What I'm going to talk about today is this notion of bioelectricity as a bridge between physics and cognition, but not a bridge in the sense of something that connects two very different things, but rather a path that transforms one into the other. And I'm going to try to argue that this is a continuum, and there's a deep symmetry between what we call physics and what we call cognition. I'm going to start with a simple claim. If I were to tell you that with nothing but the power of my mind alone I can depolarize something like almost 30% of my body cells — in other words, I can change the voltage — now you might think I'm making that up, or you might think I'm talking about some yoga practice or some rare mind-matter meditation or something like that. But actually, each of us can do this. This is completely ubiquitous, and it's called voluntary motion. So what happens is that when you wake up in the morning, you have very high-level, very abstract goals. They might be social goals, financial goals, educational goals, whatever. But you have these incredibly abstract goals in some kind of complicated social space. And in order for you to get up out of bed and do those things — in other words, to actually implement any of those goals — ions need to cross your muscle membrane so that you can walk and move around. And so one of the most amazing things about bodies in general is that they are an incredible transduction network. In other words, what they transduce are very abstract mental objects into physiology and eventually into biochemistry specifically. We have to realize that — and I know you can tell other stories about this process, and you can sort of have very convoluted theories of what's actually going on here — but the most obvious and the most useful kind of description of what's happening here is that very abstract goal states can eventually end up changing what it is that chemistry does. And the way that happens — I mean, this is kind of a minor miracle that that works — and how it happens is through this amazing multiscale network of the body.
And what I'm going to tell you is the story of bioelectricity and how it implements this kind of link. And this is a ubiquitous thing in biology — this kind of top-down control where the chemistry is actually being regulated in accordance to some very high-level kinds of targets. Now you might ask, okay, but that sounds very sort of Cartesian interactionism. In other words, exactly where is this interaction between mind and matter supposed to be happening — between cognition and just physics, or if you like, physiology or chemistry or whatever? So we start with some sort of mental construct about some goal you envision in the future that you want to undertake, but eventually we get down to physiology, chemistry, and physics. Where's the interaction happening? And some people have come up with all sorts of clever ideas about it — quantum interfaces and things like this. But for me, and the story that I'm going to tell you today is the following. If you want to know when does it click over from mind into matter, my point is never — it never clicks over, because it's cognition all the way down. There is no transduction from a cognitive medium full of intelligence and consciousness, whatever else you want to talk about, into some kind of dumb matter that only does the things that the mechanism says it should do. I don't think that ever happens. I think the reason this all works is because it's cognition all the way down, every level, and I'm going to talk about that. And so the trick then is to understand how information and how goal states and some other mental states are being communicated across scales. How the goals are being scaled from the grandiose goals of the human in whatever problem space into the other kinds of physiological state space and anatomical state space, and the goals of other body subsystems, and how these memories are remapped across substrates in the body and across problem spaces. It's a story of transformation and communication. I don't believe there are any binary categories here where these things change — I think it's cognition all the way down.
And so in my lab, the goal of my framework is to develop a way to think about very diverse intelligences, because if I'm going to tell you that your body is a set of nested cognitive agents, we better be able to start thinking about some very unusual, very unconventional kinds of minds. And so this is part of the broader field of diverse intelligence. And so what I would like to do is to be able to recognize, communicate with, and ultimately ethically relate to very weird, very unusual kinds of creatures. So of course people study primates and birds and maybe an octopus or a whale, but also we have in biology colonial organisms and swarms. We have now engineered new life forms — you're going to meet two of them in this talk. AIs, whether software or robotic, maybe someday exobiological agents if we find any aliens, and even some very strange things that I won't have time to talk about today that are actually patterns within media. Now, the goal of my framework is that it has to move experimental work forward. In other words, everything I say today is not meant to be just philosophy or linguistics or something like that. Whatever value it may have is in being able to advance research, advance discovery, new capabilities, and so on, specifically cashed out in biomedicine and bioengineering — that's mostly how we test our ideas. Also I should point out that I'm not the first person to try for something like this, of course. So here's Rosen, Bluth, Weiner, and Bigalow showing the various steps that they see of how you get from passive matter all the way up through metacognition and things like that. But I think we have kind of a new and different take on it that is very much like this electromagnetic spectrum, where we are only conscious of a tiny range of it. But with the proper technology and the proper theory — that's really critical, the conceptual apparatus that allows us to understand what all these different things have in common — it will allow us to, I think, overcome what I see as mind blindness that I think we all have because of our own evolutionary history, and be able to recognize some very diverse life forms. And the details are all here. So the way that I like to think about this is an engineering approach, and I'm not saying that this encompasses everything we want to know. There are some other things that are left out, but for the purposes of making progress I like this particular engineering approach that I call the axis of persuadability. What it means is that when we are looking at some particular system, you can't just assume that it does or doesn't have any particular kind of cognitive properties. You have to do experiments. I think this is all an experimental science. And what you're really saying when you claim that, well, this system is just a cell, or it's just a machine, or whatever you're going to say — what you're really saying is: here's a set of tools I plan to bring to that system. And the tools can range from hardware modification to the tools of cybernetics and control theory, but maybe behavioral science, learning and training, and maybe even psychoanalysis and friendship and love and things like that. Whatever set of tools you're going to bring, you're saying this is the set of tools I think is appropriate, and then we can all see how that goes for you. In other words, it's really an empirical experimental science. Cognitive claims are interaction protocol claims, and we have to test them.
Let's look at some of these kinds of behaviors. Some are pretty easy for us to recognize because they happen at the same scale and the same problem space that we operate in. So here's this squirrel. He has a pretty good theory of mind about his owners, so he's going to set up a little accident scene over here, and then he's going to check to make sure that they're watching, what happened to him, and so on. When things like this happen, we find it relatively easy to understand what's going on because it's driven by the brain, and it happens in three-dimensional space. But I will point out that even with brains, things are not very simple. And the traditional story of the mapping between brain real estate and function is actually not very good, because it doesn't cover some very interesting clinical cases. Karina Kaufman and I reviewed a bunch of these cases of human patients with actually minimal brain structure but high or normal cognitive function. And so that's really not understood very well. It's also not understood how information moves in and out of the brain. So here are flatworms. These planaria — you can train them on specific tasks such as place conditioning for these little bumpy circles. You can then cut off their head, the centralized brain is here. The tail sits there having no behaviors for about eight or nine days until they grow back a new head, new brain, and then you see that they remember the original information. So that obviously brings up the question: a, where exactly is the information, but b, how is it imprinted onto the new brain? So the idea is that it has to move in and out of the body, and it has to be able to be imprinted on the new brain as the brain develops. So there are lots of open questions even in brain-based cognition, but what we're going to focus on is biological intelligence outside the brain. What happens in the rest of the body?
So we're going to break some assumptions here that we'll talk about, and in particular I use a specific model system, meaning that we study cells as a collective intelligence having behavior and solving problems in anatomical morphospace. I'll explain what that is in a minute. So we use this system. Now, the thing is that you and I are all collective intelligences because we're made of competent parts. We are agential materials ourselves. This is the kind of thing we're made of. This is a single cell. You can see — well, this is a free-living Lacrymaria, but you get the idea. There's no brain. There's no nervous system. This thing is incredibly competent in all of its local tasks that it needs to do. And this is the sort of thing, and groups of these kinds of things are what we're talking about when we try to do bioengineering or biomedicine or anything like that. And this is a review that Jamie Davies and I wrote on synthetic morphology, with these materials which you cannot engineer in the normal mode of engineering — you have to do something quite different. So the reality is that our body at every scale of organization has agendas and problem-solving competencies, from the molecular networks within our cells up through tissues, and of course the entire organs, the entire organism, and even swarms. Every level is able to solve specific problems in different kinds of spaces. And so we're a stack of these kinds of intelligences, where what I mean by problem spaces is this: because humans are so obsessed with three-dimensional space, and we're really only good at noticing intelligence of medium-sized objects moving at medium speeds through this three-dimensional space, right? So these crows and things like that — we can sort of recognize intelligence here, not always, but sometimes. But there are all these other spaces that biology has been exploring and thriving in long before nerve and muscle appeared. So here's the transcriptional landscape, right? Moving around a 20,000 or so dimensional landscape that cells do — physiological state space. And the thing we're going to talk about is anatomical morphospace, which is just the space of all possible configurations of an organism, all the different possibilities for shapes. And so these are very hard for us to recognize. We need conceptual tools, and we need special equipment to help us understand what are the sensing, decision, action loops that living systems are making in all these weird different spaces. Now my claim is that when they traverse these spaces, they're not simply moving along a predetermined mechanical path. They're navigating it, navigating with different degrees of intelligence. What do I mean by intelligence? Well, I like this definition. It doesn't capture everything, but I think it does a pretty good job. Intelligence is the ability to reach the same goal by different means. It's a very cybernetic definition. It doesn't say you have to have a brain. It doesn't say whether you're evolved or engineered. It doesn't say what space your goals live in. It just means that in some problem space there is a goal state that you will try to achieve, and you have some degree of ingenuity to do that when situations change. So a long time ago we made this hypothesis that morphogenesis — meaning the generation of form from cellular activity — is the behavior of a collective intelligence in anatomical morphospace.
So now we would like to know: what kind of competencies do these cellular swarms deploy? And you have to make some mental twists here, but overall it's very much like neuroscience. In fact, we have a tool that you can paste in an abstract of a neuroscience paper and it will give you the corresponding developmental biology paper. All it does is every time it says neuron it says cell, and every time it says millisecond it says hours, and most everything holds because these domains are very, very symmetric. So I'm going to show you how intelligence is implemented not by groups of neural cells but by groups of other kinds of cells having other kinds of goals. And so I'm going to show you one example each of these components of intelligence. First of all, let's just look at basic homeostasis. So, anatomical homeostasis looks like this. You have an animal such as this axolotl. They're incredibly regenerative. Anywhere this limb gets amputated, the cells immediately recognize that they've been deviated from their correct location in morphospace. They will work very hard. They spend lots of energy. They will grow. They will form the structure, and then they stop. That's the most amazing thing about regeneration: that it knows when to stop. Well, when does it stop? It stops when the correct limb has been completed. So no individual cell knows what a finger is or how many fingers you're supposed to have, but the collective absolutely does. And if you try to deviate it from this normal path, this is what it will do: it will get back there, and then it knows how to stop. So that's a basic homeostatic loop. Reduce the error — it's an error minimization scheme.
Now we'll look at the next thing, which is getting to the same result but through different means. And for this I'll just show you this example which we discovered some years back, which is when these tadpoles — and so here the eyes, the nostrils, the mouth, the brain, the gut — so these tadpoles have to become a frog. In order to do that, they have to rearrange their face. All of this stuff has to move, and it was thought that this was a hardwired process: every organ just moves the way it's supposed to, a fixed amount in the right direction, and then you get your frog. Well, we decided to test that. And this is important: you can't just assume these things. You have to test them. And so we made so-called Picasso frogs. These are scrambled. So you can see the eyes on the back of the head, the mouth is off to the side, the whole thing is kind of like a Mr. Potato Head put together incorrectly. Everything is wrong. And you find that these animals actually make quite normal frogs, because all of these organs don't just move the same path in the same amount. They move through novel paths. In fact, sometimes they go too far and have to come back a little bit. They all rearrange in order to get to the correct outcome. So again, not fixed movement, but actually navigation of the space towards a known target — which leads to the obvious question: how does it know what the correct target is? I'm going to show you in a minute.
The next thing that's really important to think about is that all of these processes are hierarchical. And here you're going to see shades of what I just told you — this kind of incredible transduction of high-level abstract goals down into the chemistry, but you're going to see this in morphospace. So what these guys did in 1956 was to take some tails and graft them surgically to the flank of this amphibian. And what you see is that the tail eventually transforms in place — it transforms into a limb. Now, keep an eye on these cells up here at the tip of the tail. In fact, take the perspective of these cells. What has to happen? Well, they're sitting here locally. Nothing's wrong. There's no damage, there's no injury, there's no reason for us to change. Everything's cool. We're tail tip cells sitting at the end of a tail. But this is not a story about damage; this is a story about top-down control and holistic order, because the system as a whole knows that there's a problem. The problem is the tail doesn't belong in the middle of the animal. You're supposed to have a limb here. The cells don't know anything about that. As far as they're concerned, they're fine. So what has to happen in order for this to transform is that this very high-level abstract goal — of a creature with limbs on its side and not a tail, which these individual cells can't even conceive of, anything that complex — nevertheless, that story has to filter down to the molecular biology and the biochemistry of cell differentiation, migration, proliferation, all that stuff, gene expression, to make this happen, to turn tail tip cells into fingers. So again, you have this incredible top-down control where the local order obeys a plan that is actually developed at much higher scales. And the whole point of being a multiscale organism like this is that you have the ability, and in fact the responsibility, of aligning your parts towards specific large-scale goals in other spaces.
The next thing about this is that there's this notion of hackability. In other words, if you're a system that can maintain goals and flexibly try to meet them, then what are the possibilities of changing those goals without actually changing the hardware? Could we in fact hack this system at the software level, so to speak? And it's only certain kinds of systems that have that capacity. Not everything has that capacity. And for this, I'm going to now tell the story of anatomies and where they come from. So if you look at, let's say, the cross-section of a human torso, and you notice this incredible order — everything is in the right place, the right size relative to the right thing. Amazing. Where is this pattern specified? Well, you could also ask: here's a colony of termites. Look at this amazing structure. Where's that specified? Or the specific structure of the spider web. How about that? And we often tend to think — certainly for this, people will say, well, it's in the genome, it's in the DNA. But we can read genomes now, and we know that that isn't what genomes encode. Genomes don't encode these kind of patterns. Genomes encode proteins. They encode the tiny molecular hardware that every cell gets to have. And this structure is no more directly in the genome than the structure of these constructions. So we have to then understand how does the molecular hardware that is encoded by the genome — how does the molecular hardware figure out what it should build? How do we communicate with it when we want it to repair or regenerate? Let's say something is missing or damaged here, right? And as I'll show you in a minute, what else can we ask it to do? Could we ask it to do something else?
And so it turns out that the way that this molecular hardware figures out what it should be doing has to do with bioelectricity. Bioelectricity is a very interesting set of phenomena that underlie what I call cognitive glue. It's the thing that makes you more than the sum of your neurons. It's a set of policies and mechanisms that bind individual subunits into a larger-scale network that can think about bigger things. And so in the brain — well, we know this happens in the brain, and the way it happens is the hardware consists of a bunch of ion channels which sit in the membrane and control the voltage state of these cells. The voltage state can propagate or not to other cells in the network. This is what that looks like. So this is an amazing video of electrical activity in a living zebrafish brain as this fish is thinking about whatever it is that fish think about. And then there's this idea in neuroscience called neural decoding. The idea is that if we knew what we were doing, we could read this electrical activity, decode it, and then be able to have access to the cognitive content of that brain — that we would know the preferences, the memories, the goals, everything. We could read that out from the electrical activity. That's the idea of neuroscience.
Well, it turns out that actually evolution discovered this kind of process long before brains and muscles came on the scene. Every cell in your body has ion channels. Every cell sets a resting potential. Most of them communicate that resting potential to their neighbors through networks using exactly the same molecular components. In other words, if you ask where did this amazing trick that the brain can do come from, it is well conserved from ancient principles and ancient molecular hardware that was developed around the time of bacterial biofilms. That's when this first got going. So for almost 30 years now, we've been working on creating tools to read and write electrical information in the body, the way that neuroscientists try to do in the brain. And we want to do this kind of non-neural decoding so that when we see this early frog embryo and all of the electrical conversations that these cells are having with each other, we can ask: what are these cells thinking about? Because these cells think about how to move you through three-dimensional space. These cells actually think about your body shape, and how many eyes you should have, and where dorsal-ventral should be, and those kinds of things. So we developed tools first. We developed voltage-sensitive fluorescent dye applications that allow you to see — without having to poke individual cells with electrodes — we can see now where all the different electrical states are. Then we do a lot of computer simulations. So we create software that allows us to understand: given these ion channels, why is it that the pattern regenerates like this, or moves through tissue like that, or can do pattern completion when parts of the pattern are missing? How does it regenerate? What are the memory properties of these networks? We can ask questions like that.
I want to show you a couple of examples — one example, and this is grayscale but again the same idea. You're watching a voltage video of the early frog embryo putting its face together. So here's where the face is going to be eventually. There's a lot going on. Let's just look at this one frame here. So what you see is that at this point, long before the genes turn on to regionalize the face, already it has a subtle scaffold, a pre-pattern of where the anatomy is going to be. So here's where the right eye is going to form. Here's where the placodes are going to be. The mouth is out here. The left eye comes in a little bit after. You can see here they're not at the same time. So we call this the electric face. And this is how the cells... and this shows you that we can read the intent, the future goal states of the cells before they actually do the thing.
Now, bioelectrical patterns are not only the cognitive glue that hold individual cells towards a specific embryonic shape. They actually operate at other scales, between embryos. So here, each one of these is a separate embryo. We poke this middle one here, but you can see this calcium wave — all of them find out within minutes. The rest of them find out what happened to him. And the same thing you can see here: if we poke this guy, then this one finds out, and then that one finds out. And there's just water between them. They're not touching each other. So you can see this bioelectricity is used as a communication from cells to be an embryo, and from embryos to become what we call hyperembryos, which are groups of embryos that actually we showed can solve problems that individual embryos are not very good at solving. So this is how we read the states of this electrical cognitive glue. But what's very important is to be able to rewrite them. You don't want to just keep watching the patterns. You want to be able to do perturbative experiments to see what do these patterns actually mean. And so we don't use any kind of fields or waves — there's no magnets, there's no electromagnetic radiation. What we do is we hack the interface that the cells are normally using to hack each other. And so that is — we can control the connectivity. So through these gap junction connections we can control which cells talk to which other cells. We can control the actual electrical state of cells. So we can turn these channels up and down. And so we do this with optogenetics or pharmacology — there's a lot of tools that we basically appropriate everything from neuroscience. All of these are things that are done in the brain, but we now do it elsewhere. And that's an important point: the techniques actually don't distinguish between neural and non-neural cells. It's university departments and journals and funding bodies that try to make a distinction between them. The science doesn't make that distinction really.
So I want to show you what happens when we do this — what happens when we rewrite information into this electrical network. But first I want to show you how cells control each other. So I'll play this movie momentarily. This is again a video of voltage in cells. Here are cells in two-dimensional culture. They're crawling around. What you're going to see is that this cell right here, here he is, is going to be moving along, minding his own business. Then this group of cells is going to reach out and just touch him. And look — all it takes is that tiny little touch right there to radically change his voltage and ultimately change behavior. So here you go. So it's crawling along, doing its thing. Bang. That's it. Now he's already become like them, and he's going to join this collective and help build whatever it is that they're building. So that kind of message, that kind of electrical connection, which is so good at integrating cells towards a larger purpose — we can hijack that. And the reason you would want to hijack that, for example, is to make novel organs. So I'll show you that here. So again, here's our tadpole from the side. Here's the eye. The other eye is on the other side over there. Here's the mouth. Here's the brain up here. And this thing here is the gut. And what we can do is inject RNA encoding a particular potassium channel. We can inject that into some of the cells that are going to actually be gut, and sure enough, the message that we give them is: make an eye. They will happily make that eye. The eye has the same lens, retina, optic nerve, all the right internal components. And from this you learn several things. First of all, you learn that bioelectricity is instructive. By changing the voltage pattern to be like that eye spot that I showed you in the electric face, you can make a whole new organ appear.
This is not just about damaging things or watching patterns. You can actually call up whole new organs. The second thing you see is it's extremely modular. In other words, we didn't say how to build an eye. We have no idea how to build an eye. Lots of different cell types, tons of genes have to turn on and off. We have no idea how to micromanage any of that. We don't need to. We have a very high-level control parameter that says build an eye. The cells then figure out how to do it. Okay, this is again the hallmark of a cognitive system: multiple layers of top-down control. When I'm talking to you now, I'm not worried about rearranging all the synaptic proteins in your brain to make sure that you can understand what I'm saying. You will do that on your own. All it takes is a kind of very thin, minimal communication medium, and you will take care of arranging your biochemistry to remember these things or not. But we can count on the same thing. The material is very competent to do that. In fact, if we only target a few cells — so in this case only the blue-green cells were actually targeted — there's not enough of them to make an eye. So what do they do? They conscript all these other guys out here to collaborate with them and build the eye. They do a secondary instruction. So we instruct them to build an eye; they instruct others to participate. We didn't tell them to do that. This is something the material already knows. Lots of other collective intelligences do this, like ant colonies will call their buddies when they find something heavy, too heavy for ants to lift. What's interesting is that what goes on in the body is a kind of battle of worldviews. In other words, when we make these ectopic eye spots, they don't all turn into eye, because what's happening here is the blue cells — the cells we injected — are telling the tissue you should be an eye. But the rest of the cells out here, there's a cancer suppression mechanism which says that if your neighbor has some kind of weird voltage, try to normalize them, try to make them like you. And so the other cells are saying no, you should be skin or gut or whatever. And so depending on which view of the future wins, which pattern wins, sometimes you get an eye and sometimes you don't. So for the biomedical side of our agenda, we would really like to be as convincing as possible. When we put certain messages into the body, we want them to be really convincing, so the cells take them up, we get buy-in, we don't have to micromanage it; the cells will go on and do what they need to do.
I want to show you another example of rewriting that kind of anatomical pattern memory, because it's just such a wild thing that this is even possible. Think again of these planaria, these flatworms. They have this property that they're regenerative, so you can cut them into pieces. So typically what happens is, here's an animal, one head, one tail. We take this middle trunk fragment here, and that will regenerate into a perfectly normal one-headed worm. That happens almost 100% of the time. Now, how does this middle fragment — it has two wounds — how does it know how many heads it should have? In fact, these cells down here — well, the cells to the right of that line are going to make a head, because this little tail is going to generate a head. But why don't the cells on the left side of that line make a head? They're direct neighbors. How does it know how many heads it's supposed to have? Well, it turns out there's a bioelectrical pattern in this fragment that says make one head, one tail. The depolarized region here tells you how many heads you're supposed to have. So what we can do is take this animal and artificially, by using drugs that conduct certain ion fluxes, we can change that pattern to be this pattern. It's a little messy, but again, you can see basically it's just two heads. Now when you do this, the anatomy is still one-headed. The molecular biology is still normal one-headed. So the anterior gene is expressed up here; it is not expressed in the tail. Okay, so anatomically it's normal, molecularly, biologically it's normal, but their memory is wrong.
Their memory is wrong. This is a counterfactual memory. This is not a memory of what's happening now, or a readout of what's happening now. This is a represented state of what I will do if I get injured in the future. What you're looking at is a representation stored inside the collective intelligence of the body, not of the animal. The planarian doesn't know anything about this. It's the morphogenetic intelligence of its body that stores this pattern that says, here's what I'm going to do if I should need to regenerate in the future. So if we cut this guy, sure enough, he makes a two-headed animal. So this is not Photoshop or AI; these are real animals. And we can see that a normal body of a planarian can store at least two different representations of what to do if they get injured. Now, I keep calling it a memory, and that's because if you cut these two-headed animals, if you continue to cut them, you will continue to get two-headed animals, even though we haven't touched the genome whatsoever. So this is that reprogrammability I mentioned earlier: the ability to call up an eye or to change the mind of the collective intelligence and say that no, actually real good planaria should have two heads. They're hackable. Like a good memory, it's stable; it persists across rounds of regeneration. In fact, as far as we can tell, forever. But it isn't written in stone, and you can rewrite it.
And here you can see what these two-headed animals are doing. Not only the number of heads, but actually the shape of the head can be made to be like other species. So here's this species with a triangular head shape. We can make them be flat like a pilina, or round like an S. mediterranea, between 100 and 150 million years distance between the real species and these mimics. And not just the shape of the head, but actually the shape of the brain, the distribution of stem cells becomes just like these other animals. No changes in the DNA. That standard stock factory-setting hardware can happily visit other attractors in the anatomical morphospace — attractors that these other species hang out in normally. But you can go beyond that, because the planarian hardware is extremely reprogrammable. You can go beyond other species. You can make crazy things like three-dimensional shapes that aren't even flat. Here's a flatworm genome making something that isn't even flat. Here you can make these tubes or these cylinders, this little spiky thing. By the way, this is something that... and so we're visiting other regions of this latent morphospace.
Biology exploits this. So for example, if you look at plants, here's an acorn, and you might think that well, this oak leaf is what the acorn genome knows how to do. That is what it has been selected to do, like an instinct. It has a fixed pattern, and that's what it knows how to do. Except that along comes this non-human bio-engineer. It's a little wasp that prompts the cells with — we don't know with what exactly, probably some kind of chemicals — that prompts them to build its kids this amazing little house. It's these hedgehog galls and other kinds of galls. Like who would have even known that plant cells that make this nice flat green leafy thing reliably are even capable of making something round and spiky and red and yellow? Like we'd have no idea. So the ability — that morphospace of the material is actually not obvious at all. And so everything that I've showed you now — the planarian axial polarity, and the eye — I mean, we can make brains and hearts and many other things, and these wasps can make... like all of this stuff is incredibly plastic. Not at the level of DNA, not at the level of changing the hardware, but it's reprogrammable.
Okay, so this is what I was just showing you examples of. Now I'm going to show you a couple of other quick things. Another thing that you might want as a component of intelligence is learning. So actually it turns out that even below the cellular level, so just the molecular networks themselves, are already capable of six different kinds of learning. And you can read about that here, including Pavlovian conditioning. It's amazing — just the molecular networks themselves. And for that we're building devices to try to take advantage of this for things like drug conditioning and lots of medical applications. So there's already learning baked into the substrate, even in the molecular networks. But there's also memory at the larger scale. This is something called trophic memory in deer antlers. If a deer gets an injury at a particular point in one year, it makes a callus and it heals, and then this whole thing drops off, and then next year they build a new structure, but there will be an ectopic tine at the same location. And this goes on for five or six years, and then eventually it disappears and goes back to normal. Just think about what that would mean, and think about trying to write down a molecular pathway that explains how the body knows where the injury occurred, stores that information somewhere else because this whole thing's going to leave, it's going to fall off, these cells are gone, and then next year when the bone is growing, say, oh, by the way, you have to take an extra left turn here because last year that's where the damage was. So there are these incredible... and here you can see us scanning some of these antler experiments from Bubenik. So you can see there's at different scales this amazing memory capacity.
But it's not just about learning things that you've already been exposed to. There's also some creative problem solving. And what do I mean by that? Well, I'll show you a specific example. Again, back to the planaria: we take these worms and we throw them in a solution of barium chloride. Barium is a non-specific potassium channel blocker. What it does is it basically prevents cells from exchanging potassium with the outside world. Cells get very unhappy, especially the neurons in the head get very unhappy, and the whole head explodes. So overnight their heads sort of blow up. But if you leave them in fresh barium, what you will see is that they eventually build a new head, and the new head doesn't care about the barium at all. It's totally barium-adapted. We said that is amazing. How could it possibly do that? So we asked a very kind of naive initial analysis. We simply looked at all the genes that were expressed in a normal head, all the genes that were expressed in the barium-adapted head, and we subtracted. We said what's different? There's only a small number of genes, less than a dozen genes that are different in these cases. And so think about what this means. Planaria don't see barium in the wild. This is not something they get evolutionarily selected for, the ability to deal with barium. The cells are stressed by an exposure to something they've never seen before. They have to find an action out of a space of roughly 20,000 vectors. So the gene space — so they're stressed by a physiological stressor, the actions they can take are to turn genes on and off. You can think about it as like you're sitting in this nuclear reactor control room with 20,000 knobs, things melting down. You've got to do something. What you don't have time to do is to start flipping knobs randomly to see what'll happen, okay? Because you don't know how many, and you'll probably kill everything doing those experiments, and yet in a very short time frame this thing figures out how to do it. It's amazing. How does it find the right set of genes to solve a physiological stressor?
Here's another example. This is one of my favorite examples of problem solving in anatomical space. The previous one was in transcriptional and physiological space. Imagine taking a section through a kidney tubule. What you would see is 8 to 10 cells making a nice circular tubule like this. Now one trick you can do is magnify the amount of DNA that the embryos have. And when you do that, the cells adjust and they get bigger. So the nuclei have more DNA, and so the cells get bigger, but the animal stays the same size. Well, how can that be? If you take a cross-section, you see it's because there are now fewer but larger cells that are doing the exact same job. If you really make the cells gigantic, one single cell will wrap around itself and still give you the same structure. So think what's happening here. First of all, this is top-down control. In other words, again, in the service of a large-scale goal, this tubule, you have to call up different molecular mechanisms to get the job done. In this case, cell-to-cell communication; in this case, cytoskeletal bending — a single cell has to bend around itself. Different molecular mechanisms. It looks like many IQ tests where they give you certain objects and they say, okay, solve this problem with the use of the objects that you have. And again, the ability to do that in novel circumstances. Think about what this means when you're an amphibian coming into the world, or in fact any embryo coming into the world. Never mind that the environment is uncertain; your own materials can't be trusted. You have no idea how many copies of the genome you're going to have. You don't know how big your cells are going to be or how many cells you're going to have. You don't know any of that. You have to improvise on the spot. You have to be able to do the best you can, tell the best story that you can about what you are and what kinds of shapes you're going to be, using the tools you have with the circumstances provided. It's an amazing example of flexible problem solving in this space.
But all of the examples that I've shown you up until now have been evolutionarily selected goal states — shapes that have been around during evolution. And I've shown you some examples of systems successfully meeting those goals. Obviously there are many examples where they fail to do that. You can prevent them from doing it; it's called birth defects. And there are many ways to do that as well. No intelligence is perfect, but they do an amazing job at reaching their normal goals. But what happens when you can't reach your goals? Now you might think that, well, okay, eventually they'll just give up and do nothing. But even there, biology surprises us. And I want to show you two examples of novel beings that have not existed before. The first we call xenobots. So what we've done here is to actually — let me pause this. What we've done here is to liberate some epithelial cells from a blastula-stage frog embryo. We put them in a separate little dish. They could have died. They could have crawled away from each other. They could have formed a nice flat monolayer like cell culture. Instead, what they do is they come together and they make this little structure. How do they come together? Well, I'll play this. Each one of these circles here is a separate cell. Okay, so this is a clump of cells. They don't all look like a cute little horse, but this one did. So there you go. So it kind of moves as a unit. We still don't understand how that happens. You can see it kind of sniffing around this other region here. That little flash is a calcium signal that occurred between these two regions. So they all sort of coalesce, and when they coalesce they make this amazing thing. So these are self-motile. They swim. They use little cilia on their surface to push against the water in a coherent manner. And they swim. They can go in circles. They can patrol back and forth like this. They can have collective behaviors and interact together. You can make weird shapes out of them, like this swimming donut here. You can see them traversing different types of environments. So here it is: it takes the corner of this maze without bumping into anything, and then it decides to turn around and go back where it came from, for some reason that we don't understand — completely spontaneous behaviors.
One of the most amazing things it does is kinematic replication. So if you provide them with loose epithelial cells — so the white stuff out here is just loose cells — what they do is they, both collectively and individually, push them into little piles. Then they polish the little piles into little balls, and then the little balls mature and they become the next generation of xenobots. And guess what they do? They run around and do exactly the same thing, and they make the next generation, which makes the next generation. Right? Okay. So we call this kinematic replication. As far as we know, no other creature on Earth reproduces this way. It's a very strange thing. We didn't program them to do this. This is completely spontaneous. No one saw this coming. If you were to ask prior to, what, 2018, I think we showed this — if you were to ask what would a bunch of epithelial cells do if you take them off of a frog embryo... this is not something that anybody had in their mind at the time. One other amazing thing is you can look at a novel creature like this and you can ask what kind of gene expression do they have? We know what these cells express within the embryo; people have studied that. What kind of gene expression underlies this novel kind of being? And what we've done is to find that there are hundreds of new genes that are differentially expressed in xenobots that these cells don't express in the body. Remember, we haven't touched their genome. We haven't put in any synthetic biology circuits. There is no genomic editing. There are no scaffolds, no weird nanomaterials. This is completely native to these cells. And we found that one cluster — there's a whole bunch of genes that are upregulated — that when you do a pathway analysis, you see that they're related to sensory perception of sound. We thought, could they possibly now have hearing? So we put a speaker under the dish, we vibrated the water, and we find out that yes, in fact, during the time that you're sending them these audio signals, their behavior changes. Okay? And so they have acquired a new capacity that otherwise — normal embryos don't do this. This is just xenobots that do this. So they do all these new and amazing things.
Now you might say, okay, well these are amphibian cells and they're embryonic cells. So you know, amphibians are pretty plastic. Embryos, we know, are plastic. Maybe this is just a fluke. And so I would ask you to think about what would your cells do if we liberated them from your influence — the kind of influence I talked about on the very first slide. If the rest of the body was not telling these cells what to do, what would they be doing? What are their native behaviors? Well, I would introduce you to another creature we call anthrobots. So if you look at this thing, it kind of looks like something you might have gotten off the bottom of a pond somewhere. It looks very primitive. You might guess that it would have a primitive creature's genome. If you sequence it, you will find 100% Homo sapiens. Why? Because these are self-assembled from cells donated by adult donors who have tracheal or lung biopsies. So this is a lung epithelium that self-assembles into this motile little creature that runs around. And what do they know how to do? Well, it turns out that if you put them on a bed of human neurons where you put a big scratch through — so we took a scalpel, put a big scratch through here — they'll settle in that wound, and in fact they'll settle into a collection. So there's like a cluster of anthrobots here. And what they start to do is they start to knit across the gap. They induce the neurons to repair. It's about four days later; we've lifted off the bots. And you can see here they start connecting the two edges. Who would have thought? That your tracheal epithelial cells, which sit quietly in your airway for decades dealing with mucus and air particles, that they have the ability to form this amazing self-motile little creature. They have four distinct swimming behaviors. Not twelve, not one, exactly four. Why? No one knows. They have 9,000 differentially expressed genes. So about half the genome is radically different. They're actually younger than the cells they come from, which is kind of amazing. There's kind of an age reversal thing, which is a whole other project in our lab about aging.
Just this remarkable plasticity, and so now it's important to just take stock here. So both for xenobots and anthrobots we have a new shape, new behaviors, new gene expressions, new physiological states. Where do those come from? We know when you paid the computational price to compute a human or a frog, it was in the eons that selection has been working on that genome, bashing it against the environment. There haven't been any xenobots. There haven't been any anthrobots. There's never been any selection to be a good anthrobot, to be able to do neural healing or to do kinematic self-replication. Where do those things come from? They don't come from a history of selection. We have to think differently. There's some kind of latent space of possibilities. And I think the more straightforward, more obvious thing is that when we make cyborgs or hybrots or any of these weird things, what we're actually making is exploration vehicles to explore a latent space of possibilities, and that all the things that we see in the natural world are just tiny points in that space, but there's lots of other possibilities, and we need to explore that space.
There's a kind of a bigger idea which is sort of much more controversial and weird, and you can see we're actually having an asynchronous symposium on this. So if you go to this page you will see all the different talks that people are giving, and lots of discussion about this idea that actually what you're exploring when you're exploring this latent space of possibilities with bodies that have never had the benefit of selection for these specific patterns, is that you're exploring the same latent space that mathematicians explore when they discover facts of mathematics — when they discover patterns that are not dependent on the history of selection nor of any facts of physics. These are bare truths of mathematics that are separate and distinct, or at least the platonist mathematicians think so. And that maybe that space also contains higher-agency forms, including shapes of behavior, truths about physiological attractors and things like that, that behavioral scientists would recognize as kinds of minds — that is, complex behavioral propensities that are recognizable as kinds of minds — and that maybe this latent space is really important to understand both aspects of cognitive science and the embodiment in which these patterns navigate the physical world.
So I'm just going to start to wrap up and tell you a couple of things. First of all, that all the things that I've said today have massive biomedical implications. Basically all the interventions that we're used to are this kind of bottom-up thing, you know, focused on the hardware. Everything's focused on the molecular hardware. I haven't even talked about the stuff we're doing on cellular training and so on, and the use of anthrobots in the body as patient-specific biobots that can do healing with your own cells, so you don't need immunosuppression and all that. But all of these are top-down sorts of approaches, and I think that in the future bioelectricity is — it may not be the only, but it certainly is a uniquely tractable modality by which to access the intelligence of the body, the other layers. It is the cognitive glue. I mean, I think evolution loves bioelectricity as an effective cognitive glue. And so you can see all of that in these papers. But I think that future medicine is going to look a lot more like a kind of semantic psychiatry than it is like chemistry. I think that's where we're going. So in terms of regenerative medicine, we have stories about fixing birth defects and inducing regeneration of appendages and reprogramming cancer and things like that. But in addition to all of that, we have to ask another question. As we stand at the center of these continua, both on the evolutionary scale and on the developmental scale — we all started life as a single cell, we scaled up to this sort of adult modern human — there's another continuum here, which is the biological and technological changes that we are all undergoing. There are already plenty of people walking around with all kinds of different implants and different modifications, and even just a normal human today with their eyeglasses and their toothbrush and sunscreen and things like that is just a total cyborg compared to our hominid ancestors, the way that we're changing our bodies.
And so we have to ask ourselves when these other beings — what are their cognitive and behavioral properties going to be like? Because we're really not very good at guessing. And this kind of thing where people talk about proof of humanity certificates, where somebody's going to try to figure out whether more than 51% of your brain has been replaced and whether you're a real human or not — none of that's going to do. We're really going to have to bite the bullet on the fact that when Darwin said endless forms most beautiful and all the incredible variety of life on Earth, all of it is one tiny corner. It's an N of one example of this incredibly large space of possible embodied minds. And because biology is so interoperable, because it solves problems on the fly and does not have strict expectations about what's going to happen, pretty much any combination of evolved material, engineered material, and software is some kind of embodiment that is going to pick up ingressing patterns from this latent space of forms and behaviors. That is going to be some kind of being. And I think it's not that long from now that we are all going to be living with some very diverse embodied minds, and that's even without mentioning traditional AI and things like that, although those things are certainly coming too. And so I think we have to begin to understand that getting a handle on the diversity of possible minds, and these questions of where do their goal states come from, and being able to predict that, and so on. This is not just about bioengineering and regenerative medicine. This is also about a future for humanity where we can enter into an ethical symbiosis with beings that are not like us. You know, humans are not very good at that. But I think we're going to have to get better at it. And this new science of diverse intelligence, I think, is critical for us. It's an existential barrier, I think, for our species.
So that's it. I'll stop here. I just want to thank the people who did the work that I showed you today. So lots of amazing postdocs and PhD students and undergraduates and technical support. We have great collaborators. Very grateful for funding from a number of agencies that have supported us through the years. I have to do a disclosure: there are three companies that have licensed intellectual property from the discoveries that I've shown you today and that have supported some of this work. And as always, most of the credit goes to the model systems, because they do all the hard work. So yeah, that's it. I'll stop here.
