"Thoughts as Thinkers: the biology & computer science of inspiration" by Michael Levin

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] Right. Well, thank you so much. That was that was very kind. I appreciate the chance to share some ideas with with all of you. And what I'm going to talk about today is a really strange thing I'm trying to do, which is to develop laboratory model systems for studying inspiration and intuition. We'll we'll we'll we'll see how... We'll see what you think. And and I have two two main ideas to transmit, are really not sort of popular mainstream ideas today. One is that there is a literally a nonphysical component to this. I don't mean it's mysterious. I don't mean we can't study it. In fact, we have a research program on how to study it. But I specifically mean that it is not captured by the standard formalisms that we use in the physical and computational science sciences. And, and the other main point is that inspirations, so the specific content of inspirations, and those who are inspired, so agents of all kinds that humans and otherwise that become inspired are not actually different things. They're part of a continuum of the same thing. So this is this is what I'm going to talk about. So one reason, I really care about this is because, we're entering a world in which, this distinction between, humans and so called machines is going to be irrelevant. I don't think those terms will really survive the next couple of decades. What we're really talking about are a a spectrum of novel beings. This is this is your neighbor. This is, each each of us, are going to have all kinds of different extensions, whether for for medical purposes or for just sort of freedom of embodiment. And, for example, I gave a talk to a group of, psychoanalysts and psychiatrists and kind of pointed out that their client list is going to get very, very odd very soon. And this notion of helping beings have a better life, understanding their, their dreams, their symbolism, their potential, all of this, when you're dealing with beings that are not your standard, everyday modern humans is going to be really, really tough. So we need to develop a science of understanding where, novel minds, and their properties come from, what defines the properties of novel minds when you can't just say it was evolution and and and and and design. But but part of part of living with novel beings is trying to get a better understanding of what what what are we ourselves as... Essentially. And so this is this is what I what I want to talk about today. So and so we'll talk about, a little bit of, definitions, and then I wanna show you some some examples from biology and computer science, which are... Which I think are very relevant to inspiration and intuition. We can operationalize them as the notion of a free lunch. I'll talk about what that is and then a few ideas for the for the future. So so so this kind of thing is is is fairly common, but I like this one this one in particular, creatives that feel that, they are the recipients of particular inspirations. So this poet, described her process of, hearing a poem approaching like a like a thunderous train of air. She had to run, to get to her house to get a piece of paper. And if she if she made it, she would be able to capture this poem by the tail as she perceived it and would would write it down backwards from from back to front. That was that was her experience. If she missed it, she said it would keep going in search of another of another poet. Michael Jackson had a had a funny quote that he said, if he didn't immediately record, ideas that he got, that that he thought that god would give them to prints instead. So there's this there's this notion of, in in in creatives that there's that there's something that they're tuning into. In other words, there's a difference between these kinds of cases where on the one hand, you sit there and laboriously work through sort of turn the crank, follow the... Follow some kind of some kind of strategy. You solve a problem. You know exactly when you put in the effort, how much you put in the effort. Then you've got somebody like Von Neumann who looks at it and doesn't need to do all of this stuff. He sort of kicks it around and says, okay. I I kinda know what this is. Then then you have these other cases like Ramanujan who says, I I don't know. I I I was told this by... I was whispered these theorems by a Hindu goddess, and so here here they are. So you have this kind of range of ways in which you might imagine navigating a space of ideas, a space of answers to various questions, the space of, kinds of, things that you would like to, that you would like to find. And and we've done some some work on quantifying what it means to traverse that space in different types of, different types of strategies, whether you're laboriously, going step by step or taking leaps or, you know, sort of giant leaps. So this is a landscape that that you're, traversing. And and I'm going to then then sort of turn that whole that whole metaphor upside down in a minute. But if if we think about it that way, what we could try to quantify is the following. What is the delta between what we received in terms of competencies of some sort minus the effort that we actually put in? Okay? So if we try to quantify this, what... Was there anything left over? The standard model today says there are basically three ways that you can put in effort to get something competent. You can design it, meaning meaning you know how to solve the problem, so you engineer an algorithm or a machine, to do it. You could evolve it. So you don't know how to do it, but you generate a trillion variants, throw away everything that doesn't work, you know, sort of keep keep what's left.

[05:04] Or training, meaning that the system has had experience with some kind of a problem, took its lumps, learned learned from that experience, and eventually became good at it. So so the the one thing that Darwin and Paley agreed on, they didn't agree on how it happened, but they both agreed on one fundamental thing, that order is very rare, that most of the universe is random junk, and that if you're going to explain things like like a watch or, or or the living world around us, something something extraordinary is happening. You you need to put in a lot of effort to get something that is competent, whether it's living or nonliving. But you need you need... That that level of of competency requires explanation. Course Darwin thought it was evolution. Paley thought it was it was a rational design. Either way, they thought... They they could agree on that one thing. That order is rare. Now there were already clues going back decades ago that things are not quite as they seem. So for example, Walter Gray with his with his tortoises and, and, Breitenberg had... If you if you haven't seen this book, you should really pick it up. It's a thin little book. It's fairly remarkable. They make these he makes these incredibly simple little, little robots and sort of characterizes the the suite of behaviors that they're, that they offer. And he says, we will be tempted to use psychological language to describe their behavior, and yet we know very well there is nothing in these vehicles that we have not put in ourselves. That is a massive assumption. I think I think he's wrong about that. The rest of the book is genius. That that assumption, I think, I think is completely wrong. This is... We we need a little humility here because there actually, I think, is something in many systems that we did not put in ourselves. So so then then, of course, Stu Kaufman with his order for free. This this kind... This this was kind of really inspirational for me in the in the in the eighties when I, when I when I was reading reading his stuff. But I think we can go much further now. It isn't just order for free. There's something much more interesting that, that you get for free. And so, I'm going to talk about what that might be. So, these are the sort of, so I'm gonna talk about the biology end and then and then the computer the computer science. And so on the biology end, you know, we have bioengineers. We have natural evolution. We have learning. This is how you get competency, we think, in, in biology. So let's look at let's look at a few examples where where this fails us. What I'm going to do is show you a number of examples where I think it's it's very clear that these three standard things to look for are not telling the whole story. There is something else that the standard paradigm is leaving out. So first of all, okay, neuroscience knows quite a bit about the neural hardware, how that maps on to certain behavioral competencies. So what performance would you expect from a human brain with drastically reduced volume? So you sort of stop to think, and you say, wow. Yeah. Drastically reduced volume, should be should be total impairment. And most of the time, that's true, but it's not always true. There are, there are interesting clinical cases, and this is we... Karina Kaufman and I reviewed these in this in this in this paper, where you have situations like this, where if you were to take the... You were you were to take a a CAT scan, you would say, okay. This this individual barely has any brain at all, then then you could be surprised to find out that they have normal or above normal IQ. There are numerous cases that have been studied in literature. So some... Something is something is not is not matching up here. And it's not like you couldn't try to shoehorn this into into stories of neuroscience by talking about redundancies and so on. But this... But but the paradigm does not predict that that is going to happen. This is this is this is completely sort of out of out of paradigm data. So let's let's continue with that concept of plasticity, that kind of plasticity. Suppose I told you that I wanted you to make a tadpole of a frog that would have an eye on its tail, and I wanted to be able to see out of that eye. I wanted to functionally see out of that eye, but I don't want it connected to the brain. Oh, and by the way, when the eye dies off, when the when the tadpole begins to turn into a frog and all those... The posterior tissues get this, cell death signal so that the tail could die, I I want the eye to ignore that and and just, like, you know, continue to grow. What would you need to do? What what would be the effort that you would need to put in? So you might you might imagine that this is a really difficult bioengineering problem. You might imagine how you would how you would do it. It would take a lot of effort. Actually, you don't need to do almost anything. So here's a tadpole that has... We prevented the primary eyes from forming. That's just so we could test the vision out of this one eye. We put an eye on his tail. The eye makes a spy... The... An optic nerve. The optic nerve does not go to the brain. It goes a little bit sometimes on the spinal cord, sometimes into the gut, sometimes nowhere, but it doesn't connect to the brain. Then we built this machine to visually train these animals on visual cues so that they would have to respond respond to certain visual cues. And we found out that they see quite well out of these eyes. And, of course, when the tail dies, the eye ignores all of that completely and sort of rides back and keeps growing. And now now you've seen a frog with an eye on its butt. So so this is interesting. Why doesn't this require new rounds of adaptation, mutation, selection? We... Is a radical change in the sensory motor architecture of this animal. Why does this work out of the box? Some sort of, you know, zero shot kind of thing. Why why is there why is there no change needed?

[10:11] And it's a... We we... You know, we... I don't I don't have tons of time today to talk about all the details. I have a story of why I think it works, but we'll, we'll keep going. Okay. Here's another challenge for you. Suppose I wanted to take a shy, slow reptile turtle and give it a playful cat like speed of life. I want this thing to be like a cat. What would you need to do? We have no idea how to neuroengineer such things, so you might think, okay. Well, evolution is the way to go. You need millions of years of selection pressure to get this thing to work like a cat. So here's a piece of data from from the Internet. So so some some some... Somebody put their turtle on a little skateboard. It it basically immediately unlocks this this this new behavior. The turtle is now, he wants to play. He's keeping up with the cat. The cat, I think, is a little freaked out by this by this turn of events, but, he didn't need to do much. This is a this is a passive prosthetic that has unlocked these amazing properties. So so I don't know what we say about this. Do all the turtles for millions of years have been hanging around able to do this and just, like, not been not been able to release their other capabilities? It's amazing. And you sort of wonder what else could you release with slightly better prosthetics, and what's gonna happen when humans have smart prosthetics, which we already kinda do, and so on. Let's let's let's keep going. And, again, just to what... You know, the reason I'm showing you this this zoo of examples is to illustrate the delta between what you are led to expect from, the effort that was put in by specific, those those three specific ways and what you and what you actually observe. So suppose I wanted to, I wanted to have a chemical network that has the following properties. I wanted to learn from experience, meaning that prior... Meaning that future behaviors of the network would be different based on prior, experiences or stimuli, and these these differences would match things I see in the behavioral science handbook, you know, habituation, sensitization, Pavlovian conditioning. I want a chemical network that can do this. Furthermore, I wanted to increase its phi metric, so some surrogate of of causal emergence that says how integrated the thing is. To what extent is it more than the sum of its parts, some sort of measure of integrated emergence. I I I wanted it to to raise its five every time it learns something. You know? That that that learning would would make it to be more of an integrated agent, but I don't want it to lose the gains when it's forced to forget those memories. And this is the stuff, we we do in our in our lab is we, we try to, for example, train gene regulatory networks and so on both to to gain and lose memories for biomedical purposes. Most of my lab doesn't study this this weird stuff. We do diverse intelligence in cells and tissues and things for for biomedical purposes. So so suppose I wanted a network that had these properties, how much would I have to design or select? Or... I mean, that that sounds like a pretty tall order to make a network like this. Well, it turns out that most networks... And what we do is we model them as everyone else does as coupled, gene regulatory, coupled, ordinary differential equations, for example, GRNs, pathways, things like that. And then we test this in real in real cells, of course. So, what we found is that is that coupled, sets of, ODs, even small ones, you don't need some trillion parameter model. Like, the smallest one that does Pavlovian conditioning is just four nodes. But but these these relatively small networks, they have the property that, a, yeah, they can be trained in in five or six different paradigms. Also, when you train them, their causal emergence can go up. It it tends to... Not not in every single one, but there's a large class of these where it goes up. It... There's a couple of ways that it happens. Sometimes it just continues climbing. Other times, it's interesting. It climbs every time you stimulate the network. So while you're talking to it, the the causal emergence goes up, so it sort of pulls itself together. But it can't keep itself awake when you're not talking to it. So in between stimuli, it sort of dissolves back into the into the void, but then it wakes up again as soon as you talk to it. And every time you do it, the causal emergence goes up and up and up. So so already, you have this. Certain networks do this. They have a higher causal emergence, which which helps them learn better. Learning raises causal emergence, so this is, like, positive feedback loop. But forgetting, when you force them to forget, and so this is all experimental data that is currently in in in review, it does not erase these gains. It's an asymmetrical ratchet that points upwards. So how hard was it to find something that does this? Yeah. Actually, not find it... Not hard at all because random networks already do this. Okay? So so if you look at random networks, they're basically as as optimized as as as possible. The evolution didn't need to do much. They were already... The material with which evolution works is already primed for this positive feedback loop between learning and emergence. It's it's it's really it's really quite remarkable. This has nothing to do with replicators. This is long before replicators appear. This is a free gift for math. This does not depend on any fact of by... Of physics. This is long before selection has its, has its say... Of course, selection will shape it and use it and whatever. But but the fundamental fact that this thing has an incredible asymmetric ratchet pointing upwards in intelligence and and agency comes from the math. So so we can start thinking about, I I mean, I don't have a strong claim about the entire space, but we can start thinking about whether at least some parts of the space of mathematics is sort of fine tuned, like people have noticed about the physical universe, you know, the the anthropic principle and all that.

[15:20] So that's interesting. You didn't have to do anything. This just sort of show show... Shows up for you. Suppose, suppose we wanted to take... To to make an autonomous motile organism, a new organism, which would contain no neurons. I wanted to make copies of itself from cells it finds in the environment, kind of von Neumann's dream of a machine that goes around and and makes copies of itself. I wanted to have hearing to so to respond to sound, but you're only gonna use wild type set of frogs from cells. You can't... No no no genetic editing, even if we knew how to edit such a thing, which we do not. How much bioengineering and synthetic biology would you need to do for something like... To to make something like this? This is this is a no no... You would never get a grant to work on something like this because because it would they would... It's it's it's beyond our our capabilities, for for sure. Everyone knows this. Turns out you don't need to do anything. All you have to do is take some epithelial cells out of an early frog embryo. You set them aside, and what they do is they self assemble into something we call a xenobot. I could do a whole hour just on on xenobots. They're pretty amazing. They do many things. One of the things they do, even though they can't reproduce themselves in the normal froggy fashion because they're only, they're only epithelial cells. They don't have all the other reproductive stuff. But if you if you put a bunch of loose cells in the environment, they will run around. They will collect them into these little, little little, little piles. The piles mature into the next generation of senobots, and guess what they do? They do exactly the same thing. As it turns out, they they upregulate a bunch of genes related to sound perception, so we stuck a speaker under them. And sure enough, we found out that when you play sounds to them, their behavior changes, so they respond to to acoustic stimuli. Now you might say, okay. I want this thing to have the information architecture like a human brain, even though it's only made of epithelial cells. What would you do? You have no idea how how you would do this. Turns out they already do it. If you look at the at the calcium signaling and you apply the kind of, the kind of metrics that that people apply to neurological patients and so on to figure out if there's somebody in there versus just a pile of, pile of neurons, what you find is that xenobots are as different from the null models as human fMRI data is from their null models. And I'm not saying that xenobots are like brains. I'm saying that both brains and xenobots are doing something much more, much more fundamental. Okay? So none of this... We couldn't even be... We couldn't begin to to to engineer any of these things. And what we found out also is that, by the way, well, how about how about asking them to form memories? How about how about learning? Yeah. They... You... Again, no no clue how you would do it. You don't need to do any of that. They already do it. We found out recently that, send a bots can form at least two different kinds of memories. They remember it for at least twenty four hours. We can read out those memories. We can tell which bot has had which experience by decoding by decoding their memories. If you would like it to swim in complex patterns, so you might think it would have to have some sort of very specific nervous system. We jammed a bunch a bunch of neurons in there just randomly ask... Wanting to know what would the nervous system of a novel creature look like that's never that's never been here before. So here's the here's the neurobots swimming around. This is what the this is what the novel novel neuro... Nervous system looks like. Aside from from philosophical questions about competency and so on, there's a very practical thing going on here because, both in the life sciences and in computer science and engineering and so on, we typically have to, we we we typically... We think we know how much effort you have to put in to to to do certain things. And when you have, such things showing up that you didn't, you didn't forbid, but neither did you ask for, it becomes really critical, I think going forward for humanity becomes really critical to develop a science for understanding where these these novel, patterns come from. And I don't think it's enough to say emergence. That doesn't really, I think, even begin to scratch the surface of of what's going on here. We need to go far beyond that. So so here's a very practical question. So so we know when the computational cost was paid to have a good frog and a and a good, frog embryo. So for for eons, this genome was bashing against the environment, and and it gave you this set of set of developmental patterns and then, you know, then then these tadpoles with their behaviors. Great. When did we pay the computational cost of of of of designing a good, xenobot? They have a developmental series. He... I don't know. This is an 83 old xenobot. I don't know what the heck it's turning into, but it's turning into something. Here are here are the the the behaviors they have. When did we pay the cost for these specific competencies and features? We have to know this. So so sometimes people, say this. They say, okay. So maybe biology has some kind of, special magic to it. Maybe maybe there's something about evolution. Maybe it's just its enormous complexity. Maybe there's there's quantum, quantum, effects here somewhere. And so the idea for bio inspired design is... And then so so we we work with a lot of roboticists and and and engineers and so on. And so so there's this notion of bio inspired design that we take things that biology is uniquely good at, and we try to rep... Reproduce them in our technology.

[20:24] Now in order to reproduce them in your technology, how are you going to do that? Well, you have the same three ways. You can either try to engineer it in. So you're gonna write an algorithm. You're going to create a machine of some sort that's going to do exactly what you saw the biology doing. You're going to try to force it to do that. Sometimes people use evolutionary strategies, so so genetic algorithms, evolutionary search, things like that. And, of course, especially more recently, training. So so connectionist architectures, we actually train the thing. So so let's look at some computer science examples that basically, I think, tell us that the same situation that was going on with biology is is here as well. Because this is important because a lot of times what people assume is, okay. I think I think this is kind of a standard, common view. They say, okay. Look. Humans are not well captured by the laws of biochemistry. We are more than the mechanical laws of biochemistry. It's kind of hard to to, to really specify why, but I think most people feel that the laws of chemistry do not tell the whole story of the of the human being. But that's okay because there's these dumb machines over here, and they are fully captured by their formal model. We are not fully captured by the formal model of chemistry, but these dumb machines are fully captured by Turing computation and and and so on and these other these other kinds of things. So... Okay. For those of you who write code, especially, I I would say, suppose I told you that I want a short deterministic algorithm that was basically, assumed reliable hardware, meaning that it's not going to have any code to see whether, the things it's trying to do actually worked. Right? So no no success no success or failure checks whatsoever. But I want it to also work when the command sometimes fail with a... With an unreliable medium. I also wanted to do something else that's interesting that isn't in the algorithm at all. For example, I wanted to implement homophily, such as, this... It's a biological thing where similar similar beings tend to sort of hang out with others of of their own kind. Oh, and I wanted to do delayed gratification, something that certain human children can't do, you know, the marshmallow test. I wanted to be able to step further away from its goal. So so up a gradient in order to recoup gains later on. I'm not adding any extra code for any of that. Just just just think about that. Right? I'm asking you to write a short algorithm that does something simple, but I wanted to have all these extra features. I'm not gonna add any code for that. How how how could you possibly do it? Turns out that even a simple old bubble sort, and we now have, more more data on on other simple algorithms that do stuff like this. Even simple bubble sort, which which every CS one zero one student has studied for, I don't know, eighty years probably by now, just a few lines of code, totally deterministic, totally transparent, no quantum magic anywhere. Yeah. It sorts numbers alright, but it also does other things. It has spontaneous side quests that it does that nobody had noticed until until you actually look to see what else it might be doing that isn't anywhere in the algorithm. Okay? We have a whole research program on looking at, giving embodiments to various static mathematical patterns. So we provide robotic bodies. The controller is... Has no learning, no algorithm, no, neural net, nothing like that. Only specific mathematical structures. And we ask what what... And so so there's this there's this paper. We ask, what what what what what are the behavioral propensities that are frozen in these mathematical structures? So so I think there's a couple of lessons here. First is that I think neither, neither engineered nor biochemical, quote unquote, machines, meaning us, are only what the formal models say they are. You know, to any philosopher, that's pretty obvious so that your formal model never captures the whole thing. But I think people really have a feeling that there really are computers. There really are things that are Turing machines. I I don't believe any of that is true. I think these are these are useful, useful models that deserve us well in some cases, and they do not serve us well in other cases, not even with simple, quote, unquote, machines, neither for the competencies nor for the limitations. And I think partially, it's because our theories up until now in computer science, in biology, I think physics, although I'm I'm not a physicist, so I can't really say, are not models of the actual, system. They're models of the front end. And, and I'll tell you what I mean by that. Because there is this really interesting thing. There are... There is a source of information beyond physics and biology. For example, here's a simple equation in complex numbers. If I draw a Halley map of this equation, this is the very specific pattern that belongs to this thing. So this is a kind of pointer or an index into a space of patterns. This is what the pattern looks like. Where did this very specific pattern come from? What determines its specific features? It isn't designed. Nobody engine... Nobody designed this thing. Nobody selected it out of a, you know, a long history of selection pressures. Nobody trained it to be like this. There's nothing you can change in the physical world that would make this different. You could tweak all the constants of of... As far as I can tell, at the beginning of the Big Bang, you can tweak all the the constants of physics. This isn't gonna change. Like, the the, you know, the digits of e are not gonna change and so on. So so we know that, and we've known this since since Pythagoras and probably long before that, that there is a set of patterns, truths, mathematical patterns, that do not come from the physical world.

[25:37] They influence the physical world. You you see the influence of it everywhere in in physics and biology. But they don't come from the physical world. You can't change them in in the physical universe. You can't discover them using the standard processes of physics. They're they're really a a different set of set of things. So so, mathematicians are very comfortable with this. Not not all, of course. There are other theories of mathematics. But, many, maybe most mathematicians really don't think this is a a random grab bag of just sort of facts that happen to hold in our universe, which which is which is how a lot of, scientists that I've talked to express it. They just said, well, these are... These are just things that happen to hold. Mathematicians think that they are studying a structured ordered space of of patterns that they can systematically investigate. It's not a mysterious thing. Just because it's not physical doesn't mean it's not amenable to investigation. And and I think there is a structure to it. They... You know, they're sort of working to to fill out this this map of mathematics. What I think we can do is is something is something similar, except that why would we assume that the only patterns... Once once we've decided that physicalism is probably not gonna fly and that there is a space of truths that are not physical truths. There's a space of patterns that that are separate somehow. And we've assumed... And we've now we've now take... Taken the optimistic scientific assumption that you can study this. It's not just a random bag of surprises, emergence, whatever. It's it's actually a a set of patterns that you can study. Once we've done that, why would we assume that the only relevance turns up in mathematics? Maybe there are patterns there that would be recognizable in other sciences. So what we could do is we could formalize this notion of of of making certain embodiments, certain certain physical, whether they be robots and machines, embryos, bodies, whatever. And we could try to figure out, what what what is the extra that you got when you made this thing? Like, what what what... Can can we quantify what what this is? And so here's my here's my proposal. I am extremely skeptical of the idea that that, yes, there is this space of patterns, and it only affects math and nothing else. So I propose the following, experimental agenda, which I have now five or six people in my lab working working on this, doing experiments. The the the idea is is the following. What if this space actually contains lots of patterns? They're not static. When I call it platonic space, that's just so the mathematicians know what I'm talking about. I don't mean the specific theories of Plato. I'm not saying these patterns are all static. Some of them may be, but but but but others are easily recognizable to other disciplines. For example, coupled sets of ordinary differential equations are certainly a mathematical object, but they also can do Pavlovian conditioning, making them of interest to behavior science. So they start to sort of span the span the span the the boundary between disciplines. And so what I think we we can hypothesize is that math is just the behavioral science of that layer of that space where things are so simple that they sit still for, like, derivations and, you know, those those kinds of methods. But there are other patterns there that affect other kinds of embodiments that would be recognizable to developmental biologists, to behavior scientists, to computer scientists. And the idea would be that, basically, everything we make, when we make physical embodiments, machines, embryos, everything else, what we're making is a front end that basically immediately becomes host to patterns that that that are are are stated in this... In in in some other space that that are not captured by the three kinds of effort that I told you about. And we need to study this. We need to study this space because it's not just order for free. It's behavioral competencies. It's policies for navigating various problem spaces. I mean, we could do this for for hours. I could talk about policies for navigating the space of gene expression, the space of anatomy, the space of physiology, of metabolism. All of these things are, are specific kinds of policies that become very sophisticated. And and when they do, they're easily recognizable as kinds of minds. So you see this, like, ultimately, this is basically a a sort of ancient dualistic model where we literally have another another space that that interacts with embodiment, but we can study this. And this is something that that is is, I think, I think universal. So the final thing I'll say before I end is this. There's a there's a final symmetry that I wanna that I wanna sort of put put out there, which is which is to dissolve the category between thoughts and thinkers. So, in a very, so so just just very briefly that you can think about a sort of science fiction scenario where these creatures come out of the core of the earth, and they're incredibly dense. They you... They, you know, they live down in the core super dense. They they got the gamma ray vision and so on. So they come up to the surface of the Earth. They don't see anything. All the all the stuff that we think of as solid is basically a a fine plasma to them. They're sort of walking around. They don't even see it. And one of them is a scientist, and he has this this, this this equipment for for, like, looking at, things that to us that we consider solid objects.

[30:39] And he says, you know, our our planet is covered in this, like like, very thin, sort of gas, but there's patterns in this gas. There's eddies, like, kinda like metabolic patterns in this, and they almost look agentic. They almost look like they're doing things. They move around. They almost look like they have goals. And they say, well, that's crazy. Patterns in a... Patterns in an excitable medium can't be agents. We're we're actual agents. Those are those are just patterns. So so we've been working on this on this spectrum. Yeah. So so so whether you're a whether you're a pattern in excitable medium or or an agent is perhaps, observer relative, perhaps in the eye of of the beholder, and we've been thinking about this, this continuum where in a cognitive system, you can have a fleeting thought like a wave. It sort of comes and goes. You can have some intrusive repetitive thoughts, which sort of are very hard to get rid of, and they even do a little niche construction in your brain to kind of stick around. Then you might have dissociative identity disorder, alters. They're much more than a than a single thought. They have they have, they have agendas and goals and and and problem solving and so on. And then there's... You could have a whole human mind and maybe, maybe what... I don't know. What's what's what's beyond that? We we explore. Chris Fields and I explore that here. So the so the so the point is that maybe maybe the whole distinction between between machine and data, between thought and thinker, between pattern and agent is just too too limiting. Maybe we need to understand ourselves as patterns that can spawn off other patterns. So so, you know, large thoughts that can have they can have smaller thoughts. And, fundamentally, this is what I think, and I I know this is this is totally crazy by by conventional the metrics. The conventional view is this, that that there's this, like, universe of matter. There's a subset of this, which is which is life, and then there's this even yet smaller set of that, which are which are things with minds. I I think this whole thing is upside down. I think that what fundamentally we have are an incredible number of, patterns on a on a continuum from the most boring static thing like the, you know, these these simple mathematical constants, at least that's how they look to us, And then very, very complex, patterns that basically we recognize as kinds of minds. Some of them some of them are embodied, and some of them we would actually call living, because they do, you know, Darwinian kinds of dynamics. So this is this is how I see it. So I'll I'll just summarize and then and then stop. And if you wanna, see all the details for all this, crazy stuff I've said, it's all... I tried to work it out in in detail here. So I think that, patterns of form, meaning anatomy of behavior, physiology, of computation, are are all one class. I think it's an invariant. I think that there's a structured space, from which, currently unexpected, patterns are drawn into specific physical interfaces. I think bodies are just, thin clients or front end interfaces that host massive ecosystems of these patterns. We have a research program now to map out the space to learn to optimize and suppress the ingression of these patterns into specific media. We, yeah, we can talk about the... We can talk about this if, if if, if anybody wants to, wants to hear about that. I I think I think when these patterns show up, in, in certain agents, we we can say we've... The... There... We've received a certain inspiration, and I don't just think it's for high end human creatives. I think it goes all the way down. So don't feel don't feel too smug because I think machines get them too. I think it basically pervades this this phenomenon, pervades, systems big and small. And I don't think we are physical beings who sometimes get affected by such patterns. We we are the patterns. And and, of course, we get affected by other patterns. And somebody earlier said something about resonance, and I think that's very... That's that's probably a good way to to to think about why certain patterns interact and so on. Yeah. And so there's a symposium, where a bunch of people, have given talks on this topic. Most of them don't agree with with most of the stuff I said, but that's okay. There's lots of interesting ideas around this, and we're just getting we're just getting started. And the interesting, and important thing about all of this is that in the future, where all of the natural, life forms are like this tiny little corner of this space, and cyborgs and hybrids and modified humans and every kind of imagined combination of life forms and and machines and and software, or some kind of viable being that, we are going to have to, hopefully have some sort of ethical symbiosis with, all of them are going to be much much much like us. They're going to be hosting numerous patterns of competencies of goal states of, different kinds of minds. And we don't even understand how this works with us, and we have very little understanding of what's going to happen when we make novel beings who have not been here before through our evolutionary stream. And I think it's a it's a a kind of an existential, matter of existential importance for humanity going forward. So I'll stop here and just thank the, thank the people who did, some of the biology, that I showed you and lots of lots of amazing collaborators that that work on some of this stuff and have to do with some disclosures. So these are these are companies that have licensed some of the inventions that we've had from all this stuff.