Forms of life, forms of mind

Here is a talk I gave to an audience working on data science approaches to biomedical discoveries. They were into computer science, and concepts related to software in biology and how to infer interventions and control mechanisms. I tried to slant my talk in that direction by pointing out relevant aspects of our data and where I think the field is going.  The talk:

And here’s a transcript of the audio; it’s done by AI and not human-polished, so it’s not always exactly what I was saying:

Thank you. My background originally was computer science, and that probably colors the way I think about biology. But, today I run a group of, roughly 30 to 32 people, working on some really interesting problems that are related to, basal cognition, to computation, but also to biomedicine. And, if you’re interested in any of the details, you can go here for all the, the primary data, the, the software, everything else, this is our official site for things that, I feel strongly about. And then here are some, some things that you can find that are, ruminations that I’m not really sure about yet.

so, the main points I want to give you right up, I’m going to talk about the fact that, our bodies are a kind of, they’re built on a kind of multi-scale competency architecture that has problem solving capacity at every level from the molecular networks that build up your cells to, actually swarms and collectives.

we are composed of nested agents that collect and interpret information. And by the way, I was asked specifically to talk about, these things from a perspective of, information processing, data and software and things like that. So that’s, that, that will be the slant for today. I’m going to argue that, definitive regenerative medicine is going to require really understanding and exploiting the collective intelligence of groups of cells, not just the mechanisms, not just the molecular networks, but really their competencies. And, what we’re going to be able to do is to communicate anatomical goals in various spaces, including anatomical space, a.k.a. morphospace, and so on.

The emphasis on this approach is, away from the genetic hardware, which is where a lot of, modern molecular medicine is focused. We’re going to talk about, mostly about the software of life. I’m going to show you that endogenous bioelectrical networks in all tissues, not just the brain, are a highly tractable interface for that kind of top-down control of system level, growth and form of the body.

And we’re, tools are now coming online that, will allow you to read and write the pattern memories within tissues. literally, I’m going to argue that, cells are a collective intelligence, and that we can now write and read memories in and out of this, proto-cognitive medium which is electricity. I’m going to, I’m going to show you some examples of applications and birth defects, regenerative repair and cancer. And, these are on their way, towards, biomedicine. and so I think we’re entering now an area where some of these, some of these ideas are really hitting, biotech and, and so on.

And, really this whole field is going to be cracked open when we’re able to decode the information processing by higher level subagents within the body. That’s, that’s what I think is going to be the key to all this. So the first thing I want toshow you is a few examples of interesting biology that we don’t really think about in medicine, but I think that tell us is something very important about how bodies work. this is, this is a caterpillar. This animal lives in a, two dimensional world. It crawls around, leaves, it, that, that’s the food. It eats the leaves and has a particular brain suitable for running the sort of soft

Bodied robot that can crawl around and, and live in that environment. It needs to turn into this, this, this, this animal lives in a three-dimensional world. It doesn’t care about leaves anymore. It likes nectar, and it has a completely different brain suitable for that type of lifestyle. What’s amazing is that during the process of, metamorphosis from here to there, the brain is mostly taken apart, and most of the cells are killed off. The connections are broken and, and reassembled, refactored into a completely new brain. But what we know is that if you train the caterpillar to, for specific tasks, the moth or butterfly, that results still remembers them.

So you have here, two orthogonal, but connected, sets of information. One is, anatomical, so you have to turn this into that, but also behavioral, because the kinds of memories that are stored here have to somehow survive this incredible process that that example is even, more amplified in, these parian flatworms.

So, here’s a flatworm. One of the most amazing things about them is that they are basically, they, they, they are immortal. So there is no evidence of aging in the asexual form of these animals. They regenerate. So you can cut them into pieces. The record is, I think, 275 pieces or something like that, and every piece regrows. And so what you can do is you can train the animal, in this case, for place conditioning, to look for food on these little bumpy, kind of surfaces. And then you cut off the head and the centralized brain, the tail will sit there for about a week doing nothing. it will grow a new head.

And at that point, behavior resumes, and they start looking food again. Guess where On this same, region that they learn. So again, you have this, you have this notion that there is learned information.

It is somewhere, presumably in the body. It is imprinted on the new brain as, the brain develops. And this has massive implications. not only for basic questions of where memory is stored, what is memory and how does it move through the body, but, cellular therapeutics, someday when a patient of, seven decades of, of memories and personality and so on, has a bunch of, cells replaced in their brain for therapeutic purposes with naive, naive new cells, what’s going to happen to that, to that individual, and what’s it like to be that individual?

In fact, what’s it like to be a caterpillar slowly changing into a butterfly? And so on? So, some, some fascinating questions, but you’re starting to see the, the plasticity here and the importance of, asking about information flow in the system.

And, and the last example I want to show at the beginning is this. This is a tadpole of the frogs Xenopus lavis, here at the nostrils, the mouth, the gut. And here’s the spinal cord and the tail. And what you’ll notice is that it’s missing the normal two eyes that it would have right up here. But what we’ve done in addition to preventing these eyes from forming, is we’ve put an eye on their tail. And so then we can, we’ve, we’ve built this machine, which is a behavior testing, a training and testing device. And we find out that these animals can see quite well out of that eye. That eye makes one optic nerve.

It, connects sometimes to nothing, sometimes to the spinal cord here. It doesn’t go to the brain. And yet these animals, can see quite well, which we know, because we can train them on visual cues.

And so, this this has implications for sensory motor augmentation. basically increased human potential via this kind of, plasticity. but what’s also remarkable is that this happens in, in one generation. There’s no, there’s no need for lengthy, evolutionary selection to learn to use this different sensory motor architecture. The information that comes from this weird itchy patch of tissue on their back is recognized immediately as, as visual information processed appropriately by the brain. And they can, they can fold it into their behavioral repertoire.

So, the body and the brain are, incredibly plastic with respect to, how they process information, where that information comes from and what they do with it. That’s, that’s going to be essential here. So the future, the future goal here of, biomedicine as, as I see it, is, the end game is, is something I call the anatomical compiler.

The idea is that someday, you should be able to sit in front of a computer and draw like this, schematically the plant or animal, or organ or synthetic biobot, whatever, that you want. not at the level of molecular pathways. You should not, you shouldn’t need to know anything about molecular pathways. You should be able to, as the use, as the end user, you should be able to draw whatever, structure you want. And if we, if we had this thing, the compiler would convert that description into a set of stimuli that would have to be given to individual cells to get them to build exactly this.

Now, why do we need something like this? Well, because if, if you think about it, most problems in biomedicine boil down to the control of this very problem. So, birth defects, traumatic injury, cancer, aging, degenerative disease, all of this would go away if we had the ability to tell a group of cells what we want them to build.

Now, very importantly, this is not a three D printer. This is not about micromanaging the position of cells. This is a communications device. It’s a translator for translating your goals, let’s say your biomedical goals into the set points of the cellular collective. Now, now we’re very far away from this. We don’t have anything remotely like this anatomical compiler. And you might ask why not? genetics and biochemistry and molecular biology have been going gangbusters for decades. Why? now we have big data. We have omics, we have all this stuff. Why don’t we have this? I want to just show you a very simple example of the kind of problems we can’t solve.

So, here’s a baby axolotl These salamanders, baby axolotls have little legs, right here, little four legs. This is a tadpole of the frog, Xenopus laevis, and as I showed you, they do not have legs at that stage.

In my group, we make something known as a frogolotl. So frogolotl is just a mashup of, frog and, and axolotl cells. They get along just fine. You, you get a viable, a viable embryo out of it. Now, you could ask a simple question. Well, we have the axolotl genome. It’s been sequenced. We know all the genetic information. We have the frog genome that’s been sequenced using that information. Could you tell me if a frogolotl is going to have legs or not? And the answer is no. We can’t tell. Because while we are very good at molecular information like this, which gene controls which other gene, which proteins interact with each other, that kind of thing, we are really a very long way away from understanding how collective decisions are being made by cells.

So how do the rules of the proteins that are current, directed by our genomes, which would make legs in one species, would not make legs and another species? What are these cells going to do in a, in a chimeric context? We do not yet understand the algorithms. We do not understand the software of life yet, which is just beginning. And so, I would argue that, biomedicine is, is currently basically where computer science was in the forties and fifties. This is, this is what programming looked like back then. she’s, programming this machine by interacting with the hardware.

Literally, that was your only option you had to physically rewired. And this is how molecular medicine sees it today. But basically, all of the excitement is around genomic editing, protein engineering, crispr, pathway rewiring, those kinds of things. It’s all about the hardware.

But as we know from the information technology revolution, th th this is just, this is just the beginning of the journey, because there are amazing things to be done at the software level. So in order to understand what that is in biology, let’s go, let’s go all the way back. Where did we come from? Well, each one of us started life like this. You were a quiescent, unfertilized, cyte, a little blob of chemistry that was, well described by the laws of chemistry and physics. people would say you were just physics. I, I don’t like that phrase at all, but, but that’s something that people say.

and then eventually through this amazing process of development, if, if there is any magic, this is it right here. This, this is, this is this remarkable process of embryonic development where, we end up as something like this, or even something like this.

And, we will make, statements about, not, not being machines and, and having an internal, first person perspective and, and cognition and so on. But this, this process is extremely slow and gradual. There is no magic lightning bolt that shows up at some particular time and says, ah, you used to be chemistry and physics, but now you’re mind, okay, that that doesn’t happen. So, so we need to understand how this transformation occurs, right? Where, where does the decision making, the memory, the intelligence, the problem solving, where does that come from during this slow process of transformation?

and, and, and you might take solace in thinking that, well, at least we’re a kind of unified intelligence. Yeah, people talk about collective intelligence and ants and, and beehives and bird flocks and so on. But, but we’re not like that, that, that those things may or may not be a collective intelligence, but we, we are definitely a centralized intelligence, right?

Well, unfortunately, even that’s not true because, for example, so here’s Descartes. Descartes really liked the pineal gland because there was only one of them in the brain. And he felt that the unified human experience should be attached in some way to a single non duplicated region of the brain, right? Something that, that is, is only one of, well, if Descartes had access to good microscopes, you would’ve looked inside the pineal and said, whoa, there isn’t one of anything in here. It’s actually made up of, of a huge number of cells. And now we know that each inside, each one of those cells is all of this stuff.

Okay? So, e every intelligence is a collective intelligence. There is no such thing as a centralized, unified diamond of intelligence that isn’t made up of other parts. So we need to understand how our information processing relates to that of our parts.

And we are made of, a very unusual kind of, matter called a agentialmaterials. This is, this is a single cell. So this is a lac room area. This happens to be a free living organism, but all of the cells in our bodies used to be free living organisms at one point. It has no brain, it has no nervous system. but you can see here, it’s amazing, competency in solving all of its local problems. If physiological needs, metabolic needs, behavioral anatomical, it’s very good at it. And you can read here, Jamie Davies and I, our thoughts on, the engineering that, needs to be done when you’re dealing with genial materials.

So the thing about our bodies is that we are made of a nested structure, not just, not, not just in terms of, complexity, but really the fact that at each level, all the way up from the molecular networks and, and all the way up to tissues, organs and so on, is a kind of a problem solving agent.

They, it, it really is a multi-scale competency architecture, because each of these layers has competencies to solve specific problems in their own space. And they’re all, and there are multiple different spaces that we’ll, that we’ll talk about. But in order to, in order to have this kind of architecture, you need something, you need something important. And let’s just look at one example. So, so here’s a rat, and we all know you can train a rat to press a lever and get a delicious pellet, delicious reward. But what’s interesting to think about here is that in, in that associative memory of the lever and the reward, no individual cell here has had both experiences.

That is the cell at the bottom of the foot interacted with the lever. The, the cells in the gut got the reward, but there is no single cell that had both experiences.

So who, who owns the associative learning here? The fact that these things are associated? Who owns that? So, so that’s why we come up with this concept of a rat, which is a kind of collective made of individual cells that can do some interesting things. It’s an emergent agent. it can do credit assignment, it can figure out how the various actions of its parts lead to facts that are, only facts on the scale of the whole agent, not on the scale of its parts. And, and that requires a kind of cognitive glue. It requires some mechanism to bind these individual cells together into a system that can do that sort of, of, highly plastic, credit assignment learning and all of that.

So, the intelligence that, that these, that these collectives exhibit come in, in, different, different spaces.

So, so we know, how to recognize, we’re pretty good at recognizing intelligence of, of medium sized objects at medium speeds in three dimensional space. So primates, birds, maybe an octopus, maybe a whale. We can sort of recognize intelligence there. But, because all of our sense organs are really, just, tuned, they’re, they’re all pointing outwards. They’re tuned at a very specific size, scale range. We’re really not very good at understanding intelligence in other spaces. And this is going to be very important for, for, for biomedicine because these collectives that are bound together, are able to traverse the space of gene expression.

They’re able to traverse physiological state spaces, and the one we’ll spend most time talking about here, anatomical morphospace. In traversing these spaces, they have to make decisions. They have to remember where they’ve been. They have some idea of where they’re going, and they can solve problems and overcome barriers.

And this is really critical. In fact, this is not just, whole whole animals that can do this, but even the molecular networks. So you can see, you can see, the, the analysis in these two papers. Even the molecular networks inside of each cell are already learning agents. They, they can do, five or six different kinds of memory, including Pavlovian conditioning. So if you treat these networks as if they were an animal, and you trigger certain nodes, and you look for changes in future behavior of these output nodes or response nodes based on their history, you will see things like habitation, sensitization, associative learning, and that’s baked in.

That property is already there at the very beginning, of this kind of, complexity scale. So, so when I say intelligence, what do I mean? I like William James’ definition, which is the ability to reach the same goal by different means.

it’s, it’s nice because it’s a cybernetic definition. It doesn’t talk about brains. It doesn’t talk about what problem, space, those goals exist in. it doesn’t say, whether you were, engineered or evolved. it’s really very fundamental. It’s about competency to reach goals in some problem space. So let’s now go to, to talking about what kind of collective intelligence do we think we can get in cellular swarms? And then, and then I’ll show you why it matters. So let’s, let’s ask, first of all, where where does our anatomy come from? So here’s a cross section through a normal human torso.

And as you look at this amazingly complex structure, all the organs are in the right place next to the right thing oriented. They’re in the right direction. You might ask, where does this information come from? Where’s this encoded? And you might be tempted to say, as a lot of people say, well, it’s in the DNA, it’s in the genome.

But of course, we can read genomes now, and we know what’s in the genome. And it isn’t any of this. What’s in the genome is protein structure. And, the, the, the sequences of proteins, which are the, the tiniest, sort of molecular hardware that cells get, get to have, that hardware has to now through, through a process, which we call physiology and development, are going to build something like this, which is actually not directly described in that genome at all anymore than the structure of the termite colony, or the spider web is in the genome of the termite or the spider.

And so, so we need to understand, where does this pattern actually come from? How do the cells know what to make? How do they know when to stop? And, won’t have time to talk about this. We also make some synthetic organisms, in our lab, but they won’t have time to talk about it.

But here, let’s, let’s focus on this, this, this question of how, how would we convince these cells to rebuild something if it were damaged or missing? And, I want to show you some of the competencies of this process, because it’s not simply that, that this is a, it’s, it’s not a kind of, it’s not just a kind of immersion complexity that, well, there’s certain chemical rules, and eventually you get something like this. There’s a lot more to it. And I want to show you a couple of, a couple of amazing examples. this is, this is perhaps one of my, one of my favorites. So, so what you’re seeing here, and this was discovered in the forties, this is a cross-section through a kidney, tubule in the newt.

And, and you’ve got the eight to 10 cells out here, and then there’s an empty lumen in the middle, which is where the, the fluids go.

And, so this is what it normally looks like. Now, one thing you can do with these newts is that you can, force the early, the early cell divisions to duplicate their chromosomes, to duplicate their DNA. If you do that, the cells get bigger because the cells adjust their size to the amount of DNA that they have, so the cells get bigger, but remarkably, the, the new stays the same size, which means that the tubules have to have fewer of these large cells despite their same, diameter. So that’s what they do. So, so, amazing thing number one is that you can have two n four n, six n eight N Ns with all this extra genetic material.

No problem. You still got the same newt. So that, that’s, that’s kind of remarkable already, that, that it’s not a problem to have all this, all this genetic material around prob, the amazing thing number two is that they will actually adjust their morphogenesis to the available cell size.

And then here’s the most amazing thing of all, which is that if you make the cells truly gigantic at this, I believe is like six N, six N Ns, cells get so big that one single cell will wrap around itself, leaving a space in the middle. Now, this is a completely different molecular mechanism. This is cell to cell communication. This is cytoskeletal bending. So what, what this is, is basically a kind of top-down causation in the service of having the correct anatomical structure. Different molecular mechanisms get, called up. So if you think about, what, what the information is that’s necessary here.

So, so you’re a nude coming into the world for the first time. You are the inheritor of, of eons of, of evolutionary progress. But you can’t overtrain on that data because you actually don’t know how much DNA you’re going to have.

You don’t know, what size your cells are going to be. you, you don’t, certainly there are things about your environment that, that, that you don’t know. And then there’s also some injury that I’ll show you, momentarily that you can get around. There are all these things that you actually don’t know. You can’t assume any of this. And so, one thing to think about is that the idea that evolution really makes problem solving machines. It doesn’t make fixed solutions to things. it makes it kind of problem solving machine that has really amazing flexibility in dealing with novelty. I mean, you can’t really count on much if you can’t count on knowing what your own parts are.

And in this case, you can’t assume that your DNA is what you think it was going to be, and you can’t assume that your cells are the right size that, that they were supposed to be.

Furthermore, even though development is, very reliable, so here, we all start like this, and eventually you get a human, for example, it isn’t hardwired because if you, if you actually cut, these embryos into pieces, you don’t get half bodies. You get perfectly normal monozygotic twins. And so you get this idea of ala James that you can reach the same ensemble of gold states corresponding to a normal human anatomy from different starting positions, avoiding some local maximum. So it’s really about competency in traversing this anatomical morpho space.

In fact, some animals can do this. So that, that’s called regulative development. Some animals can do this throughout their lifespan. Here’s, here’s that, axel lot, for example, in the izo. So they regenerate their eyes, their jaws, their spinal cords, portions of their heart and brain. And so, so here it is in the limb.

The limb can be amputated at any location. They will build exactly what’s needed, no more, no less. And then it stops. It stops precisely when the correct salamander limb has been completed. this is not just for, worms and, and amphibia, human liver is highly regenerative. Even the ancient Greeks knew that. I have no idea how they knew that, but they did. deer are a large adult mammal that grows a centimeter and a half of new bone per day when they regrow their antlers. So you’ve got bone vasculature, innervation, and then human children can actually, regrow their fingertips below a certain age.

They can do a really good job, cosmetically of re regrowing, amputation wounds. So, what, what I’m saying here is that, the, this, this, this process of building a complex body is not a hardwired feedforward kind of system.

And here’s, I want to show one other example in a bit of detail. This is, this is a ta bolt here, the eyes. Here’s the nostrils. this, this guy has to, turn into this guy, and that’s metamorphosis. The tat bull has to become a frog. In order to do that, they have to rearrange their face. So the jaws have to move forward. The, the eyes, the nostrils, everything has to move around. So up until now, it was, up until a few years ago, it was thought that this was a pre-programmed process where every, organ in the face, just moved in the correct direction, the correct amount.

And, and then you get from a normal tap to a normal front, even, even that is, is, is a lot harder than, than it seems because you, you can’t actually, encode directions of movement or anything like that in a genome.

So how, how does it work? But, but, but it was thought that it was a pretty pre-programmed kind of thing. So we decided to test for new kinds of competency in that system. So what we did was we created these, so-called Picasso, TAs, everything’s in the wrong place. The eyes off to the side, the jaws are on top of the head, everything’s in the wrong place. And if it was hardwired, well, then the frog would be all messed up because you’re starting moving from the wrong position. But instead, you get quite normal frogs out of this, because every organ moves in novel paths, and it keeps moving. The whole thing keeps morphing and shifting until you get to a correct frog, and that’s when it stops.

So the genetics doesn’t give you a hardwired set of rearrangements. It gives you a system that’s, that, performs a kind of, error minimization.

It, it, it recognizes, error, and it takes corrective action. So the way we think about all this in the body is that, this, this is the story that, that developmental biology textbooks focus on, this emergence idea that there are, genes that regulate each other. There are some simple rules, perhaps like cellular automata, can capture that, that kind of thing, local rules that interact with each other. And then a amazingly, something complex comes out the other end. And that certainly does happen. There are many simple systems whose rules give rise to complex outcomes, right?

So, so for fractal, cellular, automata, plenty of that, but living things are actually much more interesting. It’s not all feedforward emergence because they have these, feedback loops where if the final product is deviated, and this is the kind of thing that cellular automata and fractals don’t do, if the final product is deviated from that set point, new activities kick in both at the level of physics and genetics to try to reduce the air and get back to where you were.

This is a basic homeostatic circuit. we call it anatomical homeostasis. Now, on the one hand, having feedback loops is nothing new in biology. So, so we all know that there are, there are feedback loops, but there’s something, well, there’s a couple of strange things going on here where they make you think differently. First of all, typical feedback loops in biology are scalers. There’s single numbers. So pH of oxygen level, hunger level, things like that. Temperature, they’re, they’re single numbers. What you have here is a complex set point that is some descriptor of a shape, maybe not down to the individual cell level, but, but you need a descriptor that, encodes complex anatomical shapes.

So how would you do that? And then the second thing is this, in biology, and especially in in kind of molecular biology, you’re really not encouraged to think about goals.

You are encouraged to think about chemistry and the way that, these systems, evolve forward with time. But you’re really not supposed to think about what is it trying to do? Is there a goal? Is there a represented endpoint for this process? that, that’s, that’s considered sort of taboo. But, but the good thing is that since the forties, and the development of cybernetics, we have had a mature science of talking about, systems with goals that isn’t magic. So, so we can, we can, make use of that, and then we can ask, okay, so where in the body is this set point actually represented in your thermostat, which does something like this, where you can point to where the memory is, you have, the interface to change that memory, and it will do what it needs to do.

So this makes a very strong prediction.

This, this weird way of thinking about it suggests that if we were to understand where the memory is encoded, we would be able to decode it, to read it and rewrite it, then we wouldn’t need to try to make all our changes down here at the molecular level, which is really hard, because it’s very difficult to work this backwards when you want to make system level repairs for the towards health. It, it’s almo it’s usually impossible to know what you need to do at the molecular level, because this process is not reversible. It’s a terrible inverse problem. but, but maybe we don’t have to do that. Maybe, maybe like with a thermostat where you don’t actually need to know what, how the whole system works.

You just need to know how the patterns are encoded, what, how the set point is encoded. Maybe we can rewrite that set point and let the cells do what they do best.

So that’s what we’ve been working on for some, some years now. And we took our, in, in asking the question, how, how could you encode a complex goal state in living tissue? We started looking at the brain. I mean, the brain is one non-controversial example where collection of cells stores gold states for an entire emergent organism, right? You have goals in your, in your brain, of, of behavioral things you’re going to do that don’t make any sense to any individual neuron. They’re, they’re only property of you as, as a collective intelligence. So, so the way the brain does it is with this architecture, you’ve got this electrical network where cells have ion channels that pass ions back and forth.

they have these electrical synapses known as gap junctions. And so you get a whole network of, of information flow through that system.

And so that’s the hardware. And then the software basically looks like this. This is, this, this group did this, remarkable video of, the, active processes in the brain of this living zebra fish. And then there’s this, research program of neural decoding. The idea is that you can extract the information from this collective agent that would tell you about the, the memories, the preferences, the goals states, the behavioral repertoires of the individual, not of the cells of the fish itself. And, and the idea would be to decode these electrical, events, because that is where all of that is stored.

It’s all stored in the electrophysiology of the brain. So it turns out that, your whole, the, this, this kind of, architecture is ancient. It was discovered around the time of bacterial biofilms by evolution. Every cell in your body does this.

This is not neuro specific. So every cell in your body has ion channels. Most of them have electrical, connections to their neighbors or gab junctions. And could we do the same kind of decoding program, basically, import these concepts from neuroscience and ask, what were these networks thinking about before the brain developed both evolutionarily and in, in, in our, developmental sequence, what do these electrical networks do? And so, we, we, we know what brain electrical networks think about. They think about behavior and ultimately, behavior in other spaces like, social spaces and, and, linguistic spaces and so on.

What does, what, what do your body cells think about? Well, we first developed some tools, so we developed, some techniques to read and write this, this electrical information. So, so we use voltage sensitive dyes. And so here’s a, a frog embryo, in time-lapse.

And you can, you can use these fluorescent voltage sensitive reporters to see all the electrical conversations that these cells are having to each other. It’s like scanning a brain. And, this was, this was first worked out by Danny Adams in my group. And then, we do computer simulations, that, allow us to try to understand these patterns and how they evolve through time. So we do a lot of computation. And then here’s an example of, what these patterns look like. Basically, this is what we call the electric face. here’s one frame out of that movie.

And you can see that even before the genes come on to regionalize the face, you can see what the, what the, what this tissue is thinking. Here’s, here’s where we’re going to place the eye. Here’s where the mouth is going to go, here are the plateaus.

and so that’s, and, and if you change those patterns, which I’ll show you momentarily, then everything changes downstream. And then, so that, that’s a required normal pattern. Here’s a pathological pattern induced by an oncogene, a human oncogene being injected in the taal. Eventually they make a tumor and the tumor starts to metastasize. But long before that, you can already tell where it’s going to be by the, voltage, by the disrupted voltage, where these cells are depolarizing and basically disconnecting from the rest of the animal. so, so, so you can watch.

And, and so, so there are some obvious, applications here, monitoring for birth defects, cancer diagnostics. We’re working on all of those things. but, but the other thing we developed are, the all important tools to actually rewrite that pattern and to rewrite the pattern.

We do, we do not use applied fields. There are no electrodes, there are no magnets, there’s no electromagnetic radiation, there are no frequencies, nothing like that. What we’re doing is taking advantage of the natural way that cells hack each other’s behavior. And they do that by controlling this beautiful interface that, that cells exposed to each other, which is the ion channels and gap junctions on their surface. So this is molecular physiology. We use drugs, we use light. we use, sometimes, protein engineering to open and close these channels and pumps, as, in accordance with our computational models.

So that’s the interface that every cell gives you. In more importantly, that tissue level collectives give you to control what information is being processed inside. This is, this is the, this is the keyboard of, of this particular computational device.

So when you do that, you can induce, you can induce, ectopic, odorless, odor or inner ears. You can make extra hearts. You can make new limbs here. Here’s our optogenetic frog, where it’s got a, it’s got a limb, growing out of its mouth. You can make ectopic four brain. And this, this embryo may or may not be smarter than this one. you can also make things like fins. Now, that’s interesting because ta bulls aren’t supposed to have fins at all. That’s a zebrafish feature. So, we’ll, we’ll talk about that momentarily. But what, what you’re seeing here is that by altering the electrical pattern memories in these cells, you can call up different organs.

And in fact, this is maybe my, my favorite example. We can take, some ion channel RNA, which, which encodes, potassispecific kinds of potassium channels.

And in that electric face, there was a particular spot that indicates this is where the eye goes. So we said, okay, could we recapitulate that spot somewhere else? And if you inject an ion channel and you recapitulate that voltage spot somewhere else else, sure enough, those cells get the message, build an eye here, they make an eye. These eyes have, retina optic nerve, all the same layers that a normal, eye is supposed to have. And so you can, so, so this is telling us here a few important things. First of all, it’s telling us that the bioelectricity is instructive. This isn’t just making defects.

This is actually passing along a message that calls up a complete, healthy organ in another location. So it’s functional, it’s instructive. It’s not an epi phenomenon. also, it’s, incredibly modular. In other words, we didn’t have to pass on all the information about how to make an eye.

This is, this is, kind of a first glimpse of this idea that this is a trigger. It’s a sub-routine call. It says, make an eye here. The tissue already knows how to make eyes, the animal’s already done it twice in development. So this is not us micro programming each cell. You’re going to be a, a retinal cell, and you’re going to be an optic nerve cell. No, we don’t talk to the individual cells. We’re talking to the collective, and we’re giving it information that makes sense for a collective. You’re an eye individual. Cells don’t know what an eye is, but the collective does. And so, so, so that’s, that’s, that’s another interesting thing. And then, kind of the last thing I’ll talk about is, is, is this, here’s a, here’s a lens sitting in the flank of a ta bull somewhere.

The blue cells are the ones that we injected. All of this other stuff is, we never touched it. There’s, there’s a secondary instruction going on. We instruct these cells make an eye. Those cells, can tell that there’s not enough of ’em. There’s too few of them to make an entire eye. So what do they do? They recruit their neighbors to help them fulfill this goal. So we instruct them, they instruct their neighbors. We didn’t have to program that in. That’s something they already know how to do. And many collective intelligences know how to do that. Ants and termites do it too. When a couple of scouts come across something that’s too big for them to carry along, what do they do?

They carry, they, they induce others to come and help. So that is that that scaling of resources to task is a, is, is, is baked into the competency of this medium.

So top-down control, kind of very, very, very modular kind of sub-routine calls at the, at the organ level, and, scaling, scaling to appropriate. And then there’s some other things that, I’m going to skip over. So we’ve been, we’ve been using this, of course, in a regenerative medicine program. So here’s a frog, which unlike salamanders do not regenerate their legs. So 45 days later, there’s nothing we can, we’ve come up with a cocktail, a bioelectric cocktail, that triggers those cells to, take on a journey towards leg building, not towards scarring.

And so immediately you can see with, within a few days, a, msx one positive blastema. You start to, by 45 days, you, you’ve got some toes, you’ve even got a toenail, and eventually a pretty, touch, pretty respectable, touch sensitive and motile leg. And it keeps growing In our latest experiments, 24 hours of exposure to the signal, followed by a year and a half of leg growth.

We don’t touch it during that time. This is not about three D printing the leg. This is not about telling stem cells well, how to grow. This is not about micromanaging it. This is about, changing the bioelectric state of the early cells with drugs. This was a, a pharmacological, reagent with, with drugs targeting ion channels and other targets that, convinced the collective to undertake a particular path through morpho space. Once you get them going, they will go on their own. You do not need to keep pushing. And so here I have to do a disclosure because Dave Kaplan and I have a, a company that we started called Morph Pharmaceuticals, which is now using this combination of wearable bioreactor and then the payload, to, to talk to the cells in, rodent models to try to push this towards, towards, biomedicine.

I will, I will switch a little bit here to a different, animal. This is back to Parian. Now, to really hammer this idea of the bioelectric, circuits as holding critical information for, anatomical structure. The memory, literally the memory. So here’s a, here’s a parian flower. If you cut off the head and the tail, you’ve got this middle fragment. And you might ask, how does the middle fragment know how many heads it’s supposed to grow? Why, why doesn’t, the, the, these cells back here at the tip of the tail, they’re going to grow a new head.

Why don’t these cells grow a new head? They’re right there next to it. They, they come from the same location. Why don’t they grow ahead? How come, how come this, this fragment only grows one head? And so if you look at the bioelectrical pattern of that fragment, you see, ah, here’s a pattern that says one head.

That’s how many heads you’re supposed to have. So, and, and so sure enough, the cells obey, and then you get a one headed worm. Well, what we can do is rewrite that pattern, much like we did for the, for the eye and for some of those, other examples I showed you. And, it’s, it’s kind of messy that technology’s still very in, its in its infancy. It’s very young. But we can make, we can make a pattern that says two heads, build two heads. And when we do that, when you, when you injure that animal, sure enough, you get a two headed worm. This is not Photoshop. This is, this is a real animal. Now, something critical is that this bioelectric pattern is not a map of this two-headed worm.

This is a pattern of this perfectly anatomically normal one headed worm.

But so, so the molecular markers, had in the, in the head, not in the tail, and so on. and then if you injure it, that is when this information becomes salient and, and builds a two-headed worm. So this is a kind of counterfactual. This, this collective, agent is able to store one of two different representations of what a correct parian is. What am I going to do if I get injured in the future? It’s not what’s my shape now, the shape now is one headed. So you can, you can, you can, put different memories of the correct, set. So, so back when I showed you, that, that, anatomical homeostasis diagram, I promised that there was going to be, a, a, a decoding of the, the actual set point.

Where, where is the anatomical pattern stored?

Well, this is where it’s stored. You can see it here, and we can actually rewrite it and it, and it controls what happens next. Why do I call it a memory? Because it has all the properties of memory. If I take this two-headed worm and I chop off the, the primary head, I chop off the ectopic secondary head. It’s got normal genetics. We never touched the genome. This is not genomically edited. There’s no synthetic biology here. This little piece in, in plain water will continue to generate two-headed worms. The question of how many heads should you have is not nailed down in the genetics. What the genetics gives you is hardware that defaults to, keeping information that says one head, but it’s reprogrammable.

It’s rewriteable. And once you rewrite it, it keeps right, like any good memory, it’s, it’s, it’s stable, but it’s rewriteable.

And so once we say, no, actually two heads is what you should have, that is what they will have in perpetuity. You can see a, a video of, what these two head guys, do in, in their spare time. And, and, and then, and then you can set it back to one headed by. By changing that pattern back to normal, not only is there a control over head number, but there’s also control of head shape. And if you remember, I showed you that you can make, tadpoles grow fins. Well, we can make this parian with this nice triangular head. shed shape generate, flatheads or roundheads like other species of planaria.

Again, never touching the genome. So this is where, oh, and, not only the head shapes, but but the, their, the shapes of their brain here and the distribution of stem cells become just like these other species.

These guys are about a hundred to 150 million years of evolutionary distance from, from these. and, what you’re seeing here is, is, is really, really to understand this, the, the, the hardware software analogy come, comes in really handy here, because without that, it’s really, really hard to understand this. The idea is that what the genetics does is set the hardware of your cells, it sets the ion channels and all the other, the, all the other pathways and so on. But that information, as, as we know, the hardware doesn’t tell the whole story. You actually can, can, change its, the various modes of the, of, of, of function.

And so, so you can, you can get the, you can get the same hardware to visit tractors in anatomical space that correspond to other species that are normally occupied by these other species.

But the hardware can go there, it’s reprogrammable. you, you don’t need to edit the genome, you don’t need to, change it, change the hardware to get it to do something different. Biological tissue is, is, is strongly reprogrammable. And so we are working now on this kind of like, full stack understanding from the, from the molecular information that tells you which channels and pumps you have all the way up to the tissue level, bioelectric, gradients, and then organ level structures. And eventually an algorithmic description of what’s going on here in the tissue so that, so that we can actually make rational change, hu human understandable, interpretations of, of what’s going on.

So all the way from the bo from the bottom up. And, I will show you a, just in the last couple of slides, I want to show you another, another application.

So, so this is repair of a complex organ. So the normal, frog brain looks like this. There’s four brainin, mid-brain and, and hind brainin, quite co complex, a very special structure. And there are many teratogens including, nicotine alcohol, and genetic mutations that screw it up. So here you see, this brain has lost the normal structure, severe defect, and we wanted to know whether we could fix this. How could we, how could we fix this using the information that we have, which is that these structures are set up by bioelectric patterns. And so we created a computer model that, described how it is that this particular structure and the size and shape of the brain is set by the voltage gradients underneath.

And there are a set of gradients that are, here, this, this bell curve is kind of the correct shape, and if you flatten it, it, it flatten it in e the direction you get terrible defects.

And so we asked this model, okay, given that you have this, given that you have this, incorrect pattern, what ion channels would you have to open and close to get back to the correct pattern? This model knows nothing about what’s downstream. it really is just talking about the voltage. How do we get back to the correct pattern? And the, the, the model predicted this one particular really interesting channel called HCN two. And it said, well, if you open HCN two with a pattern should come back to normal. And so that’s what we did. I’m going to show you the kind of the most, striking, thing that we fixed, which was a mutation of a gene called notch.

So mutated notch. In this case, we introduced a dominant, overactive notch. Mutant is really important for neurogenesis. If you screw up notch here, there’s no forebrain, the middle and, midbrain and hind brain are just a bubble.

there’s, there’s no behavior. These animals are profoundly, profoundly affected. Well, if you, if you open the HDN two channel, for example, with existing human approved drugs, which happen to be anti-epileptics, they get their brain structure back. They get their brain, gene expression back, and they get their IQs back. They become indistinguishable from controls in, in the, in learning rates. So look at what’s going on here. You can, you can repair what’s fundamentally a hardware defect, a a, a mutated notch signal. You can repair that with a temporary exposure to a drug that opens an ion channel.

So, so at some, in some cases, and I’m not saying this is going to be true always, but in some cases, there are hardware defects even that you can fix in software by understanding how to control these bioelectrical patterns. And so, again, note that we, that we fixed a very complex anatomical structure by soaking the whole animal in the drug.

We didn’t have to micromanage where it goes. We didn’t have to tattoo the thing with different, different kinds of, channel proteins or anything like that. The power of the computational model is such that, that it tells you that the circuit is where the specificity is just open. The HD two channels, the self-organizing, properties of that electric circuit will take care of itself. It takes the control out of our hands, very useful in exactly the same way that we didn’t have to say how to build an eye. We don’t know how to build an eye, but we do know how to tell cells that that’s where the eye goes. So it’s, it’s offloading, the, the, this computational approach offloads a lot of the, responsibility off of us and onto the system.

So what we’re doing now is, trying to, trying to put together, and you can sort of, you can, you can start playing with, with a, with a very, simple early form of it here, a a a kind of a, a a platform where you go in and, you, which tissues you would like, or tissues and organs you would like to fix.

You need physio data. This is the biggest thing that’s missing still. So we’re, we’re working together that, and then, from profiling studies, which channels and pumps exist in all of these tissues. So here they add, those are your targets. And then there’s this, electric, electrosurgical design environment that is going to be the, the computational model that helps us pick these drugs. So from a description of what’s wrong with the bio electrics to how, how do we fix it in terms of, ch which channels? And then, and then, drug, dr drugs that we already know.

I mean, something like 20% of all drugs are ion channel drugs. So there’s plenty of, plenty of candidate, electroceuticals. And so the last thing I want to show you is, is a, is a quick story about cancer. what happens during evolution is something really profound, where, where single cells have little tiny, control loops like this.

Let’s say pH, the, the cognitive light cone, the, the size of the goals that they pursue is very small. they’re, they’re, they’re only concerned about little, little tiny goals like this. But here, the collective is concerned about a very large goal, maintaining a proper limb. And if you deviate from that, it will work really hard and get back to it. So again, homeostatic cycle, but the size of the goal has changed, and the goal has moved from physiological space to anatomical space. So what evolution does by, by using these electrical connections, is to build networks that scale goals by electrical networks are fundamentally a kind of cognitive glue, because they allow, a competent subunits to connect together so that the collective has bigger goals in a bigger space.

And so, so this is, this is what normally happens during evolution and during development is that the goals, get bigger.

The, they, they, they, stretch into new spaces such as anatomical space, but that, that process has a failure mode, and that failure mode is cancer. And so what, this is, this is a human glioblastoma. And what happens is when cells disconnect from that network, they, they don’t need to be broken. They don’t need to, have any genetic defects. They’re not more selfish. They’re exactly as selfish as normal cells, except that they self is smaller the size of your goals, which determines the size of yourself, capital S as a collective being.

d d determines the boundary between you and the outside world. And as far as these cancer cells are concerned, the rest of the body is just external environment. They, they’re not anymore self, their self is now tiny. They, they’re, they’re, their, their boundary has shrunk.

And so that, way of thinking about it, that, that idea of, of what information do you need to determine where you end in the outside world begins? And it’s a very existential problem that every embryo has to face, suggests a new therapeutic, which is that what if instead of trying to kill these cells, whole, keeping an, a really nasty human oncogene, instead of a kra s things like that, instead of killing them, why don’t we reconnect them to their neighbors? We force the electrical connection. So that’s what we did. We inject the oncogene, we inject an ion channel that keeps, the voltage correct.

And here you go. This is the same animal. Even though the oncoprotein is blazingly expressed, it’s here. In fact, it’s all over the place. There’s no tumor because what drives isn’t the genetics, what drives is the information processing or physiology.

And that information is not allowing these cells to have a little tiny amoeba like, dreams. It’s, forcing them to have, to be part of a large collective network that remembers how to make skin and muscle and other things. And that’s what it’s doing. So we’re now, moving this, into human medicine. So this is, this is, our recent paper on glioblastoma and using some of these ion channel drugs, designed by the same, kind of, strategy that I was just telling you about to, to, to, re reduce the proliferation and, and, and, and to differentiate the cells back into, some kind of normal, normal morphogenesis.

And so the last thing, I’ll say is just my outlook for, for the medicine of the future. So, so we, we have this continuwhich I call the axis of persuadability, all the way from simple mechanical devices up through cybernetic devices, various kinds of learning agents, and then various kinds of rational, rational agents.

And who knows what’s beyond that. this continuum reminds us that there are different tools that you bring to bear at each of these levels, right? So, so physical rewiring and, and, and, changing set points and reinforcement learning and training and and so on. And so, so people, in the, in, in the, in the medical field have assumed for a really long time that cells have to be down here. But that’s an assumption. And we now know that they are, for, for sure here, and quite likely here. The evidence is now that basically you, you can’t just have assumptions about where things are.

You have to do experiments. And so we’ve been doing all kinds of experiments, and, and, and other F two on the basal cognition of cellular collectives. And the further you go along this, the more, power you get from top-down controls.

This is why we’ve been able to train horses and dogs for thousands of years without knowing any neuroscience whatsoever, because they expose this amazing interface, learning and training, and of course, underlied by bioelectricity of the brain. But, but you can train these things top down without knowing what the details are without running their neurons like a puppet. That’s not the, the, the, the, the future. And so the same thing is true for, for cells. So, so I’m using this, this, diagram, with, with permission, from, from this guy who had, some, some surgery. And I often get into discussions.

People want to talk about whether the body is or is not a machine. And I say, look, those things are, it’s not a property of the body, it’s a property of your perspective, how you want to interact with it.

If you’re an orthopedic surgeon, this is what it looks like, and you’re using chisels and hammers. It’s, it’s, it’s very machine-like, and you, you do all this stuff right here, but then you send the patient home to heal. And that healing portion is the one that we cannot reproduce in, in this way. and here it ends up, very, very nice in the end. And, this, this, this, this, this research in particular, placebo and nocebo research, research such as Fabrizio Beis, he says, words and drugs have the same mechanism of action. And that’s absolutely true because the body is a multi-scale system where the top level cognitive control when you get up out of bed in the morning because you intend to solve certain, social, financial and other kinds of goals, eventually they have to move the, the calciand, and potassium and other ions across your plasma membrane of your muscle cell.

So you can actually get up out of bed and go do those things. So your body is, is a, is a stack of different levels of control, and some of them can be, can be addressed this way and others can be addressed, via, via high order interactions. And so this is my last, my last point here is that, all of medical interventions could be divided up this way. The bottom up stuff, which is what we’ve been as a community focused on for, for a really long time. All of this. And these are the things that, try to control bottom up so they, seek to micromanage molecular states, symptoms and so on.

But, but all of this is, is uncharted territory using various kind of interfaces, including electroceuticals, including, different ways to behavior shape cells and tissues to train them towards new, kinds of, new kinds of behaviors.

And, and, and I think the, the medicine of the future is going to look a lot more like a kind of somatic psychiatry and a lot less like medicine. This is all, once, once we understand how these different levels are hacking each other and how we can, take control and, and all of this is described here in detail, so I’ll stop here. I want to thank, all the postdocs and, and grad students who, who, did the work I showed you today, are, are many collaborators. here are the disclosures. There’s a few companies that, that support our work. and, and most of all the animal model systems, because they do, they do all the hard work.

So, yeah, thank you very much.