Watch Episode Here
Listen to Episode Here
Show Notes
This is a ~45 minute talk given at the Imagining Summit (https://www.artificiality.world/the-imagining-summit/) on Diverse Intelligence and synthbiosis, approaching those topics via communication with cellular collectives in the context of regenerative medicine. The newer material is at the end.
CHAPTERS:
(00:00) Diverse Intelligence and Goals
(02:30) Beyond Discrete Natural Kinds
(06:00) Framework for Diverse Intelligences
(11:00) Scaling Minds, Collective Intelligence
(15:30) Self-Creation of Bodies, Minds
(18:30) Anatomical Order, Compiler Vision
(21:30) Agential Material, Problem Solving
(26:30) Bioelectricity as Cognitive Glue
(31:00) Communicating with Collective Intelligence
(34:30) Regenerating Limbs, Healing Wounds
(37:30) Synthetic Life: Xenobots, Anthropods
(42:30) Interoperability, Novel Agents
(45:30) Humility, Patterns in Media
(49:30) Conclusion: Expanding Sentience Idea
(52:30) Acknowledgements and Funders
CONNECT WITH ME:
Transcript
Sorry I can't be there in person. Afterwards, if anyone is interested in the details—the dataset, software, published papers—everything is at this site, and my personal thoughts on what it all means are here. The main points I'd like to share today are these. First, I'm going to talk about agential materials and, in general, the field of diverse intelligence or problem-solving in unconventional substrates. I'll try to broaden our expectations of what active agents and intelligence look like. I'll use a specific example in regenerative medicine: cellular collectives functioning as large-scale intelligence that navigates anatomical space. I'll discuss bioelectricity as a kind of cognitive glue that binds these sub-units together toward a common purpose. All of this is to show you an example of recognizing, communicating with, and actively engaging with a novel type of intelligence that we're not used to dealing with. It has many applications for regenerative medicine, which I'll show you. In particular, it serves as an example of the practical benefits of some of the ideas we're discussing. Toward the end, I'll discuss some novel embodied minds. I'll show you a couple and talk about a few that are coming for all of us. We'll talk a little about this notion of synthbiosis and the future of living positively with diverse sentient beings. Altogether, I think the whole talk could be summed up in one sentence: we can use the collective intelligence of cell groups navigating the space of anatomical possibilities to show how certain philosophical ideas can move forward to therapeutics and then to broader issues beyond.
This is a very well-known classic painting. This is Adam naming the animals in the Garden of Eden, and this is a conventional view with which most of us approach these questions. There is something profoundly wrong here, but also something deeply correct, and I want to talk about both. What I think is wrong is that by looking at this, you get the idea that we're faced with a set of discrete natural kinds. There are specific animals. We all know where they are and what the differences among them are. Here is Adam. He's able to count them and point to them, and by the way, he's significantly different from them. He's a different kind of creature. Those are the things I think need to change, and I'll show you why. But what's quite deep and correct about this is that in these old traditions, naming something meant you understood its true nature. The idea of giving something a name or discovering its true name meant you understood something fundamental about what it is and how you'll relate to it. I think this is profound because, as individuals and as a society, we will have to be able to name, in this sense, a very wide range of highly diverse embodied minds.
The first thing I want to point out is that ever since we've understood evolution and development, we now know that what we're looking at is a continuum of beings. Each of us was a single cell, both on an evolutionary time scale and a developmental time scale. Whatever philosophers might say about a human and the kinds of things that humans have—intelligence, goals, purpose, responsibilities, moral value, and so on—these are not things that snap into focus in some sharp emergence, but are actually scaled up from properties of very different kinds of systems. Neither developmental biology nor evolution offers any kind of sharp transition at which you go from being one kind of system to something magical like this. This sort of agential glow that surrounds the modern human actually spreads down into a spectrum.
Not only that, but now with the advent of bioengineering and other synthetic sciences, we see there's another continuum as well. We stand at the center of this other continuum where, using changes to biology and combinations with technology, we can produce—and in fact have and will increasingly produce—a wide range of novel beings with different kinds of embodiments and cognitive structures. Anything you want to say about a human, and you might be tempted to try to distinguish that from a machine, for example, you'll have to say something about all these kinds of beings. You'll have to ask, "What is it that makes them different, and how much do these changes matter?" Biology is incredibly interoperable, and I'll say more about why at the end. At every level of biological organization, we can introduce engineered materials, engineered algorithms, and so on, making all kinds of chimeras and hybrids.
What I'm interested in is developing a framework that will let us recognize, create, and ethically relate to truly diverse intelligences, regardless of what they're made of or how they got here. That means familiar creatures—primates, maybe a whale or an octopus—but also all kinds of unusual creatures: colonial organisms, swarms, unicellular life forms, engineered life with synthetic morphology (which I'll show you examples of), AI agents, either robotically embodied or pure software, and maybe someday even alien exobiological agents.
I'm certainly not the first person to try something like this. In 1943, Rosenbluth, Wiener, and Bigelow tried to show a scale of how you can get from purely passive matter through various cybernetic arrangements all the way up to human-level metacognition and so on. This is the sort of thing I'm looking for, and it's described in great detail in a paper called TAME, T-A-M-E, Technological Approach to Mind Everywhere.
For me, what's really critical is that whatever frameworks we come up with, they have to be useful. They can't just be philosophy. They have to move experimental work forward. They have to lead to new discoveries and new capabilities, and also to better ways to understand the ethics of increasingly complex life on this planet.
Central to this framework is the idea of an axis of persuadability. You'll notice it puts some things on the same spectrum that you might have thought were quite different, such as machines and humans. The idea is that along this spectrum, what varies is the degree of autonomy and the kinds of tools you might use to relate to that system. For some machines, you can only do hardware rewiring. You cannot punish them or convince them of anything, so those are the tools you use. For other systems, the tools of cybernetics and control theory become useful. For example, you don't need to rewire your whole heating system to get your house to stay at a different temperature; you just need to understand how the goal is encoded, set the set point, and the system will do what it needs to do. This is more autonomy and a different way for you to relate to it.
Eventually, you reach some systems where you don't need to know much about the molecular mechanisms. Humans trained dogs and horses for thousands of years knowing nothing about what was in their brains or what neuroscience was, because the system offers a completely different set of tools for interacting with it, such as training and learning. Then, at the highest end, you get into communication with cogent arguments and similar things.
The important thing is that you can't just have feelings about where certain systems sit along this continuum. It's not one of those "I'll know it when I see it" situations. You have to do experiments, because we're not very good at figuring out what kind of intelligent competencies something has until we study and test it. My claim is that anything you say about a new system—whether it's an AI, a language model, a robot, cells, or tissues—if you're saying something about its level of intelligence and agency, you are making an interaction protocol claim. You're saying, "This is how I'm going to relate to that system using this set of approaches," and therefore you yourself are taking an IQ test, because that might be a good or bad idea depending on how things turn out.
This is experimental. We are not tied to ancient philosophical categories. We have to do experiments. While we are pretty good at recognizing intelligence in animals that navigate three-dimensional space—medium-sized animals that move at medium time scales—we can sometimes see and recognize intelligence. But life explores, navigates, and struggles through many other kinds of action spaces: possible gene expression, physiological state space, anatomical configuration space. All of these spaces are equally real and challenging to the embodied minds that navigate them.
This means all of this is not just about humans. It is about all kinds of other observers that have to understand how these systems navigate those spaces: parasites, conspecifics, subunits of the system, greater systems within which they are embedded. All of these are trying to understand where a given system fits along this continuum. We have to get beyond our evolutionary firmware, which makes it hard for us to think about intelligences in these other spaces that have much different scales and ways of navigating.
The thing we all have to start with is the idea that we were all once a quiescent oocyte. You might look at that and say, "This is a system that's perfectly describable by chemistry and physics. It's a little blob of chemicals. It doesn't have any of the things we normally associate with intelligence or agency." But this process is slow and gradual, and that system will eventually become something like this, or even something that makes statements about not being a physical machine and having extra properties. This is the process we need to understand: the scaling of minds from very simple things that are amenable to chemistry and physics, all the way up through systems that require all kinds of new tools. This is not the end of the story, because this simple subunit can scale up to something enormous, but it can also scale back down again and fall apart into pieces or transform into something completely different.
While all of us make this journey across the so-called Cartesian cut, we now have to understand the scaling. How does the cognitive light cone of these simple systems inflate to something like this? The first thing to realize is that we're all slowly amplified versions of much simpler biochemical systems, but at least we're a unified intelligence, right? We have a brain and enjoy a pretty unified perspective on the world. We know we are a self. We're not like a collective intelligence such as ants or beehives, right? We're different. We're a truly unified intelligence.
Descartes liked this idea and liked the pineal gland inside the human brain because it was a singular organ, and he felt that the unified human experience needed a single structure in the brain to correspond to. But if he had access to good microscopy, he would have found out that inside that pineal gland are many cells, and inside each of those cells is even more complexity. There really isn't one of anything, and that reminds us that all intelligence is collective intelligence. We are all made of parts. This is a single cell, called a lacrimal area. You can see the kinds of building blocks we are made of; they have their own agenda and competencies. Each has a small cognitive light cone that extends a little in time and space around this tiny region, and that is where its goals lie. It has physiological, metabolic, and behavioral goals. We need to understand how this scales up into the goals of much more complex agents.
Even inside that cell, there are simple molecular networks that can do things like Pavlovian conditioning. Intelligence does not only appear in the brain and nervous system. If we go back to the cell level, or even before that, aspects of memory and learning are already present when you have a small network of chemicals interacting. We're trying to use that process by developing applications in drug conditioning, where we can train cells to respond to drugs in particular ways because they are already made of agential material—not passive matter or simple machines, but machines with learning capacity inside every cell.
Now we see that evolution uses this multi-scale competency architecture. Every level of organization in our bodies and beyond, up to groups and collectives, has its own ability to solve problems in different spaces. They all have various competencies to navigate their spaces in an adaptive way. Intelligence does not only show up at the level of brains and whole organisms; it is present all the way down to molecular networks.
Alan Turing was very interested in different types of embodiments of mind and machines that could think, and other ways to do the kinds of things humans do, such as problem-solving and intelligence through reprogrammability. This is what he studied. Interestingly, he wrote a paper about how chemicals pull cells together and generate order during embryonic development. Why would someone interested in the intelligence of mathematics and reprogrammability look at chemicals in development? I think it's because he realized a profound truth: the mechanisms involved in the self-creation of minds are very similar to those involved in the self-creation of bodies. We can learn from development, and we have to, because that's where our minds arise, how minds self-assemble. I think he was onto something important.
If you look at an embryo—say, a piece of a flat chicken or mammalian embryo—you might say, "Okay, there's one embryo." Normally, that's what comes out of it. But if you ask, "How many selves are actually in there?" the fact is, if we make little scratches in this blastoderm (and I used to do this as a grad student with duck embryos, using a needle to put some scratches in), you end up with little fragments that, before they heal together, each one decides it is the embryo and self-organizes. You get twins, triplets, and more. So the real answer to how many selves are inside an embryo could be anywhere from zero to probably half a dozen, at least for the kinds of embryos we know about.
They have to do this amazing self-construction. They have to know where they end and the outside world or other beings begin. This is not set by genetics. This reveals the self-construction that any living agent has to do to pull itself out of a background of disorder and establish its boundaries between itself and the outside world. This has many implications for issues of individuation, including split-brain patients, dissociative identity disorders, and similar conditions.
What we're looking at here is the ability of an excitable medium in which the cells act. What makes this one embryo is that they all share the same plan of what they are going to do—the path they will take through anatomical space. They're all aligned and committed to the same story. That's what makes it one embryo. These little islands are committed to slightly different stories because each one will do its own thing. It's a matter of alignment to become one of these beings.
So let's ask, where does this information come from? We all start life as a collection of embryonic blastomeres. Here is a cross-section through a human torso. Where does this order come from? Look at this incredible pattern. I could just as easily show you brain architecture. It's an amazing pattern. Where does it come from? You might be tempted to say it's in the DNA, but we can read genomes now, and even before that, we knew that DNA doesn't directly set any of this, any more than the DNA of termites sets the structure of their colony, or the DNA of spiders sets the construction of their web. What the DNA encodes is the hardware—the proteins, the tiny hardware that every cell gets. After that, it's the physiology, the real-time working out of this physiological software, that allows the cells to figure out what to build and when to stop.
We're interested in understanding how to rebuild—if something is damaged, for example, how do we convince these cells to rebuild? As engineers, we want to ask, what else is possible? What else can these cells build besides their native course through anatomical space?
Imagine in the future, this is my vision of where this field is going. We want to get to something called an anatomical compiler. That means someday you will sit down in front of a computer and draw the plant, animal, organ, or bio-bot you want. It can be the standard one for transplantation, or something completely new. You should be able to draw anything you like. If we had this, it would compile your target description into a set of stimuli to give to cells to get them to build exactly that. In this case, we've drawn a three-headed flatworm, and here it is. This is the end goal for the field, but we don't have anything like this yet.
Why do we need it? If we had it, all of this would go away: birth defects, inability to regenerate after traumatic injury, cancer, aging, and degenerative disease. If we had a way to communicate to cells what they should build, all these problems could be solved.
This anatomical compiler is not a 3D printer or anything that tries to micromanage cell construction onto a scaffold. It's a communications device, a translator. It tries to get your desired specification, as the bioengineer, into the recorded set points that the cells use to decide what they're going to do. We don't have this yet, despite genomics, biochemistry, and molecular biology making great progress for decades. It's because we have not yet understood how to communicate to cells.
The state of the art is very good at understanding genes and proteins interacting, but we're still far away from being able to regenerate a limb, fix a birth defect, or create a different structure. That's because biomedicine has focused on the hardware: genomic editing, protein engineering, pathway rewiring. All those things look like this: this is where computer science was in the 1940s and 1950s. If you wanted it to do something different, you had to physically rewire it. The assumption is that what we're dealing with is merely a complex machine. That's the standard assumption of molecular biology and much of biomedicine. But I think this is wrong. We're fundamentally dealing with agential material. Once we realize this and use the tools of behavior science and cognitive neuroscience, we will be able to do for medicine what computer science has done for information technology. On your laptop, when you want to go from Microsoft Word to Photoshop, you don't get out your soldering iron and start rewiring, because we now understand that if your hardware is good enough, you can do much better than that. You have high-level ways of communicating with your device, and the biological hardware is absolutely good enough.
Let me show you some examples. We started out talking about intelligence and navigating anatomical space. What are the capabilities of biological hardware? First, here's development: a human egg gives rise to a human body. This is an amazing rise in complexity and it's very reliable, but that is not why I call it intelligence. It's not just because complexity rises. It's easy to increase complexity with no intelligence. That's not it. It's because this process has amazing problem-solving competencies. One is that if you cut this early embryo in half, you don't get two half-bodies; you get two perfectly normal monozygotic twins. As I showed you with the duck embryos, the whole thing is a decision-making process of finding borders and navigating toward your anatomical goal state. It is not a mechanical, hardwired step through a bunch of biochemical steps. That's not what's going on. Once you've cut it in half, each side realizes that a part is missing, and they can all get to the goal state from different starting positions, avoiding certain local maxima. That is the navigation I'm talking about.
In some animals, this process continues throughout their lifespan. The salamander, an amphibian called an axolotl, regenerates its limbs, eyes, jaws, portions of the heart and brain, spinal cord, and so on. When they bite each other's legs off, no matter where it is amputated, the cells will rapidly grow and eventually give you a perfect axolotl limb. The most amazing thing about this regenerative capacity is that it stops. When does it stop? It stops when a correct salamander arm has been completed. What you're seeing here is a kind of anatomical homeostasis, like the thermostat in your house, which continuously seeks to reduce error relative to a set point. It has a memory. It has to remember what the correct point is. What it's remembering is this particular structure. The cells keep going until that structure is rebuilt. It's the ability to get to the same point from different starting positions, even when you're deviated from it.
You can think of all development as just regeneration. Development is regenerating an entire body from just one cell. All of this is about error minimization and navigating that space to get to where you need to be.
This is one of my favorite examples. You can take a newt and make sure that during the earliest steps of cleavage, the DNA multiplies extra so the cells become bigger to accommodate those extra copies of the genome. When the cells get bigger, the newt stays the same size. Taking a cross-section through the tubule, you can see that instead of eight to ten cells, there are now fewer of them that still make exactly the same structure.
When you make newts that have so many copies of their genome that the cells become enormous like this, then one cell will wrap around itself and leave a hole in the middle. So what you're seeing here is this incredible problem-solving capacity where if you're a newt coming into the world, you can't count on how many copies of your genome you're going to have. You don't know what the size or the number of your cells are going to be. You have to figure out how to solve your problem and get to where you're going using the tools that you have.
Now, this is one set of tools, cell-to-cell communication. This is cytoskeletal bending. It's using a completely different set of molecular steps to achieve your goal. This is textbook intelligence. It's using the tools you have to solve the problem under completely novel circumstances. So not being able to assume, and not being able to take your evolutionary past literally, is really critical for problem-solving.
Now we see that what happens in the creation of bodies is this incredible problem-solving capacity where you can do all sorts of things, and they will readjust and get their goal met. So how is that possible? What is doing all these computations, and in particular, where is the memory stored? How do these things know what shape they're supposed to be?
A couple decades ago now, we started in my group developing some of the first tools that were used to discover this. The idea is we took our inspiration from the nervous system. In the brain and in the nervous system, which is the one non-controversial example of systems that have goals and have clever behaviors to pursue those goals, this is how it works. The hardware relies on ion channels to set voltage states in these neurons, and these electrical voltage states do or don't get propagated to their neighbors through these electrical synapses. As a result, you get this very dynamic electrophysiological activity.
Here's a zebrafish brain recorded by this group showing you what happens when the fish is thinking about whatever it is that fish think about. You could imagine, as people are now doing this kind of project of neural decoding, could we read and decode these physiological events to understand what are the memories, the preferences, the goals, the behavioral repertoire of this animal? All that cognitive content is here in this electrophysiology. That's what neuroscience is committed to.
It turns out that this system, this kind of bioelectric system for processing information towards goal-directed activity, is ancient. It is not something that was new with brains. Every cell in your body has these ion channels. Most of them have these gap junctions that allow the networks to form. We can actually then use these tools to do a kind of non-neural decoding and ask, "What are all these cells talking to each other about? And what does the collective remember, and what does it know that the individual cells don't know?"
We had developed some of the first tools to molecularly read and write this electrical information. Here it is, using voltage-sensitive fluorescent dyes on a bunch of cells. You can see all the amazing electrical conversations that these cells are having. Some of them are fast, some of them are slow, but this pattern is, just like in the brain, this pattern encodes the collective processing of information in this group of cells. And then, of course, we do a lot of computational modeling to try to understand how that works.
Let me show you some specific dynamic examples. Here is what we call the electric face. This is again a voltage-sensitive dye showing you what happens when a frog embryo is trying to put its face together. This is one frame from that video, and what you see here, the thing looks like a face. This is why we call it an electric face. It's very easy to see that these bioelectric states in fact map to here's where the eye is going to go, here's where the mouth is going to go, here are the placodes to the side. So what we're seeing is the pre-pattern. We're seeing the informational scaffold that these cells are going to use to turn on specific genes and ultimately create a vertebrate face.
As I'll show you in a minute, this pattern is essential to being able to reach that goal. There are also patterns that are pathological. If we introduce a human oncogene into these tadpoles, the oncogene will make a tumor, but the way it does it is by disconnecting cells from the electrical network. Here you can see this is where the tumor is going to be, and there's a bunch of escapees, straggler cells here, that are going to basically disconnect from the network, roll back to their unicellular lifestyle, and just treat the rest of the body as external environment. That border between self and world, evolution scales it up, but during this cancer process, it can actually scale back down to the tiny cognitive-like cones of individual cells, which then do not recognize that they're part of this large network. And that has many implications for therapeutics, which we are pursuing.
So I've shown you how we detect these patterns, but of course, more important than detecting is actually being able to write them so that we can show what happens. We don't use applied fields, electrodes, waves, frequencies, electromagnetics, nothing like that. What we do is we manipulate the native interface that these cells are normally using to hack each other. Those are the ion channels in the membrane and these gap junctions that allow the cells to talk to each other. We use all the tools of neuroscience, which is pharmacology and optogenetics and all these other tools to control the electric state of these cells.
Now, the million-dollar question: Having posited that this electrical connectivity is the glue, much like in the brain, it is the glue that binds individual cells together into these grandiose pattern memories of making limbs and eyes and brains and so on, what actually happens when we perturb these pattern memories? Can we test this idea? How do we know this is right?
I only have time to show you a couple of examples, but here's one. I showed you that little spot of voltage depolarization that indicates where the eye is going to go. If you actually inject some potassium channel RNA into other cells that are going to make other structures, this is in the frog model, this is a tadpole here, you can set up an ectopic little domain of voltage that says, "Build an eye here." And when you do that, sure enough, that's what the cells do. So there's an eye sitting in some gut tissue. If you section them, this eye has all the lens, retina, all that stuff that eyes are supposed to have.
All of this was done not by micromanaging. We did not control the stem cells. We don't have any idea how to build an actual eye from scratch. We didn't need to do that. We communicated to those cells, "You should build an eye." This is an example of using that bioelectrical interface to communicate with a new, well, with a different kind of collective intelligence. It doesn't navigate three-dimensional space the way that motile animals do. It navigates anatomical space. And we were able to pass it one message, "Make an eye right here." That's it. It did the rest. It was competent to do all the downstream steps.
By the way, if you only get a few cells here, what they will do is recruit all their neighbors. The blue ones are the ones we injected. All this other stuff, we never touched it. These cells induce their neighbors to help them complete this construction project, much like I'm sure you've seen ants communicate and recruit nest mates to help them move a larger piece of food or something like that. All of this is part of the competency of the tissue. We did not have to build that in.
Another example of communication occurs in our regenerative medicine program. Frogs, unlike salamanders, do not regenerate their legs. So what can we say to these cells to get them to rebuild a leg, even if we don't know how to do it? Well, we discovered a bioelectric cocktail where one day, so 24 hours of treatment, gives a year and a half of leg growth, during which time we don't touch it at all.
Here you can see what happened immediately. Within 45 days, there are already some toes. There's a toenail. Some pro-regenerative genes are coming on, and you can see the leg is touch sensitive and it's motile. The animal can feel the touch and can use that leg to swim. This is again the idea of not micromanagement, but actually communicating that you should go not the scarring route, but the leg-building route and letting the system take care of the rest.
At this point, I have to do a disclosure. Dave Kaplan and I have started a company called Morpheceuticals Inc., which is trying to move this towards clinical applications, first in rodents and then hopefully in human patients, to regenerate organs.
I show you this. The details aren't super important, but what is important is the idea that when we take seriously that the dynamics that we're seeing are not just a mechanical clockwork, but are actually a set of decision-making processes by a competent system that is trying to minimize error relative to set points, it unlocks all kinds of discoveries that are inaccessible if we treat the thing as a dumb machine. And I'm sure this is just scratching the surface.
The next thing I wanted to show you was, when we're looking at these examples of repair, and there are many examples of birth defects and many other kinds of things that we've done, it's very clear where the pattern comes from. Presumably, it comes from evolution. In other words, that is the correct shape for that species. And so that's where the pattern comes from, or so the conventional story goes. But we wanted to push that idea.
The way we did it was to try to make a synthetic novel life form that is in a different configuration than anything in its evolutionary lineage. The way we do that is we take early frog embryos. We take some of the epithelial cells from the top here. We put them by themselves. We dissociate them and we put them by themselves in this little depression.
Many things could have happened. They could have died. They could have crawled away from each other. They could have made a flat monolayer like cell culture. But we wanted to know, what would these cells do if we extract them from the embryo and give them an opportunity to reboot their multicellularity? Well, what they do, it turns out, they gather together. They don't do any of these things. They gather together and they make this thing we call the Xenobot. Xenobot because Xenopus laevis is the name of the frog that we get them from, and we think this is, among other things, a bio-robotics platform.
They start to swim. They're autonomous. They use little hairs on their outer surface. These hairs are normally used to move mucus down the body of a frog. But here, they're using it to row against the environment. They can go in circles. They can patrol back and forth like this. They can have collective behaviors where they interact with other bots. Here's one moving down this maze-like structure. You'll see here that it takes this corner without having to bump into the opposite wall. And then here, it spontaneously turns around and goes back where it came from. Remember, there are no neurons here. This is just skin. This is a novel protoorganism that's made entirely of skin cells here.
It has all kinds of interesting behaviors. If we look at its electrophysiological activity, so this is a calcium sensor, there's tons of it. This actually looks like the kind of thing that people record from brains, even though there's no nervous system here. If we were to actually functionally analyze these signals, there are some really interesting features of them that, you know, this is just a pre-print for now, but there are some interesting comparisons that can be made between the physiology that goes on here and the kinds of things we would read from a brain.
They have some unexpected capabilities. Here's something we call kinematic replication. We've made it impossible for these bots to reproduce in the normal froggy fashion. They don't have any of the organs that are needed to do that. But what they can do is if you sprinkle a bunch of loose skin cells into their field, so this white stuff is just loose epithelial cells, what they'll do is they'll run around and they will collect them into little piles and they'll polish the little piles, both collectively and individually. And then because they're working with not passive matter but an agential material, these little piles mature into the next generation of Xenobots. And guess what they do? They run around and make the next generation, and they make the next generation, and so on.
So what has the frog genome learned here? What do genomes do? Over the years of evolutionary selection, it's certainly learned to do this, so it can make a specific developmental sequence and then some tadpoles and so on. But it also makes xenobots with their own really weird developmental sequence and things like kinematic self-replication. As far as we know, no other animal on earth reproduces via kinematic self-replication. There's never been any selection to be a good xenobot. There's never been any xenobots.
This immediately raises the question, the goals that these collective systems have, and now this is working up towards the broader thing we're going to talk about, AIs and so on. The goals of these systems are not obvious, and the goals of novel systems which do not have a history of evolutionary selection are even harder to predict. So what's really necessary now is a science of understanding what are the goals that novel collective systems are going to have. By the way, their behavioral intelligence, over the next few months, we will have a couple papers out that actually characterizes that.
I'll show you another quick example, which is this. If I just showed you this video, you might guess that this is some sort of primitive organism I got from the bottom of a pond somewhere. We have no idea what its goals or its competencies are. Turns out this is something we call an anthropod. It's made out of human cells. If you were to sequence this, this is 100% human, normal human genome. This comes from adult patients. These are tracheal epithelial cells that come together to be an anthropod instead of a xenobot from the frog.
So this is a total reboot of human multicellularity. Again, doesn't look anything like this standard organism. But they have some amazing properties. If you put them on a Petri dish where you've grown a bunch of human neurons and you put a big scratch through it, here's the defect, a neural wound here. What you'll see is that these bots cluster. They form something we call the superbot, and underneath, within four days, you lift it up, what they've been doing is taking the two edges of the neural wound and healing it together.
Now, who would have known that the cells that sit quietly in your trachea for long periods of time, when given the opportunity, will rebuild themselves into a self-multi little creature that has the ability to go around and heal neural wounds? Who knows what else it can do?
To start winding down here, I want to point this out. Because of this ability of life forms to decide on the fly how to solve various problems, they are not hardwired, at least most of them are not. They are constantly using fairly ingenious problem-solving competencies to get their needs met. It means that not only can they regenerate after damage, but life is incredibly interoperable. Almost any combination of evolved material, engineered material, and software is some kind of viable agent. So we have cyborgs and hybrids and chimeras. Many of these things already exist. They will increasingly be more prevalent.
All the variety of biology that Darwin saw when he used this phrase "endless forms most beautiful," that's one tiny corner of the space of possible beings, and we are going to be living with all of these. This will increasingly be a part of society, not just to make weird creatures and see what we can do in an engineering sense. The changes that certain people will make to themselves and to other systems, both biological and technological, are shedding critical light on understanding what are we. What does it mean to truly have a mind, to have goals, to have intelligence, to understand when we say, "Well, this thing doesn't really understand. Well, I really understand." What does all that mean for a collective of cells to say that they understand something and they can navigate a specific space?
All of this is about understanding us better, but critically it's about being able to ethically relate to a wide range of beings that we have never seen before and are not on the evolutionary tree with us.
Just two very quick things I will say is that a lot of people have strong feelings about what certain types of AIs can and can't do and what living things can do and so on. I want to point out that it is really essential at this early stage of the game to have some humility about this. Because we do not know what it is that allows biological machines, constructs of biochemistry and so on, to enjoy the kinds of cognitive properties that we all have. We actually don't know. Therefore, we have to be really careful about saying what other kinds of constructs can and can't do.
In fact, there's been all kinds of research in minimal matter and our own work on very simple sorting algorithms. If any of you are coders, it's something you study in your first year of algorithms in computer science curriculum. We found novel problem-solving capacities in simple sorting algorithms that nobody had noticed for a really long time. So it really doesn't take much to start to climb that continuum of agency.
I will argue that for the same reason that you can't look at the laws of chemistry and think that you've understood everything that's important about being a cognitive being in the world, you really can't look at the construction of a machine either in its hardware or its algorithm and think that you've understood everything that it's doing. Unexpected, not just complexity, not just unpredictability, but unexpected cognition starts to crop up extremely early, and it doesn't even require cells, never mind brains or anything like that.
The very last thing I want to say, I'm a couple minutes over, but the very last thing I want to say is this. I just want to tell you a quick story. This is a kind of science fiction story, but I think it makes an important point. Imagine that some creatures come out of the center of the Earth. They live in the core of the Earth. They are incredibly dense. They come up onto the surface. What do they see? Well, they don't see any of our stuff. They don't see any of this. They're using gamma rays as vision because they have to see through rock. They don't see any of this.
What they say is that, "Well, this planet seems to be covered by a very tenuous, very ethereal plasma." There's some kind of gas that's on top of them, but none of this is solid to them. They are solid. All of this is basically gas. They're sort of walking all over everything. And one of them is a scientist, and he's been watching this gas, and he says to the others, "You know, I've been watching this gas. There are patterns in this gas, and you would almost swear that these patterns are actually doing something. You'd almost think that they are agents of some kind. They seem to have goals, and they solve problems." And the others are saying, "Well, that's ridiculous. Patterns in a gas can't be agential. We're real. They're just patterns in an excitable medium. And by the way, how long do these patterns hold together?" He says, "About 80 years on average." "Well, that's nuts. Nothing interesting can happen in 80 years."
This kind of thing reminds you that, and this has been understood by people working in metabolics and so on, we too are temporary patterns in metabolic media, like whirlpools and solitons and gliders in the game of life and so on. All agents are actually patterns in some medium. So we have to be really careful in our commitment to expanding our ability to recognize novel minds, whether they be in physical embodiments like ours or something very alien.
So I'll end here and just point out that I think intelligence is probably very widely distributed throughout the world. Learning to rise above our evolutionary limitations and recognize it, I think, is critical to expand this idea of sentient beings. I don't think we've even scratched the surface of what that really is.
We now have the field of diverse intelligence, which has a research agenda: principled search for frameworks that avoid teleophobia, but also avoid unwarranted animism, painting up hopes and dreams on every rock. We don't have to daydream high or skew low. We can get it right, or at least we can optimize way past what we have now by dropping untenable, ancient categories that are not serving us any longer, these things that sound like we know what they are, machines and organisms and so on. It sounds like we know what we're talking about, but we really don't.
So this is the beginnings of a research program that can use ideas of a continuum of agency, observer-relative models, and I think AI as universal translators to these diverse intelligences to really have a much better future than is possible using today's frameworks.
If anybody's interested, I can send any number of papers where we go in depth on all this stuff. These are the people that contributed to all this work, lots of post-docs and grad students and undergraduates and tech support. We have many amazing collaborators here, our funders that I thank for supporting this work, and again, disclosures. There are three companies that are supporting us and are commercializing some of our IP.
So, I'll stop here, and thank you so much for listening.
