Thank you, and thank you very much for having me here to speak to you. This is kind of a brand new talk that I created around this topic. Let's see how it goes. If you want to look up any of the actual papers, data sets, software, everything is available at this lab website here, and this is my personal blog where I write what I think some of these things mean. So what I'm going to try to talk about today is a slightly different way to think about motivation in a wide variety of substrates. And the way that I think about these things is through an axis of persuadability.
In other words, what I'm interested in is a spectrum of cognition, all the way from very simple systems to very complex metacognitive ones, to ask what kind of tools would we use if we wanted to persuade these systems to do something different, right? And so here you might use hardware rewiring, and here tools of cybernetics and control theory, and then here behavioral science, and then here psychoanalysis and friendship and love and things like that, right? So the tools are quite different along this spectrum. But this is typically how I think about organizing a very wide and very diverse set of possible agents.
Importantly, all of these are empirical questions. In other words, which set of tools is appropriate for any given system cannot be decided linguistically or philosophically. We have to do experiments. We have to try different tools from different disciplines and see what gives us the most purchase. So this tends to be, in terms of motivating systems to do new things, is how I think about them. But today we're going to talk about, in particular, intrinsic motivations and where do they come from. So in this brief talk, I'm going to try to give you these ideas.
First of all, I'm going to generalize patterns in terms of forms of both structure and behavior. So I want to emphasize the symmetry between patterns of the body, so anatomies and shapes and things like that, and patterns of behavior, and I think both of those can be dealt with as one kind of thing. I'm going to briefly talk about our favorite model system, which is morphogenesis, that is groups of cells creating or repairing some kind of structure. And this is interesting for today's topic because we have argued for years that that process is not a mechanical, kind of hardwired process, but is actually best understood as problem-solving navigation.
And when you have a system that does problem-solving navigation of a particular space, it becomes interesting to ask what are its motivations in doing that. And then we specifically will talk about where do goals, that is the targets of certain kinds of motivation, where do they come from? In biology, we love selection and genetics and physics and things like that, but the question is where do they come from in novel systems that have never existed before and have not been the target of selection for those specific goals?
And towards the end, I'm going to suggest a kind of very old but really out-of-favor idea of a structured latent space. You could call it a Platonic space, that's what mathematicians do, but a structured latent space of patterns that are important both for biology and physics. And this idea that some motivations, in fact, come from this space, and we have to understand the space if we are going to understand intrinsic motivation. And my claims towards the end are going to be that even very minimal systems that are not living and not biological are getting some of the magic, and by magic I simply mean some order for free, so patterns that you did not have to evolve or design.
And even minimal systems can be interfaces to those patterns from this latent space. Not just brains, but also chimeras and simple algorithms, all kinds of things that can do that. And at the end, I'll sketch a research program of how we are studying that space. So the first thing to remember is that whatever motivations you might have, as an uncontroversial example of what it means to have motivations, we were all a little bag of chemicals once.
So both evolutionarily and developmentally, here's a single cell in the developmental case. It's a quiescent oocyte, a little group of chemicals, and then there's this smooth, continuous, slow process of forming something, maybe even something as complex as a human, that is going to have motivations, is going to talk about pursuing goals, about not being a machine that is mechanically forced to do certain things, but to have goals that it pursues with reason and with motivations, basically. And let's just remember that we all came from a simple chemical system.
So whatever we're doing here, we have to be able to tell a story of scaling and origins, both developmentally and evolutionarily, timescale. And so the first thing that I'm going to talk about is to try to unify this notion of forms of body and forms of behavior and argue that they're really parts of the same class. So what are forms of behavior? Well, these things can be quite obvious in the case of animals that are like us. So here, this little squirrel is setting up a little accident scene here, and he's waiting for his owners to notice this, and then he's going to even look to see if anybody's paying attention to him here, right? So these kinds of high-agency behaviors are easy for us to recognize.
There are others that are actually not obvious. You know, when you see something like this, it's entirely unclear what exactly is going on here, and you can't just decide based on observation the level of agency of any of these behaviors. You have to do perturbative experiments. But in any case, we know what forms of behavior look like. Of course, we have to remember that this is not just for fancy large animals. Even the materials of which we are made, so the cells that make up our body, like this, this is a Lacrymaria, no brain, no nervous system, but very competent in its local environment to do the things that it needs to do.
And in fact, even the material it is made of, meaning chemical networks, itself has the ability to do at least six different kinds of learning. So just chemical networks alone, and this is a free gift from mathematics, it doesn't require a cell or evolution or anything like that, but just the properties of networks like this allow them to do habituation, sensitization, associative conditioning, and so on, where the future behavior of the system is different based on its past experiences. So now we understand we are part of an agential material at multiple scales that gains the ability to have motivations.
But these things start very, very early on. And so now I just point out that while we as humans evolved to be good at recognizing motivated behavior of medium-sized systems moving at medium speeds through three-dimensional space, so this kind of thing, we're pretty good at noticing that. But biology has been doing this in other spaces long before brains and muscles appeared. So cells and tissues navigate the space of gene expression, they navigate physiological state spaces, they navigate anatomical morphospace.
And while these spaces are difficult for us to recognize, I think we can argue that they are no less real than the three-dimensional space that we typically take as the actual space that we live in and the space in which we define motivated behavior. And of course, mathematicians and computer scientists are well-versed in visualizing all kinds of other crazy problem spaces that are difficult to draw. So what I'm going to introduce is this idea that morphogenesis, ontogenesis, meaning this process of how you get from here to here, is actually behavior.
It is behavior of a collective agent, meaning a collective intelligence, a group of cells. No individual cell knows what you're supposed to look like, but the collective does, and I'll show you what I mean by that. And so these outcomes, right, this cross-section through a human torso, this fly, which is running a stripped-down two-dimensional antero-posterior morphogenesis program on its wings to scare off predators, or actually building things outside the body, such as these termite mounds or spider webs.
All of these outcomes are the result of a kind of physiological software that runs on the genetically specified hardware. And these are all very specific patterns, but it's really interesting. The standard view in developmental biology often is that you have two things. You have genetic information, and then you have emergence. And complexity science is supposed to explain how you get from individual low-level rules that every subunit executes to then something complex emerges like a body.
But what's really interesting here is not the hardwired sort of reliable embryonic development that usually makes the same thing every single time. That isn't what needs to be explained, actually. The real trick here, and what's relevant for us, is to explain the flexibility and the ability of cellular collectives to pursue goals by diverse means. This, by the way, is a definition of intelligence. This is William James's definition of intelligence, the ability to pursue the same goal by different means. And what I mean by that, I'll just give you two examples. We could talk for hours about this.
This is an example of regeneration, and actually even development is a kind of regeneration. You're rebuilding the whole body from one cell. But the key here is this anatomical homeostasis, that you have this, you have the structure. If you amputate anywhere along the line, the cells will work very rapidly. They will rebuild the correct structure, and then they stop. When do they stop? They stop when the correct structure has been rebuilt. So you deviate from your goal state in anatomical space, and then the tissues work very hard to get back to where they are. When they get back with an acceptable error, then they stop.
And so what you're seeing here is not just a single feed-forward hardwired process, the way that cellular automata do it and so on, but actually it's an error minimization scheme. It's a feedback process with a set point that guides activity. It makes the activity stop when you reach the set point. So this is important. This is a system that expends energy and effort to get back to a specific goal state. It's not just kind of an open-loop process. Whatever happens, happens. It's a closed-loop system that is functionally motivated to reach this state.
But this is not just about damage, or repairing damage. And you see that when they take a tail and they graft it surgically to the side of the animal, to the flank, you see that over time it transforms into limbs. Okay, there's no damage here. These cells aren't injured in any way, but they turn into fingers. Why do they turn into fingers? Because the local order here obeys a large-scale plan. The individual cells don't know what a finger is or what a tail is, but the collective does.
And the collective has a set point that says in the middle here there should not be a tail, there should be limbs. And the whole point of this kind of architecture, this multi-scale biological architecture, is that the higher levels deform the action space for the lower levels so that even though there's no error at the small scale, the higher level knows that there's an error and it actually forces the cell behaviors and the molecular biology within, so turning on different genes and so on, to dance to this new kind of large-scale structure.
So this is important. Being an active agent means motivating your parts. It means having control over your parts to get them to do things that they otherwise wouldn't do, and to have system-level goals that can reach down to the molecular machinery. Now, what's important is that what I'm showing you now is that this process of morphogenesis has goal states. It has represented goal states, which we can actually now read and write. I don't have time to show that to you today.
But one of the things we can do is we can use those methods to actually rewrite those goal states. And when you rewrite those goal states... So here's a flatworm. Initially, every fragment will make a perfect little worm. So they're extremely reliable regenerates. But what we can do is because we know how the goal state is encoded, and it's encoded basically the way that your goals are encoded bioelectrically, meaning electrical memories in the network, we can rewrite that pattern and we can create two-headed worms.
And if you amputate these two-headed worms, the fragments will again give two-headed worms, and again and again forever. We haven't touched the DNA. There's nothing genetically wrong with them. The hardware is completely stock, evolutionarily produced hardware. But like all reprogrammable hardware, it can store multiple set points towards which to work. And so what's interesting is that these morphogenetic goals are rewritable, and this is the hallmark of a persuadable system. They're reprogrammable top-down.
When we make two heads, we don't have to control the gene expression. We don't tell what stem cells to make what tissues. We have no idea how to build a head. It's very complex. We change a very simple, large-scale, top-level bioelectric pattern memory that says make one head. And we say, no, actually a correct worm should have two heads. And then we leave it alone and let the cells build. In fact, not only the number of heads, but actually the shape of the head, right? So this is a nice triangular head of a Dugesia dorotocephala.
We can ask it to make flat heads like a P. felina or round heads like an S. mediterranea, or of course its own normal head. And this is not only the shape of the head, but the shape of the brain, the distribution of the stem cells. So there are other attractors in this morphospace, in the morphospace of head shapes, and normally they're occupied by these species. But this one, 100 to 150 million years distant, is perfectly happy to visit these other attractors if the target of its motivation is changed.
So if the set point is rewritten, it's happy to do this. Now, when we look at something like this that has these very specific attractors, we can say that, well, where did they come from? So selection, presumably, right? There's an evolutionary history that established these specific shapes for these specific species. And the same was the case for the other things that I've shown you today. But what I'm going to show you now is that it actually goes much deeper than this. It is not just the history of selection. And we have to ask, as we make hardware that finds truly novel forms, we have to ask where these come from.
Now, in computer science, where do patterns come from? Well, sometimes they come from a designed algorithm that a human programmer or maybe another high-level AI wrote an algorithm. Sometimes it's from learning. So there's some sort of reward function and a system has learned to meet that function. Sometimes it's from selection. So there's an evolutionary algorithm that threw away everything that didn't match a fitness criteria and whatever's left. So this is how we typically think about it. Biology, well, we've decided this is no good anymore, but we can certainly learn.
So biological systems are very good at learning towards specific functions and, of course, evolution. So selection for specific criteria. Biology leans even more on physics. Computer science does now to some extent, but biology even more so, thinking that certain patterns are dictated by specific aspects of the physical world. And so we want to now start to look at how well does this all do for us when it comes time to novel patterns.
So first of all, just to show you what I mean, this is standard development. So here's an acorn, very reliably it makes this oak tree with a very specific pattern. And this is misleading because it gives us this idea that that is what the genome knows how to do. This genome knows how to make this particular shape. But along comes this, it's a non-human bioengineer, it's a little wasp that hacks the plant cells. These galls are not made from the insect cells, they're made from the plant cells. It leaves prompts for the plant cells to build something completely different.
Instead of a nice flat green thing, it builds these round, spiky, like some amazing structures here. So what this is telling us, and we would have no idea without this kind of biohacking, we would have no idea that this is even possible because all we've seen is one point in that latent space, which is the shape of the leaf. We didn't know that this point exists, but now we do. And so I want to expand this idea that this kind of hacking is not just by different organisms hacking each other, but by wholes, bodies, collectives hacking their parts, cells hacking each other, and we as engineers trying to persuade cells to do different things. I'm going to show you a couple of examples.
The first thing is this. So I showed you those planarian flatworms. What you didn't see in that movie of those two-headed flatworms moving around is a structure called a pharynx. The pharynx is this like motile tube. It normally stays hidden inside the worm. When the worm wants to eat, it extends the pharynx, eats the food, and then that's it. What you're seeing here, I'm going to play this video, this little thing right here, that's a pharynx that has been removed from the animal, and whoops, oh man, sorry, this video for some reason isn't playing, that's too bad.
What you would see if the video played, what you would see is that the pharynx actually has a life of its own. What it will do is move around, and this stuff is liver, this out here is liver, which the planarian likes to eat. The pharynx is going to move on its own, it's going to burrow into the liver, and it's going to come out the other end, because of course the food just falls out the other end. It's quite amazing that when released from the rest of the body, it has its own behaviors that you normally don't see, right? The rest of the body keeps it quiescent most of the time.
The default behavior of this thing is to run around and try to dig into things that seem like food. I'm going to show you another example. These are, here, let's pause this. So these are cells, and the color here, these are cells in a dish moving around, and the color here represents bioelectricity. In other words, this is not a model. This is measurements using a voltage-sensitive fluorescent dye that measures the voltage of the cells. And you can see here that this collective is at a different voltage than this single cell.
But notice what's going to happen. The single cell walking around minding its own business, this guy's going to reach out and just boom, and just touch it. Just a single touch here, like a tiny little, look at, bang, that's it. Just a tiny little touch already converts the cell to the same voltage as the collective, and now it merges with the collective and it's going to do other things. So this is an example of cells hacking each other by physiological signals, and we can take advantage of this.
And this is what we've done here. In an early frog embryo, we've injected some ion channels that change the voltage of cells that are normally going to be gut. So this is a tadpole from a side view, cells that are normally going to be gut. And what that's done is reprogram those cells to be an eye. Again, we didn't control the stem cells or the gene expression that's needed to make an eye. We didn't have to micromanage any of that. We just produced a signal that says, make an eye here.
That is sufficient. That top-down resetting of the goal states of these cells is enough for them to make an eye. In fact, we only got a few of the cells here. This is a section through this kind of thing. The blue cells are the ones that we affected. They then secondarily told all the other cells what to do and sort of convinced them to participate in this eye-making project. So there are multiple levels of goal resetting going on here.
So in all of these cases, we can think about evolution setting different kinds of attractors for eyes and for heads and for limbs and everything else. And this idea that the cells can control each other, bodies can control their own parts, and we as engineers can mimic some of that and probably and hopefully have applications in regenerative medicine and things like that. But what happens when these kinds of things start showing up, and this is already happening, we already have cyborgs and hybrids, and there's only going to be more, when standard living things modify themselves, both technologically and biologically, to become forms that have never existed in evolution, what kind of motivations are they going to have?
What kind of cognitive capacities? What kind of goals? What kind of behavioral repertoires are they going to have? If we can't rely on genetics and history as a tool in those cases, what else is there? And I want to show you a couple of model systems that we use to study this sort of thing. First of all, these are epithelial cells at the apical end of the frog embryo. And we can ask, what happens when we liberate them from the embryo? What happens if we give them a chance to reboot their multicellularity? Do they die? Do they crawl away from each other? Do they form a boring flat layer like cell culture?
Well, what you can see here is that when we dissociate these cells, what they do is they come together, right? They form this little structure. This is kind of... oops, sorry. This is kind of cute. You can see this is a group of cells. Each one of these things is a cell. It sort of wanders over here. It looks like a little horse. They don't all look like that. They look like different patterns, but they move as a collective. The little flash you're going to see here is calcium signaling as these groups of cells talk to each other. But then they assemble into this.
And what happens is they become self-motile. So we call this a xenobot. They swim. They can go in circles. They can go back and forth like this. We can make crazy shapes out of them like this swimming donut. They have collective behaviors. One of the amazing things they do is called kinematic self-replication. So when you provide them with loose material, such as these epithelial cells, xenobots will push them into little piles. They will polish the little piles like this. And guess what those turn into?
They turn into the next generation of xenobots, and what do they do? The same thing, and they make the next generation, and that makes the next generation, and so on. So this is kinematic replication, and so what you're looking at here is a novel creature that has never existed before. No other creature on Earth, as far as we know, replicates by kinematic self-replication. They have all kinds of, I don't have time to tell you today, they have all kinds of amazing new behaviors, gene expression patterns, and so on, very different from what they do inside the body.
So all of this calls for an explanation. Now, you might think that this is some feature of amphibian or embryonic cells, that they're special and plastic somehow. Well, what about your cells? What about cells from your body? What would they do if we freed them from the influence of the rest of the body? So what you're looking at here is not a primitive organism that we got from the bottom of a pond somewhere. We call this an anthrobot. If you were to genetically sequence this, you would see 100% Homo sapiens. These are cells taken from adult human donors that give epithelial biopsy samples for diagnostics and so on.
You can buy those cells, and we have a protocol by which they self-assemble into this self-motile little creature. These things don't look anything like any stage of normal human development. Again, and they've never existed before, and you might wonder, what behaviors does it have? What does it do? Well, one thing it does is if you sit it onto a plate of human neurons into which you've made a big scratch, so we take a scalpel, we make a big scratch here, the anthrobots, which are green here, they drive down the wound, they pick a spot to land in, they assemble, sometimes in this sort of cluster of superbots, and you can see what they start to do. They start to knit across the gap.
They start to heal the damage. And this is what it looks like four days later. If you lift them up, here's where they were sitting. You can see that we're healing this. So they have this weird intrinsic motivation for healing. We didn't make them do this. There's no genetic change here, no synthetic biology circuits, no weird drugs, no scaffolds, no nanomaterials, nothing like that. This is just what these cells are doing together.
And no one would have any idea that your tracheal cells, which sit there quietly in your airway for decades, have the ability to form a self-motile little creature that, by the way, heals neural wounds and who knows what else it does. This was the first thing we checked, and this is what we found. They have a drastically changed transcriptome, over 9,000 differently expressed genes, so about half the genome is expressed completely differently in these constructs, in these anthrobots, than they are when they're back in the body.
So again, freed from the body, right, this configuration that has never existed before has new behaviors, new capabilities, new gene expressions, four discrete behaviors, and you can build a little transition diagram, like an ethogram. Why four? Why not more? Why these genes? Why not other genes? All of this is completely not obvious from the genome or from the history because there haven't been any anthrobots. And so here's what we need to ask about things like this. They're a tool to focus our attention on something very special.
That we know when the computation was done to shape the genome to do this. In other words, in the eons of making a frog, for example, the Xenopus laevis or the frog genome was selected in particular ways to have this developmental sequence, to have these behaviors. But it turns out it can also do this. These are xenobots. This is a xenobot developmental sequence, completely different behaviors. When did we pay the computational cost to get this? Where did this come from? There's never been any xenobots or anthrobots.
There's never been selection to be a good xenobot or a good anthrobot. Where did their goals come from? We certainly didn't make them do this. Selection hadn't made them do this. How did it happen? And so the last thing that I want to point out here is that this kind of amazing aspect that is having competencies, goals, and so on that were neither directly engineered nor apparently directly selected for is not just for complex living things, even very simple things, and I don't have time to get into all the details, but you can find it in this paper, even very simple things like sorting algorithms.
So bubble sort, yeah, five or six lines of code, completely deterministic, no magic, everything is there, fully transparent. Even that thing is doing additional things like delayed gratification, like clustering. It's these weird side quests that are not in the algorithm. We know because the whole algorithm is six lines of code, like you can see it, there's nowhere to hide, unlike in biology, there's nowhere to hide. And when you find new behaviors, specifically goal-seeking behaviors, behaviors that would be recognizable by any behavior scientist that are not in the algorithm, or in the case of these bots, are not engineered or selected in, we really have to ask where they come from.
And just realize that that is not something that life certainly improves it and scales it up, but it is not something that is unique to life. It goes all the way down to incredibly simple minimal systems. So here's kind of the most speculative and controversial part of the talk is I'm going to take a stab at answering this question because I think now it is very timely. Now, where do the goals of novel systems come from as we think about the hybrids and the cyborgs that we're going to live with and the AIs and the language models and everything else? The question is, can we develop a science of predicting their motivations, of interacting with them to hopefully shape their motivations in a way that's aligned with ethics and so on? Where do these goals come from?
And so I think we have two choices. The first choice, which I think is what most people today go for, is something that we call emergence. So simply the idea that we know that when simple systems do something in a repetitive way, oftentimes complex outcomes show up. And sometimes, like in cellular automata, the game of life, these kinds of things, sometimes people say, well, there's emergence of complexity. And when you say, well, why this particular thing, you say, well, that's just facts that hold in our physical world. That's it. We just have to know that sometimes there are these emergent things.
And I really don't like that way of doing it because I think it's very mysterian. I think collecting sets of surprises, things we didn't see coming, is not very productive for future science and discovery. And so I prefer what mathematicians often do, which is to assume that the spaces of these non-physical forms come from a structured latent space. That is, it's an ordered space. It's not a random bag of surprises. It is an ordered space, and you can do a research program to map out the contents of that space.
So, for example, if you look at this shape, this whole shape is a Halley plot of this very simple mathematical function. That's it. In complex numbers, if you plot the Halley map of this, this is what you get. You can also, I've made movies of this thing by varying the numbers slightly, so you can take kind of a flyover of this amazing world. It doesn't hurt that these things are very organic looking, you know, this sort of very, very biological looking. But all of these forms come from these mathematical shapes. Now, these mathematical objects, very simple.
If you want to know why this particular shape, the answer is not going to be found in the physics department, nor in the biology department, because there is no fact of physics that determines this. There is no fact of history. This was not selected from a random set. This is the way it is because of reasons that come strictly from mathematics. It is not biology or physics or anything else. And in fact, many mathematicians think that what they're building is a map of this space. They think that they systematically discover truths or facts or patterns in that space. They're not inventing them from scratch.
And so I think this is a very helpful strategy for biology and cognitive science going forward. Evolution exploits an incredible number of very powerful, what I call free lunches. They're patterns from mathematics, they're free gifts from mathematics that evolution can use, but does not have to create from scratch. And so rather than assuming that this space of patterns is just random emergencies that we just have to catalog, you know, you get to keep a sparse ontology, you get to be a physicalist, but it doesn't help you move forward.
What I think we're better off doing is biting the bullet, as mathematics did, and just assuming that there is this latent space that we can study. And in that case, all of the things that I've been showing you, the anthrobots, the xenobots, the worms with two heads, all that stuff, are vehicles. They're exploration vehicles. They're new interfaces for new patterns that help us map out the space. And so my proposal is that that space contains not only the low-agency, probably low-agency, we don't actually know, mathematical truths, so facts about prime numbers and things like this, but also patterns that we would recognize as kinds of minds.
In other words, behavioral propensities. So that space contains the familiar mathematical objects, but also types of behavioral competencies. And so my proposal is that when Hawking asked what breathes fire into the equations, maybe he had it backwards. Maybe it's the mathematical truths, both the very simple and the very complex ones, that actually breathe fire into the biological interfaces.
And certainly there have been both philosophers and physicists who were pretty clear about this, about the fact that there is a symmetry that goes through all of these sciences that are informed by patterns that eventually, if you keep asking, but why? Why is it this way? Why is it that way? You end up in the math department eventually. And there's this, I won't have time to say much about this, I have to wrap up, but there's this, we've discovered an incredibly interesting thing that might even suggest a kind of intrinsic motivation to the whole evolutionary process, which is a mathematical relationship between causal emergence and learning.
That they basically potentiate each other, forming this loop and this ratchet that aims upwards in both the metrics of causal emergence and learning. So the fact is that because biology is so interoperable, almost any combination of evolved material, engineered material, and software is some kind of viable agent that forms an interface for all kinds of complex ingressing patterns. The natural evolved beings are here, but the option space for the future is massive.
And if we make new interfaces, we're going to meet some new minds, and it is up to us to figure out how to ethically relate to all of these beings. And one of the key messages is that because of the patterns that ingress from this latent space, neither living things nor, quote-unquote, machines are what our formal models say they are. As Magritte was telling us, they are just that, formal models. They capture some things, they leave many things out, and most people are comfortable that the minds of living things and their motivations are not fully described by the laws of biochemistry, but most people think that that's not true for computers or machines or so on.
And I think that's not true. I think those things also receive the benefit of these kind of ingressions. And so I'm going to stop here and just say two things. First of all, that I think that there are patterns of form and behavior that are very important to guide the motivation of even minimal agents, certainly biologicals, but even minimal agents. I think that physics is what we call things that are constrained by these patterns. Biology is the study of things that are enabled by these patterns and that actively exploit the patterns.
I think this kind of thing has nothing to do with quantum effects and so on. Even kind of the classical, Newton's classical world was already showing evidence of this. Plato and Pythagoras already knew this. And I think our attitude towards novel beings, because of these intrinsic motivations that I think are prevalent in systems, should be less Skinner, so less reward functions and so on, and more Piaget, learning to recognize and hopefully align intrinsic motivations that come through all of these systems.
And I think the mind-brain relationship is basically the same thing as the relationship between mathematical truths and physics. And we can say many interesting things that I don't have time for here. And so just the last thing I'll say is that this is not just philosophy. There's a rich research program here. So building new interfaces to work out the mapping between the physical embodiment and the patterns that come through, quantifying how much do engineers and evolution get from this, and understanding the contents of this space leads to some very interesting questions.
So I'll stop here. I'll thank the students and the postdocs who did the work that I showed you today, lots of amazing collaborators. I'm not claiming that any of the people listed here part of my talk, this non-physicalist model, it's pretty unpopular nowadays, so I'm not saying that they subscribe to this, but the work that I showed you was done by all these people, the interpretations are mine.
And a quick disclosure, there are three companies that have licensed some of the discoveries that we've made and are trying to push them towards clinical and other applications. And they support the work as do our many other funders. So thank you very much and I will stop here.
