"Neuroscience Beyond Neurons: communicating with unconventional minds"

Watch Episode Here

Listen to Episode Here

Show Notes

This is a ~1 hour talk given at a conference on the philosophy and future of neuroscience, so I oriented it around extending the concept of neuroscience: not the study of neural phenomena, but more broadly around how neuroscience of neural and non-neural cells shows a model system for understanding collective intelligence, and using morphogenesis as a stepping stone model system to show us how to do behavioral experiments in unconventional cognitive agents. I didn't have time for any of the biomedical applications of these ideas (lots of other talks about that) but alluded at the end to some deep questions around the properties of novel cognitive beings and our framework for addressing it. There was a really good Q&A after which will be put up by the original conference organization when they post their version. I'll link it here when it's up.

CHAPTERS:

(00:00) Defining unconventional minds

(10:05) Scaling cognitive light cones

(15:59) Morphogenesis as collective intelligence

(25:46) Bioelectricity as cognitive glue

(36:56) Reprogramming biological pattern memories

(48:59) Xenobots and synthetic life

(55:15) Platonic origins of order

PRODUCED BY:

https://aipodcast.ing

SOCIAL LINKS:

Podcast Website: https://thoughtforms-life.aipodcast.ing

YouTube: https://www.youtube.com/channel/UC3pVafx6EZqXVI2V_Efu2uw

Apple Podcasts: https://podcasts.apple.com/us/podcast/thoughtforms-life/id1805908099

Spotify: https://open.spotify.com/show/7JCmtoeH53neYyZeOZ6ym5

Twitter: https://x.com/drmichaellevin

Blog: https://thoughtforms.life

The Levin Lab: https://drmichaellevin.org

Transcript

This transcript is automatically generated; we strive for accuracy, but errors in wording or speaker identification may occur. Please verify key details when needed.

Main Episode

[00:00] Thank you so much. That was an incredibly kind introduction. I appreciate very much the opportunity to speak to all of you. I'm sorry. I cannot be there in person. What I thought I would do today is talk about our work, trying to, visualize neuroscience as not the science of neurons, but actually, something much more fundamental around how we are going to understand and communicate with unconventional minds. And I will describe what I mean. If anybody's interested in diving deep into the primary papers, the datasets, the software, everything can be downloaded at our website. And this is my personal blog where I discuss what I think some of these ideas mean. So in our lab, we try to take some very deep questions that have been around for a really long time, so so deep philosophical questions, and we try to drive them towards applications. In other words, ideally, the the big ideas that we have should make contact with the physical world. And specifically, we do this kind of loop where we think about what is really fundamental about biology, what are the lessons that biology teaches us, and how we can expand on that. So so a life or life as it could be. And we use these principles to help us design and understand artificial novel kinds of intelligence or minds as it could be. And then, of course, we use those to help us, again, to understand life. So we hope there's some kind of a some kind of a cycle there, and we we do this in many contexts that sometimes look like regenerative medicine and sometimes, like AI research and sometimes, like, behavioral science and so on. What I'm in particularly... In particular interested in today is this this concept of when we make novel minds. In other words, beings that have never been here on earth that don't have a specific evolutionary history that we usually, rely on to, to make these, to make these decisions. How do we know what kinds of goals, what kinds of competencies, what kinds of preferences? What are they going to be, and where do they come from? It's a more more deeper philosophical question is where do the properties of novel minds come from? So, this is going to contact us in many ways. First of all, of course, the definitive regenerative medicine in humans leads directly to, various kinds of augmentation and hybridization technologies. So no doubt, humans are not going to be satisfied with simply improving the damage to their bodies with with novel technologies. They're going to extend. And so your neighbor at some point is going to have some kind of implant, maybe some percentage of their brain is now replaced with, something. And, we're going to have to figure out not the difference between humans and machines or AIs or language models or anything like that. This is this is what we're talking about. We're talking about your neighbor who now has very different sensors and effectors that you do and how how to relate to them. And then, for, mental health professionals and other people in cognitive and behavioral science, I point out that your clientele list is going to get very, very weird very soon. We... You you know, you're going to have beings that are not the standard everyday humans, and we're going to have to help these beings to have a better life to interpret their their dreams, their... You know, the the the symbolism of their unconscious, all of this. You know? What what is that like for creatures that are not the standard humans? And all of this requires us to address our own mind blindness. I I I call mind blindness the fact that we, like all, finite observers, are limited by our own evolutionary history to the ability to really detect intelligence at very specific scales of time and space. Things are very large or very small or very fast or very slow. These things are difficult for us to recognize as embodiments of mind. And so much like with the electromagnetic spectrum, you know, for the longest time, we didn't realize that we were only seeing a tiny a tiny amount of it. So so we need to we need to help this. So, today, these are the points that I'm going to try to get across. First of all, I think that neuroscience isn't really about neurons at all. It's about scaling of cognition. It's about a cognitive glue, which I mean by the the policies and mechanisms that allow collective intelligence, that bind individual competent agents into a larger a larger kind of intelligence. And the way that we can use that information to recognize, to understand, to communicate with, and to ethically relate to minds in unconventional guises. And we're talking... We'll talk about our own mind blindness and the use of, basal cognition, meaning minds in more simple life forms, on the, the the web of life with us to try to understand truly alien minds. So that's the first thing I will talk about. The second thing, is I'm going to show you a bunch of examples where we used a specific model system that is morphogenesis, the ability of cells to construct anatomical, structures. We're going to use that as a as a model system to understand what what it might be like to communicate with a novel intelligence. This is sort of a stepping stone between us and real aliens because these are our own cells.

[05:05] We should be able to understand them, but they're actually a very different kind of mind. Examples of, of their different kinds of, intelligence, and we'll talk in particular about bioelectricity or electrophysiology for the neuroscientists as the cognitive glue that much like in our brains holds together the mind of the body. So we'll be talking about that. And then at the end, I'll go some impacts of this, the applications and, the impacts on the ethics of the forthcoming diversity of beings. So so the goal of my framework is to recognize, create Mhmm. And be able to relate to truly diverse intelligences no matter what they're made of or how they got here. So this means the familiar biologicals, weird biologicals such as colonial organisms and swarms, synthetic new life forms, AIs, whether embodied or, or purely software, and maybe potentially exobiological agents. Of course, I'm not the first person to try for something like this. So here is, Rosenbluth Wiener and Bigelow trying to describe the the the the set of steps that lead from passive matter, all the way up through the the science of cybernetics to understand how they go up to the human level self reference and metacognition and things like that. So what I want is is a is a system that will move experimental work forward, not just philosophy, not just definitions, but new capabilities, new experimental capabilities and better ethical frameworks. Now I I would... The the first sort of unusual thing that I wanna point out is that we we shouldn't think of the different kinds of systems that are out there as in discrete differences of intelligent versus not, cognitive versus versus just physics. I I think those kinds of things are incredibly unhelpful because if you take bio, biological development, so embryonic development and evolution seriously, what we see is that we we are at the at the at the center of a continuum starting off with with chemical machines, to speak, like molecular molecular networks and then cells and so on, both on a developmental and an evolutionary scale. So what's happening here is is a is a transformation, meaning that the null hypothesis should be continuity. It means that if you think there are emergent phase transitions or sharp categories, these need strong evidence. You have to argue for them. The baseline, the null hypothesis, should be slow, gradual change. And this is what happens to each of us. We all took this journey from being an unfertilized oocyte through, the different disciplines. At first, you were the the subject of physics and chemistry, then developmental physiology, and then eventually, behavior science, maybe eventually psychoanalysis or something like that. Maybe some of your cells take a detour through one of these other possible, branches. But the point is there is no magic line where you go from being just chemistry and physics to being a mind. This means that we need models of scaling and transformation. We need to understand what it is that scaling, what kind of intelligence, how much, not these old categories of, is it or isn't it. Now, to take a closer look at at, at our own origins, at embryonic development, this is where we come from. This is an early embryonic blastoderm, there might be, let's say, hundreds of thousands of cells here. And we look at that and we say, there's one embryo. What what is there one of? What what are we counting as one when we say there's one embryo when you see all these cells? And each one of those has incredible number of chemical reactions inside. Well, what we're counting is commitment. We're counting the fact that all of these cells are, in agreement. They're all aligned, both physically and informationally, aligned to the same journey in anatomical space. They're going to move from being an oocyte to whatever kind of body, a giraffe, a snake, whatever it's gonna be. So... And and and in fact, if we make some cuts in this blastoderm, you will find out that each island that doesn't feel for some time the presence of the others will organize its own embryo. So actually, the number of individuals in this excitable medium is not obvious. It's not set by the genetics. It is it is worked out in worked out in real time. And so the the... What what what we see when we say there is one, being here or or one agent, we actually see a kind of a self reinforcing model of commitment of of where these cells are going to go in anatomical space. And there are many interesting questions in neuroscience. For example, so how many individuals, like, with developmental biologists will ask, how many individuals can you can you pack into a single, embryonic blastoderm? Well, in neuroscience, you might ask, how many personalities can fit in a brain? And we and we still really don't know the answer to that, that question. The the the the neural tissue supports, what what number of personalities per cubic millimeter, you know, something like that. So so the first thing is this kind of individuation. The next is the scaling of the cognitive light cone. What I mean by the cognitive light cone is not the range of sensors or effectors.

[10:09] I mean the size of the biggest goals that a system can pursue. And you can find lots more details here. But, basically, single cells have a tiny little cognitive light cone. They're really pursuing, scalars such as pH, metabolic level, things like that, around a very small region of space time. A little bit of predictive capacity, a little bit of memory, but really very, very simple kind of goals. But a collective can have huge grandiose goals. So for example, here's a group of cells making a limb of an amphibian. If that limb is amputated, these cells will very quickly get back to the original goal... To the original set point. They will regrow the limb, then they stop. Like any homeostatic process, it stops when the error from set point becomes small enough, then it stops. So the goal state here, the anatomical set point is very large. No individual cell knows what what a limb is or how many fingers you're supposed to have, but the collective absolutely does. And so the cognitive light cone of a network of cells is much bigger than that of a single cell, We can actually plot different different size of cognitive glycones of different kinds of systems and compare them. We'll talk about how this scale up happens. So I think it's important to to keep in mind that all intelligence is collective intelligence, not just beehives and ant colonies, but all of us. We are made of things like this. This is a single cell. Now this is a free living organism, but, you get the idea. There is no brain, no nervous system, but extreme competence at its own tiny little little agendas, And it's these kinds of, unicellular beings that at some point, had to come together to to pursue much larger, much larger goals. So this means that it's, it's really important to have some tools to understand where a system fits on the scale of intelligence. So so I propose this thing called an axis of persuadability. What I mean is that we cannot simply guess. We can't sit back in our philosophical armchair and have feelings about, what should or shouldn't be called intelligence. We have to do experiments, which means that we have tools. We have tools of rewiring. We have tools of cybernetics and control theory. We have behavior science. We have psychoanalysis and psychology. We have all these different tools. And what we have to do is empirically determine for any system. We have to empirically determine, what set of tools is going to be appropriate for that system. So we have to do experiments. We don't know ahead of time. So how do you find a novel kind of mind? We have to hypothesize a problem space that it works in. We have to hypothesize what its goals might be. Then we do experiments to see what degree of competency it has to meet those goals when circumstances change. Meaning you have to do perturbative experiments, put barriers and change sort of the circumstances. We have to try all the tools of the behavioral science handbook. We have to try training it. We have to try communicating with it. We have to try rewriting its goals. And then we all get to see as an empirical outcome, we all get to see how that worked out and where where along the the continuum your your system optimally lands. So this is what we do in our lab. We take all of these approaches and rather than have pre pre sort of conceptions about where things are, we do experiments. And what it... And it turns out, and and and many other labs in the field of diverse intelligence do this as well. It turns out that, we are, we consist of a multiscale competency architecture, meaning that at every level, the molecular level, the subcellular level, the tissue, and so on, at every level, this... There are systems that have agendas. They have competencies, in various kinds of, problem spaces. Okay? And we have to understand how all of these things work together in order to really, understand what what cognitive systems have in common. We, as humans, find it relatively easy to identify intelligent behavior for medium sized objects moving at medium speeds through three-dimensional space. So birds, some other mammals, maybe a whale or an octopus, They can solve problems on our time scale and our spatial scale, and we recognize that. But there are many other spaces in which biology, for example, has been operating for for for millions and millions of years that are very hard for us to recognize. For example, the high dimensional space of gene expression, the the anatomical space of possible shapes. This is what we'll spend most time talking about. Physiological spaces, metabolic spaces. If if we had, for example, I I think that if we had evolved, for example, with senses of our inner blood chemistry, I think we would have no problem recognizing that, our body is full of symbionts, such as your liver and your kidneys, that navigate spaces all day long to try to, help us and keep us alive. But it's very hard for us to imagine and visualize because we're so obsessed obsessed with vision vision and three-dimensional space and so on. So so we're interested in how other systems intelligently navigate these other spaces. Not... Embodiment is not just in three-dimensional space. It's in many other spaces, which, by the way, once you see this, you should be very suspicious of, discussions such as nonembodied AIs.

[15:15] They're nonembodied because you don't see them rolling around the the the, you know, the floor or moving around with, like, the way the way that we do. Should be very suspicious of that. These other spaces are hard for us to to recognize. So what do I mean by intelligence? I like William James' definition. It's it's very sort of cybernetic. It's, substrate agnostic. He defines it as the ability to reach the same goal by different means. And the best part about this is that it's very experimentally tractable. Once you have a hypothesis about what the goal is, you can try putting barriers. You can try to move it. You can make, different changes in the system, And you can ask what degree of competency does it have to reach its goal under under, novel conditions. And and all of this is is easily quantifiable and measurable. So so we made the hypothesis many years ago that morphogenesis, meaning the creation and regeneration and, upkeep of a complex body, is the behavior of a collective intelligence. The collective intelligence is that of cells. The behavior it's doing is in the space of possible shapes of anatomical morphospace, and that, we... So the hypothesis is that we can use certain tools of behavior science to understand how your body comes to be. And and this means that, in order to understand what kind of collective intelligence these cellular swarms deploy, we have to remap some familiar behavioral science concepts. So instead of milliseconds in neuroscience that you're used to, we have to think in minutes or hours. Instead of three d motion, you have to think of more of a space. The mechanisms, interestingly enough, stay exactly the same. Ion channels, electrical synapses known as gap junctions, neurotransmitters, all that stuff you can keep because that's that's exactly what, what underlies this intelligence. And, many of the algorithms, all the tools of cognitive science work great. Perceptual bistability of of, you know, all the kind of, Frestonian sorts of, sorts of approaches, active inference, all that. All of these were great. In fact, we make tools. So here's a tool you can play with, called FieldShift, where you paste in the abstract of a develop... Of a neuroscience paper, and it just makes the changes and gives you the corresponding developmental biology paper. It actually works really well. So, for the next, ten minutes or so, what I'm going to show you are, the components of intelligence in, in morphogenesis. Because understand, I've made a I've made a very grandiose and testable claim here. What I've said is that morphogenesis is a kind of intelligence. That is not what most people learn in developmental biology class. What you normally learn is that there's a genome. The genome, in some way, has information about the body, and there's a mechanical, sort of hardwired unrolling of of the laws of chemistry doing what chemistry does. And eventually, there's this notion of emergence and complexity, then you get the... You get a you get a body. I'm telling you something completely different. I'm telling you that that the process of construction of bodies by cells has many of the hallmarks of intelligence, because they are not hardwired consequences of the DNA. In fact, they interpret the DNA in creative ways, and we can talk about that after. So let's look at some of the... So so if I'm going to make that claim, I have to show you what are the competencies that should be recognizable to any behavior scientist. What are the competencies that this thing has? So now I'm going to show you a bunch of them. First of all, homeostasis. Okay? The basis of goal directed activity. So being able to reach a goal when you're deviated from it, the the simplest version of that is just is just homeostasis. So I already discussed this creature such as this axolotl. If you amputate the limb, very quickly, it gets back to the correct position, and then it stops. So it knows what to do. It knows what... When to stop. You can cut it anywhere along here. It it it has the ability to to get there from different starting positions, and it has a set point. This is very simple anatomical homeostasis, so the simplest kind of goal state. Now the next thing is, the same end by different means. Can it do this by... From different, scenarios? So this is an early, let's say, human embryo. When you cut it in half, you don't get half bodies. You get two perfectly normal, monozygotic twins. In fact, can cut it into a a number of pieces, each piece gives rise to the same thing. So that means that you're able to navigate to this ensemble of gold states from different starting positions. It is not hardwired. Here's another example of, why it's not hardwired. We discovered this, around 2012. When you have these tadpoles that have to turn into a frog, they have to move their craniofacial organs around. So the eyes, the nostrils, the brain, the mouth, all these things have to move around to go from this kind of face to this kind of face. It was thought that this is a hardwired process. Every organ moves in the right direction, right amount, and then you get a normal frog from a normal tadpole. So we decided to scramble the initial positions. We made the so called Picasso tadpoles where everything is in the wrong place. The eye is off to the side, the the... On the top of the head here, the mouth is out here. Everything is just in the wrong position. And we found out that these guys give rise to pretty much normal frogs because it is not hardwired. All of these pieces move, and they keep moving until they get to the correct pattern, and then they stop.

[20:23] Sometimes, actually, they go a little too far, they have to come back a little bit. But eventually, they settle down, and they stop in the correct pattern, which leads you to ask the question, how do they know what the correct pattern is? I'm going to show you that in a few minutes. But but notice that that's, that's a question that you are definitely not encouraged to ask in developmental biology class. In behavior science or neuroscience, maybe that's okay. You can ask how does this animal know what to do. So it's okay to ask that question for neurons, for networks of neurons. You know, brains do know things and they have goals and so on. But you're not supposed to ask that question about other, other kinds of cells. But but that's that's... I think that's profoundly wrong, and and I'll I'll explain why. So so again, we see that you get the same anatomy via different paths in novel circumstances because it has this, endpoint. It is not simply rolling forward to whatever emerges. It has a specific endpoint that it will self correct to get. The next thing that's really interesting is hierarchical non local control. So here's something, kind of everyday magic that happens in neuroscience that people don't think about very much. When you wake up in the morning, you might have extremely abstract goals, maybe financial goals, maybe social goals, maybe research goals. These are these are incredibly complex abstract kinds of things. But in order for you to get up out of bed and do those things, to execute those goals, ions have to cross your muscle membranes. In other words, chemistry has to change. The way chemistry works has to change very specifically in order to execute behaviors to help you meet those goals. So what's happening here in the human body? Of course, there's a there's a there's a neural system that is, its its whole purpose is to, convert convert across levels. In other words, to take very highly abstract kinds of goal states and eventually make ions cross muscle membranes so that you can walk. So the chemistry obeys these kind of, mental mental structures. That that amazing, aspect of cognitive science is incredibly ancient, and it goes all the way back. Like, all of these mechanisms, they go all the way back to the control of morphogenesis. So here's an example. They took a... This is a old work from the fifties. They take a tail. They graft it to the flank here to the middle of a... Of a... Of this animal. And what happens is over the next, few weeks, this tail converts. It it remodels into a limb. And, what it does, these cells up here at the end of the tail, they become fingers. Now why? There's no damage up here. There's nothing wrong up here. There's no reason for anything different to be up here. But the whole system is a top down control so that the individual cells, they don't know what a finger is or how many fingers you're supposed to have, but the collective does. And when it realizes, when the large scale real cellular swarm realizes that this thing doesn't belong in the middle of the body, that it should be a limb, it issues the commands downstream that then will coordinate all the molecular events to turn tail tip cells into fingers. Same exact thing that's going on here. Top down control. It's not about damage and regeneration. It's about achieving your goals, which are set by a higher level. Okay? Your goal... Your goals are set by a higher level, which means there has to be a mechanism for aligning those mechanisms. That's that's what that's what we'll talk about. So this this leads us to the next, kind of capability, which is hackability. The idea that cognitive systems are fundamentally hackable. They are more like software than they are like hardware. They are trainable. They change the way they act based on prior experience. They can be exploited, by others. They can be trained. They can be communicated with. They can make they can make decisions, and, they can be extremely flexible about their about their work. So so is it possible that, you you know, that that cells in in in the process of morphogenesis can do this? Mean, we're told that the morphogenetic outcomes are, the product of of of DNA and so on. Where where does the, where does the hackability enter it? So, in order to, to explain this, we have to start with this question. Where do anatomical specifications come from? So we we know that the standard, human starts off as as a bunch of embryonic blastomeres eventually make something like this. This is a sort of cross section through the, through the human body, through the torso, and you see all this, like, incredible structure. Everything is in the right place next to the right, you know, next to the right thing. Where does this come from? Well, you might think it comes from the DNA, but, of course, we can read genomes now, and we know that the genome doesn't say anything about this directly. There's nothing in the genome that talks about what an eye is or where it should be or how many you should have. That's not what's in the genome. The genome contains proteins and some timing information about when those proteins become available to cells. But where does the anatomy come from? How do groups of cells know what to make and when to stop? Right? The genome doesn't tell you that any more than, the genome of the ants has encodes this, the structure, this the structure, this termite mound, or the genome of the spider tells you exactly what the spider web is supposed to, supposed to look like. All of these things are created by the by the software that runs on the hardware that the genome encodes. So the genome tells every cell what what is the hardware, that it gets to have. But after that, the software takes over.

[25:24] And by the way, when I say hardware and software, I don't mean that, cognition and living systems can be fully encompassed by our computational metaphors that we have right now. I don't mean that. I just mean that the concept of software has some really interesting components surrounding reprogrammability that that we do need, that that that are important. So how how do these cells know? Okay? How do they know? Well, we have one example. Now now people usually tell me, sort of in in the developmental biology... Molecular biology community, people will say, how can you have a set of cells that know things? Cells don't know things. But, of course, in neuroscience, we know that absolutely we have we have a science of groups of cells that know things, and that's neuroscience. So, in your brain, you have a variety of of these of these neural cells that, have ion channels on their cell surface. They acquire voltage potentials by allowing, charged molecules to go in and out, and, they can they can transfer that that voltage information to their neighbors through electrical and chemical synapses. And in... It is the commitment of neuroscience that the electrophysiological dynamics on the brain... And and you can see here, this is, this group made this amazing video of a zebrafish brain, thinking about whatever zebrafish think about, that that the... This electrophysiological activity actually encodes or contains all of the cognitive content of this animal. So its goal states, its preferences, its capabilities, all the things it's thinking about, everything is somehow encoded in this electrical activity. So we... This is a... We we already have a science of, that is that is committed to the idea that that, mental content is, in some sense, encoded in electrical activity in groups of cells. And, neuroscientists also try to do this thing called neural decoding, which says that, okay, if we could image all of this and process it appropriately, we we should be able to guess what the animal is thinking about. Or or or or also we should be able to incept new information, you know, write new information into the mind of this animal. Okay. Well, the key is that it it... And and so and so this this electrical electrical network that can store goals, can guide the system towards, implementing those goals is, is all encoded via, this kind of hardware, ion channels and and so on. Well, it turns out that that amazing architecture for being able to store goals, store memories, know things, and let your knowledge and your past experience guide your future actions towards, adaptive, goal directed behavior, that system is incredibly ancient. It did not, just suddenly arise when neurons and muscles evolved. It's as old... It's at least as old as bacterial biofilms. So every cell in your body has ion channels. Most cells in your body have these electrical synapses, these gap junctions. And, this idea of using using electricity to integrate information across space and time to form larger structures that think about bigger things, this is this is absolutely ancient. And so what we've been doing in our in our lab is to basically basically steal all of the tools and concepts from neuroscience and apply them outside of the brain, outside of the nervous system to see what does it look like in other contexts. Again, because this is an this is an example of trying to understand an alien kind of mind that functions at different scales of time and space in a different, different kinds of problem space. So just as as as people are are doing this kind of imaging of of the nervous system, we developed the first molecular tools to read and write the, electrophysiological information in non neural cells. So for example, using voltage sensitive dyes, we can track the the dynamic electrical state of individual cells in vitro. This is a frog embryo, and you can see all the cells are having various electrical conversations with each other as it tries to, understand, how many, eyes do we need? How... Where does the head go? How many tails do we have? All all these kinds of things. So so we have ways of of of listening in to the bioelectric, states that underlie morphogenesis. We have, computational platforms where we start with the expression of ion channels, and then we can, sort of infer tissue level states and understand what's going to happen. We can do simulations. We can, try to integrate it with, connectionist ideas and machine learning where things like, inpainting and outpainting, basically, the restoration of memories from from partial stimuli and all of that, understand memories as attractors in this kind of, electrophysiological network. All of all of this rich, body of work is applicable outside the brain. Interestingly enough, while we have different departments, different journals, different funding bodies for developmental biology and neuroscience, the tools themselves don't, they they don't make the distinction. We use exactly the same tools. We use, optogenetics. We use ion channel drugs, all of the same stuff that that, neuroscientists use, and they work perfectly well outside the brain. They don't make the...

[30:29] These artificial distinctions that we make. So, I'll show you, just one ex... I'll show I'll show you one example of a of a of a pattern memory. This we call the electric face. So here's an early frog embryo. This is a time lapse. What it's doing is putting its face together. And there's a lot going on, but you can see in this one in this one frame from from this video, here's where the eye is going to go. Here's where the mouth is going to go. Here are the placodes are gonna be out out here. Basically, what you're seeing is a, a preview of a, of a of an of an informational scaffold laid out in voltage and resting potential, that tells these cells what a normal frog face is supposed to look like. And it turns out that if you change any of these patterns, you change the gene expression, then you change the the anatomy of the face. So so bioelectricity, much like in the brain, is the excitable medium that allows tissues to store set points. This is the gold state for craniofacial morphogenesis, and we can see it long before any of those processes have actually kicked in. Not only is bioelectricity this kind of cognitive glue of single cells making up an entire embryo, it works at the next level too. So you can see here, these are embryos. Each one of these is a separate embryo. And if we poke this one in the middle here, these guys find out about it. You can see the, the injury wave spreading. These guys find out about it, you know, in in in about thirty seconds. So, this is this is the kind of, this is the kind of information spreading in coordination across scales that electricity is really good at. But observing... I mean, observing all of these things is is nice, but you have to go beyond that, you have to do perturbative experiments. How do we know that any of these things are actually functionally important? May... Maybe they're all epiphenomena. How do we know they're functionally important? How do we know what the information content actually is? So, again, we use, all the same tools, of neuroscience. We do not apply electric fields. We don't have any magnets or waves or frequencies or anything like that. What we do is we manipulate the interface that cells are using to communicate with each other. Meaning, we can control the ion channels. We can open them. We can close them. We can control the gap junctions. We can open them and close them to control the topology of the network. And, but in this way, we have some degree of control over the synaptic plasticity, so the connections between cells, and the intrinsic plasticity, meaning the the the the change in voltage states of individual cells. So this is this is how we do it. And and now I'm going to show you what happens when you actually use this interface to communicate with the cells. And here's a natural example of communication. I'm gonna play this video in a minute. The colors here correspond to voltage states. So what you're seeing is a bunch of cells out here. They're all green, so they have one one particular, voltage. This cell has a different voltage. It's sort of mining its own business out here. As soon as this cell reaches out to touch it, look, Just that tiny little touch here, already the voltage is different. So look. It's it's moving along. It's moving along, and this guy is gonna reach out to touch. Boom. That's it. It turns green. So all it takes... And now and now it joins the collective, and it is going to do whatever this group of cell is going to do. So that tiny little contact is enough to basically hack this cell, change its voltage, and get it to get it to come over and and start participating in this. So this is what we would like. We would like to be able to send very convincing messages like this. We would like to be able to control cell behavior. So I'll show you a couple of examples. First, I showed you a little spot of of voltage that determines where the eyes are going to form. So we used an ion channel to establish that same voltage somewhere else in the embryo, for example, on the gut. So here you're looking at a side view. So here's a side view. The tail is out here somewhere. Here's the gut. Here's the the animal's right eye. Here's the the brain. The mouth is over here. So what we did is inject in, in the early embryo, we injected some RNA encoding a particular potassium channel. It it establishes a certain kind of spot, and the cells interpret that spot as make an eye here. They... Those eyes have lens, retina, optic nerve, all the same stuff. We don't have to tell it how to make an eye any more than I have to tell you how to rearrange your synaptic protein so that you understand what I'm saying to you. You will handle all of that. Again, because our bodies are this multiscale transduction system where high level prompts or high level goals, right, goals are your own prompt, they're prompting you, We'll eventually transduce into the chemistry that that needs to make things happen. Same thing here. We provide a very high level prompt, make an eye. Cells do do all of the rest. In fact, if we only target a few cells, another neat trick they have is they can they can, co opt other cells in their vicinity to help them. So if we only get a few of these, of these green cells, they will get all of these other cells, which we never touch. They can... They try to convince them to help because they can tell there's not enough of them to make a proper eye. There's many other collective intelligences that do this. Of course, ants and termites will scale, effort to... You know, they'll call their neigh... Their their buddies to come and and help if, if something is too big. So what's actually interesting in that system is that you can you can watch a battle of worldviews take place in this collective intelligence. If you look early on...

[35:36] So so we inject this embryo with a potassium channel, and we use, we we look for the expression of early eye genes. If you look early on, I mean, there it is, of course, in normal eye, but you see this one, two, three, four, five, you see these other regions that are going to make additional eyes, and you say, great. This thing's gonna make so many eyes. Cool. When you come back the next day, there's only two. You come back the next day, maybe there's only one. Because what's happening here is that the cells that we injected are saying to all the neighbors, okay. Let's make an eye. But the neighbors, because of the... There's a cancer suppression mechanism that's very ancient, the neighbors are saying, no. You should be like us. You should be gut or skin or muscle or whatever. And they have this battle, and depending on who's more convincing, the collective of cells would either go towards the eye fate or they will stay, then this this will disappear. They will they will be... The cells won't disappear. The the the eye markers will disappear because the cells will, decide to to do what they originally wanted to do. So what I'm showing you is that, these bioelectrical pattern memories are instructive. They they they change what happens. They are high level controls. In other words, we're not talking to individual cells. We're certainly not talking to genes. We're talking about organs. These are very high level control prompts to the collective intelligence, not to individual cells. And so now, let me show you another example of this pattern memory, and we did this in planaria. Planaria are this little little flatworm here that you can see. Here they are. Normally, they have one head, one tail. If you amputate the head and the tail, this middle part regenerates very nicely, and it makes a normal worm, one head, one tail, every single time. How does it know? How does it know how many heads to to make? For example, if you cut this if you cut this off, this wound here makes a head. This wound here makes a tail. Okay? In fact in fact, this wound on this side also is gonna make a tail, but this one will make a head. How do they know what to do? We looked we looked using our our bioelectrical imaging, and we said... We saw that there's a bioelectrical pattern that we thought might mean make one head. Here it is. And so then, basically, we we used the ion channel, drugs to change the pattern to say make two heads. It's a little messy, but, but you can see, make two heads. And that animal, if you cut it, will in fact grow two heads. So here here they are. This is not Photoshop or AI or anything like that. These are real animals. Of course, until that happens, they have one head and one tail. They have the normal molecular markers. So here are the head markers in the anterior, not in the posterior. What this is is a latent memory. We have changed the internal representation. So again, this is taken straight out of, neuroscience. We have changed the internal representation of what the correct goal state is, But that memory is latent until they get injured. Until they get cut, nothing happens. Once they get injured, then the cells consult this pattern, and then they build two heads. So the pattern here is not the map of of this animal. It's a it's it's a it's a it's a memory of a future state that it will try to implement if it gets injured in the future. Again, a counterfactual counterfactual memory. So the body can store at least two different representations of what a normal planarian should look like. So when I said that at this point there are groups of cells that know what their goal is, they actively try to implement that goal, I can call it a goal because we can change it and the cells will do something else. It is not merely descriptive. It is a set point, like on your thermostat. When you change it, the whole system will do something else because that's the whole point of a of a set point. By the way, there's an amazing, memory phenomenon here. If you take these these two headed worms and you amputate the heads, the middle fragments will continue to regenerate as two headed. So that tells you right away the information on how many heads to have. It's not in the DNA. There's nothing wrong with the DNA here. The... If you were to sequence these, you would have no idea that they have two heads. The genetic sequence is is the same. We didn't touch the genome. The genome encodes hardware that by default is, has a default memory of having one head, but it's rewritable. It's reprogrammable. Like any good memory, it is, labile. You can you can change it. You can rewrite it. And by the way, we can take the two headed worms and rewrite it to... Back to one headed. So here they are. You can see a little video, of these guys. Sometimes the two heads cooperate. Sometimes they don't. But again, there's this there's this memory, and, and and it is not is not genetic. In fact, you can go further, and if we understand that, this tissue takes a journey through anatomical space to be... To grow whatever head it's supposed to grow, we can ask, could we push could we push it to a different species target morphology? And the answer is yes. So here's a nice triangular shaped, duratocephala. If we amputate the head and we use, ion channel modulation to confuse the cells so that they're not exactly sure what the correct pattern should be, Sometimes they make a a flat head like this p. Falina. Sometimes they make a round head like an s. Mediterranean. Sometimes they make a normal head. All of this is, again, without touching the g n... The DNA, so the hardware is perfectly capable of visiting the attractors of other species that hang out there normally. If you do this, it's not just the shape of the head that changes.

[40:46] The distribution of stem cells and the shape of the brain becomes just like these other species, about a 100 to a hundred fifty million years distant from this. Okay? Again, it's a bit pure... Purely purely software change. In fact, you can make some really strange things. You can make some worms that aren't even flat at all. You can make these crazy spiky things. You can make cylinders. You can make hybrid forms, like this. And so and so there's this very interesting question of this. Can we map out this latent morphospace? Where else are these cells willing to travel? What else can they show us of the of the morphospace? We're using these things as a as a kind of exploration vehicle for exploring this space, which is certainly what biology does. It's not just for for animals. You know, plants do this too. Like, you might think that this, oak leaf is what the oak genome knows how to do. Right? So so so every single time these these, these acorns become, become oak plants, and they make this exact kind of leaf with this shape. So you think, okay. Well, that's what the genome knows how to do. And that's because development is so reliable. You know, it lulls you into a false sense of, security here because because it's so reliable, and you think that's what it knows how to do. But along comes a bioengineer. This is a nonhuman bioengineer. This is a a wasp that lays down some chemical prompts, and those chemical prompts force the cells of the plant, not of the wasp, but the plant, to build something completely different. So these are the kind of structures that the wasp embryos like to live inside, and so, evolution has found some, some prompts that caused the plant cells to build this. You would have absolutely no idea that the... That these cells could even do this by just by looking at the leaves. They give you no no clue. You have to understand that the the space of, behavior shaping prompt that that any system, can obey. Okay. So, just a couple more a couple more components, and then I'll I'll start to kind of bring this all together. So so so learning. Right? Learning is an important component of intelligence. How far down can we go? What what what can learn? We know certain kinds of brains can learn. What else can learn? It turns out that, and we could talk about learning in cells and slime molds and things. I wanna I wanna be, like, go all the way down to a to a really shocking example. Gene regulatory networks or molecular pathways. Okay? If you have the right choice of which node you're going to treat as the conditioned stimulus, as the unconditioned stimulus, as a response, you can find... You can then, find out that even molecular networks, never mind a cell or or a neural network or anything like that, just molecular networks alone, and they don't have to be large networks. The smallest network that can do Pavlovian conditioning has only four nodes. So they can do about six different kinds of learning. They can do habituation, sensitization, associative conditioning. They can count to small numbers. We're still trying to figure out what else they can do. This is already at the molecular level. And so you can you can read about you can read about all all of that here. But One of the things we're trying to do, of course, is to build devices to take advantage of this and actually train the networks in your body for applications such as drug conditioning. There are many cases where you would like your molecular networks to forget certain physiological experiences. So maybe maybe you would like to do a placebo where you take a very powerful drug that human patients can't really take for a long period of time, and you pair it like in Pavlov's experiment. You pair the very powerful drug with some kind of neutral thing that barely does anything. And then after a while, it turns out you can just use this alone. Right? That's that's the idea. That's drug conditioning. So there's all sorts of potential applications to trying to understand what is the the competency of the material that goes into into the evolutionary process. Already at the level of molecular networks, it already has certain aspects of, of cognition, such as simple forms of learning. This this goes, sort of... Kind of learning properties go up. I mean, I showed you the planarian case. Here's a here's a case in deer. So certain kinds of deer, every year, they lose this antler rack, and then they regrow it. Well, Bubenik found after about thirty five years of of experiments that if you make a damage at one point of the of the antler, next year, this whole thing falls off. The animal grows a new a new antlerac, and at that same location, a new ectopic tine will grow. Okay? Eventually of it, you know, eventually it disappears. So so here's... You've made the damage, then it then it grows. And for five or six years, it will remember where the damage was. Just try to imagine writing a molecular, mechanism kind of, a model. You know, the kind of thing that's popular is, you know, figure seven in a cell paper. A bunch of arrows, like this thing turns on that thing, you know, these molecular pathways. How would you write a molecular pathway for something like this where the whole... All the tissue drops off, goes away, and it has to somewhere in the body presumably store the information about where the damage was in this highly branched structure. But you can see you can see these kind of learning effects also in all in all sorts of aspects of morphogenesis. Okay. So so the last things I wanna show you is is around creative problem solving.

[45:51] So first, creative problem solving towards default goals, the kinds of evolutionarily understandable goal states. So so I have two favorite examples. One is, again, planaria. If we put planaria in a solution of barium chloride, barium is is a nonspecific potassium channel blocker. Their their cells in their in their head are very unhappy. Their heads basically fall off. They just sort of... They can kinda blow up and fall off. Well, in short order, the planaria regrow a new head, and the new head is completely barium insensitive. No problem. Lives in barium just fine. We looked at the at the transcriptional differences between original heads and barium, adapted heads. We said, what's different? There's only about a dozen genes that are that are up and down regulated to make this possible. But planaria never see barium in the wild. This is not something they have a built in evolutionary response for. So, this means that they just, like, visualize yourself in this nuclear reactor control room. If the thing is, melting down, you don't have a lot of time to combinatorially try one of, combinations of maybe 20,000 genes, all these switches to to see what'll happen. You have to very rapidly figure out which of those 20,000 options will solve your stressor. And by the way, you don't know what is causing the stress. You have no idea which genes to use. No problem. This thing figures it out. So there are, really interesting problem solving capabilities in physiological space and transcriptional space that we don't understand. If we had models of this, I think AI would be way further along, and I think biomedicine would be way further along. We have to understand how cells solve problems like this. Another example is, what happens if you make radical changes to the animal's parts. So this is a cross section through a, a kidney, tubule in a newt. So lots of little cells, make make this kind of structure. If you make the cells... If you if you add, you make polyploid newts, so you add genetic, copies of the genetic material, The nucleus gets bigger, the cells get bigger to accommodate. But the animal stays the same size. How is that possible? Well, when you look, it turns out that, oh, the cells are much bigger, but there's fewer number of cells that still do this. So the cell numbers automatically scales to the size of the cell. And if you make the cell so gigantic, one single cell will bend around itself to still make the same thing. So look at what's going on here. The new didn't know that you were gonna mess with the chromosome number, that you were gonna make it to have gigantic cells. It adapts on the fly, and it still makes the... It it still finds its goal of making a proper proper new body by using different molecular mechanisms. This is cell to cell communication. This is cytoskeletal bending. This is every IQ test you've ever seen. They give you, here are five objects, use them to solve this, this novel problem you haven't seen before. That's what's happening here. The cells have different affordances in the genome, and they can pick and choose those as needed to solve a wide variety of problems. So, so for the... So for the... For this last part here, I mean, I've shown you all these different types of different types of of of protocognitive competencies. Now now we get to the final one. Everything else I've shown you was in the service of goals that you could say were established by evolution. Making the right structure, repairing the right structure, surviving under novel poisonous conditions. All of those things, are set points that were given by evolution. What happens when you're a novel being? You're new in the world. You you were not given, goal states by a lengthy process of selection. What happens? Well, what doesn't happen is cells don't give up. Okay? Cells... If you can't meet your original goals, you will find new goals. So I wanna introduce you to two two new beings. The first is called xenobots, and we call them that because xenopus lavis is the name of the frog where these cells come from. Early embryo, we we scrape off some cells from the epithelial layer up here. It's going to form skin eventually. We put them off by itself. The cells, what could they do? They could die. They could crawl away from each other. They could make a nice flat monolayer like cell culture. Instead of that, what they do is, they they come together. So each one of these things is a cell here. This is a clump. This is a clump. You can see kinda cute that it looks like a little horse. They don't all look like that. I just really like this this video. But they move around, and they sort of join up like this. They have... They they have the little signaling events here. They get together, and they make this. This is a self motile little creature. It uses cilia, little tiny hairs to sort of coordinate... In a coordinated way to row against the the water. They can go in circles. They can go back and forth. They can have collective behaviors like this. You can make them into weird shapes like this swimming donut. Here's one navigating, some kind... Navigating its way down the maze. I'm not saying it it has any goal state in mind. This is just interesting behavior. It it takes the corner without having to bump into the opposite side. Then at some point, it spontaneously turns around. So you can see what happens. So it's it's it's moving along in here. It will take the it will take the turn, and then spontaneously sort of reverses and goes back where it's came from.

[50:55] If we look at the, something that neuroscientists really like to do is they like to track the computation in cells by looking at calcium signaling. Calcium is like this generic, sort of indicator of of of computational ability. So here are two xenobots, expressing a calcium, reporter. So you can see lots of calcium activity. You might even ask some questions about whether there's, like, mutual information here. Are they talking to each other? But one interesting thing is that if you apply, if you apply information theory metrics, the same kind that people apply to, neurological patients to see if they're dealing with a a bunch of cells, or whether they're dealing with a brain in which there is a human observer sort of trapped inside there. Right? Is there someone home? You know, that that kind of decision that that neurologists often have to make. So they have these they have these tools, like phi metrics and things like that. If you apply those metrics, you find out that this this... The signals in these xenobots, which I remind you, have no neurons. This is pure... This is purely skin cells, basically, epithelia. They are as different from their null models as human fMRI data are from their null models. Sorry? So so I'm not claiming that these things are like brains. I'm actually saying something different. I'm saying that both them and brains and many other things Yeah. Have a specific kind of dynamic that is associated with being more than the sum of your parts, which is basically what what a lot of neuroscientists think, cognition is. Now they have some they have some interesting capabilities. For example, if you give them loose, epithelial cells to work with, they will push them into little piles. They will polish those little piles, and those little piles become the next generation of xenobots. And guess what they do? They go and do exactly the same thing, and that makes the next generation and the next and the next. So we've made it impossible for these guys to reproduce in the normal froggy fashion, but they found another way. It's called kinematic self replication. They do many other things. It turns... They they express hundreds of genes differently than in... As xenobots than they did in in the body. Turns out they express a whole cluster of genes related to hearing and sound perception. We put a speaker under the dish. Sure enough, they respond to they respond to sound. We also found out that they can form memories. They can form at least two different kinds of memories of their experiences, which we can read out. So we can read their mind twenty four hours later and ask, which ones of them had which experience, we can tell. Because because at the level of behavior, calcium signaling and transcription, you can see the engrams. You can see the memory engrams. Again, no no neurons anywhere. Now you might think this is some weird aspect of frog embryonic cells. So I would ask you, what do you think your cells would do if I liberated them from your adult body? So I introduce you here to anthrobot. So this is... This this little guy that looks something that came out of a out of a pond somewhere. If you were to sequence him, you would see nothing but, Homo sapiens. 100 normal Homo sapiens genome. These cells are taken from the trachea of adult, human patients, usually elderly patients. They're undergoing biopsies. We buy the extra cells. They they... The cells turn themselves. We don't touch the genome. There are no synthetic biology here. No scaffolds. No weird nanomaterials. The cells have another type of being they can make, which you would have absolutely no idea about by looking at the genome. They have all kinds of interesting properties. For example, if we put human neurons in addition, put a big scratch through them, here, we put a bunch of anthrobots down. They come. They form this thing called a super bot cluster, and then what it does is it heals the the the damage. If you lift them up, here's where they were sitting. You can see four days later, they they start to heal across the gap. They have over 9,000 differentially expressed genes. They have four specific behaviors that they do that we can look at the ethogram of transition probabilities between the behaviors. So now you can start to you can start to ask these questions. What determines their transcriptome? Why do they express certain genes rather than others? Why do why do, xenobots, react to sound in a way that frog embryos don't? Why do they heal neural wounds? Why four behaviors, not 17 or one? Where where do these things come from? So so in the last five minutes, I want to address two topics, and this is, like, the most, sort of, unusual and controversial part of the of of the talk having to do with where did all this come from and where do I think it's all going. So origins. The first thing I wanna say is that, we have a story of how we paid for the the computational cost for making a good frog. So so this set of developmental stages, the behaviors of tadpoles, people say that that the way this this developed was by having the frog genome, bash against the environment for millions of years. So selection. There were a set of environments that, that that caused changes in the genome such that you get this particular animal. That's great. Where do the competencies of xenobots come from? There's never been any xenobots. There's never been selection to be a good at kinematic self replication. There's never been selection for xenobots to hear or for anthrobots to heal neural wounds. There is no stage of early human development that looks like an anthrobot.

[56:04] We don't have an ancestor that lived inside somebody else's body and healed their neural wounds. None of this ever existed. There were no selection pressures for it. Xenobots have a developmental sequence. This is an 83 old xenobot. It's becoming something. I have no idea what it's becoming. They have different behaviors. So when did we pay the computational cost of this? It is not enough to use the familiar, sort of, the the crutch of emergence. You can't just say that, well, at the same time that we selected for a really good frog, somehow we also selected for being a xenobot. No. The point of evolution is to have some tight correspondence between the series of environments that that an... That a lineage has gone through and the final outcome. That's supposed to be the whole point. You're supposed to be able to explain the properties of a creature by the series of environments that it had. What do we say for xenobots? How do we learn to predict all of this kind of stuff? So I would just point out that, in biology, we we like two sources of of order for paying for competency. There is selection, so history of certain environmental influences, and there's physics, aspects of the environment that that cause certain things to happen. But there's another source of order, and that's the truths of mathematics. So if you look at something like this, this is called a Halley plot of this very simple little little function. This entire complex structure, and it doesn't hurt that it looks very organic. I I kinda like it, is is is associated with this little this little prompt, this little seed, and the specification of this particular pattern, not just general complexity, but but the fact that the anatomy, it really is this. It's not anything else. It is exactly this. You... There is no fact of physics that explains it. There is no fact of selection or biology that explains it. There is something else going on here, which Plato and Pythagoras knew about two thousand years ago, that that, there is a space of important facts which are not physical facts. Facts you can't change by doing things in the physical universe, like tweaking the universal constants. You're not gonna change the value of e, the natural logarithm. You're not gonna change these fractals. None of the truths of mathematics are changeable, by things you do in the physical world. So there so so there is this interesting source of order that doesn't come from physics, and it doesn't come from biology slash evolution. So I'm gonna... I I could give a... You you know, we could we could talk for an hour on this. So I'm just gonna allude to this, in case anybody's interested. The the framework that I'm working on now has the following idea, that there is there is a structured space of mathematical truths, but that isn't, but that isn't the only thing that, that is there. It's not just the patterns that we study in mathematics. I think that same space contains other patterns that we recognize in developmental biology, in behavioral science, in computer science, and in physics. Well, I don't have a lot to say not being a physicist. But I think but I think all these other disciplines are basically just behavioral science of other kinds of patterns in that space. I think there are sort of the low end, sort of low low agency patterns that we call the objects of mathematics, but then there are higher agency patterns like, certain behavioral propensities, certain policies for traversing problem spaces that we recognize as kinds of minds. And I think, basically, that the mind body relationship is exactly the same as the relationship between the truths of mathematics and the facts of physics. I think that, basically, the space has a... It has a lot of useful useful structure that we need to study in other disciplines besides besides math. And I think that when we make things, be they embryos, biobots, cyborgs, machines, robots, you know, human bodies, whatever, when we make these things, what we're really making are front end thin clients. We're making embodiments for a very rich ecosystem of patterns, that are, going to, going to ingress into this functionally into this into this thing. And so I'll just I'll just finish here by pointing out that we are going to be living with a tremendous number of diverse novel beings. Okay, that every combination of evolved material, software and engineered material is some kind of agent, cyborgs and hybrids and chimeras of every description. We need to be able to recognize them, communicate with them, and enter into a kind of ethical symbiosis. So what I've told you today is that, improved communication and collaboration with the agential material of life is going to give rise to, all kinds of, useful applications in in regenerative medicine and AI and so on. But all of this is really a part of cognitive science. And that bioelectricity happens to be a good interface to all this as an example model system, but the world is much richer than this, and we need to, expand our mind blindness. And, yeah, I think I'll just stop here and thank the the students and postdocs who did all the work. Lots of lots of collaborators. I thank the funders who have supported the various work I've showed you. And the four disclosures, these are companies that have a license, some of our discoveries that I have to do. So thank you very much. I'll stop here.