Thoughts as Thinkers: biology & computer science of inspiration within & without us

Watch Episode Here

Listen to Episode Here

Show Notes

This is a ~35 minute talk titled "Thoughts as Thinkers: the biology and computer science of inspiration within and without us" that I gave right before Chris Fields, at a recent philosophy conference. Still working on different ways to present the concept of patterns as behavioral policies, their origin, and applications of this idea.

CHAPTERS:

(00:00) Policies of navigation

(04:40) Biological free lunches

(10:15) Molecular learning ratchets

(14:00) Competencies of xenobots

(20:48) Latent space of patterns

(27:12) Embodied mathematical minds

PRODUCED BY:

https://aipodcast.ing

SOCIAL LINKS:

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

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

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

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

Twitter: https://x.com/drmichaellevin

Blog: https://thoughtforms.life

The Levin Lab: https://drmichaellevin.org

Transcript

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

Main Episode

[00:00] Thank you so much for having me here. It's a massive pleasure to do this together with Chris, because we work very closely on these things. It's gonna be really interesting to see our two perspectives.

[00:00] What I'm going to try to transmit today is basically this. I'm very interested in the origin of cognitive properties, especially of novel beings that have not been here before. One of the things that I do is try to generalize the concept of patterns as policies of navigation. Once you do that, you can think about policies of navigating all sorts of spaces. Of course, three-dimensional space and the conventional idea of behavior, but also anatomical morphospace, physiological state spaces, gene expression space, metabolic space, linguistic space, so on. And so we have this notion of patterns as policies for navigating those spaces.

[00:00] I'm going to make the claim that the three ways we have of basically paying for competencies or good patterns that do something interesting are design, selection, and training. I'm going to argue that these actually don't predict or explain what we see. More importantly, they don't suggest the important research programs that need to happen now. There's some additional source or sources of order that our conventional paradigms have not been capturing, and that's what I'm going to talk about.

[00:00] These patterns are on a spectrum. At the very low end of very minimal systems, it just looks like what people call emergence. I think this notion of emergence is too anemic for what we really need to do here, but at least it's a start. At the medium scale, what we see are unexpected problem solving competencies that were not paid for in the conventional three ways. And at the high level, I think the same phenomenon is what we call inspiration when we talk about human creatives.

[00:00] So what I'll do is show some model systems and just a few examples. I can go on for hours on some of these amazing examples, but I'm just gonna show a few. And then mention our research agenda, which is to map the space from which these patterns arise and try to understand better the interfaces that allow some but not other patterns to come forth.

[00:00] The first thing for us all to realize is that we are heading into a new world, which is going to completely blow up the distinction between living things, machines, humans, robots. I think none of these categories are gonna survive the next decade. What's going to happen is that, first of all, our neighbors are going to be some kind of cyborg with certain smart prosthetics, whether for medical reasons or for reasons of freedom of embodiment. We're going to be extended in various ways. Our mental health practitioners are going to have to deal with clients and patients who have very different kinds of minds. They have different sensors. They have different ways of engaging with different worlds that the rest of us can't live in right now. We're going to have to figure out how it is that we can share a world together and understand where their symbols come from. What is their collective unconscious like, if there is one, all of those things. That requires us to not only understand the novel beings, but actually to understand ourselves and where we come from and what we really mean when we say, here we are. There's a body and the mind and there's contents to the mind and so on.

[00:00] I'm also interested in this notion of inspiration, and I'll just read one simple example. This is from the poet Ruth Stone, who basically said that growing up, she would feel a poem approaching like a thunderous train of air. What would happen is she would have to run to her house to get a piece of paper and literally beat the poem to the house. If she was too slow, the poem would keep going and continue across the landscape looking for another poet. But if she caught it, she would be able to write it down backwards from tail to front, pulling it by its tail. That was her experience of what it's like. Most significant creatives have stories like this where they talk about receiving downloads and inspirations and so on. For example, Michael Jackson used to say that if he didn't immediately record certain musical ideas, he felt that God would give them to Prince.

[00:00] So this idea, both from certain mathematicians like Poincare and lots of different creatives, that there's a source for information that you don't pay for in conventional ways. What I'm interested in is quantifying this. I'm gonna call these free lunches, although I don't think they're actually free. I think they're heavily discounted. We do pay, but we don't pay, via conventional ways, all of the cost here. The free lunches are defined as what we got, meaning the competency of some particular system, minus the effort that was put in. And the effort is via design, evolution, or training.

[05:05] It's interesting to note that both Darwin and Paley were in complete agreement on something. They were totally in disagreement about what the source of the order is. Of course, Darwin thought it was selection. Paley thought it was designed by God. But what they were agreed on is that order is rare. It requires extraordinary effort and that most of the universe lacks competency, and that we have to explain either the haystack or the watch. All of these need to be explained. And I think there's fundamentally something incorrect about this view.

[05:05] Now there have been people studying this notion of order for free. For example, Stu Kauffman did amazing work in the nineties pointing out why and how random networks have certain interesting ordered properties. Even earlier, we have Grey Walter with his mechanical tortoises and this amazing book called "Vehicles" by Braitenberg, showing behaviors implemented by extremely simple kinds of artifacts. And he writes this: "And yet we know very well there is nothing in these vehicles that we have not put in ourselves." I think he's fundamentally wrong about this assumption, and it is an assumption. This is not a result. It's an assumption.

[05:05] So I'm really interested in this notion of what could we get for free that isn't just order. Order is nice, but it's a low-end kind of thing. But what do you actually get for free? So, again, these are the three ways that we know how to put in effort to get a good controller. So let's look at some biology examples. And, again, these are just a few examples. I could show lots and lots more.

[05:05] First of all, neuroscience knows quite a bit about the neural hardware and how that hardware is mapped onto behavioral competencies, or so we think. So based on standard neuroscience, you might ask yourself, what performance would you expect from a human brain with drastically reduced hardware? If you take a human brain and drastically reduce the amount of tissue, what would you expect? Standard neuroscience says you would expect drastically reduced performance. And most of the time, that's true, except not always. There are really interesting clinical cases that you can read. We review them in this paper where the majority of the brain tissue is simply missing. It's just not there. And yet the patient has normal or, in some cases, above normal IQ. So there's something wrong here. There's a disconnect. And I'm not claiming this happens all the time. These are very rare cases, but there's enough of them that we now see that there's something we're missing about the mapping between how much neural real estate you need and the actual functionality. There's something that we don't understand.

[05:05] Continuing with this theme of neuroplasticity, suppose I said to someone, I want you to design an animal. I want you to design a tadpole that would form an eye on its tail. I want it to be able to see out of that eye. I don't want the eye connected to the brain, but I still want it to be able to functionally see out of that eye. And, by the way, when the tail dies, on the journey to becoming a frog, I want the eye to ignore all of that and to do its own thing. What would we need to do? And you can imagine what grant reviewers would say about a proposal like that.

[05:05] The answer is, you don't need to do anything. It already works. If I take this tadpole, I prevent the primary eyes from forming. We put an ectopic eye on its tail. The eye makes an optic nerve. It does not go to the brain. It stops somewhere, maybe in the spinal cord, maybe in the gut, but stops somewhere here. And then we put them in this machine that we built that trains them for visual cues. And it turns out that, absolutely, they can see out of this eye. They can be trained on visual kinds of assays.

[05:05] So this is also very strange. Why don't we need lengthy adaptation, rounds of mutation selection to radically change the sensory motor architecture of this creature and maintain function? Why does it work out of the box? Machine learning people would call it zero shot. You don't need to do anything. You make this radical change, it works. Of course, we don't have a lot of, or maybe any, technology that works like this.

[05:05] Suppose I wanted to take a turtle, a shy, slow reptile. I wanted to work like a cat. What would we need to do? Well, millions of years of evolution would be required as effort, or perhaps some sort of neuroengineering that we have absolutely no idea how to do at this point. It's actually a much more simple path. Somebody put their turtle on a little skateboard, and immediately, it unlocks this amazing behavior. So the turtle is now keeping up with the cat in terms of the speed of their interaction. It wants to play. The cat looks a little freaked out by it, as you would imagine. As should we be, because this raises a lot of interesting questions. Why is a simple passive prosthetic like that unlocking these amazing capabilities? That isn't typically where the properties of these beings come from. It usually takes a lot of evolution, or perhaps engineering if we knew how to do it.

[10:12] So let's look at the material that all of these things are made of. The examples I showed you so far were cellular kinds of organism level. Let's look down at the molecular level. Suppose I wanted to design a chemical network that could learn from experience, meaning that prior events would affect its future behavior in the way that behavior scientists understand. I wanted to increase its causal emergence. For example, we could use phi metrics. So a measure of integration, by which the system is more than the sum of its parts. I wanted to increase the phi every time it learns something. I wanted to not lose those gains when it's forced to forget the things it's learned. What would I have to do? How much would I have to design and select from the space of possible chemical networks to get this to happen? It turns out that lots of networks already do this.

[10:12] So molecular networks, as it turns out... And in this case, what we're doing is modeling them with sets of coupled ordinary differential equations. It turns out that if you do assays familiar to behavior scientists, like, for example, associative or Pavlovian conditioning, we find the following. We find that many networks can in fact learn. They can learn in roughly six different kinds of behavioral paradigms. When you do train them, the causal emergence goes up either continuously or in these spurts. When the causal emergence goes up, they actually become better at learning. And if you force them to forget, and we do this for biomedical reasons, it's important to have some of your molecular pathways forget things, in certain disease states. When you force them to forget, it does not bring them back down.

[10:12] So this is a remarkable, asymmetric ratchet that means that every time they learn something, they become more of an integrated agent. Every time their integration goes up, they're better learners. But if they are forced to forget, they do not drop back down. So it's an asymmetric feedback loop. It's a positive feedback loop between learning and integration that points upwards towards higher agency and higher intelligence.

[10:12] Now, if I were to look for networks that do this, it's pretty easy because already, random networks, not designed by us, not selected, not any of the typical ways that we get such an incredible thing, random networks already do this. It turns out that random networks are incredibly close to being optimal for this kind of ratchet. Evolution can tweak it and exploit it, of course, and, you know, apply it in various ways, but it doesn't need to create it. It's already there. It precedes replication. It precedes Darwinian kinds of processes. It did not need to do this. The properties of this ratchet are a free gift from mathematics. It doesn't come from physics, meaning it doesn't depend on any fact that physicists discover. It doesn't depend on selection or a history of biology. It is already there in the material that evolution gets to use. And in fact, not just the material, but the properties that come from mathematics. And this is gonna be very important in a few minutes.

[10:12] And we can talk about whether at least some portion of the space of these kinds of properties is fine tuned, like people have suggested for the parameters of our physical universe, because they're so optimized for something that basically drives this spiral of learning and intelligence.

[10:12] So that's the material. And now here are the consequences. We'll jump forward to what this means for beings that have never been here before. What I've told you up until now is that there are competencies that do not seem to rely on the physics of the environment, meaning you can't blame it on some aspect of the world they live in, nor on the history of selection forces for specific outcomes. So now we can talk about novel organisms and how these properties that go all the way down to the molecular networks affect what happens when you make new beings that have never been here before.

[10:12] So suppose I told you that we wanted an autonomous motile organism. I want it to contain no neurons, but I do want it to be able to make copies of itself from cells it finds in the environment. So von Neumann's dream of a machine that makes copies of itself from stuff it finds around. I want it to have hearing, basically to respond to sound. But I'm gonna insist that you use wild type frog cells. No genetic modification. No synthetic biology. Just wild type frog cells. How much bioengineering would we have to do to make something like this? If you think ahead of time of a project that you might be funded for, you think how much effort, how many people over how many years it would take to create something like this from scratch?

[15:09] It turns out that actually you don't need to do anything because if you just give some skin cells, so these are epithelial cells that are going to be the skin of the animal, in a frog embryo, you give them freedom, meaning you extract them, liberate them from the frog embryo, you get something called a xenobot. So here it is. You get this motile little creature. And if you sprinkle a bunch of loose epithelial cells in the environment, what you see is that they run around. They collect these cells into little balls, then they polish the little balls, and they become the next generation of xenobots. Guess what they do? They do the same thing, and that becomes the next generation and the next generation and so on. They express tons of genes differently than these cells would if they were still in the embryo. If you look at those genes, you find out that a bunch of them have to do with hearing. So we decided to check. We put a speaker underneath the dish. Sure enough, they respond to various kinds of sounds.

[15:09] Having seen all that, we might say, okay, but now I want it to learn. What would I have to do to this thing to make it learn? Adding some kind of a memory circuit, sensory machinery, connect it all up together. It turns out you don't need to do that. They already have learning capacity. They can form memories of at least two different experiences that we can give them. In this case, chemical experiences. This is just scratching the surface. We're just beginning to understand what their preferences are, what kind of things can they learn, how smart they are. We don't know. This is just the very beginning. They can form two different kinds of memories and hold on to them for at least twenty four hours that we can then read out afterwards.

[15:09] If you would like it to swim in more complex patterns, you might think that you would have to put in some kind of a nervous system. I mean, already here, the fact that they learn... And remember, these are just skin cells. There is no nervous system here. You might think that you would have to either design or engineer or select some kind of a nervous system architecture. Actually, you can just jam a bunch of neurons in there. That's what we did. We just shoved some neural cells in there. They form some kind of a nervous system that begins to answer the question of what do nervous systems do when they don't have a history of sculpting by selection? And this is what you get. You get this organism with much more interesting complex patterns. Of course, they don't have any muscle. So what you're seeing here is not the conventional thing where the neurons are running muscles in a fish or something like that. These neurons are actually affecting the cilia. They're affecting the little hairs on the outside. So this is a very strange kind of circuit.

[15:09] What if you wanted the exact same thing, now made of human cells? Not embryonic, so no more embryos. So you can't rely on embryonic plasticity. You want the same thing out of human cells, and you want it to be able to heal neural wounds. What would you need to do? Again, if you think about the amount of effort that would have to be put in either by you as the engineer or by selection, you would have to have a long history of situations where this kind of thing was beneficial, such that you would then evolve such an organism. Turns out, again, that you don't need to do anything. If you liberate some adult tracheal epithelial cells from humans, again, not embryonic, adult, they too will form this little creature that runs around, does various things. It has a wild type human genome. If you were to sequence it, you have no idea that this is anything but a human. And it turns out that if you put it on top of a neural wound, so we plated a bunch of neurons here, took a scalpel, put a big scratch through it. If you put them down in this environment, what they'll do is form a superbot cluster and start to heal the neurons across the gap. If you lift them up four days later, this is what you see underneath.

[15:09] So let's step back and look at what's going on here. We have to ask the question of when we paid the cost, the computational cost of all of these competencies. We know, at least we think we know, when we paid the computational cost of making good tadpoles and eventually good frogs. So this genome has been bashing against the environment being selected to make this developmental sequence and then these tadpoles with specific functionalities. But when was the cost of making xenobots paid? There's never been any xenobots. There's never been any selection for kinematic replication to our knowledge. No creature does or ever did kinematic self replication. They do have a developmental sequence. This is a week old xenobot. This is an 83 old xenobot. It's turning into something. I have no idea what it's turning into. These things have never existed before. They have their own behaviors that they do. Lots of weird things that they do.

[15:09] If you do a kind of reverse ecology to take their transcriptome and work backwards, what kind of environment would have given rise to those kinds of genes being expressed, it isn't this world. It's a somewhat related world, but it isn't this world. So there's no straightforward evolutionary backstory to this. In other words, these cells existed all the way back. But evolution is supposed to explain the complexity and the specific competencies of a creature with high specificity for a past history.

[20:16] That was the whole point of the Darwinian story: you're supposed to be able to explain what creatures are doing by reference to the past environments that push them towards these outcomes. You can't tell that story here. Okay? You know? So you can try. Some people sort of wave their hands, and they say, well, you know, complex things will do complex things. But that isn't enough. We need to understand the specifics of it. We need to understand why these 600 genes, why these competencies, and the standard way of predicting those does not work.

[20:16] So what I propose is simply this. There's basically two options when faced with all of these examples where the thing you got out, the competencies you got out, and the amount of effort that was put in that we can see according to our three ways that we know. When they don't match, there are two things that you can say, basically. It seems like the majority of you today, this is what most people say, is that it's emergent. And when I ask what that means, some people say, well, it's a set of facts that hold. There is no upstream explanation. If there was, it wouldn't be emergence anymore. It would just be a normal thing. But there are certain patterns that are simply emergent, which means there is no upstream set of facts that explain it. These are just facts that hold. Okay? That's it.

[20:16] And the benefit of that, of course, is you can have a very sparse ontology. You don't have to have another quote unquote realm. I don't really know what that means, but I know people really don't want another realm. Everybody wants monism. But first of all, sort of by definition, I think you buy it on the cheap. Meaning, you just insist that we're going to shoehorn everything into one kind of world. And second, I find it a very mysterious and pessimistic view. I don't want to think that there's a random set of facts that hold, and we'll just sort of write it down in our big book of emergence when we get surprised by interesting things. I would prefer the more optimistic assumption, which I take to be the basis of all science, that these things are not just random facts at all. There's an ordered structured latent space of patterns. Again, patterns being various policies for doing things and perspectives that a being might take on the world. There's a latent space of patterns, and we should be studying this systematically. There's a research agenda to understand what patterns exist, what patterns do not exist, when do we see them. And I think that these synthetic morphology beings that we make and various other hybrids and cyborgs and all this stuff, I think all of these are a kind of vehicle for exploring this latent space.

[20:16] Now one of the things that people sometimes think is that, well, biology is amazing at this. Life is very special. For whatever reason, living things are subject to these unexpected free lunches. But over here, we have these dumb machines that are in fact perfectly described by algorithms, and these dumb machines only do what we tell them to do. So biology is magic. Maybe it's subject to some of these free lunches. But at least the machines only do what you tell them to do.

[20:16] So we have a research program on this, and the reason is that the biological examples, while super impressive, are very complex. And in biology, you can never really prove anything because you never know all the mechanisms. There's always some kind of quantum microtubule or something somewhere that you don't know about. So it's important to also have minimal models where we know exactly what we put in. It's a classical deterministic system. There aren't any quantum magic effects. We know exactly what we have. So we have some simple models from computer science that, unfortunately, I don't have time to really delve into. Also, there's a much bigger paper coming on all of this soon.

[20:16] But, again, we see the same thing. In engineering and robotics and computer science, we really have three ways of paying for competency. We can design it. If we know how to solve the problem ourselves, we can design it. We can run an evolutionary search to select for it, or we can train it if the system can learn and if it has access of a past experience with something. For example, in the case of the xenobots, I haven't been training any xenobots to practice their kinematic self replication.

[20:16] So I'll only mention one example. There's a lot more coming. If you wanted to have a short deterministic algorithm... And just imagine as a coder, your boss tells you, she wants a short deterministic algorithm. The algorithm needs to be written for the assumption of reliable hardware, but we also want it to work if the commands occasionally fail. So we write it for reliable hardware. We want it to work in unreliable hardware. Also, we wanted to do something else that's interesting at the same time, not just complexity, not just unpredictability, but actually to do something else that's interesting that is not in the algorithm. For example, maybe it will implement homophily, which is this biological tendency for beings to hang out with others of their own kind, which I think is quite relevant to this imperial model of multicellularity that Chris and I wrote about many, many years ago. And I wanted to do delayed gratification.

[25:24] I wanted to do something that even some human children have trouble with, which is the marshmallow test. It's the ability to go further away from your goal in order to recoup gains later. So I want the code to do all this stuff, but I'm not adding any extra code for any of this. I want it to just be there, and I don't want you to write any code for that. That would be a crazy thing to be asked to code up. And I think most people would say this is impossible. Turns out that even very minimal algorithms already do this.

[25:24] The first paper in the series looks at sorting algorithms, and it turns out that not only do these classic sorting algorithms work on a nonreliable medium, even though there is no code to actually check whether any of their commands were carried out or not, it still works perfectly well. Turns out that they do other interesting things. You could call them side quests that are basically not asked for in the algorithm. And if you look at it the right way, you can actually see that in many cases, it's doing delayed gratification. So we have these minimal examples of this.

[25:24] We also have a project where we actually give bodies, so in silico and eventually real robotic bodies, to various mathematical patterns to ask the question, what kinds of behavioral policies, what kind of intelligent minds can be frozen in various data patterns that are not obvious to us because they don't have a body yet. Can you give them a body and see what is it that they actually do? And some of these patterns are mathematical structures, for example, that nobody designed in accordance with the problems to which we are putting them. But they have competencies that nobody anticipated. And so I think this is really interesting.

[25:24] In the last couple of minutes, I'll just say what I think all this means. First of all, I think everybody should think about this continuum of change. If you do think that human creatives have access to something that is interesting and beyond the typical... And I'm not saying it's mysterious. I'm saying the absolute opposite. I'm saying that we should be studying the latent space of these things. If you think that humans have access to this, it doesn't make any sense to then make this categorical distinction with dumb machines that are forever barred from intuition.

[25:24] It's kind of popular in some circles to say that humans have intuition and inspiration, but machines are just algorithms. They'll never be able to do that. This is the situation that we really need to worry about. I mean, do people think that this starts to wink out once you start augmenting yourself with engineered components? I think that's a very implausible view.

[25:24] I think that while it is true that the formal system of chemistry does not tell the story of the human mind, these low level descriptions do not properly describe what the biology is doing. So for example, computationalism and so on, I think, is insufficient to understand what the biology is doing. But we shouldn't feel too smug, because I think these formal models are not fully describing what the machines are doing either. We should just remember that all descriptions are basically metaphors that are useful, hopefully useful in some circumstances, but they fail in others. And I think it is a big mistake, even if you accept that biologicals are doing something interesting, to assume that these so-called dumb machines aren't going to be doing that simply because we have this formalism that says that they don't do that.

[25:24] So I'm arguing for experiments over assumptions here, and our experiments have been quite interesting. I think minds are not defined fully by our models, neither for the limitations nor for their competencies. And I think partly because our models in all these disciplines are models of the front end, the thin client, not the thing itself.

[25:24] And what do I mean by that? I'm saying that there is a well known source of patterns beyond physics and biology, and that's mathematics. We know, and we've known for centuries, that there are things like this. This happens to be a Halley map of this kind of simple function. The specifics of this pattern are not given by any truths of physics. It cannot be changed by changing the constants of physics, as far as I know. It wasn't selected by anybody. The specifics of this belong to a different class of explanations than we use in the standard physical sciences. And Chris will say much more in-depth things about that. I'm simplifying it greatly for someone like me who's basically an engineer trying to understand how to have relationships with various systems.

[25:24] So the thing is, we know this about mathematics, and we know that most mathematicians don't think it's a random grab bag of facts that hold. It's a structured space that they're investigating.

[30:31] But I don't think we can assume that mathematicians have a monopoly on these kinds of things. We should be asking what other disciplines are affected by this. And one way is, as I said, when you give these mathematical structures bodies, you get to learn things that are relevant to behavior scientists. These free lunches, I think, have to do with patterns that are not limited to conventional mathematical patterns that provide competencies for navigating spaces.

[30:31] And what I would argue is that there's a structured space. There's a research program here. We can formalize the degree of effort versus payoff. We can make new interfaces, and we can try to generalize these patterns into novel problem domains.

[30:31] The last thing I'll say is this. What I think is going on here is that the space of patterns that has been called the platonic space in the past contains the things that are low agency kinds of patterns that sit still long enough to have formal derivations and proofs and so on applied to them. But then it also has much more complex high agency patterns that are recognizable to other disciplines. For example, there are cybernetic truths and theorems that are somewhere between math and behavior science. And as we've shown, even groups of ordinary differential equations are mathematical objects, but they are of tremendous interest to people doing cognitive science because they can already do Pavlovian learning and so on.

[30:31] So I think some of these patterns are recognizable as mathematical truths. Others are basically recognizable as kinds of minds, kinds of policies for navigating different environments. And so the big picture for me is that whereas, conventionally, we say that there's a lot of dead matter out there. Some of it is alive, and a tiny corner of furry mammals, and maybe a whale or an octopus, have some kind of intelligence to it. I fundamentally think that what we have here is an enormous space of patterns that are in fact kinds of minds. Some of them at any given moment have physical interfaces, robots, machines, embryos, living bodies, cyborgs, and so on, through which they ingress into the physical world and have various experiences and so on. And some percentage of those we would consider alive. They're wet and squishy, and they undergo Darwinian dynamics, but certainly not all. So I turn this whole thing upside down.

[30:31] So I'll just summarize and stop here to say that I think the invariant across biology, cognitive science, and computer science are patterns that are fundamentally policies for problem solving. I think we need to study the structured space from which very specific patterns ingress once you make an interface for them that allows us to recognize them. There's a research program here. We have, I don't know, five or six people working on all of this stuff.

[30:31] And I think this notion of inspiration is really important. It works on all scales, inspiration all the way down. Even molecular networks are the beneficiaries of these free lunches that are important, but that were not paid for by conventional means in the physical world. That would be my claim.

[30:31] And much of this is described in this paper, I mean, in Philosophies, and there's this symposium where a number of people here have given talks. And by the way, it should be very clear, most of them do not agree with all of my views here, but they do have very interesting, relevant thoughts on aspects of this.

[30:31] And I think this is all critical because all of the biologicals that we see here on Earth, Darwin's endless forms most beautiful, are a tiny little region of this space of possible combinations of embodiments from evolved systems, engineered machines, and software. All of them are subject to ingressions from various kinds of properties that we cannot predict in advance. We don't have yet a science of this, but we are all going to have to live together, hopefully, in some kind of ethical symbiosis. And so trying to understand these other beings and where they come from, I think, is fundamental to understanding ourselves and where we come from. Because if the conventional evolutionary and physics story is not sufficient for these guys, it probably isn't sufficient for us as well. And a lot of the assumptions that we've made about natural forms are probably up for grabs too.

[30:31] So I'm gonna stop here. Thank the students and the postdocs who did all the work. Lots of amazing collaborators. And as a disclosure, there are three or four companies that have now licensed some of our inventions, so I need to say that. And thank you so much, and I look forward to seeing what Chris has to say.