lineage that's been through that's been through the evolution and has accumulated all this useful information. There are some animals and at least as far as we know, maybe nematodes like C. Elegance, that are extremely hardwired. We know exactly how many cells are going to have, all the cells of the same position that's determined by lineage. They're very hardwired. But the majority of living forms aren't like that. They take that information and they reinterpret it in novel ways. I think that now now in normal development, we always see the same thing happening, so we kind of assume that it's some kind of hardwired mechanical process, but it's not that at all. For this reason, we can take we can make a tadpole which has no eyes in the head, but it has an eye on its tail, and that those those animals can see right out of the box. They don't need new rounds of selection or evolution or adaptation or any of that. They can see immediately. We can take cells from an early frog embryo and they become zenobots, and they do interesting things. There have never been any zenobots. There's never been any selection to be a good xenobot, right, so that plasticity can you can you explain to the audience with zenebot is sure? Yeah, So so we make we make in our group, we make zenobots and anthrobots. These are these are living systems that we call them biobots because we use them for, among other things, bio robotics kinds of applications. They are living uh organisms made of cells in the In the case of xenobots, they come from frog cells, and the case of anthrobots, they come from adult human tracheal epithelial cells. They are self motile. They will move around a dish. They sort of swim around the dish on their own. And they have lots of interesting capabilities. For example, the anthrobots, if they find a neural wound, they will heal They will heal the peripheral innervation by taking the two sides of the neural wound and kind of connecting them together. You know, who would have known that your tracheal cells that sit there for you know, quietly in your body for decades, have the ability to form a self motile little creature that runs around and do these things, and does these things so the plasticity and people have been noticing this, this kind of thing for a very long time. Developmental biologists is that living systems will play they hand their depth. They don't just automatically, at least most of them don't just automatically do the same thing. They will try to as much as we try to telecoherence story in confabulation and linguistic space, they will confabulate in transcriptional space meaning gene expression space, in physiological space, and in anatomical space to put together some kind of a coherent lifestyle. Given novel, novel circumstances, the environment can be novel. You can interface living tissue with all kinds of weird materials. They will always try to make something, And I think that's because they never assume that you can take the past literally. They have to on the fly put something together. One of my favorite examples is what happens in the new to kidney tubules. You have a newt If you take a section through the kidney tubule, you see eight or ten cells are making like a circle, and they make this. They make this two well, one of the things that the people have found is that you can treat them in a way that makes multiple copy number of their chromosomes so they have the genetic material. Instead of two n they will have four and five and six end and so on. If you do that, the cells get bigger, but the nude stays the same size. So that's kind of amazing. So you take a cross section through the two bule and you see, oh, the cells are bigger, but there's fewer of them and they still form the exact same structure. Well, you can make a highly polyploid neud like that that has gigantic cells, and in that case, one single cell will bend around itself, leaving a hole in the middle, which is a completely different molecular mechanism. Right, it's not cell to cell communication. That's some kind of cideoscalarle bending. So now think about what this means if you're a nude coming into this world. You can't count on how much genetic material you're going to have. You can't count on and never mind not being able to count on the environment. Right, who knows what the pha your water is and all that. Forget that you can't even count on your own parts. You don't know what your chromosome number is going to be, you don't know how many cells you're going to have, you don't know the size of your cells. You have to do something coherent in that case build an actual neud when everything changes. And that's why I think that that's the fundamental thing about confabulation is that if you commit to the idea, which I think biology has to. Unlike our computer technology, which relies on a highly reliable hardware, right when you code, you don't worry about your you know, cpu doing something weird. You just assume it's going to do what it needs to do. You don't think, you know your copper is going to go off or something. In biology, that's not the case. The medium is completely unreliable. You have no idea what you know, how many proteins have we given the type you have, or if they're going to get a little bit teen natured, or you know what's going to happen. If you assume that your medium is unreliable, then instead of this kind of hardwired here's how we do it every single time. Idea, what evolution is going to produce are sense making problem solving agents in different spaces. It can be very simple things bacteria and you know, but already you're off to the racist because you can't count on your environment being the same, and you can't count on yourself being the same. You're going to mutate, right, your parts will mutate, Everything will change.
So this is one of the first things that I was absolutely intrigued with in biology when I was very young, which is, how in the world does a mouse's heart and an elephant's heart do the same thing? When you know these two cases, you've got totally different number of cells making this, and yet it makes the same structure that does the same thing. So what is the way to understand how biology can code for these higher order structures.
Yeah, I think that there are key elements of understanding what's going on that come from behavioral science. This is we are not going to get to this purely by the concepts of chemistry and physics, although those are crucial to understand. What we have here are problem solving collective intelligences. So when you have a bunch of molecular networks that make a sell that's a coherent organism like a like an amebo or a lachrom area or something that that do all these interesting things. They are making a next They are contributing to a next level collective intelligence that does that, that operates in some kind of space and has a small cognitive light coone work and do certain things that have a little bit of predative power forward, a little bit of memory backward. When those cells come together and form an organism, once again, you have a collective intelligence that now projects into a new space. Whereas the cells we're solving problems in physiology and metabolics and gene expression, you know, now have a system that solves problems in anatomy. So so when you take an early embryo, let's say, an early mammalian memory, and you cut it in half, you don't get two half embryos. You get too perfectly normal monozygotic twins because each side has to figure out, oh way this is missing, why I have to rebuild and so on. And so my point is not that we attribute to uh, you know, high order human level self consciousness to these things. I'm not saying they have the metacognition to know what they're doing. What I'm saying is we have a simplified version of intelligence, which we know there had to be because we came that that is our origin. We know there has to be a version of intelligence that is, you know, sort of on the on the left end of that spectrum going all the way back to primitive cells and before that actually, And that's that's how we need to think about this as as as problem solving, continuous dynamical problem solving.
Great, So let's take where we are now and return to the issue of memory. So how does memory work in a brain?
So I'll just address to you know, a couple of things that I can speak to. What one is that I think the conventional story that memories are some sort of fine tuning of synaptic connections. I think that story is very incomplete, and there there are many people, you know, like Landsmen and sam Gershman and many others that are that are working on that. I I tend to think if I had to guess, I would say that there probably isn't one substrate of memory. I would look at memory as an interpretation process, which I think neurons are very good at this of interpreting a reservoir. That reservoir is everything else the cell is doing. The side of skeletal states, the molecular networks. I mean, some people pick up have picked up transcriptional uh signatures of certain memories that mice have had, and so on. Every everything in the cell, all the complexity that is going on, can be used as a reservoir in a sense of reservoir computing, to be used as prompts to reinterpret these those prompts as memories that are useful and with so again maximizing salience, not as early fidelity, but salience useful in their novel context. So I think I think memory is a lot about creativity. I think it's a lot about uh, having prompts that that that push you into new new kinds of problem solving, you know. And and if if your if your body and your environments stay extremely constant, then it just looks like the old version of memory, where you store a piece of data, you read it out and that's it, right, That's how it looks like. That's what it looks like from the outside. But I don't think that's what's going on, you know.
In my in my book Live Wire, I make the argument that even though in Silicon Valley we think about everything as being a trim and efficient layer of hardware, and then you build uh trim and efficient software on top of that. That's not at all how the brain's working. Instead, you've got this constant reconfiguration. And I know that you also reject that dichotomy between a computation layer and a passive code layer.
So how do you think about that? Yeah, well, I think I think there's a couple of major differences between how how we build hardware now you know, comput computational devices, and what biology is doing. The first thing we've just talked about, which is the reliability the idea that in in in the computational where you have levels of abstraction and you try to screen every layer from all the vagaries of the level below. So if you're coding in you know, c or something, you're not worried about what the copper is doing and what the silicon is doing. You you you assume that the function calls you have are going to do what they need to do and you go from there. Biology isn't like that. All all the all the layers are somewhat unreliable, and you need to be interpreting it at all at all times. Josh Bongard and I are working on a framework called polycomputing, and the idea and this is this is partially based on some amazing work that is student to Saparsa had done showing that the same set of physical events can be interpreted as different computations by different observers the exact same set of physical events. So give us an example. An example is I mean in her work, they were looking at the vibrations of particles and you look at them in one way and you see an end gate, and you look at them a different way and you see an ore gate. That's one example in biology. What it means in biology is that. And by the way, he and I wrote this paper called this plenty of room right here kind of riffing off of finements, a comment that there's plenty of room at the bottom because because biology has this thing where every level is already occupied, there is no room at the bottom because every level is occupied. How do you as if your evolution, how do you put in novel functionality when every level already has something. And by the way, when you make changes, you're going to screw up. If you make changes in the given subsystem, you're going to screw up all the other systems. That depend on what it's doing. So one thing that I think happens in biology is this poly computing where you don't necessarily change the system. You add other systems that see what's already going on in a different way and make use of it as a computation but from a different perspective. So, if you're some kind of chemical pathway that mitochondria are using as part of the metabolic path way, some other system can look at that and say, well, I'm gonna use it as a as a clock, I'm gonna I'm gonna take I'm gonna use it to regulate my timing, or I'm going to use it, you know, in some other, some other signaling capacity. And so so I think what we have in biology is not this linear stack first of all, not a linear stack, but a kind of a super a society of multiple nested, cooperating and competing agents which all have their own perspectives and they all interpret everything that goes on around them in whatever way they can. Uh so, And and you know, we're used to the fact that in a computation, we supposedly know what a given algorithm is doing, right, you can and if you don't know you can ask the person who wrote it and they'll tell you this is what this thing is computing. But in biology, I don't think there is any one fixed answer to this. It's doing whatever you as an observer can usefully think it's doing that. That doesn't mean anything goes. If you have a story that doesn't help you get around in the world and thrive, then you don't know what it's doing. But but but multiple observers can have different stories about the about the same thing.
So give us if you can't give us another specific example of that.
So the cite of skeleton, on the one hand, is used by the cell to get around, and so you might say, well, this is this is my my movement machinery. That's that's that that I'm counting on to maintain certain cell shapes and so on. But at the same time, there's other data showing that the site of skeleton can actually be can be storing memory. It's also serving as a scaffold for other molecules to find where they need to go. They're moving around with motor proteins and things like that, and they're just there's just lots of lots of different uses that any given mechanism is performing at any one time, and there are multiple different readout systems and this is this is why you know, there's the same molecule induces eyes in one context, that induces UH. You know, it might induce a kidney rudiment in a different UH contest. There are transcription factors that have that have many different roles depending on the context. In our work on bioelectrics, the exact same stimulus induces a tail to regenerate on a tackle, but a leg to regenerate on a froglet, and they never get confused, so that specificity is not in the in the treatment, it's in the surrounding cells being able to interpret that exact same signal in whatever way makes sense for them. The ability of these subsystems to cooperate in UH, in in in groups and solve problems together is is really like a fundamental thing in which biology is is different than the kind of you know, control systems that we have now in computer science. You know, Stephen J.
Gould wrote about exaptation, where you have something that develops and then it turns out to have a use in another way. But what you're talking about is even more sophisticated in that in the sense that it can retain its first use and be used for a second thing and the third thing all the same time, just by reading that data out in different ways.
Is that right? That's that's exactly right. And you know, some of some of the latest the stuff that I've been thinking about really tries to turn this whole thing on its head. And you know so, so in the standard touring computing paradigm, you have a machine and you have the data. Right, so you have a you have the process. You have this machine that reads the tape and it and it records, you know, the byproducts of the computation onto the tape and so on. So typically we look at this from the perspective of the machine. That is, we are the whether the where, the cell or the human or whatever. We're forming memories and we're writing it down into some memory medium. The memory medium is passive. The memories themselves are passive. They're just marks on a tape, and then we can read them out when we want. Some of the latest work that we've been doing, it starts out by thinking about it backwards and saying, well, what does this look like from the perspective of the data, right, data that are not passive. They're not passive patterns within some medium. They're actually active patterns. And from the perspective of the tape, the tape runs the show that machine is going to do things depending on what is written on the tape. So if I'm a pattern on this tape, I can make the machine do things. From my perspective, I'm in control. And so now it sounds it sounds a little crazy to say that that these patterns are doing things and that they're agential and whatever. But let's keep in mind we are patterns too. We are temporary metabolic patterns within an excitable medium the way that people study you know, other temporary patterns like solitons and whirlpools, and you know, all different kinds of all different kinds of systems. And from from that perspective you can you can see that different kinds of patterns persist in different media, be they cognitive media or just computational media. And asking what does the world look like from their perspective and how much, how much problem solving capacity, how much agency in fact, might those patterns have has massive implications not only for new computational architectures, but also, for example, for regenerate medicine, where you want to understand what are the persistent information structures that cause cells to do or not do various things in disease states, and you know, pro regenerative states and so on. So let's double click on that.
So what would that mean for a memory to be like an agent to be doing something.
I'll tell I'll tell a story that I read. I'm sure at least part of this was motivated years ago by a science fiction story that I'm not exactly sure what it was. I think it was a it was an Arthur Clock story, but I'm not one hundred percent sure. So let's just and I'm sure I've also twisted it in a different way. But let's just let's just let's just visual it's because your memory is creative to totally. There may have been no story. I have no idea, so, you know, I just want to give credit in case there was. So so let's just let's just visualize this. So from the center of the earth come these creatures, these core creatures, right, they live at the center of the earth. They come out onto the surface. They are incredibly dense because they live at the core, they have vision that operates, let's say, in them in gamma rays. And so they come up to the surface. What do they see, Well, pretty much nothing, because everything that we see here is like a fine ethereal plasma. To them, they are so dense. All of the stuff that we think of as real objects are basically not even within their within their ability to perceive directly. So they're walking around, stomping through everything. And you know the same way that when we walk past, you know, some kind of flower bed, there's all kinds of like fine you know, patterns of ascents and so on, we just sort of walk right through it, mix that all up. So they're stepping all over everything, and well, one of them, one of them, is a scientist, and he's taking some careful readings of what's going on around him, and he says to the others, Hey, you know there's this there's this like fine invisible gas around our planet. This is like plasma around and there's patterns in this there's regular patterns in this plasma that kind of hole together. And he say, so, what, well, I've been watching some of these patterns, and you know, they seem to be almost like they do things. They almost seem agential, they almost seem like they have goals, and like they you know, they're not they're they're they're sort of like you would see waves or solitons moving through water, you know, and they look like they hold together for a period of time. And they say to him, well, how long do these hold together? Well about one hundred years. Well that's crazy, nothing, nothing interesting can happen that that quickly. You know, They're just temporary. They're temporary, fleeting, you know, sorts of sorts of patterns. And and by the way, we've been watching the ecosystem here, and some of these patterns are really like not conducive to the health of the ecosystem, you know, these patterns are really are really like screwing things up. So they're they're kind of like these these recurrent but the unhelpful patterns. So so I have a blog post which has this this this fictional dialogue between that creed, that that that core scientist, and he tries to talk to one of the patterns. We of course are the patterns, and so the human says to him, it's really imperative that you guys understand that we are alive and and we have we are, we matter, we you know, in a moral sense, we have goals, we have memories, We persist. And he says, well, I feel like I'm crazy. I'm talking to a pattern and gas. You know you can't be real. He says, I'm real. I'm solid. I live for you know, millions of years. You're a temporary pattern in this gas. How can I take you seriously as a coherent intelligence? And so just thinking about it that way reminds us that all of this is relative, and that we two are patterns, and what other patterns around us have a degree of coherence and live and strive and have different kinds of degrees of problem solving competency and other kinds of things that we don't know. And so once we think about that, once we realize that this distinction between you know, real solid beings like us and the temporary pattern like we are all on that spectrum. We are all patterns. So once you think about it that way, it unlocks the ability to take the tools that we use to understand real embodied beings and ask ourselves, how do some of those tools and concepts from behavioral science and so on, how would they apply to certain other kinds of patterns in other media. So what are patterns in media? Well, thoughts within the cognitive system are patterns. You can have fleeting thoughts that sort of come and go. You can have persistent thoughts, you know, thoughts that are hard to get rid of. Right, then it's all many examples of that, and some of those thoughts actually do a little bit of niche construction. Niche construction in biologies, when an animal modifies its environment that makes it easier for them to persist. So you're doing something to the environment that makes it easy for yourself to stick around. Well, there are data that depressive thoughts, persistent thoughts, those kinds of things actually modifying brain issue in ways that makes it easier to keep having those kinds of thoughts. Right, So you got your fleeting thoughts, you got your kind of persistent thoughts. Then maybe you have some dissociative personality alters which are way more coherent than a simple persistent thought and in fact somewhat agential. So they have a is and they have memories or whatever, but not a full on human personality. So then you have you have that, and then who knows what's beyond that? Right, trans personal psychology will say that maybe maybe there are there are bigger things past that. So so I think that, uh, you know, this this idea of having patterns within a medium and maybe within a cognitive meaning, but also a computational medium. If you're data in a database of being being shuffled depending on that architecture, maybe you can take the perspective of that data and ask yourself, what does the world look like from my perspective, right from the perspective of the pattern, and what is the pattern doing or not to facilitate its own persistence and to facilitate its own transformation that usually is required if you're going to persist over long periods of the time, you may need to change. So that's that's you know, that's some cutting sort of cutting edge stuff as far as what we're thinking about to understand some of what goes on in these kind of complex biological cases that we want to be able to control in medicine and so on.
And presume will you think about that in a Darwinian context in terms of if I'm a thought and you know, so I'm some pattern that is a thought and I'm trying to keep myself alive. There are certain mutations perhaps that I can have, or certain things that I can do that give me an advantage in that domain.
That's part of it. But I think that the bare bones Darwinian paradigm, which is short term self interest, competition, and random change, those three things I think are woefully incomplete as a story both of actual evolutionary change in biology and the kinds of things that we're talking about here. The alternative to this, of course, is that a system that changes with some sort of foresight. Now that doesn't mean long term purpose. I am not saying that there's some sort of human or above level of a plan that is executing the changes that are happening. What I'm saying is that we cannot necessarily assume that the change that is happening is completely blinde and we cannot assume that there isn't some computational process done at the level of the lineage that is actually guiding the changes that are that are happening. One way to think about this is to think about the whole lineage, you know, I don't know, fifteen million years of alligators or something. Think about that that whole lineage as a giant, single agent distributed over time, bigger than we're used to thinking about, where every each individual animal is a hypothesis of that agent about the outside world. Some of those hypotheses are good, some are not. The thinking evolves as time goes on, right, So again, this is cutting edge stuff, you know, this is this is I'm not at all saying that we have all this worked out. This is just these are things that we're working on and some ideas going forward. But there are lots of people thinking about how much and including Richard Watson, how much and what kind of computation is done by populations like this that is not captured in this very simple uh competition for resources random change model. The way that you think about memories in the brain as being like their own agents, patterns that stay alive, and the recollection of memory as sort of the creation from some physical evidence that's there, recreation into your current world. Does this tell you anything.
About Ribou's law, which is the oldest rule in neurology, which is that older memories are more stable than more recent memories.
Have you thought about that at all? I've not thought about that specifically. It sort of makes sense that you're If you're a pattern that has managed to stick around for a really long time by interaction with the surrounding cognitive system in a way that causes it to keep you around and for you to persist, it makes sense that you have now picked up on whatever properties, residents, whatever, that allows you to be pretty stable in the system. There's this term that people use sometime about a breakthrough where you reinterpret a number of things that happen in your life. If you find out that this person, maybe that you were mad at, had some other problem going on. They knew that they had cancer, but they didn't tell you that, and suddenly they're lashing out at you, you have a totally different interpretation of it. You're going back through your memory and recasting everything in a different light. Do you have any interpretation of that in your framework? I mean that sounds to me like a very sophisticated human level cognitive version of a process that happens all the time, going all the way back to our simplest ancestors, which is that circumstances change and it forces you to reuse whatever information you had from the past, whatever tools you had from the past, to make sense of what's going on now, and that I think is the fundamental basis of intelligence. That's why I think this requirement to confabulate because everything changes is an intelligence ratchet. It requires cells to get good at solving problems in their spaces, which eventually bubbles up as collective intelligence scales and the cognitive light gones expand. It then eventually starts to look like the kind of intelligence that we we're used to seeing. But that that that fundamental process I think is is very ancient and fundament and basic. Mike, does this change anything about how you think about yourself? For me? I think it's it's very very important to face this, this this paradox, right, the paradox which which we face this as cognitive systems, but also species face this as well. If you don't change, you will likely die out when when circumstances change. But if you do change to meet those circumstances, you're no longer the same, You're not you anymore. So So what does that mean? Right, that's the paradox? How how can you possibly persist in this idea of persisting as a as a as a pattern and uh realizing that because things change all the time and this, this, I think is is fundamental. What is in our control are not the thoughts that we have right now. What's what's in our control is the long term application of effort to modify our own cognitive system to have different thoughts in the future, the thoughts you would like to have more of, and behaviors you would like to have more of versus something else. So this idea of committing to a consistent, long term process of self change, you know, the Buddhists, you also call it the body step of a vow, This idea of enlarging your cognitive light cones so that you're able to have the goal of compassion, you know, beyond our current limited human kind of scale that we can actually you know, work towards the goals of a certain size. Yeah, that's that's that's what motivates me. And the plasticity is really I find it incredibly hopeful and positive, this idea, this this incredible plasticity that has intelligence at its core, that every single cell is intelligent within and it's exerting intelligence in its cooperation and competition with others to form larger scale structures that can be molded top down, molded over time, to be better and to improve over time.
That was Mike Levin, a professor and biologist at Tufts University. So wrapping up this two part episode about the self, we saw that everything in your biology is changing all the time.
Your cells are.
Constantly turning over their pieces and parts, but we have memory to bind the use together. Now, I've talked in several episodes about how memories change in their character. They're not like a file of zeros and ones that are written down in a computer and then read back out perfectly.
And the way Michael Levin.
Thinks about this is that memories get compressed. They get encoded down into the neurons or the connections between neurons, or the inner cosmos of proteins and side neuron and then when these memories get reinflated later they find themselves in a different world, they get it interpreted by the new brain that is looking at them. So I want to make this model clear. So here's my analogy to capture that. Imagine that the world out there has lots of things that need to be bolted down, and so you create a wrench, and your metal wrench is in some sense a compressed representation of the world out there, a world full of bolts. So when you see the wrench, that reminds you.
Of all the bolts that are out there.
Okay, Now, imagine that you bury that wrench, and some other creature, some future human creature, digs it up in a thousand years and she doesn't see it as a wrench, but to her it's maybe a weapon, or it's an instrument for conducting electricity.
On her spaceship.
Or she takes it to be a ceremonial artifact, or she uses it for physical exercise, or she looks at its clean, balanced design and uses it for a piece of art. The point is that what you buried is not what gets exhumed in a new world of the future. And that's what happens to memories too. You bury something that has some meaning in the now, but what you dig up is interpreted through the eyes of.
The future you.
And if there's one thing we can count on, it's that, despite all your intuitions to the contrary, that future you will not be the same as the you now. It'll be someone you don't know, who doesn't share all your values and opinions, and is someone you can't accurately predict the ship of theseus with all those changes does not in fact remain the same ship. I was recently talking with my friend Lisa Joy, and she said she thinks it's strange that the longevity community cares so much about extending their lifespan by decades, because that future person will be somebody potentially very different from who they are now. So who are you saving if you go through a lot of trouble now to extend your life. Whoever you're saving, it's a stranger to you. You're doing all this work for someone you don't know. So, coming back to the question of why we have this notion of an unchanging self, Michael's answer is that the job of the brain is to make models of consistency, like this is what a chair is, this is what a backpack is, this is what a bicycle is. And even though there may be a lot of variety in the specifics that you come across and things might change, you nonetheless are good at summarizing things as objects. You lump them into unchanging categories. And so Michael argues, the same cognitive machinery is turned on to our selves. Even though there's a lot of fluctuation of what that refers to. We lump the self into one object that we call me, and that high level cognitive model just doesn't change much. And so as we close, we are left with this remarkable paradox that we move through life carrying memories and stories and beliefs about who we are, and we carefully preserve them like relics in the soil of our minds, and we expect them to stay the same. But with each retrieval, every time we unearth them, they are interpreted afresh. They're reshaped by the hands of a self that is itself ever shifting. But the illusion our brains create for us as a model of the self as unchanging, a fixed point in a fluctuating world. And it's a comforting thought that we're a single thread woven through time. Maybe there's also a beauty in realizing that each of your future selves is a stranger unto you, an explorer who picks up that wrench of memory, holding it to the light and interpreting something new each time.
Who we are, what we hold dear.
Maybe these aren't artifacts that are meant to be saved or preserved perfectly.
They are living stories.
They're reimagined and repurposed by every future version of us.
So when you.
Think of your future self, who you will be tomorrow or a month from now or a decade from now, think of that stranger that future you, and maybe smile at the mystery of what that person will even remember, what they'll care about, what they'll let go of. After all, part of the adventure of life is not just holding on to who we were. It's also about meeting time and again who we are becoming. Go to eagleman dot com slash podcast for more information and find further reading. Send me an email at podcast at eagleman dot com with questions or discussion, and check out and subscribe to Inner Cosmos on YouTube for videos of each episode and to leave comments.
Until next time.
I'm David Eagleman and this is Inner Cosmos.
