# Might we be surrounded with undetected minds? with Michael Levin

**Channel:** Inner Cosmos with David Eagleman
**Source:** https://www.youtube.com/watch?v=un9yp7MQFlo
**Transcript page:** https://www.withtranscript.ai/video/un9yp7MQFlo

## Chapters

- 0:00 — Introduction: Intelligence in Unexpected Places
- 5:43 — Defining Intelligence and Testing Problem-Solving
- 13:45 — Plasticity, Xenobots, and Anthrobots
- 22:58 — Collective Intelligence and Mind Blindness
- 31:24 — Reading and Rewriting Cellular Memories
- 35:20 — Patterns as Agents: Thoughts as Thinkers
- 40:48 — Rapid Fire: Ethics and AI Implications
- 44:43 — Outro: A World Full of Minds

## Transcript

**[0:00]:** What is intelligence? And if we look hard, might we find it in very weird places, not just in brains, but in all kinds of structures in our universe? And how would we even recognize it? And what does any of this have to do with a dog born without front legs learning how to walk bipeedally or making new little organisms out of single cells? Or how Wikipedia might be like an axelottle and why we are so blind to the vast variety of minds that might surround us. Welcome to Inner Cosmos with me, David Eagleman. I'm a neuroscientist and an author at Stanford. And in these episodes, we sail deeply into our three-pound universe to understand some of the most surprising aspects of the world around us.

**[1:03]:** Today's episode is about intelligence. Not in the way that I've talked about in earlier episodes about how the structure of the human brain gives rise to intelligence and how we could measure whether AI is intelligence. Today's episode is way beyond that. Today I'm going to talk with one of my most brilliant and creative colleagues, Michael Leaven, about how we might find intelligence all around us in ways that we don't typically intuit it. So let's start at the beginning. What is intelligence? It's a word that we usually reserve for something abstract and cerebral, something associated with problem solving and planning and passing IQ tests. We tend to picture intelligence as a property of brains and especially big human brains. We're generally willing to grant some intelligence to dolphins and chimps and clever birds like ravens, but it's hard to know how to think about so many other things happening in the world.

**[2:04]:** For example, my skin cells heal a wound. Is that intelligence or is that just biochemical cascades? A plant grows towards sunlight. Intelligent. A worm gets its head cut off and it regrows it. That's amazing. But we don't tend to call that cognition. But what if we've been looking at the whole notion of intelligence too narrowly? What if intelligence isn't just about neurons and genes, but it's about goals? And specifically, it's about the ability of a system to navigate towards an objective, to adapt to its circumstances, to make decisions along the way. That's a broader definition of intelligence. And if we apply it, suddenly intelligence doesn't just belong to creatures with brains. It becomes something that shows up in places we didn't expect. Think about really simple creatures like a tadpole. It's millions of cells collaborate and communicate and organize into an eye and a spine and a heart without anybody orchestrating the whole thing.

**[3:14]:** There's no central planning. It's just a kind of emergent intelligence at work. Or think about a flatworm that can be cut into pieces and each piece regenerates a complete properly shaped body. How does each fragment know what's missing? Where exactly is that information stored? What is guiding the process? And as we ask these questions, that leads us to ask how we can learn to talk to these systems in the language that they understand like voltage gradients or bioelectric circuits or chemical signals. Can we start reprogramming the goals of tissues? Can we tell a clump of cells to build something new? And can we use this kind of knowledge to regenerate organs or repair birth defects or create entirely new forms of life? These are the kinds of questions that today's guest has spent his career exploring. And his work leads us to the conclusion that we're probably surrounded by minds almost all of which we don't recognize.

**[4:19]:** Minds are pervasive and they're not restricted to brains but spread across all kinds of levels of organization from single cells to societies. My guest is Michael Leaven. He's a distinguished professor of developmental and synthetic biology at Tufts University. And I've had him on the podcast before because he's really one of my favorite thinkers in the field. He's massively creative and always pulling off amazing results at the frontier where biology meets information theory or computation or philosophy. And as you're going to see his work always challenges our deepest intuitions about agency and memory and selfhood. So you've heard of SETI the search for extraterrestrial intelligence. Recently Leaven proposed suti the search for unconventional terrestrial intelligence. as we're about to hear, his position is that right here on Earth, there are already aliens among us that stretch and sometimes break our typical ways of thinking about minds.

**[5:25]:** So, if you've ever wondered where intelligence begins, how far it reaches, or whether you might share more in common with blobs of cells than you think, this episode is for you. Here's my interview with Mike Leaven.

**[5:43]:** So Mike, let's start by telling us how you define intelligence. Okay. Uh what I use is a definition that helps us move forward in the lab. I do not claim that it's the only definition or that it captures everything there is to uh to capture about intelligence. But I like William James' definition, which is some degree of the ability to reach the same goal by different means. So it's some level of ingenuity to get your goals met when things change. That doesn't capture play. It doesn't necessarily capture creativity, things other than problem solving, but but this is what we focus on experimentally. So, typically when we think about intelligence, we think about brains and nervous systems, but you think it doesn't even necessarily require that. Correct. Because because if you're looking at it in this in this way that that it's basically a functional capacity to navigate some kind of problem space and meet specific goals under changing circumstances.

**[6:32]:** There are apparently a wide range of architectures that can do this. And in order to see that, what you need to do is to relax some really constraining assumptions that we often have about the problem space that we're working in. And so you often describe intelligence as scale free. So just give us a sense what you mean by that. Yeah, I mean that you know as as humans, we are because of our own evolutionary firmware, we are so obsessed with three-dimensional space and moving around in three-dimensional space to the point where if people see some sort of AI that isn't rolling around on some sort of robotic body, they're going to say this is not embodied, right? because they're expecting a body. People are expecting a body that moves through three-dimensional space. But actually the biology for example has been solving problems in navigating all kinds of spaces that are hard for us to visualize.

**[7:15]:** So the space of gene expression states, the space of physiological states, the space of anatomical possible outcomes and so on. And so in order to understand uh how we navigate those spaces, it you have to think in other scales. Some of these things happen very slowly. Some of these things happen incredibly quickly. Um some of these things h are very small. Some of them are are very large and we we are just you know focused on medium-sized objects moving at medium speeds but you know I'm not claiming it's entirely scale free but I'm saying that there are very deep scale and variant principles that operate at many different scales besides the ones we're used to. So how have you gone about looking for intelligence at other scales? So one thing that you that you can do once you've bought into the fact that uh we can't assume how intelligent anything is or what kind of intelligence it has but you have to do experiments.

**[8:03]:** Then then what it turns out you have to do is you have to posit some sort of problem space you have that that the system is working in. You have to posit some sort of goal that is these are all hypotheses here some sort of goal that it's um it's trying to reach and then what you have to do is perturbational experiments to uh prevent it from reaching its goal. And then you see what how you know how how smart the system is in getting around whatever you did to it. So you knock it off track and then you see how it comes back or if it comes back. Knock it off track is a good one. Um add barriers of some sort in whatever space you're working. Doesn't have to be a physical barrier, but whatever space you're working in, add a barrier or in fact manipulate the system itself.

**[8:39]:** So change the system, right? It's not all about the environment. It can be about the system itself. And so uh we we've done this now in many different contexts. Um here are a couple of favorites of ours. The biggest one that we do most of our work in is morphagenesis. So we all make a journey from a single cell to whatever you know we're going to be a human a giraffe a plant whatever and that journey as it turns out as a matter of experimental fact it turns out that that journey is not a kind of openloop uh what you know the way people model emergence and complexity lots of simple things happening again and again and ultimately some sort of complex event happens that isn't how it works it's uh it's very contextsensitive if you try to prevent uh let's say embryos or regenerating limbs or or you know any you know metamorphosis any of these processes, you try to prevent them from reaching their goal.

**[9:24]:** They often have extremely ingenious ways to get there anyway. Okay? And you can quantify this and you can say what are the kinds of um perturbations that it's able to deal with and does it have delayed gratification? Does it have planning? Does it have a representation of the f of counterfactual states? Does it have learning and memory? Does it store you know a map of its environment? You can you can test all of these things empirically. So you mean in terms of let's say an embryo developing what we think traditionally in textbooks is that the genetics somehow gives a blueprint and the whole thing just unpacks. But what you're asking is how is the system intelligent if we knock it off track or put barriers in the way? How does it figure out how to come together correctly? So what's what's a specific example of something you've done?

**[10:08]:** Uh here here are some of my my favorites. These these first two are not my work. These are this is like classic classic work in the field. So, if you take um normally imagine cutting a cross-section through the kidney tubule of a of a noot, normally what you'd find is like 8 to 10 cells and they work together to uh to build to build this this this tube-like structure. So, what you can do is you can make polyloid nudes that have multiple copies of their chromosomes, which means their cells have to get bigger. Those nudes are still the same correct size. So, that's the first interesting thing. Wow. When the cells get bigger, the thing scales down. How does it do it? by using fewer but bigger cells to make the exact same structure. So that's an adjustment, right? Never mind the environment. Your own parts are changing and this thing is figuring out how to get to the exact same goal, the same nude, same shape, same size, fewer of these bigger cells.

**[10:54]:** So let me ask you a quick question. Is this analogous to the fact that a mouse's heart and an elephant's heart are doing the same thing, but they're made of a completely different number of cells? It's a massive heart in an elephant. Very tiny in a mouse, but it's doing the same function. It's similar. Uh but but there's only but there's one major difference. Both in a mouse and in an elephant, what people will say is, "Well, both of those have had long a long long history of being what they are, and so this is just mechanical. It just does what it does. My my example is is different because you've done something to this N that it does not normally do. You've given it a completely novel uh uh circumstance and then it adjusts in a new way. And the craziest thing happens when you make the cells really gigantic.

**[11:37]:** Okay, these are like I think six or eight nes. So massive polylity. What happens is the cells are so big. There's no room for more than one cell. One cell will wrap around itself and give you the lumen of the tubule in the middle. Now now this is crazy because because that is that's a different molecular mechanism. That's cytokeleletal bending. Before it was cellto cell communication and tubularogenesis. So think about what this means if you're a nude coming into the world. Never mind the environment. You don't really know what your environment is going to be. You don't know how many copies of your chromosomes you're going to have. You don't know how big your cells are going to be. You don't know which of your many genetic affordances you can use, right? You have different molecular mechanisms you can use. You have to figure out what to do in a novel circumstance and still get the job done.

**[12:18]:** I mean, this sounds like uh every IQ test you've ever heard of when people show you, here's a little box, some tax, and a candle and I want you to, you know, solve this this particular problem, right? Yeah. you have genetic affordances and then that morphagenetic process is not just doing the same thing every single time. You have to solve these problems. That's one of my favorite examples. Another one that that we discovered is tadpoles become frogs and in order to do that they have to rearrange their face because their face looks actually quite different. Um and you know so the eyes the jaws everything has to has to move. What people thought was that this is a hardwired process. Basically somehow the genetics just tells every organ how far to move in what direction and then you get from a normal taple to a normal frog.

**[12:56]:** So we decided to test that because you can't assume these things. You have to test. And so what we created was something we called um Picasso tapos. So basically uh scrambled all the initial positions. So the the the the mouth is off to the side, the eyes are on the back of the head, like everything is completely scrambled. And what we find is that they make pretty normal frogs because all of these things will move in novel paths to reach the correct goal and then they stop. Actually, sometimes they go too far and then have to double back and stop. This is another example. You start them off in the wrong position. And they don't just blindly go the same distance. They actually go until they meet their goal. And you know it's a goal. And when I say goal, I don't mean it's a human level like I know what my goal is.

**[13:33]:** That's a kind of metacognition that I'm not claiming here. I'm saying it's in the cybernetics like your thermostat has goals. It's a it's a set point. And now how clever are you to to be able to reach that set point when things change?

**[13:45]:** As an interesting analogy, what's going on at the level of brain plasticity? We tend to think that let's say a dog's brain is pre-wired to drive a dog's body. But one of the examples that I talked about in my book Livewired was this dog who was born without front legs. And so she just walks bipedily and she moves all around and uh gets by that way. Why? Because she needed to get to her, you know, her dog bowl and her water and other dogs and so on. And so she just figured it out. It turns out it's not that hard for a dog to walk on back legs. And the question is, could all dogs walk on their back legs? Presumably, but they don't have the proper motivation uh to do so. But the point is that the the dog's body is very flexible. It meets the goals of the world.

**[14:33]:** Another analogy is the world's best archer, as in he's got the world record for the longest accurate shot in archery, uh, is a guy named Matt Stsman, who happens to have no arms. Uh, and he got interested in archery and figured out how to pull the bow with his legs and so he shoots with his legs and became a great archer that way. Amazing. Amazing. Yeah. Yeah. Yeah. The plasticity is is incredible. And you know the the earlier the earliest example that I know of of this hind leg thing is is slipper goat which was this I think it was in the 40s this guy slipper um published a study of a goat who again born without four legs learned to walk on its hind legs when they dissected the goat they found out that a lot of the uh adjustments that you need for bipeedal locomotion right so things about

**[15:20]:** the hips the back you know the spine all that kind of stuff were all there right as opposed to what you normally think of for for the evolution of modern humans is you know how however many hundreds of thousands of you need for that. And this is what's really interesting about this plasticity is that you can project it into other spaces. So, so as you pointed out, you know, can a can a dog brain run an upright body? Right? Now, look at at individual cells. Can the same genome run a completely different anatomy and set of behaviors? And this is what we've this I mean other people have shown other other examples of this, but for example, in our lab, xenobots, anthrobots, right? These these living constructs that have a completely different body than than what they normally do. They have a different behavioral repertoire, no genetic change, same gene regulatory networks are running a completely different body.

**[16:04]:** For the listenership, could you define anthrobots and xenobots which you've built? Sure. Um, let's start with the zenobots. So, so in the case of xenobots, uh, what um, our team did and this is in collaboration with Josh Bongard's lab at University of Vermont and this is Doug Blackston and my group and uh, Sam Criedman did a lot of the computational work for it. What happens is that when you liberate some epithelial cells from an early frog embryo, normally what they do is they form this like two-dimensional outer covering of an embryo and the you know the the outer skin layer. And they do that because they're induced to do that by the other cells. Well, if you get them away from the other cells, you sort of liberate them. Then you find out what they really want to do on their own. And what they do, they could do many things.

**[16:40]:** They could crawl away from each other. They could die. They could, you know, make a flat layer like cell culture. What they actually do is they form this little ball with with psyia that are on the outside. these little moving hairs and they organize them so that the thing can swim and it starts swimming around. It has all sorts of interesting behaviors. A couple years ago we showed that they do kinematic self-replication which is that if you sprinkle a bunch of loose skin cells in their environment, they will collect them into little balls and guess what? Those become the next generation of zenobots, right? So they can do this weird kinematic replication that as far as we know no other creature does. They express hundreds of genes differently than than they do within the embryo. No genetic change by the way, right? This is this is we're not adding anything.

**[17:18]:** There are no scaffolds, no no synthetic circuits, but they use their their transcriptional affordances differently. They they hundreds of new genes and among other things, it turns out they're sensitive to acoustic vibrations. That's the latest thing that just came out a month ago is that we can because we found they were turning on a bunch of genes related to hearing and we said, is it possible that these things could hear? And so Vipai, my group, um put a speaker under them and showed that yeah, there's actually sounds you can send them that they will that they will respond to. So that's Zenobots. Anthrobots are a similar story because when we first did it, some people said, "Well, you know, they're embryionic cells and the amphibia are plastic. Maybe that's why this is like a frog embryology thing." You know, this is specific to Zenobot. So, I said, "Okay, what's the furthest you can get from an embryionic frog?

**[18:00]:** Well, that would be an adult human." And so we went and we took um tracheal epopithelial cells from from adult human patients and we showed and this is Gizm Kumushka's work a PhD student in my group who developed a protocol whereby again simply by taking the cells out of their normal context you get to release their uh the various possible outcomes that they can do and they make anthrobots. It's a little um round little thing that that that zips around. It has a couple of interesting properties. First of all it it can heal neural wounds. So if you plate a dish of human neurons, you put a big scratch through it with a scalpel, they will when they find the scratch, they settle down. A bunch of them, we call it a superbot cluster, they they they settle down and within about 4 days, if you lift them up, you see that what they did for what they did, meanwhile, is they healed across

**[18:44]:** the they healed across the gap. Okay? Who who would have thought that your trachithelial cells that sit there quietly dealing with, you know, mucus and and and air particles um have the ability to to to to heal neurons. And these guys express about 9,000 genes differently than right. So about almost half the genome they express differently than uh than they do in the body. Um they're by the way younger than the patient than the than the cells that they came from. So so actually that process of becoming an anthropot actually rolls back the epigenetic clock. So um so they're they're they're a bit younger. There's some fascinating you know behaviors and and all of this is run by that standard controller. So, so that's kind of my point is that is that there's an amazing plasticity in the brain and nervous system, but this goes all the way down.

**[19:26]:** This is not just for, you know, fancy fancy brains. So, we think about this as problem solving by the system. And what's interesting, let's just come back for a second to the dog or the goat without fourlims. We generally assume, okay, look, if you're born with the typical structure of the animal, then you just develop in this way. But otherwise, there's a lot of deep problem solving that has to go on. But I know that you think about it as, hey, maybe the system is always problem solving. Maybe it's problem solving no matter what, if you have front legs or not. It's just figuring out what to do to get to the goals. Yeah, I think I think that's right. And and you know, in the last couple years, we've really um emphasized this and and started to develop this idea that, you know, you can think about it as as beginner's mind.

**[20:15]:** Basically the way the the reason that all these incredible plast the plasticities exist, you know, when when we make a Doug Blackston years ago in our lab made a made tadpoles with eyes on their tails and these these guys could see. They were not connected to the brain. They make an optic nerve that connects sometimes to the spinal cord, sometimes to the gut, sometimes nowhere. They can they can see and they can learn visual tasks. Why does that work out of the box? Why don't you need, you know, new rounds of selection, mutation, you know, basically adaptation? All of these things, plasticities work, I think, because it never expected everything to be in the right place to begin with. It has to solve the problem from scratch every single time. And that goes back to the idea that biology is fundamentally dealing with an unreliable medium.

**[20:56]:** Think about the way we build computers today. So, we have um error correcting codes, we have abstraction layers, right? The the reason that we you know, our microchips can't can't scale down easily is because you don't want the the the data interfering with each other, right? when you get to that atomic limit, you know, the the the memory um the bits that are in there are starting to, you know, interact with each other and you don't want that. All all of all of our current computer technologies built around the fidelity of the data and that's because the interpreter of that data is us the user. We don't you know the computer has no issues inter it doesn't need to interpret the data. We interpret the data. So all the computer has to do is is keep the data still. Biology is exactly the opposite. First of all, you have no hope of keeping anything still in biology.

**[21:36]:** You have no idea, never mind your environment, but you're going to mutate as a lineage. You're going to mutate. You can't count on your parts. You can't count on on, you know, knowing how many copies of any protein you're going to have. Things degrade the environment, you know, internal millure, eh, it's plus or minus, you know, whatever homeostasis, but things are always changing. So, I think what biology really cranks on and we've done computational simulations showing how this happens that the minute you have this kind of uh problem-solving material, I call it an agential material because it's not just a computational material, it's actually an agential material. And the minute you have that material, evolution, it it starts to hide information from selection because you're not looking at the genome. You're looking at the the what's going on after you've you've solved the problem using whatever tools the genome has given you.

**[22:17]:** And that means that evolution starts to spend a lot of its time cranking on that problem solving capacity. It spends uh you know less of its time um uh on the on the hardwired mechanisms and more of its time on that creative confabulatory problem solving. So I see all of these things, you know, behavioral memories, genetic memories, meaning, you know, your genome of your lineage, all of these things are basically messages. They're messages from your past self. They're prompts. But at any given moment, it's up to you how you going to interpret them. And and the biological material has eons of of of pressure to learn to tell good stories with whatever it's given, whatever information it's given. And that's morphagenesis and behavior and so on.

**[22:58]:** Yeah. You know, as as an analogy, this is exactly the argument that I made in LiveWired is that the genes do not specify the blueprints for making the brain. Instead, it's just specifying how to build this problem-solving organism. And as you know, you know, children can get a hemispherectomy, which means half of their brain is removed, for example, if they have a an epilepsy, that affects an entire half of the brain. So the surgeon removes half the brain and the kids grow up to be just fine because the other half that remains takes over the missing functions. So So actually I wanted to ask you about about that. I want to see what what your take on it is. So So we reviewed recently I've got this um Karina Kaufman and I reviewed these cases where people people have massive amounts of brain missing like like the part that's left is incredibly small.

**[23:47]:** So most of them of course have very reduced function but but the interesting cases and and there are some amazing cases where it's a massive reduction on both sides right so it's not a hemispherectomy and yet they have normal or in some cases above normal intelligence what do you think is going on in these in these you know fairly unique but still have to be explained cases what's going on there yeah so this is the uh this is the magic of live wiring I think one of the cases that you used in that paper was um people with hydrophilis which means you get this buildup of this pressure in the ventricles these fluid fil spaces in the brain and the whole brain gets squished up against the sides of the skull. So when you look at it on MRI, it looks like it's essentially empty space and the little bit of brain is squished up against there.

**[24:29]:** What that demonstrates is exactly what you and I both love, which is how flexible this material is because, you know, you can't run over half your laptop and expect it to still function, but you can squish this stuff any way you want and it just figures out how to accomplish the goals, in this case, the cognitive and movement goals that that it needs to do. There's one of the the cases in the the medical literature of this guy who was 40 years old and he went to the doctor because he was having a little bit of leg pain and uh the doctor couldn't figure out why the guy's leg was hurting. So he said, "Hey, why don't we just take a brain scan?" And and that's when they discovered that most of the brain scan just looks like uh empty or fluid fil space. But you know, he was married, he had a job, normal IQ.

**[25:16]:** It's it's quite remarkable how different this liveware is from the way that we think about building things in Silicon Valley. And of course you're what what you do which is so remarkable is is look at all the cells the whole system and this massive flexibility the and the collective intelligence of all the pieces and parts all the way up. Tell me if you think this is a good analogy about collective intelligence. I was thinking about I was just trying to think of an analogy and I was thinking about with Wikipedia everybody's doing their little thing depending on their own expertise they put in some things and and nobody who's doing this knows the giant shape of the full Wikipedia it's much too big for any given human um but nonetheless everyone's doing their things and what you get is this collective intelligence out of it and and I was thinking

**[26:07]:** about whether I could stretch this analogy You know, if some part of Wikipedia got cut off, like an axelottle's limb, it would grow back and it would take the right shape again because that knowledge is somehow stored in all the individuals who are writing the stuff. But again, nobody knows what they're doing. Everyone's just contributing where they see a gap. Does that seem like an interesting analogy to you? It is. And what it suggests to me and I'll actually I'll actually talk to Eric Hull about this and see if see if this is a good analysis to do because there are now computational tools uh from from information theories and and Eric Hull was one of the one of the key developers of some of this where you can actually in a specific given circumstance you can actually ask whether the higher level uh has more causal power than

**[26:54]:** the lower level. And so so it's actually an amazing advance because it means that questions that before used to be philosophy and people argued about this for you know probably thousands of years whether right whether reductionism you know was was all you need or whether sometimes you have these higher level things that are causally powerful and now now there's actual math to answer that question. It's it's it's quite amazing. And so so you know there are Python toolkits now to estimate in your given system. Is everything explainable by the lower levels or is there a a higher level that does something that the lower levels don't do? And actually people like Julio Toni and um and Lissa Alpantakis in his group they apply this to all kinds of human patients. So so coma you know locked in asleep anesthetized awake right are you dealing with a pile of neurons or is there a human being in there?

**[27:36]:** You know some kind of collective. Right. So, so, so now what you're making me think is this whole this whole um process of Wikipedia, we could we could in theory apply those tools and and and really empirically ask the question, is there a collective there that's bigger than just the individual processes that go on when people get on and and and edit those uh you know, edit those um those Wikipedia entries? Let's do this experiment. I love it. I want to make sure that we have enough time to talk about diverse intelligences. So one of your interests is in not just looking at human brains and thinking about okay how is this intelligent and so on but saying what are other systems that are intelligent. So let let's dive into that. Tell us about how you think about diverse cognition. Yeah. Uh so so because of some of the things that we've already discussed meaning that problem solving is something that biology has to grapple with at the very beginning you know at the very origin of life and in fact

**[28:30]:** probably long before that you know you can't afford to be like a lelassian demon that's going to track micro states you have to coar grain your environment you have to start uh telling uh telling yourself kind of agential stories about what's going on in other words you have to make models of the environment where you coar grain a whole bunch of different things that are happening and you say okay I'm going to treat all those as one thing and this is danger or this is food or this is a consp specific or this is you know low pH you know for an amoeba or whatever it is. So, so, so that kind of thing, having to having to tell these kind of agential stories that you can then eventually turn on yourself and say, "Wait a minute, I also am an agent that does things."

**[29:04]:** And, and the need to improvise, continuously improvise meaning for the information that you get because you're not told nobody's going to interpret anything for you. You have to interpret your own genome, your own physiological states, your own memories. And so, I'm really interested in the different ways that this gets amplified in evolution. And of course, you know, brains, you know, the familiar brains are one way that that happens, but there are many other ways that that happens. And I want to uh just briefly give you two two quick analogies that uh I think illustrate some of the aspects of what the field of diverse intelligence is about at least the way I see it. First of all, think about the electromagnetic spectrum. So back in the day when we didn't have a proper theory of electromagnetism, we had lightning and static electricity and light and magnets and various things like that.

**[29:46]:** And we thought those were all different things. We thought they were all categories like sharp crisp categories. Nobody thought that light and magnets were the same. And those are are you know we we had categories for all those things and also so that's the first thing we we we thought these were all distinct and because of our of our own evolutionary uh history we only sensitive to a tiny part of that spectrum there were huge examples of this phenomena that we were completely blind to and and then we eventually we ended up with a good theory of electromagnetism we did two things first of all unified it says no these are all actually in a very meaningful way in a they they are all examples of the same underlying phenomena okay so a deep unification so that's great And two, they allowed us to make technology, useful technology that allows us to operate across the spectrum to be able to detect and modulate things that before were completely invisible to us, meaning we didn't think they existed, but now but now we

**[30:35]:** know better. So something like this is what I think is going to happen for cognition. I think we are sensitive to an extremely narrow spectrum among the gigantic space of possible minds. I think uh they are all around us, but we are totally mind blind to to most of them, you know. I think I think that's a good term. mind blindness is is that we just don't recognize these things because we don't have a good theory that explains why the problem solving in of an amoeba of a thermostat of of a you know of an organ of a human of a collection of humans doing Wikipedia whatever why these are all actually on the same spectrum we don't have a good theory yet and and the second thing is we don't have the technology and this is something else that um I think we have a lot to talk about in terms of prosthetics okay cognitive and and physical bodily prosthetics that would allow us to interact with these other beings that are all around us.

**[31:24]:** So let's dive into some examples of diverse intelligence. Sure. So let's just start from the from the beginning and work our way up. So so so brains. Okay. We all know brainy uh uh kinds of animals exist. Then because of what we understand about navigating other biological spaces, we can think about plants and we can think about cells and we can think about tissues and organs which also solve problems. They store memories. They can learn. They can be communicated with. This is what all of the biomedical efforts in my lab are are um pointed at which is learning through in particular a bioelectrical interface. They're all oriented towards communicating our goals to cells and tissues. So for full-on regenerative medicine, it is not going to be sufficient to try to micromanage receptors or or genetic states. We are going to have to get the buyin of the cells, resppecify their goals at a high level and get them to do these complicated things that we can't possibly micromanage.

**[32:16]:** So give us some specific examples. So one of the things uh that we have learned to do is much like neuroscientists read electrical patterns in the brain and they try to decode them. So this is the neural decoding where people um want to uh read the electrophysiology of your brain and say here's your memories, your goals, your preferences and be able to read that out. We've learned to do that and and we developed the first tools to do it in the early 2000s for the rest of the body. So when I say that the early embryo navigates anatomical morphospace to the shape of whatever it's going to be and that it is an active agent that has a memory of where it's going. It has a representation of where it's going. That's a very big claim. You then have to say well what's the mechanism for storing the representation?

**[32:57]:** Where is it? Can you decode it? And can you rewrite it? And so this is what we've done. We've developed tools to to to read the electrical memories of collections of cells. This goes right back to what you said. No individual cell knows what a face is or what an eye is or how many fingers you're supposed to have. But the collective absolutely knows. And we can read this out. Now in a few cases, we can literally see the the in images and videos the memory uh the electrical pattern that is of the future shape that is guiding the cell activity. Moreover, it serves as a kind of cognitive glue that binds all the cells towards one story. One story of what a correct embryo is supposed to look like. This is why you you say it's an embryo and not a pile of cells because they've all committed to the same journey in that space.

**[33:36]:** This actually this this idea is at least as old as as Harold Burr in the 30s. He he without anything other than a good voltmeter he was able to kind of already figure this out. Amazing. And so now we can we can read those memories. We can decode those memories and we can rewrite those memories because if I take a plenarian flatworm and I say oh look this is where it says that you should have two heads if you're injured. We can rewrite that. And this is um Fallon Durant's work when she was a PhD student in my group. we can uh rewrite that electrical pattern. No genetic modification, just just brief application. It only takes about three hours of brief application of ion channel drugs that we've chosen specifically in tune with a computational model of how how you would do that. And we change that pattern.

**[34:18]:** Instead of saying one head, now it says two. Now that becomes a false memory because the worm currently doesn't have to. It has one and it'll sit there perfectly happy. The anatomy does not match the memory. It's a latent memory until you injure the thing. And when you cut it, bang. That's when the cells consult the memory. And memory says, "Build two heads." Well, that's their ground truth. They don't know any different. And so, they will go ahead and they will build this new vision of what a worm is. And it's a memory because it is permanent. If you take two-headed animals and keep cutting them, they will continue regenerating as two-headed even though their genome is a perfectly standard genome. If you were to sequence it, you would have been none the wiser that this thing has two heads. And so this this kind of thing, the ability to to put new goals into the mind of the collective is the the kind of an earliest example of communicating with it because we can we can in some cases

**[35:02]:** we can train it. Another thing we're really working on is to actually ask it questions. That would be really cool because cells have all kinds of problem solving capacities. I would love to be able to actually ask them questions in a in a way and AI is a very powerful tool that we're now using to start to communicate with these things. Yeah. So so that's so that's kind of the first weird kind of mind.

**[35:20]:** Meaning in our body we have I I can't you know I don't think you can count them. I think you know it's not probably not really infinite but but an a very large number of um of different cognitive units inside your body solving their own problems in their own time scales and so on. But you can get weirder than that which is which is this um you know I'll start with a very quick story and this this goes back to a to a sci-fi story from from that I read years ago. Imagine these creatures come from the core of the earth. They live they live in the center of the earth. They're super dense. They come up to the surface. What do they see? Well, they don't see physical objects. As far as they're concerned, everything here is like a thin gas. It's like a plasma.

**[35:55]:** They are so dense they walk right through us the way that we walk through, you know, patterns of pollen in in a garden and we don't even we don't even notice it. And so, one of them is a scientist and he's looking, he says, you know, this gas that we're that we're walking through, I kind of if if you actually look at patterns within the gas, it almost looks like they're doing something. It almost look like they're agential. They like these patterns, you know, they walk around, they have behaviors, they're doing stuff. And and the others say, "Well, that's crazy. We're real physical beings." patterns can't be agents. Pattern, you know, patterns in an excitable medium can't have, you know, their own their own memories and their own goals. And by the way, how long these patterns last? He says, well, they dissipate after about 100 years.

**[36:27]:** He's like, yeah, no, it's not anything, right? So, okay. So, so what that reminds us of is that the distinction between, you know, we we too are patterns, right? We're metabolic patterns that hold ourselves together for some amount of time and then we dissipate. And this distinction between patterns and objects is in the eye of the beholder. And so that leads you to ask what are the things that we think of as mere patterns in an excitable medium that might be agents themselves. And so that's the second that's the next kind of level is can we communi can we can we recognize and communicate with patterns patterns of gene expression patterns of bioelectric state you know this this whole thoughts are thinkers idea from from William James. So you look around, you see these patterns everywhere and you think which one which of these are agential which have what we might call intelligence.

**[37:15]:** Unpack the thoughts or thinkers idea for us. Yeah. Yeah. So so this is uh and and I admit I haven't I haven't looked for the actual reference but but I'm pretty sure I saw this in in James' book where uh what he's pointing out is that look, you have fleeting thoughts. They come and they go, right? to sort of run through your your your memory medium and then they and then they go. Then you have persistent thoughts and these are a little harder to get rid of. They do a little niche construction as you know they they kind of change some of your brain to enable to be to make it easier for them to to persist, right? these these intrusive persistent you know kinds of thoughts. And then you have you go further on the spectrum and you have personality fragments like from a um you know like from a dissociative identity kind of situation and then you keep going and then you have a full

**[37:57]:** coherent human personality and you say okay well that's the thing we're we're kind of used to but but it's on a spectrum and then and then who knows right some people claim there's like a superhuman you know sort of um a larger larger super you know mind and so on. I don't know. So that's the idea. And so there are two there are two ways to to think about any of these situations that were sort of given to us by by the by the touring paradigm. You can say that this the the cells those that's that's your touring machine. That's your that's your machine. That's the real agent. And the patterns that move through it, the information the energy/information patterns that move through it, they're they're just they're just patterns. They're passive data and and and it's the agent that that processes the data, right?

**[38:36]:** we, you know, our brain moves around information and it moves energy and and and our body does the same thing. Okay? Or you can flip the whole thing, which is what we're working on now, which is to say, what if it's the patterns that are the agents and everything that happens to the machine, meaning all the outcomes of of gene expression, of protein movement, of cell behavior, of morphagenesis. What if that's those things are just a scratch pad? It's the it's kind of a stigm that an ant colony will eventually, you know, particles and pherommones and things will move around because the ant colony mind is kind of doing its thing as the ants, you know, send messages to each other. What if the the the anatomy and physiology that we see and and and the and the body of the touring machine is the scratch pad of of the actual agents which are the patterns, you know, working out their

**[39:18]:** their dynamics in the in the physical world. And it actually has um some some real implications just very quickly. For example, in our program on aging, right? So, so we're trying to um understand and address aging. So, imagine the the classic way of thinking about aging from a biomectric standpoint is we know that during embryogenesis there's a biomectric pattern that guides morphagenesis. And so, probably what happens is that those memories become fuzzy in adulthood. And as you age, they just get fuzzier and fuzzier. Their cells have no idea what to do. The memory degrades and the and the agent, the physical body doesn't know what to do anymore. That's the standard approach and that's what, you know, that's one thing we're doing. But you could flip it and you can say what if the agent is actually the pattern that it's trying to um the vocabulary kind of fails us here but it's trying to uh ingress into the physical world through

**[40:04]:** our medium. And maybe what happens as we age is that the cells become less and less able to implement it. They become unresponsive. The machine slows down. Maybe the the the the the mind of the agent of the morphagenetic intelligence is perfectly fine, but the machine doesn't respond. And so those are experimentally distinguishable. And and we're doing those experiments. We actually have some data for this now. And so those are those are just very different. And the way you then would address aging from two different pers from those two different viewpoints is is quite different. So that's what we would love to do is to is to learn to recognize and communicate with other kinds of agents that are not even physical objects as such. They are they are persistent patterns that may have all kinds of um inner you know their own agendas.

**[40:48]:** So let me ask you a couple of rapid fire questions. If there are diverse intelligences everywhere, if we can start understanding these patterns around us as being cognitions of their own, what does this mean for ethics? Yeah, this is this is a huge pro this is an absolutely huge problem. I think that it is foundational to the development of ethics as a mature species to learn to recognize and ethically relate to um minds that are nothing like ours that are basically not on the you know um at least in some cases because you can actually but believe it or not you could get much even even weirder than this than this pattern thing that I'm talking about and so it's certainly above my remit to try and formulate the ethics but what is very clear is that we need to learn to recognize them we need to learn to communicate with them and we need to start thinking about what do we owe other beings that live

**[41:36]:** with us that ha that live in spaces and have goals that are really hard for us to visualize. What are the implications for AI at this moment that we're in? People tend to uh have a very kind of a binary view on this. They will they will either say, "Oh yeah, it it talks like us and therefore it's like a human brain." Or they'll say, "Oh, no. This thing is a machine and therefore it's nothing like us." So I think both of those are are are terrible. And f first of all, because in order to be intelligent and and have meaningful cognition and maybe moral worth, you don't need to be like a human mind. There are many minds that are nothing like a human mind. You don't have to be like humans. And I don't believe at this point, as far as I know, we don't have any AIs that are like a human mind, but that doesn't mean they're not minds.

**[42:17]:** And and the other problem is that there is no such thing as a machine. And if you believe that uh algorithms and um the facts of of physics around you know silicon and copper and the kinds of things we make computers out of if you think that those things tell the entire story of artificial intelligence then you should think that the story of biochemistry is is any everything you need to know about the human mind and that's you know I think that's blatantly false and so for both in both cases I think we have to be extremely open to the idea that we do not understand how different kinds of minds ingress into the through different interfaces. And I realize this is a a weird way of putting it. This is not the standard kind of neuroscience way where intelligence is created by the hardware.

**[42:58]:** I don't actually believe that's true. I think I think consciousness is is is separate and what we see what we provide when we make you know AIs, robots, embryos, biobots, all of this stuff. We make interfaces, different interfaces for it. And we are currently very bad at guessing ahead of time what is going to appear when we make certain kinds of interfaces. And you know I think I think one of the most relevant um uh pieces of our work for this is the the stuff that that we did Taming Jiang and um Adam Goldstein and I wrote this paper on uh unexpected competencies in sorting algorithms like bubble sort. These are things that that computer science students have been studying, you know, in first year CS for, I don't know, 60 years, I guess. And nobody had actually looked at it the way that we had looked at it.

**[43:44]:** And we found this thing has delayed gratification and has these weird little side quests that it goes on that are not in the algorithm at all. In other words, if you just stare at the algorithm, you know, it's six lines of code. It's a deterministic algorithm. There's no magic. There's no new biology to be found. You know exactly what it's doing. And yet, it does things that we do not expect it to do in the algorithm. Not just randomness, not just complexity, not just unpredictability, but things you would recognize as cognitive competencies. And that means that if we don't if if we can't, you know, sometimes people say to me, uh, well, I build I build language models. It's just linear algebra. I know what they're doing. There's nothing. I said, look, we we don't even know what bubble sword is doing. If you can't if you don't know what bubble sword is doing, you sure as hell don't know what these language models are doing.

**[44:24]:** And so, we need to treat all of these things as empirical questions, not philosophical decisions that we can make. And we have to get much better at understanding how new minds ingress even in tiny interfaces like low complexity very um simple kinds of interfaces.

**[44:43]:** That was my interview with Mike Leaven, biologist at Tufts. Every time I talk with Mike, it's hard to look at the world the same way. So we started by asking what is intelligence? But instead of finding a crisp singular answer, we were handed something far more powerful, which is a new lens, a new way of thinking about intelligence, not as a static property that some lucky creatures have and others lack, but as a multi-dimensional space of goal-directed behavior shaped by evolution and context and purpose. And this is of course a reframing of life itself. Because through this lens, intelligence isn't only confined to a cranium. It's not restricted just to animals with brains. Instead, it's something that shows up. Wherever you have systems that are solving problems and adapting to things they didn't expect and correcting errors, wherever there are goals, there may be something in play that is like a mind.

**[45:47]:** And when we look through this lens, the universe becomes alive in strange and beautiful ways. Cells are more than dumb building blocks. We can see them as decision makers. Organs are more than machine units that are chugging along. We can see them as negotiating parties and their own societies. A regenerating flatworm is more than a textbook collection of cells. It's an entity that knows what it's missing and takes action to restore itself. In a sense, it remembers what it used to be and it holds that shape in its future and moves towards it. So, Mike Leven's work suggests that the basic machinery of cognition like memory and problem solving and preferences, this all might emerge way earlier in evolution than we've assumed because cognition might not require neurons. It might not even require consciousness in any familiar sense. What it does require is something more basic that you can achieve with lots of architectures, a capacity to act in service of a goal.

**[46:54]:** And this raises all kinds of great questions. If we accept that intelligence exists in a multi-dimensional space, what else around us counts as intelligent? How about a tree sending resources through its root network? How about a colony of ants adjusting its foraging behavior? How about your immune system adapting to a virus? How about a cluster of engineered cells navigating a maze? And one thing I think is really important here is thinking about what all this means for the future of AI. At the moment, we're only building machines inspired by brains. But I think when we look back in 20 years that will seem quaint and we will be seeing a lot more emulation of the distributed adaptive self-regulating qualities of other more spread out and sometimes more creative biological systems. Could we design machines that do physical things not just like minds but like cell assemblies and bodies?

**[47:55]:** And finally, with everything that we talked about today, what does this all say about who you really are? Because when we're being honest, we are not individuals in the traditional sense. We are collectives. We are billions of cells and trillions of microbes all operating with partial autonomy and some goals. And this vast ball game which is much larger than we can conceive is somehow coordinated into the illusion of a unified self. The story of you is a kind of consensus reality emerging from many smaller parts most of which have no idea you exist. To my mind this is a call for awe. I don't know why we only talk about this stuff occasionally on a podcast. Why aren't airplanes flying around with banners celebrating this kind of stuff? Why aren't we talking about this on CNN instead of local political cycles? Because the lesson from today's episode is that intelligence is probably not rare but common.

**[49:02]:** We always look at it as a strange exception to the rules of nature. But maybe it is the rule. And if this is the right lens to look through, what it means for us is that the world is full of minds. Strange and ancient and in many ways alien minds. Some fast, some slow, some huge, some microscopic. Some we've built ourselves. And most that have been here all along waiting for us to notice. We're just starting to map this territory. And I think one of the lessons from Levven's lab is that the boundary between mind and matter is more porous than we generally assume. And the more we study this, the more we're going to need to update our science textbooks. But more importantly, we're going to need to update our intuitions about what it means to be alive and to be intelligent. We'll need to tune into the fact that the whole world around us might be more alive, more curious, more goal seeking than we thought to imagine.

**[50:01]:** In that light, the story of intelligence isn't a peak that we have reached, but a vast landscape where agency is common and every living system, no matter how small or strange, might be solving problems that we have yet to understand.

**[50:23]:** Go to eagleman.com/mpodcast for more information and to find further reading. Join the weekly discussions on my Substack 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.
