Watch Episode Here
Listen to Episode Here
Show Notes
This is a 1 hour 10 minute presentation of ideas around the architecture of biology (and how it differs from today's computational systems). Focusing on morphogenesis as a model of collective intelligence, I talk about the intelligence ratchet that results from life's need to creatively interpret information on the cognitive, developmental, and evolutionary scale. I end with some speculative ideas about Platonic space and cognitive patterns therein, which radically enlarge the set of beings. Th...
CHAPTERS:
(00:00) Introduction: Body, Mind Symmetry
(05:00) Diverse Biological Intelligence Examples
(12:50) Multi-Scale Competency Architecture
(19:00) Collective Morphospace Intelligence
(24:00) Bioelectric Communication Interface
(29:30) Rewriting Anatomical Memories
(38:00) Selfhood and Collective Goals
(46:00) Unreliable Substrate, Creative Problem-Solving
(55:30) Exploring Novel Forms, Patterns
(01:06:30) Summary and Future Research
CONNECT WITH ME:
Transcript
Thank you so much for having me here, and I look forward to our discussions. Today, I'm going to have a mix of some things that I think we have pretty strong evidence for and then some very conjectural things at the end that we can all talk about.
These two links: this is my academic site. All of the papers, the datasets, the software, everything is here. And then this is a blog where I put some more speculative ideas about what I think this all means. What I'd like to do today is show you some biology that I think is relevant to intelligence and bio-inspired computing and things like that.
I always thought it was really interesting that Alan Turing, who obviously needs no introduction here, who was interested in problem-solving machines and reprogrammability and things like this, also wrote this amazing paper which was about the origin of order in chemicals during morphogenesis. I think that what he was onto is extremely profound. It's a deep symmetry between the self-creation of bodies and the self-creation of minds. I think that those two things are very tightly linked, and we learn a lot about both of them by bouncing them off of each other.
What I'd like to do today is basically four things. I'm going to show you some unconventional examples of intelligence in biology, which I think argues for a view of mind all the way down. We can discuss how "down all the way" means, but basically, I want to show you some biology that's different than the kind of biology that normally underwrites thoughts in cognitive science, in bio-inspired computing, and so on.
I want to tell you about some recent progress that we've made with actually communicating with this agential material. I'm going to argue that it's not just a computational matter or active matter; it is actually an agential material, and we now can access one of several interfaces for communicating with it. Then I want to talk about why that works and what I think is a driving force of this process and the kind of creative aspects of the architecture of life. And then at the end, I'll throw around some ideas about what I think that might mean.
This is a well-known piece of art. Here Adam is naming the animals in the Garden of Eden. I think there's something about this that is very rate-limiting for progress in the field, but there's something else that I think is actually very deep and correct. The thing that's rate-limiting here is that in this system, what's pictured is a very fixed idea of natural kinds. It's very clear here that you have very specific, discrete animals. And then you have Adam here who is actually different from them, and it's supposed to be sort of very objective what everything is. I think that's a fundamental problem that we're going to have to get around.
What's good here, though, is that in the original story, it was on Adam to name the animals. God wasn't going to do it. The angels couldn't do it. It was actually on Adam to name the animals. What is cool about that is that in ancient traditions, naming something means you've discovered its inner nature. And it was on Adam to do this because he was the one that was going to have to live with these creatures.
I think that's very deep because what we now understand is that both evolutionarily and developmentally, this thing that we normally take in, let's say, philosophy of mind discussions as a normal, adult human, is actually at the center of a very continuous set of transformative processes where we all start life as single cells, and then there are various things that lead up to this. Whatever agential glow you think the human has, you have to have some sort of a story about how it shows up and where it shows up and so on.
But not just the natural story. There's also this axis that tells us that we can start making modifications, as we already have, both technologically and biologically, and then we end up with some other things where it's going to be quite difficult to say whether what we have is a human or not. That tells us right away that we're dealing with continuous variables, not hard categories.
What I'm interested in is developing a framework that lets us recognize, create, and relate to truly diverse intelligences, regardless of what their composition or origin story is. I'm interested in unification. I want to understand what all of these things have in common: the familiar creatures, colonial organisms, swarms, new bioengineered lifeforms, AIs whether software or robotic, and maybe someday exobiological agents.
Obviously, I'm not the first person to try for something like this. Here's Weiner, Rosenblith, and Bigelow back in the 1940s, trying for something like that. And then this is kind of my recent exposition of how I think about these things. But the most important thing about this framework is that it needs to move experimental work forward. Not just philosophy, although philosophy I think is important. It needs to actually impact things like biomedicine, synthetic bioengineering, and so on, with new capabilities, and I think it should be leading to improved ethical frameworks.
For the talk today, I'm going to break it up into three parts. I'm going to show you some unique features of the biology, and we'll talk about how it works and what it means.
Here's one example. You're looking at a tadpole of the frog Xenopus laevis. Here is the brain, it's got some nostrils, here's the mouth, here's the tail, the gut. What you'll notice is that we prevented the primary eyes from forming. But what we did do is put an eye on its tail. This is an animal with a radically different sensory motor architecture from normal tadpoles. If you track the optic nerve, here it comes out like this. Sometimes it synapses on the spinal cord here, sometimes on the gut, sometimes nowhere at all. But if you make a machine to train them in visual assays, as we have done, what you find out is that these animals can see. The machine trains them to respond appropriately to visual cues, and it turns out they can see.
Now, this is fairly remarkable. Why is there no evolutionary adaptation needed here? Why don't you need new rounds of mutation and selection when you suddenly mix up all the circuitry? These eyes are not connected to the brain. They're putting their information, at best, on this sort of centralized bus. Why does this work out of the box? I'm going to make the argument that this is because the standard tadpole never assumed where the eyes or the brain was going to be either. I think this is an active problem-solving process.
Here's another example. These are planaria. We'll talk more about them later. These little flatworms, they've got a brain, they've got a central nervous system, and they've got some interesting properties. They're the only animal in which you can study learning and brain regeneration in the same animal. What you can do is train them to recognize these little bumpy spots as their safe area where they're going to eat. So this is place conditioning.
Then with these worms, you can cut them into pieces. Every piece gives rise to a normal worm. So you take these trained animals, you cut off their heads with their brain, you have a tail. The tail sits there doing nothing for about eight or nine days. Eventually, it grows back a brand new brain from scratch, and then you find out that these animals actually remember the original information. You can read more about that here.
What you've got here is two things. One is the storage of learned information outside the brain. But I think even more excitingly, what you have is the ability of that tissue to imprint the information onto the new brain as the brain develops, because in the absence of that, there's no behavior. So you're watching information. There are two things going on here. One is information moving across the body, and the other thing you're seeing is a kind of interplay between learned information and behavioral information and morphogenetic information, or the kind of memory you need to actually grow back the correct head. We'll talk more about that.
Then we have some interesting instances of what look like a virtualization. This is not AI or Photoshop or anything like that. This is a real fly. What it's doing is running a kind of stripped-down, two-dimensional morphogenetic program on its wings. These are ants. The reason it does this is because when it moves the wings around, predators who don't want to deal with stinging ants stay away from it. So it's like a protective kind of thing. But it's kind of remarkable that it's able to just run this coarse-grained morphogenetic ant program on the wings.
So that's wild, but certainly not more impressive than the actual fly itself. So now we have to start thinking about where did the rest of this come from, the high, the three-dimensional high-fidelity version. As we'll get into the development momentarily, I just want to point out that some of what we see in biology has capabilities that you don't know just by looking at it.
All over the world, there are tadpoles turning into frogs. In order to do that, it has to rearrange its face. It has to move the eyes forward and the mouth and the nostrils. Everything has to kind of rearrange. If you were to watch that, you could think that this is just a hardwired process. Somehow the genetics makes every organ move in the right direction, the right amount, and then you get your frog.
We actually decided to test that, and we created these, what we call Picasso tadpoles. We scrambled all the organs like a Mr. Potato Head doll. We put the eye on the back of the head. The mouth is off to the side. Everything is scrambled. What you get from that is actually quite normal frogs, because all of these things will continuously move in novel paths relative to each other to give you a normal frog. So the genetics does not actually specify a bunch of hardwired rearrangements. What it gives you is a system that can do a kind of error minimization scheme, and it can get to where it's going from different starting configurations. We will talk a lot about that.
There is also this interesting phenomenon. By the way, most of what I'm telling you, you've probably not seen in standard reviews of biology and so on. These are all, I think, some interesting cases that are informative for things that are not captured by the current paradigms.
This is a phenomenon known as trophic memory. Deer, every year, grow these antlers and then they shed them. One thing that this researcher named Bubenik found over 40 years of doing these terribly difficult, lengthy experiments, maintaining deer herds in Canada and doing these multi-decade experiments, is that if you actually one year make a little notch, then the whole antler falls off. Then months later, they regrow a new one, and what you'll see is that it's basically the same pattern, and you'll get an ectopic tine at the location where you made the injury. Eventually, some years later, it'll go away.
So this is information of injury that remains somewhere in the body of the deer and then, months later, is re-expressed and forces new growth. We actually have all the original antlers because Bubenik retired and needed to do something with them, and he sent them all to us. So we had all these things CAT scanned so you can start to actually study this process. But you can already start to think that it would be pretty challenging to come up with a typical molecular biology explanation of how the location, the three-dimensional location of injury, is stored within the stem cells of the scalp, and then eventually when the growth happens, you get to this point and they say, "Oh, by the way, take a left turn and do an extra tine here."
This plasticity, this ability to deal with novelty, this kind of morphogenetic memory, is really critical to understand what's going on in biology because morphogenesis itself is incredibly reliable. We see this again and again. We see acorns turning into oak leaves, and we think that this is what the oak genome has learned to do. This is what it encodes, which is how a lot of people talk, that this thing encodes this kind of shape.
What you wouldn't know unless we had the benefit of this little non-human bioengineer, this little wasp, is that with appropriate signals from the wasp, those exact same cells which reliably, billions of times every year, make exactly the same flat green kind of structure, these same cells are actually capable of building this or this or many other kinds of incredible structures. This is not the insect cells making this. These are the plant cells hacked by signals from the wasp to build something that is completely different than what it does by default. And we would never know that they're capable. So there's a lot of this kind of reprogrammability and plasticity.
Now, I'd like to talk about why these amazing examples work. We're going to talk about this idea of the multi-scale competency architecture and the plasticity of boundaries between agents and how we communicate with what I think is the intelligence that underlies all this.
This is the basic life cycle that all of us have gone through. We started as a single cell, which people will often say is quote-unquote "just physics." So it's a quiescent oocyte. It's got some chemicals in it. You might look at it and say that there is no intelligence there. It's just a little blob of chemicals. But then there's this remarkable process of embryogenesis, which leads to being one of these things or maybe even something like this, which is going to then make statements about not being a machine and so on.
So, what we really need to understand is how you got from here to here because development offers no special point in which you tick over from chemistry and physics into mind, psychology, psychoanalysis, and whatever. This is a very slow, gradual process of development. There is no bright line. So that means we really need to understand this transformative process that scales us up from just chemistry and physics.
By the way, this is not even the end of the story. I'm going to talk a little bit about what can happen after that, which is some breakdowns of the collective intelligence known as cancer and also some radical transformations into anthrobots, which can happen even after the patient is deceased. So we will talk about all of that.
This is the kind of thing that we are made of. The basic subunit is something like this. This is a single cell. This one happens to be a free-living one called the Lacrymaria. It's one cell. There's no brain. There's no nervous system. Everything that it's doing here as it hunts for its food is handled within one cell. So we are already made from a very sophisticated material that has a lot of competencies and agendas on its own.
In fact, you can even go below that. We recently found that the gene regulatory network models, so not even cells, not even groups of cells, certainly not neurons or brains or any of that, but just gene regulatory networks, which are models of genes turning each other on and off, if you do the experiment and you actually train them in various paradigms, you can find six different learning capacities, six different kinds of memory, including Pavlovian conditioning. We are now building some devices to actually do this for applications like drug conditioning and other things. So not only are the cells that we are made of quite competent at all kinds of things, but actually the material itself, the molecular networks inside of them, already have learning capacity.
But some people will argue, at least we are true unified intelligences, right? We have this nice big brain. And it's a single, it's a singular organ, with maybe at least we're different. We're not just, we're not like ant colonies and beehives and things. And it's just a metaphor when people call these things liquid brains and so on, right?
Well, Descartes thought so. And he really liked, in particular, the pineal gland because there's only one of them in the brain. And so he thought that was a reasonable place to centralize the human experience, which is kind of unified most of the time. But if he had access to good microscopy, he would have looked at the pineal gland and realized that there's not one of anything because inside the pineal gland is all of this stuff. And inside of each one of these cells is all of this stuff.
So, we are all collective intelligence. We are all made of components, made of pieces. And whatever we have is the result of some kind of alignment of the competencies of our parts. And this is what we need to understand.
When I talk about intelligence, obviously there are lots of different definitions. I don't claim this to be the best one or the only one. But I like William James' definition: some degree of the ability to reach the same goal by different means. It's a very cybernetic definition. He's not talking about brains. He's not saying what kind of goals, what problem space. But basically, it's a kind of navigational competency to reach your goal when things are different.
It turns out that what we are made of is this multi-scale competency architecture where we're not just nested structurally, but actually every layer solves problems in different spaces. What kind of spaces? Well, we, as humans, evolved over time in our environments. We're reasonably good at noticing the intelligence of middle, sort of medium-sized objects moving at medium speeds in three-dimensional space, right? And so dogs and octopuses maybe and birds and things like that. We're okay with that.
But we're actually really bad at recognizing similar kinds of navigational skills in other spaces. For example, the space of possible gene expression, the space of possible physiological states, and most of all, we're going to talk about this, the space of anatomical possibilities. I tend to think that if we had evolved with a primary sense of the blood chemistry of our bodies, for example, so let's say, another ten different sensors that could look inwards and tell us the physiological states of our body, I think we would have no trouble recognizing our liver and our kidneys as this kind of autonomous symbiont that navigates these spaces, is intelligent, and helps keep us alive by the decisions that it makes. But these are hard for us to recognize.
I will also point out that once you start thinking in this direction, we can realize that actually this perception-action loop, which a lot of workers in robotics, in AI, claim that it's really important to have embodiment and to have this feedback between the agent and the environment, I'll just point out that a lot of things that we think of as not embodied are actually perfectly embodied because they're just carrying out this loop in other spaces that are not obvious to us. So paying attention to these other spaces in which engineered agents are actually working in is, I think, really important. Embodiment, I don't think, is what we typically take it to be.
So let's talk about the intelligence of the cellular collective as it navigates anatomical morphospace. What I'm going to claim is that groups of cells form a collective intelligence, and their behavior plays out in morphospace. It plays out in the space of anatomical possibilities. So just like we are a collective intelligence of neurons which allows us to navigate through the three-dimensional space and linguistic space and other things, all of that started when cells were working together to navigate anatomical space.
Now, why do I call it intelligent? Not because it's reliable and not because there's a massive increase in complexity from this stage to this stage. That's not it. It's not about reliability and it's not about the complexity. It's about the creative problem-solving capacities.
The first thing you see, and I already mentioned an example of this in that frog face, is that many kinds of embryos, including mammals like us, you can cut them into pieces as young embryos, and you don't get half bodies. You get perfectly normal monozygotic twins and triplets. So that means they're navigating the anatomical space to get to their goal state, the sort of ensemble of states that represents a normal human target morphology. They can get there from different starting positions. That's interesting. If you chop off half each side, immediately recognizes what's missing and will get to where it needs to go in that space.
Some animals can do this throughout their lifespan. This is an axolotl, and these guys regenerate their limbs, their eyes, their jaws, portions of their heart and brain. You can see right away that when they lose portions of the limb, and they bite off each other's legs fairly frequently, when they lose a portion of the leg, the cells will build exactly what they need to build and then they stop. That's the most important thing about regeneration is that it knows when to stop. When does it stop? It stops when the correct salamander limb has been completed. So not only does it do exactly what it needs to reduce the error between this state and that state, but it recognizes when it's done to some tolerance of comparison and then it stops.
All of our bodies are this ship of Theseus. This old philosophical puzzle of continuously replacing the planks in a ship and when is it, how, when is it still the same ship and when isn't it? Because all of our cells continuously turn over and the material within our cells continuously turns over. So we are not a stable object. We are a continuous construction project. And you can think about the actual ship of our body as not the material. It's not the physical self. The ship is actually a model in the minds of the replacement machinery that guides their activities. That's the actual ship, right? The actual ship is the model that they're all following to know what to do next and where the pieces go.
In fact, in some species such as in planaria, especially in these asexual strains of planaria, that project can last forever. So planaria do not age. These asexual strains do not have any evidence of senescence. They just sort of go on forever replacing their cells.
I want to look at two issues here. First, can we rewrite this plan? So this machinery, meaning the cells inside the body, have a plan of what they're constantly upkeeping. And here, as I said, for very long periods of time. Can we rewrite that plan? What would it take to rewrite that plan? And then we'll talk about how the cells actually align.
So the first thing to think about is that collective intelligence, such as the cells that make up something like this, it needs something I call a cognitive glue. And there's a variety of cognitive glue mechanisms, including stress sharing and some other things, but we're going to talk about bioelectrics.
So here's a rat. The rat presses a lever, gets a reward. But notice that no individual cell has had both experiences. The cells at the bottom of the feet interact with the lever. The cells in the gut get the delicious sugar. There is no single cell that had both experiences. Who owns the associative memory? Well, it's the rat. And the reason that there is a rat is because there is a process, electrophysiology, which binds all of these cells together into a higher level agent that can know things that the individual parts don't know. So it has goals, preferences, memories, and other things that the individual parts don't have. That's what binds it together into a higher level intelligence.
We know something about how this works in traditional intelligence and behavior. You've got some hardware which is the brain and central nervous system. The way that works is you have some cells in a network. The cells have ion channels which are these proteins that allow charges to go in and out, and that sets up a voltage, and that may or may not get propagated through the network through these electrical synapses. So that's the machinery. And that gives rise to some electrophysiological dynamics, which you can think of as the software that runs on this hardware. I'm certainly not claiming that our current paradigms of writing software are how this works at all. But anyway, the idea is that there's massive reprogrammability here and plasticity and so on.
The way that this system works is that it issues commands to your muscles to move you through three-dimensional space. And neuroscientists have this project of neural decoding, where they try to read out the electrophysiology, decode it, and try to extract the cognitive content of your mind. So the commitment of neuroscientists is that if we could just understand how the encoding works, we could retrieve your memories, your goals, your preferences, all of that stuff. We could retrieve that from the electrophysiology because that is where it's encoded. And in fact, write in new memories, as people have done in mice, incept false memories into these animals.
Well, it turns out that this amazing system is not new to brains and neurons. It is incredibly ancient. In fact, it was discovered around the time of bacterial biofilms. Evolution was already doing this when bacteria were making biofilms. And every cell in your body has these ion channels. Most cells have these gap junctions, these electrical synapses. You have exactly the same parallel kind of scenario.
But instead of issuing commands to your muscles to move you through three-dimensional space, what this system was doing, and you can ask, okay, what, before there were any brains and neurons, before we could move around, what were these electrical networks thinking about? They couldn't be thinking about motion. What were they thinking about? Well, they were actually thinking about traversing anatomical morphospace. And what they do is they issue commands to all of your cells to change and move the configuration of your body through morphospace, so that you can make the journey from being a single-celled fertilized egg to whatever we end up being.
So what we've been doing over the years is developing a very parallel research program, basically neuroscience beyond neurons, to try to learn to decode this somatic electrophysiology, understand what it's encoding, and try to read the mind of the anatomical decisions that the body makes. Again, this is not a fixed hardwired process. As I have shown you many examples, there are lots of decisions to be made, because you need to be able to reach these outcomes despite all kinds of novelty. And I'll show you more momentarily.
We've developed some tools to directly observe the bioelectrical states that go on in these tissues. This is an early frog embryo. These are some explanted cells deciding whether they're going to stay together or leave. The colors represent voltage, so this is just like imaging a brain. This is done with voltage-sensitive fluorescent dyes or genetically encoded reporters. So we can now read the electrical states of these living systems, which before was not possible because you'd have to poke every cell with a separate electrophysiological electrode. So now we can get these amazing time-lapse movies.
We do a lot of computational modeling, so different scales of models from the molecular biology of these ion channels, to the tissue electrophysiology and how the patterns spread over time, and properties like pattern completion and the various attractors in the state space of this physiological network and so on. Here's what these patterns actually look like. I'm going to show you just two phenomena so you can see what this is.
This is a frog embryo putting its face together, and you see there's a lot of complicated things happening. But if you look at one frame out of this video, we call this the electric face because it actually looks like a face. It's telling the cells exactly where the organs are going to be. Here is the mouth. Here's the animal's right eye. The left eye comes in shortly thereafter. Here are the various placodes off to the side. This bioelectric pattern is what determines the gene expression and downstream anatomy of the face. You can actually read out the plan of what it is going to do later by tracking the electrophysiology.
Not only does this set of bioelectrics dictate the global order within a single embryo, it turns out that there are similar phenomena that function across embryos, so multiscale. Here, we poke this one here, and you can see this calcium wave propagating where these two guys, they're not even touching. There's salt water between them. They find out about what's happening here because these waves are propagated, and you can see this. You can see it here. There's a communication at this level just like there is here, and here you can see the slow bioelectrical changes, and then the more rapid kinds of things that go on in a few of the neurons here.
So watching and tracking these things is all well and good. But more importantly, you've got to do perturbational experiments. That's because you can't judge the intelligence or learning capacity or anything else of a system just by observation. You have to do perturbational experiments to see how it pursues its goals under various circumstances.
So what we did was adapt the tools of neuroscience to rewrite these bioelectrical pattern memories. We don't use magnets. There are no electrodes. There are no applied electromagnetic fields. There are no frequencies. What we are doing is exactly what neuroscience does, which is interact with them through the interface that these cells are normally exposing to each other, which is the ion channels and the gap junctions on the surface. So we can open and close these things. We can use optogenetics. We can use drugs. We can use molecular biology to control the patterns, the spatial patterns of bioelectric state, which do for the somatic decision-making machinery what brain bioelectric patterns do for our behavior.
So this is the communication interface that we are trying to hack, and that's the only reason that the things I'm about to show you actually work. It's not because we're that clever. It's because this system is exposing a very powerful API to us to enable us to send these kinds of signals.
I'm just going to show you two examples of what the material is capable of. One is that I showed you that electric face. And when you look at the electric face, there's a little spot of depolarization that says, "Build an eye here." So we wanted to know, what happens if you produce that signal somewhere else, let's say on the gut or in the tail?
We inject some ion channel mRNA, which encodes a few potassium channels. They set up that little voltage spot, and sure enough, the cells get the message. They build an eye. So here's an eye sitting on the gut. If you section them, these eyes have all the lens, retina, optic nerve, all the stuff they're supposed to have.
Now, we've learned a few things. First of all, we've learned that the architecture is... Well, first we learned that the bioelectricity is actually instructive. So by giving it a specific stimulus, we actually had it make a new organ. Not just screw up what normally happens, but actually create a good new organ. So it's instructive.
Number two, the architecture is incredibly modular. The eye is a very complex organ, dozens of different tissue types. We did not... We don't know how to make an eye. We didn't say, "Do this with the stem cells," or, "Position the retina this way or that way." We didn't do any of that. We gave it one very simple top-level subroutine call which says, "Build an eye here," and then the system did everything else.
Notice that that's exactly what happens in neuroscience. When you give somebody a message, for example, if you have language or some other behavioral example, you don't have to worry about them going in and micromanaging the synaptic weights in their brain. They do all that for you. You're providing a very simple signal. It rearranges its own lower levels of molecular biology to make it work.
The other thing that's kind of cool is that if you... This is a lens sitting out in the flank of a tadpole somewhere. If we only inject a few cells, the blue ones, what you find is that the eye itself is actually made of a bunch of cells that we never injected. It's these guys that can realize there's not enough of them to build a whole eye, so they recruit their neighbors. So there's a little communication tug-of-war going on here. While the neighbors... This is a cancer suppression mechanism. The neighbors are all saying, "Your voltage is wrong. Change it. You should be skin." These guys are saying, "No, no. Our voltage pattern is that of an eye, and you should help us make an eye." And so this goes back and forth, and sometimes the skin wins, and sometimes the, or whatever organ. Sometimes the eye wins. But here's what you get. It's sort of a battle of commitment, of which path down in morphospace are we going to go? And then they all sort of make a decision, and they all go together. And we know other collective intelligences that recruit their neighbors to handle a larger task when need be.
So that's the first thing I wanted to show you, that we can actually now communicate through that interface to control its behavior in anatomical space, and make it build whole organs, and exploit competencies such as recruitment and so on. We didn't have to teach it to do that. It already does.
I'll show you one other example of this, which is how to rewrite these anatomical pattern memories. So I told you earlier that when you cut a planarian, let's say we're going to cut off the head and the tail and take this middle fragment. They are incredibly reliable in building a one-headed worm. You might ask, "How do they... How does this piece know?" So this piece has two wounds. How does this piece know how many heads it's supposed to have and where the head goes?
It turns out if you look at these animals, there's a voltage gradient that you can see that actually is interpretable to... And you could see one head, one tail. And the molecular biology shows you here are the anterior markers are up in the head, and when you cut them, that's what happens.
So what we were able to do is to change this bioelectrical pattern by exposing these animals to specific ion channel-modifying drugs. With the use of a computational model, you can say, "Okay, which drug, which channels do I need to open and close to change this?" And so then you can make something like this, which has this pattern that says, "Actually, no, two heads, one at each end." It's a little bit messy. The control of this is still being worked out. But what happens there is that you can then make these two-headed animals. Again, not AI, not Photoshop. These are real live animals.
Here's the most important thing. This bioelectrical map is not a map of this two-headed animal. This is a map of this perfectly normal-looking, one-headed animal, which, by the way, not only the anatomy is normal, but the molecular biology is normal too. The anterior markers are up in the head. They are not in the tail. So anatomically, there's nothing wrong with this animal. Structurally, the hardware is completely fine. Genetically, everything's fine. We haven't touched the genome whatsoever. But it has this weird memory. The collective intelligence of the cells are storing an altered version of what we are going to do if we get injured in the future. It's actually, I think, the beginning of counterfactual memory, that amazing ability of brains to think about things that are not true right now, either the past or the future, but not true right now.
So here's what you have here. We can put in a pattern that is not true right now. And that's okay. It's a latent memory. Nothing happens until you cut them. And when you cut them, this is what guides... This is the state towards which they build, and when they... And they stop when they achieve it, and that's how you get these two-headed worms. So the normal body of a planarian can store at least two different representations of what a correct planarian should look like. So we have now this material that is able to store memories of what it's going to do in the future, and those memories are readable, and they are rewriteable.
The reason I call it a memory is because if you take these two-headed animals and you continue to cut them, no more drugs or any manipulation of any kind, just continue to cut them in plain water, you will continue to have two-headed animals, because the material has a memory, and once you've changed that pattern, it holds. Again, keep in mind that there is nothing genetically wrong with these. We haven't touched the genome. And so the question of where the number of heads is encoded is kind of tricky. What the genetics encodes is a machine that by default reaches a pattern that says, "One head." So that's kind of the default, instinctual pattern. But it can be overridden by experience, by only a few hours of physiological experience with these drugs. And after that, the memory changes and then it holds.
Here you can see these two-headed guys hanging out. It has all the properties of memory. It's long-term stable, it's rewriteable. I showed you conditional recall, and it has distinct behaviors. And we can actually now take the two heads and turn them back into one head. So we can flip it back and forth.
Interestingly enough, it's not just about the number of heads, it's also about the shape of the head. Because, for example, this guy with a triangular, very characteristic species with a triangular little head, can be, if we amputate the head and then confuse the cells with a gap junction blocker, they are able to actually make flat heads, like a P. felina. They can make round heads, like an S. mediterranea. Or of course, they can make their normal heads. This is not just about head shape, but also the shape of the brain becomes just like these other species, about 100 to 150 million years of evolutionary distance, right? So the exact same hardware, no genetic change, the exact same hardware is perfectly willing to visit other attractors in the anatomical state space that belong to these other species. That's where they normally live. This one normally lives somewhere else. But it can actually visit these other attractors if it wanted to.
We can go much further and we can make things that don't look like planaria at all. We can make these crazy spiky things, these kind of cylinders, hybrid forms. So the latent space of possibilities for this genome, much like what I showed you with that plant gall on the oak leaves, the capabilities of the material is far, far beyond the reliable version that you normally see.
So the last piece of this that I want to talk about is the creation of the self and the plastic borders between the self and the outside world. Specifically, if we are going to be a collective intelligence where the pieces are somehow aligned towards specific goals, and I've shown you how they store the goals, and I've shown you how we can rewrite some of those goals. But how does the alignment work? How do we get all the pieces? The individual cells have no idea what a head is or what an eye is or any of that. It's the collective that knows. So how do the cells align towards this?
I want to come back to embryonic development, and let's just look at an early blastoderm here, this early kind of thin layer of embryonic cells. We look at that and we say, "There's an embryo. There's one embryo." But what is there actually one of? There are maybe 100,000 cells. What are we counting when we say there's one embryo? And what we're counting is alignment of goals. We're counting the fact that all of these cells, under normal circumstances, all of these cells have bought into the exact same plan of where they're going to go in anatomical morphospace. They are all going to work together to build this.
The only reason we are an "I" as opposed to a "we" is because of these gap junctions that keep all of these guys connected into an electrical, electrochemical network. Because if you take a little needle, and I used to do this as a grad student in duck embryos, and make little scratches in this blastoderm for the four hours or however long it takes for these guys to heal back up, each of these little islands can't feel the presence of the other. And what it does is self-organize an embryo, and then eventually they heal and you get these conjoined twins and triplets and so on. Each of these things can be its own embryo.
So there are a couple of interesting issues here. First of all, how many selves, how many individuals are actually inside of a single embryo? Well, the genetics don't set that. It's an excitable medium that can self-organize multiple individuals. It can be anywhere from zero to probably half a dozen or more. Then you have this issue that in a case like this, once they heal up, every cell is some other cell's environment. How do they set the borders, right? Where do I end and somebody else begins? Where is the outside world?
So what this is reminding us is that this material, these agents, self-organize from the very beginning. They have to... The first thing they have to figure out is, "How do I align my parts to have a common goal?" And the next thing is, "How do I set the borders between me and the outside world so that I can work on goals that are scaled to the collective, at the expense of other things that happen in the world?" This actually has lots of interesting parallels that we could talk about to issues in psychiatry and cognitive disorders, because there are various kinds of dissociative disorders that introduce multiple selves in the same material, in the same biological hardware.
So one of the ways I think about these things is with something I call the cognitive light cone, which is the size of the biggest goal that a system can work towards. So individual cells have little tiny cognitive light cones. They only care about metrics right inside that single cell. So spatially very small. They have a little bit of anticipation capacity, a little bit of memory going back, but they remember tiny things like what pH should I be at? What's my hunger level? These are some very, very small kinds of goals. But the collectives can have these grandiose construction projects.
So here's a single cell with its tiny cognitive light cone, but evolution and development allows them to work together, and individual cells have no idea what a finger is, but the collective absolutely knows because if you damage it, they will work to get to the same thing and the right final pattern.
Now that electrochemical network, the network that allows the system to store and pursue bigger goals than these systems can, that has a failure mode, and that failure mode is known as cancer. They can break down. When cells disconnect, electrically disconnect from the rest of the network, they can no longer remember this massive construction project they were working on. Basically, they roll back to their ancient evolutionary self, and they become like amoebas. This is human glioblastoma. At that point, these cells, they're not more selfish than other cells. They just have smaller selves. All they're working on are little tiny goals at the level of single cells, which is proliferation and migration, which is metastasis.
Just to show you kind of a practical application of those ideas, if you inject human oncogenes into a frog embryo, it will make a tumor. With voltage imaging, you can tell ahead of time where the tumor is going to be. Here are the cells that are already disconnecting from this electrical network, so their voltage is all wonky. And then instead of chemotherapy, instead of trying to kill these cells, if you forcibly reconnect them to the electrical network by injecting an ion channel that sets their voltage correctly so that they stay part of the network, here's the oncoprotein. It's blazingly strongly expressed. Same animal, no tumor, because it's not the genetics or the hardware problem that's going to drive this. It's actually the decision-making of the collective.
So in some cases, some of these hardware problems, this genetic mutation, can actually be ameliorated, not by removing the cells, not by fixing the genetics, but actually by convincing the cells that they are all part of one collective and keep working on the nice skin and muscle and everything else.
So what I think is happening here is that evolution is pivoting some of the same competencies, this idea where you have multiple subunits. The subunits are themselves not only active, but they have goal-directed competencies. They're able to achieve either simple homeostatic or homeodynamic ends. But what the collective is doing is aligning them so that while they're working on their local problems, actually together it ends up solving problems in other spaces.
Evolution has pivoted this across first metabolic spaces and then physiological spaces, and then genes came along and it became transcriptional space, and then eventually multicellularity with anatomical morphospace, and eventually nerve and muscle came on the scene, and so you got behavioral and then linguistic, and who knows what else is after that.
So the last two things I want to talk about is, first of all, why does this work? Why are these things progressively being scaled up into higher level spaces with bigger cognitive light cones? Why is that actually happening? In order to do this, I want to talk about the unreliable substrate and the commitment that biology has to creative problem-solving.
Let's just start with this example. Caterpillars are a soft-bodied creature. They crawl around and they eat leaves, and they have a brain that's suitable for that purpose. Now, when caterpillars become butterflies, in order to do that and become this kind of creature, they have to undergo a process of metamorphosis, which basically causes them to completely refactor their brain. Most of the cells die. All the connections are broken. They completely refactor it, build a new brain that's suitable for this kind of lifestyle.
If you train the caterpillars, and that's all described here, the butterflies still remember the original information. For example, this is the early work of Doug Blakiston, linking a particular color disc, a stimulus, to go find your leaves. Now, one problem is how do you keep information when the substrate is being completely taken apart and refactored? That's fascinating.
But even more interesting is the fact that the actual memories of the caterpillar are of no use to the butterfly. The butterfly doesn't move the way the caterpillar moves. The butterfly lives in a three-dimensional world. It also doesn't care about leaves. It wants nectar. So you can't just keep the memories. You have to actually remap them onto a completely new architecture, and that means you have to generalize, and that means you have to not just hold onto the fidelity of the information, but you have to actually remap them for continued saliency.
This is kind of a paradox of change that was seen by evolutionary biologists. If you're a species that refuses to change, that does not change, you will eventually die out when you fail to meet the needs. Conversely, if you do change, then you're not the same species anymore, so again, you're gone. So what does that say about us as a continued process, right? This idea that we cannot remain the same. We are not a fixed object that's trying to maintain itself. We are continuously trying to remap ourselves onto new scenarios, and that's all described here.
One way to start to think about this. First, I'll do the cognitive version first and then the morphogenetic version. The cognitive version is that at any given moment, none of us have access to our past. What we have access to are engrams, memory traces that were formed in our body or brain by past experience. And at any given moment, our job as active agents, which are sort of future-facing and needing to make decisions, our job is to interpret the messages left by past selves.
So this is a view of memory as communication, and all messages need to be interpreted. And we don't have to interpret them. In fact, we can't guarantee to interpret them the way that our past self interpreted it because we don't know what that was. Everything will be different. Now a butterfly is a drastic case. For us, the change is less drastic, but over puberty and those kinds of things, we actually do undergo some massive changes.
So what we have here is this kind of architecture, which you will all recognize immediately, which is that there are experiences that get encoded, compressed into engrams, and then on the other side, the past self is doing this. Future selves or our current self has to pick up these compressed engrams and reinflate them into whatever reasonable story about ourselves, about our external environment, is going to be best suited for us now adaptively. So this part is probably algorithmic and deductive. This part is creative because you've lost information here. There is no algorithmic way to do this.
So the biology has to commit to an unreliable substrate. You know that everything is going to change. The environment is going to change, but even you're going to change because as a lineage, as an evolutionary lineage, you will be mutated. You cannot assume anything.
Now remember the very first thing I showed you. Why does that eye on the back of the tail, why does that work? Because in all of these cases, I think that biology is making problem-solving agents, not fixed solutions, and it's always ready to reinterpret the information it has. So this is the cognitive version of this, of reinterpreting our memories.
Here's the morphogenetic version. Imagine the amazing problem that biology faces. So here, this is a cross-section through a kidney tubule of a newt. Normally, there are about eight to 10 cells that work together to form this thing. But one thing you can do is you can make early newt embryos with extra numbers of the genetic material, right? So these are called polyploid newts. Instead of 2n, they can have 4n, 6n, and so on.
When you do this, first thing you find out is that you still get a normal newt. Amazing. So I guess you can have extra copies of your chromosomal material and it's fine. Then you find out that the cells get bigger to accommodate the new large nuclei, but the newt stays the same size. Well, how can that be? Well, that's because it automatically adjusts the number of larger cells to the same structure. And then the real kicker is that when you make these cells completely gigantic, you find that only one cell will wrap around itself and give you the same exact structure.
The reason that that's amazing is that this is a different molecular mechanism. This is cell-to-cell communication. This is cytoskeletal bending. So think about what you're facing as an embryo, in this case, a newt, coming into the world. What can you count on? Well, you can't really count on the environment. You know that's going to change, so you have to have all kinds of physiological systems to make up for different environments. But you can't even count on your own parts. You have no idea how many copies of your genome you're going to have. You have no idea what the size of your cells are going to be or the number of your cells. You have to make it work with different affordances, different molecular components that you have, right? These mechanisms or these. You're going to have to make a normal newt no matter what under a very large set of different perturbations.
Being able to use the tools you have in novel circumstances that you've never seen before to solve a problem is literally intelligence. It's a kind of, in anatomical space, it's a kind of creative problem-solving. Josh Bongard had something back in 2006, I believe, around robotics that actually didn't know what shape they were either and would have to discover it on the fly. So this is extremely powerful.
The fact that biology is always dealing with an unreliable substrate means that the genetics that you get passed on from prior examples of your lineage cannot be taken literally. They are a bag of tools that you are going to have to utilize however you can under different circumstances to get the job done.
This is very different from how we do a lot of computing today, because instead of these nice, sharp demarcations between the abstraction layers that we have in our architectures, we have redundancy and error-correcting codes to make sure that our data stay fixed, that everybody knows what the data mean. They don't go floating off. The higher levels don't need to worry about the lower levels being unreliable. That is exactly the opposite of what biology does. Biology assumes that everything is up for grabs, that the material, you never know how many copies of anything you have or what your situation is going to be. It doesn't overtrain on its evolutionary priors. It reinterprets on the fly.
What it's doing is producing agents that solve problems. My gut feeling is that a lot of consciousness is around this idea of continuously having to interpret your own memories and build an active story of what you are cognitively, the same way that your body replacement machinery is constantly continuing to build up a story of what your organism is.
This has huge implications for evolution itself, because actually, we've done lots of computational modeling of this. If you actually model evolution operating on a competent material as opposed to a hardwired mapping between genotype and phenotype, some very interesting things happen in terms of the pressure that comes off of the genome. Eventually, all of the work gets done on the actual competency as opposed to the structural genes. And you end up with this amazing intelligence ratchet, that basically the more competent the material, the less selection can actually see the genome. And the more trouble it has distinguishing good genomes from bad genomes, because the material is continuously making up for it, rearranging the frog faces if they start out wrong and all that kind of thing. Then basically what ends up happening is that all the work of evolution ends up being done on the competency mechanism itself.
You end up with something like planaria, which I think of that ratchet has gone all the way to the end for reasons we can talk about, where their genome is incredibly noisy, and yet they are highly regenerative, cancer-resistant, do not age, and they have some other amazing properties specifically because of this. So I think the climb of intelligence, as we think about what is intelligence, as we think about bio-inspired computing, what's at the basis of all this is the continuous need to interpret information from scratch with an unreliable material and not have allegiance to the fidelity of it, but to optimize for saliency in an architecture that has this crazy, self-modifying hardware with multiple observers at different scales, all trying to hack each other, constantly trying to generate their own meaning of an interpretation of what's going on.
Okay. So the very last thing I'll just show is a few, kind of more recent thoughts around where I think this is all going. So all that plasticity that we've been talking about, this ability of life to make sense of novel scenarios, has a consequence which is that it's very interoperable. Pretty much at any level of organization, from cells, tissues, and so on, we can introduce engineered material. So we can make these hybrids, these chimeras, which is one reason why I think a lot of this discussion of, you know, proof of humanity certificates and what real humans are and so on, I think are a complete lost cause because we're not dealing with simple machines. This is what we're going to be dealing with. Trying to figure out whether this is 50% or 51% human is hopeless.
So we need to start to ask ourselves, all of these different kinds of beings, what is their behavior, what is their mind going to be like if they are not on the evolutionary tree of life with us? I think that all of Darwin's endless forms most beautiful, all the variety of life, is like a little tiny corner of the option space of possible bodies and minds. Cyborgs and hybrids and just every combination of evolved material, engineered material, and software is going to be some kind of agent, and we are all going to be living together, and we need to be able to understand each other.
There's one other piece that I've added to this, which I'll just close on, which is there's another input here which is what Whitehead called, and I think is a good term, is ingressing patterns. So where do these patterns come from, right? When you make these novel creatures that never existed before, we know where biological patterns come from, we think. People will say, "Well, it's evolution," right? You have a long history of selection. That's where the electric face and all of these goals, the goals of these collective intelligences, they were set by eons of selection.
So okay, so now we have all these creatures who have never been here before. Where do the goals of collective intelligence come from? Where are they written, where are they specified? Biology likes two things for when you claim that you know where information comes from. You have to be able to rewrite it. In the case of DNA, you have to be able to rewrite the DNA and show that the anatomy changes. And so then you can say that's where the information was. And then it likes a history. So it wants to know, well, why that information as opposed to some other kind of information. Okay? And there are different mappings, direct and indirect, from the genome to what happens. But basically, this is what you want.
So I want to just point out that there are a lot of patterns. What you're looking at here is something called a Halley plot, and then a couple of videos just made by tweaking this formula. This whole thing right here comes from decoding this tiny little formula in complex numbers. That's it. There's a simple algorithm, about 10 lines of code, that takes this simple generative seed and makes something incredibly rich like this. There are some other patterns that come out for other formulas, just lots of richness.
So now we can ask, if you're a biologist, you want to ask, "Well, where is this encoded?" This is a very specific pattern. Doesn't hurt that it's kind of organic looking, but it's a very specific, complex pattern. It's not, it's not, this is not a compression of it. We can't recover things like this for arbitrary patterns. Where is it encoded? Well, there's no genetics, so there's no history, there's no historical story to be told about this. There doesn't seem to be any piece of physics that you can blame on this particular pattern. I would argue if the rules of physics were different, the mathematical structure would still be what it is.
So now we ask the final question, like, okay, if these patterns exist outside of a historical evolutionary track and some kind of a physical property, what does that mean for biology, and can we find novel forms with no history? I'll just show you two quick things.
If you look at this little organism, I might ask you what this is, and you might say that this is a primitive organism I got from the bottom of a pond somewhere. And I would ask you, "What do you think the genome looks like?" And you might say, "Well, it would look like one of these primitive organisms." If you actually sequence this, you'll find Homo sapiens. This is 100% human cells. These are not embryo cells. These are adult, in fact, usually elderly patient cells that we have allowed to reboot their multicellularity, and they make something we call an anthrobot. This self-motile little creature looks like no phase of human development. It is an entirely new reboot of morphogenesis, no genetic change, no scaffolds, no drugs, no synthetic biology circuits. This is just something else that human cells will do when given the opportunity. The original patient may or may not still be alive, but some of its bits will have their own little life here.
They have amazing properties. They have a massively altered transcriptome. Half the genome, in fact, is expressed differently, even though, again, the genetics are still the same, but they turn on different genes in their new lifestyle. They have discrete behaviors, so we can make an ethogram of different kinds of behaviors. And they have weird capabilities. If we take these neurons and we put a big scratch through them, then these anthrobots will sit down and they make, they actually start to knit the wound closed. So they'll actually repair neural scratches. Who knew that... they come from tracheal epithelia. Who knew that your tracheal cells which sit there quietly in your airway for decades, when given the opportunity, can make a completely different little motile creature with new transcriptomes, new behaviors, new capabilities?
Where did this come from? There's never been any anthrobots. There's never been any selection to be a good anthrobot. This situation never comes up in life. We don't know where this comes from, but as I showed with those mathematical forms, we do have a precedent for things that, that forms, in fact, complex forms that do not require either an explanation at the level of physics or evolutionary history.
The same thing is true with... These are xenobots. These are frog cells. Something cool that they do is kinematic self-replication. So when you provide them with loose skin cells, they pretty much do von Neumann's dream of a machine that builds copies of itself from material it finds in the environment. They run around and they collect these cells into these little piles, and the cells mature into the next generation of xenobots and so on and so on. Again, never existed before. No other creature to our knowledge does kinematic self-replication. This stuff is new and not predictable from the genome.
So then we can ask, where do these specific goals and competencies come from if not a lengthy history of selection? So we know that there are all kinds of shapes, behaviors, properties of networks, features of prime numbers, facts about computation and so on, and so there are two options. A lot of people, because they want to be kind of sparse in their ontology, will say that, "Well, these are just facts that hold about the world." Okay? And if you do that, what happens is I think that you then end up with a system where every once in a while you come across these new facts that hold, you're surprised and amazed by them, you write them down, and that's it. And that seems to me a kind of mysterianism. I don't like it.
I prefer this idea that there's an ordered non-physical latent space of patterns which can be studied systematically. It has a metric to it. We are not just random. There's not just a random grab bag of these patterns, but we can study these systematically, and we have a research agenda. It means that these things like when we make embryos, when we make biobots, when we make AIs, when we make, you know, hybrid constructions and galls and things, what we are actually doing is building vehicles to explore this latent space. We're able to sample that latent space and see what other patterns are going to come through. We make interfaces, right? This platonic space of patterns is, we can poke little holes in it systematically and try to understand the mapping between the things we build and the patterns that come down.
So, you know, of course, biologists don't tend to like this because who wants another non-physical space, and people like to be physicalists about this. But I think that horse has left the barn. I mean, mathematicians mostly, I think, do think that they're exploring a space of preexisting patterns.
Then my hypothesis, so now this is the totally kind of conjectural part of the thing, is that, okay, if we already know that there are free lunches around types of rules of geometry, rules of computation, of prime numbers, all these kinds of things that evolution can exploit, then and we already know that morphological competencies are a kind of cognition, so morphogenesis is a kind of intelligence, then maybe the patterns in that space are not all low-agency kinds of boring things about triangles and truth tables and prime numbers, but actually maybe some of them are actually kinds of minds. Maybe some of these patterns are determining morphogenesis. Some are determining behavior, and maybe these patterns are in fact kinds of minds. And we're currently working on understanding the cognition of these xenobots and anthrobots which, again, have never been here before.
The last thing I'll just say because I'm way over time here is that I think the Garden of Eden 2.0 that we're all going to be living in is going to be expanded in two ways. First of all, a much wider range of beings in all kinds of embodiments that go way beyond anything that biological evolution has seen, and that all of these things are fundamentally pointers or interfaces. When we make embryos, I don't think we make intelligence. I think we facilitate its ingression by pulling down patterns into functional bodies as pointers or interfaces the same way we do it when we build, you know, when we build triangular objects and things like that.
Trying to understand different patterns, different patterns in excitable media is actually really part of this process to radically expand the set of beings that we can... the cognitive beings that we can think about, because we too are patterns. We are temporary metabolic patterns, and to the extent that we are agential, we have to start thinking about what other patterns in what spaces may have different competencies.
I often think about a fictional scenario where creatures from the core of the Earth, extremely dense, come up to the surface. They don't even see any of us because we're like a fine gas to them, a plasma. They might be really unwilling to see patterns in this gas that only live for about 100 years, unwilling to see those patterns as agents. To them, we're just disturbances, temporary vortexes in this space.
So we really need to expand our definition. We started by saying it's not just brainy animals. In fact, slime molds can learn, and cells can learn, and maybe your organs and the morphogenetic agents. But it's even bigger than that because then you actually have hybrid agents and so on. Now maybe we have to start thinking about the space of patterns that includes not just physically embodied beings, but a much wider space of patterns. We can think about expanding these categories the way that numbers have been expanded.
Okay, that's it. I'm going to stop here and just summarize. If we want to understand biological intelligence or make bio-inspired technologies, these are the things we have to really think about. Intelligence, I think, is pretty much everywhere. We have to get better at being able to recognize it.
A large part of that is going to be to understand the input that comes neither from the hardware/genetics nor from the history of evolution. I take seriously the idea that what we are going to need to do is explore the patterns that we get from this platonic space. This means that a lot of humility is needed here because, A, we too are patterns, so we need to start thinking about what other patterns are hard for us to recognize as agents.
And I think this idea that when we make something, we know exactly what we've made because we've made it, I think is completely wrong, because there's always this extra input. You get more out than you put in.
So the research agenda that we have here in this field of diverse intelligence that's emerging is that we can now develop tools, including some AI tools, and protocols for recognizing intelligence in very unfamiliar guises, communicating with them, and this drives practical applications.
In biomedicine, this means using the interface to control collective intelligence for regeneration, for cancer suppression, for fixing birth defects, for aging, and so on. We've got some other stuff cooking that will be here in a few months to study the cognition of really surprising things, like patterns in minimal media.
I will just thank all the people who did the work that I showed you today. Here are our post-docs, our PhD students, some of our amazing collaborators, all the funders that have supported the work, and a disclosure of three commercial entities that have funded some of this stuff. And of course, the animals that do all the heavy lifting to teach us about these things.
Thanks.
