In Continuum with Mike Levin: Research Conversation Part Two
Transcript:
In Continuum: Research Conversation 2 Michael Levin & Andrea Hiott
Andrea Hiott: [00:00:00] Hey everyone. I'm posting the second conversation that Michael Levin and I had. Sorry, it took a little bit of time to do so, I'm sure you're all busy too with work and end of the year. Uh, and I just wanted to wait until I could really listen back to it. While I was working on a relative paper and I did just listen back to it.
We had this conversation quite a few months ago. Uh, so it was wonderful to look, listen to it again and rediscover. Some. Really important points that I hadn't remembered from it. We really get into the idea of the binary and the continuum. And we talk about way making an cognition as navigation and try to kind of. Get into a little of that nuance. So. I'll just go ahead and post it and leave it as it is for the most part. And the next one will come in the next year.
So. Thanks for listening. And I look forward to your comments and. I hope you have a wonderful end of the year and a wonderful 2024.
I remember that we were talking, [00:01:00] just had mentioned the xenobots. Should we just start there or? Sure, yeah, what I remember that I wanted to ask you about was when you, I think What I'm interested in is do the xenobots create their own shape or do you choose the shape? Do you roll them up or do they roll themselves
Michael Levin: so there's two types of xenobots. And what, what you can do is you take that, you take off the, um, the ectodermal cells off the embryo, and then you can dissociate them.
Yeah. So, so you, you separate them. They're kind of like a, just, you know, floating there. And then, and then you put them a little in the little pit and they kind of coalesce together. So, there's two versions of that. Uh, by default, if you do nothing other than that, They will arrange into a Xenobot on their own.
You do not have to roll them up. So they roll themselves. They, they do it themselves. But now, now what they will do is they will make a spherical, a spherical bot. If you want to manipulate that, there are two basic paths that you can go. One which we've already published and the other one is what we're working on.
[00:02:00] The thing that we've published is that you can sculpt them. So this is what I would call a bottom up intervention because you're trying to micromanage it. So what you do is, and this is my my, uh. His staff scientist in my center did this Doug Blackiston. And so he's a micro kind of a microsurgeon.
What he'll do is he'll go in and like, uh, basically sculpt out ablate certain parts of it. And, and at that time you can also put in other kinds of cells if you want. So the very first paper actually had, he sculpted it to have little, little stubby, little legs, like a little Ottoman kind of thing.
And then, and then he put, um, some, some muscle cells in there. And the muscle cells could make the legs go and this thing actually walk. That's not the xenobot that, that I spend most of my time talking about, but, but that was actually the first paper.
So, so you can sculpt it and he's, he sculpted it that way. He also in the, in the last paper, what he did was he, um, he made it like a little Pac Man with a little triangle taken out of it. And what we found is that.
That shape is really good at gathering up a bunch of loose cells and packing them into the next generation of Xenobot. And that was, that shape [00:03:00] was actually designed by an AI that was designed by a, by a, uh, an artificial intelligence algorithm running in Josh Bond guards group. So, so you can do stuff like that.
Now, what we want to do in the future is we want to get beyond that. And we want to be able to communicate, uh, morphological goals. To, uh, to those cells. And we want to say, I need you to be flat like a pancake, or, or I want you to be, uh, you know, like a football or, or, or, or have this more complex shape, like whatever it's going to be.
So ultimately down the, down the road, that's what we want, but, but we haven't, uh, we haven't done that yet. We haven't published that yet. So we're, we're working on that.
Andrea Hiott: , I think shape is so important, right? And the morphology, and I'm trying to kind of connect it to all these other bigger ideas that I'm working on in terms of cognition. Um, and you already sort of do it in your work, but I'm trying to find the link.
So I guess what I'm, what I was asking is like, Is, I'm wondering, like, is the, um, I guess first I would ask you, like, where's the chemistry and where's the mind and the way that the xenobot forms itself, in your kind of opinion? [00:04:00] Um, but behind that is what I'm trying to get at is, are, are they kind of working on the patterns of their own shape?
Is there, is there a continuity between the shape that they are and the shapes that they create in a sense?
Michael Levin: Yeah. Um, well, I want to say, I want to say one thing first, which is that. The, uh, question of, of that, um, continuum of, of shape and mind and everything else, uh, xenobots are not yet a great, uh, example for that because there's so little we know.
So I'm, I'm of the firm. I mean, many, many people have, have, uh, have opinions as to what, you know, what what the cognitive level of xenobots has to be, but just because they have sort of philosophical commitments to specific viewpoints. I don't believe that is the way to do this. I think it's, I think it's an empirical question.
I think we have to do experiments. And so I have not yet made any claims about, uh, the, the cognition of these xenobots because we haven't actually published the data showing what it is that they can do. [00:05:00] A few things, but, but, but we have not, so, so I haven't made any claims about xenobots about that. No, of course not.
Andrea Hiott: I'm jumping ahead. I'm jumping ahead because the way I define cognition is already movement and it's a completely different than, you know, I'm already starting at a different place. So that's why I'm asking. But you're right. That's good to point
Michael Levin: that out. Well, only because, right. So, so, uh, because, because ultimately we want to be able to be very specific and we want to be able to say xenobots have this level of competency in navigating space.
They can, uh, you know, what can they do? They may, maybe they, so, so we're testing, you know, Do they have preferences about where they go? Do they have memories about where they've been? Do they have, uh, ability to overcome obstacles? Do they have the ability to work together you can, you test all these, all these things, right?
So, so we don't know yet. And so, right. So that's why I'm not making a claim, but if we, if we sort of step back to this more general notion of way making or, or navigation or whatever it's going to be. Then absolutely, they're not only doing that in three dimensional space, they're doing that in morphous space.
I mean, they do. And in fact, if you cut them and injure them, they will [00:06:00] in fact, try to regenerate back to their original shape. So, so there is some of that there. It's just, it's just, there are other systems where. We know more and thus it's easier to kind of make that connection in a rigorous way Um, you know, otherwise you're fighting two battles at once, right?
You're fighting this this general tendency of people to try to separate uh cognition from morphogenesis and also you're fighting the fact that we actually don't know Enough about what they specifically can do. So that's the problem. So I try to I try to
Andrea Hiott: kind of separate those Yeah, this gets really messy.
But I think maybe it's okay to just like let it be messy for a second because yeah um if we start if we Could we can, of course, erase this. Usually when we hear cognition as humans, we think of mind and thought and memory and all this kind of awareness of cognition really, right? Um, but if we could kind of erase that in a way that I think you're definitely both of us are kind of trying to get at this different understanding of cognition that I think is actually very similar, which is why I'm so excited about your work.
[00:07:00] Um, and actually I think your work is kind of maybe. an example of what I've been trying to formulate, I guess. Um, and I think there's other examples in different areas of science. Like we're all sort of trying to talk about this, but if we could just like start over for a second and just say, okay, any making way, or I also use the term navigation, but just as we know, that always means you have a goal.
And sometimes maybe making way doesn't have a specific goal, which is something else I want to talk to you about. But if we just think of it as any agent from any position. Making its way through its encounter, or it's, I think what you might call like a surface, computational surface, I started to realize that your light cone idea is a little bit like what I mean by an encounter, right?
So if we just determine a position, an agent, what I say, an agent base, and I've heard you call a gentle, I think we also use similar terminology. If we just say, okay, this is from where we're going to measure it. And however that position makes its way is cognition. Like if we just pretend like that's what cognition is, has nothing to do with that creature.
That agent knowing [00:08:00] that it's cognitive or anything else, then, um, when you create this kind of xenomorph, or you don't even create, when you've changed the frog, let's just go with the frog, when you've taken the skin and you've, you know, separated the cells and they start to reform, that is a kind of making way.
So for me, that is already a cognitive. Act, and it's also, um, physics, like, and it's not either or of those things. It's both of those things, but we would call them differently depending on from where we're going to measure it. So if you take it like that, would you just say, okay, then it's, it is a cognitive thing that the way it comes together and forms this, uh, new, new shape and then begins forming other shapes that look like it and so on.
Michael Levin: So, so yeah, yes. But to me, um. Because, because I treat, uh, cognition as a continuum, not a binary variable, I think the real question is going to be where on that continuum is it, so what is it capable of? So I completely [00:09:00] agree that this, this like, um, metacognition that I know I'm aware and I know my goals and I can change my goals and stuff.
Yeah, that's all, that's all really advanced, uh, kind of stuff that's way on the right of this continuum. Things are on the continuum long before they get to that point. So, so I have no trouble saying that very simple systems. And even bacteria have some some primitive metacognitive loops about their metabolism and things like that.
So, uh, so I have no I have no problem saying that we are somewhere on the much lower sort of part of that spectrum, but we're somewhere on there. I'm sure I'm sure we are. I just don't know exactly where there are other systems where I can tell you where I think they are, because the experiments have been done.
I, you know, that's, that's why I, the only reason I harp on this point and it's not, I'm not trying to get too specific, I'm, I'm fine with letting things be messy and percolate while I'm completely fine with that. But, but, um, I'm trying to fight this, this, this, uh, pervasive. Uh, tendency for people to make claims about these things, uh, just based on [00:10:00] their philosophical pre commitments, not, not on actual data.
And so, so I keep, for that reason, I keep saying, as much as I like all kinds of things to be cognitive, I don't know where they are on a continuum until we, we test them. So, so we don't know, you know, how much associative conditioning? I don't know. Can they, right. Can they, you know, we, when we have evidence that they can form memories now, but.
We don't know how complex these memories can be. Can they chain them together? I have no idea. These are, these are all things that, that we need to figure out. And, and the other interesting thing is, um, I didn't realize how, so, so it's funny, you know, you just, you said a minute ago that, um, uh, navigation presupposes a goal.
And I never really thought about that. I realize now that people do think of it that way. I didn't realize. I didn't
Andrea Hiott: either. I don't think it does, but it does in the literature.
Michael Levin: Yeah. It's, it's so strange. Even, even, we even, we even had a paper where. We talk about one of these synthetic things, traversing a landscape and traversing to me, just, it could be, it could be a random walk.
It just means that it's moving over the landscape. Right. [00:11:00] But a reviewer got back and said, Oh, traversing must mean it's this like, um, uh, you know, this, uh, very directed kind. I did not realize that people, people have those connotations. I mean, I think, I think for, for navigation.
You could have a very generic goal of exploring a landscape. It doesn't mean you have the goal of knowing where you're going. I don't think that implies at all. So I don't know here. This is one of those things where I think we have to decide, are we going to use words in the way that everybody uses them because we want to avoid those battles and, and have, uh, you know, have more adoption.
Or do we bite the bullet and say, no, actually, here's what those words should mean otherwise. Right. And just sort of lead the, the hopefully a change in the terminology, you know? So sometimes both are, are useful, right? So, I don't know, cer certain things I try to insist on and another things like
Andrea Hiott: whatever.
Yeah. There's a big debate you probably know, uh, in neuroscience now about language and the terms we use and, you know, Buzaki [00:12:00] who've, um. You know, advise me a little bit on some of my own stuff talks a lot about, how we just sort of inherited all these terms like memory, for example, from James, and we sort of assume what it is before we even look for it in a sense.
Um, so I think that's what I mean by the messiness is like we're all assuming terms mean a certain thing based on our. past trajectory. Um, and we don't realize, I mean, you write about this a bit in that paper you sent me last time. Um, you know, some of the confusion is just that we don't have a kind of an understanding of what we're talking about where there's a lot of misinterpretation going on that we don't even really know about.
Um, I think that's what I'm actually trying to do in formulating a philosophy or a theory of cognition, which is really all I'm doing is. Gathering all this other data and research from science and trying to show that, Oh, actually we're all talking about the same thing. And that's all I'm really trying to do.
It's not even that there's some new thing. It's just to try to put it into a framework that actually works across species and scales and disciplines because [00:13:00] nobody's done that because it's so messy. But, um, maybe we can talk about a more kind of concrete example before I start asking other questions.
Cause yeah, the xenobots, I know you, it's still very new and also a lot of people. In a similar way are coming with different references of what they think you're doing so it can get very Difficult. So what's an example that you think where we could actually talk more about physics mind kind of
Michael Levin: software.
Yeah. Yeah. I mean, for example, we could talk about, uh, plenaria uh, fragments, figuring out how many heads they're supposed to form, or we could talk about. Uh, tadpoles rearranging their face, uh, from abnormal positions to become frogs. Some of these examples where, where, we can talk about the, um, the new kidney tubule where, uh, it solves a really amazing, uh, kind of a defect in its own parts to still get to where it's going, which is to make a proper newt with a proper kidney tubule and so on.
So, yeah, any of those talk
Andrea Hiott: about planarian because they seem to be the answer to everything [00:14:00] in a way. Um, so if we think about. them, which regenerate in a completely different way, or would you say regenerate or reproduce in a very different way? They regenerate. Okay. Um, so I've heard you say something like, uh, that I think in talking about those that bioelectricity is this kind of, uh, strange boundary or layer between the mind and the physics when it comes to Changing and this is something I'm still not so clear about, is what the electricity is in the bioelectricity and how we're kind of distinguishing between matter and it matter in that pattern.
So I don't know, maybe that's kind of a very broad, but in terms of, of those, where would you start to think about software and hardware or mind chemistry?
Michael Levin: Yeah, I start to think of it. Actually, let's let let me let me try a slightly different way that I haven't I don't think I've talked about it this way in these talks yet.
Think about a think about a [00:15:00] computer. Right? When you first turn on the juice. For some number of, I don't know, probably nanoseconds, what you're dealing with is a piece of physics, a dynamical system well described by Maxwell's equations and all of that, right? But at some point, once it starts grabbing, once it starts executing the thing that we recognize as, um, formal computation in the von Neumann architecture, so it starts, so it starts grabbing instructions off of this memory location and doing something with them.
It's still a physical machine. It's still obeying the laws of physics. It's still doing all that stuff, but we can look at it and say, wow, it's executing logic, the, the, the sort of basic principle of rational thought it's doing, you know, let's say it's calculating the and, um, you know, the and function or the or function or something like that, or it's calc, you know, some, some kind of calculation.
So. So you so here you have a system. You can look at it a certain way and say you can be a reductionist and say none of that none of that exists. All [00:16:00] I see is electrons moving according to Maxwell's equation. That's all that exists. Or you could say, I mean, yeah, that's true. But that's the most boring part of this.
The more interesting part of this and the more useful part of this is that it's Computing functions. It's it's it's calculating. Maybe it's reasoning. You know, it's inference. It's calculating inferences and so on So so I think that that is a like a minimal simplistic before we get into life and all that That's a kind of a simplistic system where you start to see that It's both.
It's yes, it is a physical system. Yes, it, uh, yes, it, it, uh, it, it obeys the laws, but also it's doing this amazing thing. That's, that's, that's getting close to cognition. Like it's actually generated, you know, it's following truth tables and, and, and, and, you know, and those, and those kinds of things. So, so I think the same thing happens in, in living organisms.
You can look at the bioelectricity as just the flow of ions in and out of membranes and what that does to the arresting potential and all of that, but fundamentally, much like [00:17:00] with the computer, there's an, there's a, it's, it's computation because there's an interpreter, there's somebody that looks at that, at that whole physics and acts on it based on what they see, and in the case of the computer, it's us, in the case of the biology, it's cells and tissues watching each other and making decisions based on what they do.
Okay. And the first part of what you said, you know, the, my point about, um, changing scales, it's like, sometimes, um, I'll start a talk by saying, uh, did you, did you know that I can control 30? I can, I can depolarize 30 percent of my body cells just by thinking about it by force of thought alone. And people say, well, that's crazy.
And then a couple of people say, well, maybe it's some sort of weird, um, uh, you know, yoga thing or something. Right. My, some kind of mind body miss, like, no, no, this is 24 7. This is what happens in your body when you, when you decide to get up out of bed in the morning and go to work. Mm-Hmm. . Mm-Hmm. That's because you have a high level cognitive, uh, in fact a metacognitive system that has social goals and, and whatever, and decides that you're going to work.
And in order [00:18:00] for that to happen, the muscles in your body have to become depolarized in a certain pattern. So you can walk out. Walk out. Yeah. And that means that the, our architecture is such that. The thing we call cognition, which are these high level, the kind of mental processes and goals has to filter down to the molecular biology of your cells.
It has to, otherwise there would be no voluntary motion. So, so that's, and so you say, well, how in the heck does that happen? Well, bioelectricity, so neural and non neural that that's, that's why bio, that's why I think bioelectricity is so exciting because it shows us that it is the layer that enables cognitive processes.
To, uh, to control chemistry in your in your cells. I mean, that's 20. That's what happens all the time. It's not some rare, you know, specialized thing.
Andrea Hiott: And at the same time, it is chemistry itself in a way, right? I mean, doesn't it depend on where we're going to measure or look at it, what we're going to call it?
Absolutely.
Michael Levin: You can, you can choose as an observer and observers are scientists. [00:19:00] Observers are parasites. Observers are conspecifics. Observers are all kinds of things. You are your own observer yourself too. As an observer, you can choose the level that you work on, but we have to be clear that not all choices are equal.
So, so if you are, just imagine, um, I kind of imagined it this way, uh, You are, you are running a, a software company and you're interviewing candidates and somebody shows up and, and, uh, to the interview and they say, Hey, um, I'm a, uh, I'm a strong reductionist. Uh, I believe in electrons and I believe in Maxwell's equations.
And that is it. There is nothing else going on in the computer. It's not that they're wrong exactly. Would you hire that person? They're not going to code a damn thing because, because, because
Andrea Hiott: there's no dynamic way of change there.
Michael Levin: Well, they don't believe that the algorithm makes the electrons dance. If you do not believe that this algorithm, what, what's an algorithm.
It's a, it's a, it's not a physical thing, but if you don't believe that the algorithm functionally makes the computer do [00:20:00] things, you will never write any code ever and and so and so and that's it and it's not that you're wrong and you can have a great life in physics, but you're not going to invent software.
So, so that's what you know, that's my point. And I think evolution is very, very quote unquote aware of this in the sense that these observers do not, for example, um, when cells watch each other's voltage, and they do they make decisions based on each other's voltage. They are not watching whether that voltage got there by sodium by potassium by chloride.
They're ignoring those details. They're watching the voltage. Why is it good to watch?
Andrea Hiott: by the way, because that makes it sound like they're looking at it or something, but it could be they're just sensing it too, right, aren't they? I mean, it's not like with eyes necessarily.
Michael Levin: No, no. There's no eyes. I'm saying, yeah, yeah.
I don't mean eyes. I mean, I mean, they're, they're sensing it. They have, they're taking measurements. Yeah. It's a sensory measurement. It's a sensory measurement. That's, that's what I mean. Uh, they're paying attention to the voltage. And in doing so, they specifically ignore the molecular details of what was it, sodium or potassium that got you to that voltage.
They ignore that the same [00:21:00] way that when you're a good coder at a particular level, you have to ignore the levels underneath. If you always jump down to the lower level, you're very limited in what you
Andrea Hiott: can do. That's what language is. Giving you in a way. I mean, that's why we create those kind of scaffolding systems in a way.
So does that, I think just to go back to this idea of navigation, right? And the way it's used in the literature is that there's a goal. And so it's as if, as if the agent itself knows that it has a goal or the agent has the goal. But if we think of it in the way you just described that kind of stuff, it could also be that the person measuring it has, um, seen where, what its goal is or something.
Um, I mean, you could also kind of, yeah.
Michael Levin: Um, yeah, no, 100%. That's in fact, that's that's the definition in my in my TAME paper. I say that goals are in the eye of the beholder. Now you are also a beholder of your own goals. So it's not just you know, I'm not saying that we are nothing except what people make of us.
I'm not saying that. [00:22:00] But but any of this when when when we say a system has a particular goal, that means that I as an observer have picked a problem space. I picked a bunch of states that I think this thing, uh, this thing has as goals. And I've done some experiments to see that if I perturb it, it still tries to get those goals met.
It's mostly, it is, and, and somebody else could, could have a completely different, true, you know, um, a set of choices about that. And, uh, in, in principle, multiple viewpoints could be right. But they're not all equivalent because some give you much better, uh, outcomes than others. So, but yeah, very much so. I think it's very observer, you know, it's, it's observer
Andrea Hiott: dependent.
That's what ends up causing the trouble in a way sometimes is that we, we're still stuck in that either or dichotomous model where we think it's either software or hardware. Um, it's either mind or physics always across all, uh, positions of assessment when it's because it's very hard to hold in your mind the fact that depending on from what agent you're [00:23:00] measuring or assessing, what is physics at one level can be look like mind at the next or be mind.
I mean, because you know, that's a really, I think that's a very hard thing to articulate or to step back enough to. To see,
Michael Levin: well, another way to put it is, is this, uh, I think the reason physicists see, you know, uh, chemicals and particles and so on, and not mind they see matter and not mind is because they're using low agency tools.
So when, so you need some sort of, uh, um, an impedance match between the tools you're using and what you hope to see. So if you're using rulers and voltmeters and, uh, all of these kinds of things, Like as physicists do those are all low agency systems. All you're ever going to see is low agency matter that you're only ever going to see that.
Uh, if you want to see mind. Yeah. If you want to see mind, you, you have to have a mind to be able to recognize mind. Um,
Andrea Hiott: but even then it sounds like we're still talking about mind as if it's something [00:24:00] else. So in a way, it's like you have to have a mind that has similar regularities to whatever it is you're measuring.
I mean,
Michael Levin: Yeah, well, that's a good question, right? I mean, this is this has been dealt with in science fiction for 100 years or more. It's like, are we going to be able to recognize minds that are very different from ours? And I think it'll be hard for sure. But I don't think it's impossible. And I think you get there by stretching your own mind.
And I think this is what the field of diverse intelligence is all about. It's about stretching your own mind in new directions. To be able to recognize a wider set of
Andrea Hiott: beings. Let me push you on that a little because if we think of mind as the way we make way and basically this gets to memory and everything because it's not just that you're making way.
It's that you're Aligning with certain regularities, which then you kind of habitually use in the same way you were talking about coding. Right? You can't keep recoding everything every time. I mean, so you're sort of building this framework by which you can move better and [00:25:00] farther and explore more. Um, so.
If we really think of mind as that, you know, then more or less what we're saying is we just don't recognize certain forms of making way or movement. For example, if something's using some kind of electrical physics to, to move that we haven't been able to observe and clarify yet from our sensory position, we wouldn't see it.
I mean, it happens already all the time, right? With different kinds of animals and things.
Michael Levin: Yeah. Yeah. I, I, I a hundred percent. And I think if I remember correctly, and I'm not sure that I do, but if I remember correctly, I think that at some point was maybe it was the ancient Egyptians or somebody thought that um, the brain was a way to cool the blood because, because you know, and I mean, yeah, from a certain perspective, that's great.
It's a radiator fan and it cools the blood, but you've, but you've missed, you've missed, uh, the importance. And, and I think this happens to us all the time. I think we look at stuff and in science, this happens all the time. You look at stuff and you have a level of, of, of explanation and you say, you know, great.
Now I understand what it [00:26:00] and, and what, what. really harms the, the ability to discover these things is this kind of, um, this, this idea that you're always supposed to skew downwards, that, that the lowest level explanation is always preferred, that if you can do it in terms of chemistry instead of cognition, you're better off.
That, that, that I think is absolutely pernicious and it keeps us from discovering these other very useful perspectives.
Andrea Hiott: Absolutely. Because that itself is one, it's, it's like a linear. Instead of a spectrum, I think a spectrum can be understood as linear, but that's not, I don't think that's what you're saying, this continuity, I don't, when you talk about scale free cognition or something, you're not necessarily talking about it's linear with a beginning and end or something, but when we talk about reduction, that's what we're kind of talking about, that everything can be reduced in the same way, rather than there's all these kind of fractal, uh, positions from which everything can be decomposed, or the same thing can be decomposed in many different ways, right, depending on.
Where are you going to measure from?
Michael Levin: Yeah. Well, and also it's a [00:27:00] very, um, it's a very specific, actually a very strong claim and it's not often made, um, explicit, but I think people, most, most scientists, uh, work as though they believe this, which is that they say you really should reduce it. To let's say chemistry, um, but, but they don't really mean the full reductionist program because if somebody then said, Oh, you mean like quantum foam, like we'll reduce it.
They say not quantum foam. That's too far chemistry. It's all chemistry. I'm like, well, then you're not a real reductionist. Are you? That just means you've picked chemistry as your, you know, as your preferred level, but just the stuff underneath. And if you want to, you know, I don't know what, if it's in favor now or what, but, but, you know, if you want a string theory explanation of, of, of genes, they're going to say, ah, you're crazy.
That's too far down chemistry. Like chemistry is the way to go. And I'm just saying, that's not good reductionism. That just means you've picked chemistry and, uh, that's a very strong. Yeah. And that's a very strong claim that chemistry is, is the level for all of these things. And I don't think it's remotely true, [00:28:00] but, but I think we need to understand that that's, that that's a commitment that, that a lot of people have.
Yeah.
Andrea Hiott: And I think that's important too, that because you can, it can quickly sound as if you're, we are, I am trying to reject, um, these binaries, but actually these binaries chemistry or mind, for example, where we could take other things, you know, uh, they're very important. They actually help us to understand the system and they are literal and they, they are meaningful and all of this.
It's just that for me, at least it's like, how do we understand that those are just, uh, those are not, not just, those are ways we've, we've developed to understand something that can be understood from. Other positions through other categories without one or the other being false or true specifically.
Yeah,
Michael Levin: I, I think I'm, I'm even more radical on this point. I think I, I don't like the binaries because, because they lead to pseudo problems, right? Once you've, if, if you've assumed, if you like binaries, then, then what do you make of embryonic development? [00:29:00] Because, because you start as just physics as a little on, on, you know, quiescent And then before you know it, you're a, you know, nine months later, you've got these cognitive capacities and you're recognizing shit.
So, so if it's binary, then you owe a story about where it clicks over. There is
Andrea Hiott: no place. I don't think it is binary, but I think it can help us understand, like, for example, how would we, now we, for me, I would say. From the beginning that's making way. From the beginning that's cognitive. This is very hard because I also mean from the beginning, like I mean cognition is making way.
I also mean making way is cognition. So I'm not reducing walking or swimming or crawling to cognition and I'm not reducing cognition to walking and swimming and crawling. I'm saying those are two different ways which we can measure a process that is not binary. So in that way, the binary becomes important just because of contrast and the role that contrast plays and how, so [00:30:00] far, we've proceeded to develop a kind of, a kind of knowledge because, I don't know, maybe you disagree, but in a way, I feel like now we can see that process of that nine months as cognitive and as making way or navigating.
And those things aren't opposite, but somehow we use this binary categorization to develop this, the science that allowed us to take that step back now and go, Oh, actually, this process is not binary.
Michael Levin: Yeah. Um, so, so I agree with you. I think it's cognitive from the very beginning, but, but that has an interesting implication because the very beginning is a little blob of, of, of that's, that's well described by chemistry.
So, so once you say that, that that's cognitive, Then, then, then people say, well, well, then what isn't I mean, basically then chemical reactions and I say, yeah, look at the minimal matter research in, uh, in, in diverse intelligence, you know, active, active matter droplets. I mean, yeah. And so, and so [00:31:00] then, then, uh, that basically pushes you to this idea that.
Uh, there may not be anything that isn't described by some metaphor from from behavioral science somewhere, and that may well be true. And people like, um, uh, Chris Fields and Karl Fiston, who are working out a kind of, uh, you know, Carl calls it a physics of sentience and the sentience of physics. It's basically redoing the fundamental equation that your body is able to compute in terms of the environment all, Uh, way we look at physics from a active inference and from an active inference perspective, uh, yeah, that's great.
And, and I think, I think in many ways that, uh, undoes the binary distinction that people like to make, but I also think, you know, I think these, like, if we step back and we say, okay, what are, what are these words for, you know, what are these categories even for, right? Why are people using them? I think what it all boils down to is it's a, it's a protocol claim.
It's a claim about if you tell me. This is a, this is a system that is, uh, here on this continuum. It's, you know, it's a mechanical clock or [00:32:00] it's a, it's a learning agent or it's a thermostat or whatever. All you're really saying is I don't need to know if you think it's mine or matter or whatever. What I really, what you're really telling me is here is a collection of tools that you should.
used to interact with the system. And those tools might be rewiring. They might be rewards and punishments. They might be, um, a complex, uh, a conversation that we're going to have. They might be, uh, just, uh, resetting a set point. That's what I hear when, when I hear somebody claim that this system is, or isn't that what I want to know is.
Great. What are the, what are the set of tools? You're telling me that rewards and punishments won't work, that I have to use rewiring. Well, let's find out. You could be right. You could be wrong. That's why I think it's, it's an empirical matter because because these, these are all protocol claims in the end.
Um, what techniques are you going
Andrea Hiott: to use? Yeah. And that's why grounding it in engineering, like looking at it from that framework can be the most helpful instead of from a more theoretical sort of standpoint. And also, I mean, it can get, um, if we would just say, oh, everything is cognitive [00:33:00] and everything is way making, um, that's really not true.
You can't actually say that because you're always. Speaking from a position of measurement or assessment. So, um, I, I mean, that would be like a whole, a whole discussion within itself about how we have to sort of apply the same idea to our own, uh, our own, our own position, you know, it becomes very meta in that way.
And, and really you're just trying to understand something. And when I start to think about why are we doing this, right? Like. Or when people start saying, Oh, does that mean that life is whatever moves if it doesn't make its way, then it's not life. Okay. Yeah, yeah, maybe. But that life itself is an assessment from our position.
Um, so, yeah, I think it's great that you said, like, what's the point? Because we could just. Maybe right now even go to a really practical level, like in our practical position as agents in this, uh, existence that we're in, there are particular challenges and difficulties and urgencies, whether you want to look at it on a planetary scale, or you want to look at it [00:34:00] In terms of your own psychology and how to get through the day, you know, for people who have certain, um, uh, uh, challenges like physical challenges or mental challenges, whatever you want to, however you want to assess that and, and, and name it.
We want to do something about that, right? We might not need to understand why so much as just alleviation or better way making or Better navigation, right? Which I think you're, by, by focusing on engineering. Are you sort of framing it towards that too, a bit when it comes to your own work?
Michael Levin: Yeah. And I mean, I, I see engineering really broadly. So, so one way to, I don't just mean. Engineering because we're going to treat humans and have human relationships the way we engineer rumbas and things like that. Um, engineering to me is much broader. So, so I think of it as. More of, um, the way, the way that, uh, you can think about hacking.
So hacking, so if you're a hacker, you might be on [00:35:00] one level, uh, where, where you're down rewiring some wires, you might be writing code and things like that, or you might be doing the social engineering part where you're, you're relating to other people and getting them to tell you their passwords or whatever, you know, whatever it's going to be, it's all, you know, you're, you're comfortable, uh, at different places along that continuum.
And I think that. Yes, on the left side of that continuum, it's all engineering because you are mostly exerting your own agency to get the thing to do whatever you want it to do. But on the higher level of that continuum, when you're interacting with animals, with other humans, with other high level minds, or whatever they may be, however they may be embodied.
Um, it's much more of a, of a dance where their agency is giving you something. It's not just a one way kind of a thing, but that's again, that's a. That's, that's continuous, you know, people like to have pet lizards and, uh, and things like this. It's not just, you know, uh, you know, my vacuum cleaner versus, versus my spouse.
I mean, there's all kinds of stuff in between, right. Where you know, uh, you can have different degrees of [00:36:00] relationships depending on the match or the, or the mismatch between your cognitive light cones. Um, and so, yeah, so engineering is all the way from like very, very mechanical kinds of stuff all the way up to.
Complex human relationships where there's multiple loci of agency involved. And it's not just about control. It's about getting something back because, because the system, the other side is also agential like you.
Andrea Hiott: Yeah, that's a great point to go back to the planarian from that position. So this is something I don't understand totally when, when you change.
Uh, or, or some other, whatever, anything that you've talked about in your work, when you're going to, for example, change the position of the eye, or you're going to grow something in a different way, or, um, and you're basically that's bioelectric, right? You're changing the pattern. You're not changing the genes.
It's not happening in transcriptional space. It's happening in this. So what's happening? Is it that you're changing the communication between? The ions. I mean, are you changing the sodium? What's actually happening [00:37:00] there on that level? That's changing that.
Michael Levin: Yeah. So, so, so physically the way you do it is by, uh, there, there are four or five different techniques by which you can change the pattern of voltage around across a set of tissues.
You can do, you can use, um, ion channel modulation, you can use optogenetics, you can change sodium potassium chloride. None of the details matter for the reason that I said before, as long as you get the voltage, right. I mean, the details matter when you're trying to get the voltage, right. But, but, but there are many ways of doing it and it doesn't matter how you got there.
The, what makes the change is the change in voltage. Voltage is super cool because it's a, it's a, um, it's a high level entity. Voltage doesn't know if it got there by sodium or by voltage. It's like this coarse grain feature. And that is what the cells are tracking. They're not tracking the individual ions.
They're tracking the voltage. So, uh, so you make that change. The reason you make that change is because. That electrical pattern is literally a memory in the mind of the collective [00:38:00] intelligence of cells. That, that group of cells is storing a memory of what does a correct, for example, what does a correct planarian look like?
And that information, they have to have that information because when they're damaged, they have to rebuild the correct planarian. So
Andrea Hiott: it's kind of the trajectory or all the nested coding. You're shifting that, but you're shifting it all at once. We're shifting
Michael Levin: the set point. So, so, so, so when you have a thermostat right in your house.
Um, you may not know how anything works. You don't need to. You don't need to rewire it. You may have no idea what the details are, but all, but if you know, you have to know a couple of things. You have to know it's a thermostat. If you miss that part, then, then no good. So you have to know that that's what it is.
And once you've guessed that that's what it is, you have to know how to reset the set point. And we know that for a thermostat, there's a particular register in memory that has a number in it. That number is what the whole system is going to use as its representation of what is good and versus bad.
That's it sets the baseline valence for that system. So if it's good, then nothing happens. If it's bad, then it's got to exert all kinds of [00:39:00] energy to try to get to wherever life is good. So, uh,
Andrea Hiott: So it's very specific to that agent base. That position that.
Michael Levin: Yeah, yeah, yeah, it is. Although it is, although it is, although people have done, um, experiments in memory transfer, and there is some ability to take that medium.
And stick it into a different agent and have them interpret it the right way. That that's, that, that's a very big open area of research for the, for the future. Like how, how much of that, um, my, my favorite example of that is caterpillar butterfly. So if you train caterpillars to look for, uh, the leaves that they eat on a disc of a particular color.
They become a butterfly. The body is completely changed. The brain is completely refactored. Cells, most of the cells are killed. And you're right. That the whole thing, and then the butterfly, not only does the butterfly remember the original information, but it's actually generalized because the caterpillar said Leaves are to be found on this particular thing.
[00:40:00] Butterfly didn't eat leaves. Butterfly doesn't want leaves. It wants nectar. But what a butterfly remembers is food. It generalizes the notion of food from leaves and nectar. And so, so amazing. Uh, but, but it's a completely different body that's interpreting that memory. So not only does it have the memory, not only can it still interpret that memory, but it can also generalize from that.
So, so I, you know, to some extent it seems very specific to the agent, but on the other hand, we have these examples of these memories sort of porting across agents, which I think there's something here that, that will be huge. I think we just don't. Yeah. I
Andrea Hiott: love that example of the caterpillar and butterfly, but I also think it's kind of tricky.
I mean, when I was, I think I was. Listening to one of your talks, I was thinking if we could zoom, zoom, zoom, zoom, zoom into that process of whatever is happening when the caterpillar is changing to the butterfly, of course, there's going to be continuity, isn't there? Like, from my perspective, if I'm thinking of it as a trajectory, right, if you're kind of mapping a trajectory, so you're going to just start with any particular agent base within that whole thing.
It's not that everything becomes mush and then [00:41:00] reforms. I mean, you could sort of, if you're going to, if you could zoom, zoom, zoom, could you? Could you kind of follow one position as a trajectory that's been changing in and because I guess that's what I was getting at too with the xenobot like does it create its own shape or is the shape created from where it's coming from isn't there is there some kind of continuity between not necessarily the physical shape but you know the same way you talk about all these other different spaces and I do too I think this is a big part of of this.
new understanding of cognition is all these different spaces from which we can, again, measure or assess, like, wouldn't there be some space in which that is a continuous trajectory between caterpillar and butterfly?
Michael Levin: Well, uh, I, there may be, but I mean, look, uh, We don't, for example, uh, if, if it were that simple, we would have computers that, that were robust to that.
We don't. If you, if you take a computer, uh, a normal computer architecture and you swap some [00:42:00] memory cells around, nothing works. There's, there's physical continuity, but the thing is completely gonna, gonna not work. Yeah, but that
Andrea Hiott: could just be that we haven't, I mean, in the same way that we can't zoom in to the degree that I'm talking about, then we also wouldn't be able to represent it in a computer.
I
Michael Levin: mean. Well, except that in the computer, we've have zoom, we can zoom into the, to all the way in the computer. There's nothing preventing it. That's why I choose the computer because unlike biology, where things are hard in the computer, we build it from, from copper atoms. Like we know exactly.
Andrea Hiott: Yeah. But that's only one kind of particular.
Uh, representation or something that we've actually created, um, I think I like that you say it as a companion or a partner, right, to in this kind of journey. So there could be, I mean, we could think of quantum computers or we could think of some kind of living computer built from bacteria and fungi or something that might actually Then then you probably could switch around a couple things if it's a living.
Yeah. Yeah.
Michael Levin: Yeah, for sure. No, I'm not saying it's [00:43:00] impossible to do that. I'm saying it's entirely possible. I'm just saying, uh, we have no idea how that would work because we do not have any architectures that have that property. I mean, the butterfly does. So we know it's possible somehow. Somehow it gets done.
Um, but, but also the question of what is their continuity of because yeah. Because some kind of an engram that said leaves, uh, is not what the butterfly is, is picking up, right? So, so That's what
Andrea Hiott: I mean by a different space. It's not the space that we've, that we, that we assume. Yeah, we don't know Yeah, way of looking at it.
Michael Levin: Um, you know, there's, there's also this issue of, and I think, I think this is probably what we're getting out when you were talking about, um, uh, you know, what, what are the xenobots really doing? Like one of the things when, when we, when we ask, where do biological shapes come from? The typical answer is, well, selection.
You look like a frog because over the years, everything that didn't look like a frog and tried [00:44:00] to live in this pond died out. So now you look like a frog. Okay. That's fine. Except that, um, there's never been any xenobots and there's never been selection to be a good xenobot. And so it isn't that. And so, and which is one of the reasons we made xenobots is exactly to probe exactly this issue.
Where do these patterns come from? And I think we have to be really, um, we have to redefine, uh, what we're asking. Because I, I, this is, this is my kind of favorite, stupid example that, that I always give you ever seen them a Galton board, you know what that is, they have this, right? So, so, so you get, you get a toy of this on, on Amazon.
It's like a, it's a, it's a board. It's got a bunch of nails banged into it. Uh, you take a bucket of marbles and you sort of dump it on top and the models go boom, boom, boom, boom, boom. And if you've got enough marbles, the outcome is always the same. It's a bell curve makes us beautiful, beautiful, you know, binomial distribution, right?
Um, and so you can ask the question, all right, that's a great curve. Where is that pattern encoded? So you start staring at the nails. I don't see it there. And you start looking at the wood. Well, I don't see it there. And is it the [00:45:00] distribution of nails? Not really. Is it how I dumped the marbles in? No. Um, where, where does it come from?
And so I think, and there are many other examples like this. I think we have to, uh. I mean, basically, this is what what Plato and Pythagoras were on about when they were talking about this platonic space of forms, that there are laws of physics, laws of computation, and so on, that are not to be found in the physical world, like the details of all the physics could be different at the Big Bang, all the laws of the physics could be different.
The distribution of prime numbers is still what it is, it's not going to change. So, so a lot of the answer to the question of where do these shapes come from, that you have to look there in addition to, oh, it's because, you know, in the physical world, somebody, somebody scribbled on, on this particular scratch pad, but, you know, before it's, it's partially that, and it's often sometimes like somewhat that.
But it's often much of it is not that
Andrea Hiott: makes me think of Schrodinger's cat or these typical experiments of, you know, [00:46:00] quantum physics with the light going through the, the discs that make these kind of patterns and stuff like, again, it could go back to that idea of From where we are assessing it and this whole kind of scaffolding that we've built towards measuring and assessing, um, there is this kind of regularity, which gets at another thing, which is not that you're just changing the environment, right?
When you change, when, for example, let's talk about regeneration just because I know we have to go soon and I want to at least. Bring up that word, but, um, you're not changing the environment when you regenerate, are you? Is it just that simple that you're changing the electrical pattern and we could just call that the computational surface or we could call that the light cone that you've somehow now changed the pattern of the encounter and therefore The agent itself will change or it's not
Michael Levin: that simple.
Is it? No, no, no. And these are all these words are not interchangeable. So, so like, um, the cognitive, we, we, we don't know how to change the cognitive Lycon. The cognitive Lycon is the scale of the [00:47:00] biggest goals that a particular, uh, uh, system can, can pursue. Actually, that's not quite true. And I think in the cancer treatments, we do, we do expand the cognitive Lycon, but, but, um, that's not what we do for regeneration.
So, so in regeneration, yeah. Uh, we try to, we, we, we, we, we change the environment in two ways. One is like, like for example, for limb regeneration, we have this, um, wearable bioreactor and, and part of that, and it gives us like protective aqueous environment. Part of that is to convince the cells that it's safe and it's worthwhile to put energy into regenerating.
If you're a mouse or something, and you're running around outside and you're, you know, your arm gets bitten off. You're not gonna be able to regenerate anything because you're putting weight on it. You're getting infected and stuff like that. So, so the cells, there's no point in trying to regenerate anything.
So, so you're trying to convince the cells that it's a safe environment. Also, you're providing a payload of drugs that. Target, uh, the electric state of the cells that convince them that the journey through morphous space that they should [00:48:00] take is the one that goes towards making a limb and not the one that goes to scarring and, and, and stopping growth.
There are multiple paths that they could take. So you're trying to point
Andrea Hiott: happening at once the, so you are actually changing the chemical kind of ion, the ion channels via chemicals too.
Michael Levin: Not just yes, you Yes, you have to you have to but but but but if you get the problem is if you get I mean, and people do say they say, Oh, you know, you're talking all this stuff about these high level controls, but I see you're just putting chemicals on on ion channels.
If you if you focus on that part. It's like when you're showing your thermostat to somebody who's never used one, they sort of staring at how your finger goes on the button and go, Oh, I see it's the, it's the, it's the physical interaction between your, your skin and the button of the, of the thermostat.
And you say, I mean, yes, that occurs, but, but you're missing the whole point. Like that isn't the important part. And I'm saying, no, no, I'm tracking the physics. That's what you. done. You've, you've put your skid on the button. Like, yeah, yeah, I guess. But, but you're missing what's, what's important about this.
And I think that's, , it's the same thing here. I mean, yeah, we use [00:49:00] these chemicals to get the channels to open and close. But that isn't the point. The point is that, uh, the reason we did it is to control the decision making of the collective intelligence. And that's the hard part. The hard part is figuring out what signals are going to convince that group of cells to do A versus B.
And that's happening at
Andrea Hiott: a different level.
Michael Levin: Yeah, that's at a completely different level. And that means that you really have to have a model of the internal perspective of the system. With a bowling ball, you don't have to do much of that. It's just you, you, you look at the landscape and you kind of know what's going to happen with a living system or various other complex systems that have to be alive.
You really have to pay a lot of attention to what are the internal beliefs, the goals, the competencies, uh, the memories of the system, otherwise you have no idea what your signals should be.
Andrea Hiott: Mm. Okay. So why is, how is it different, something like, um, how is this different from something like, uh, talk therapy or something?
Like, what would be different, right? [00:50:00] When we do talk therapy, we are kind of literally, of course, people get really mystical about, oh, the vibration, and we're changing the vibration, and you do literally, when you meditate or something, or when you talk, you change the vibration, um, is that, like, people might say, oh, that's changing the bioelectricity field, too, and I mean,
Michael Levin: it is, in a certain sense, if somebody talks you into, uh, taking up an exercise program because they sort of point out how good it is for you and all that.
They've absolutely changed the bioelectrical properties of your cells because you're going to flex your muscles and all this kind of stuff. I mean, so on the one hand, yes. And one of my favorite quotes is Fabrizio Benedetti, who says, uh, words and drugs have the same mechanism of action. And I think that's true in the end.
That's absolutely true. But like with everything else, It's an empirical question. What is the best level? Is the best level. So, so here's, here's my claim and, and it is an, an experimental claim and I could be proven wrong. And, and, uh, we will see, here's my claim. The best way to do regeneration is not at the low level of managing molecules, [00:51:00] and it is not at the high level of talk therapy.
It is in the middle of addressing the bioelectric, uh, uh, memories of, of collectives, of cells. Now, somebody else could say, guess what, dude, you, you missed, uh, you missed the most, the most exciting discovery. The most exciting discovery is that I can, I can talk people into regrowing their, their, their limbs.
Fantastic. I'm not saying that can never happen. Maybe there's a way to go from that high level, uh, of verbal interface to the, I mean, there is the, there is this field of, um, uh, hypnodermatology. Where where people people can can make changes in your skin cells by going through the verbal, maybe at some point, somebody will figure out a way to do that for healing and regeneration.
Fabulous. At this point, I think it's I think the right level is kind of in the middle. It's the it's it's it's tweaking directly the memories of the cells, but Yeah,
Andrea Hiott: and that's not necessarily, again, it's not that you're saying that one or the other of those has to be true. It could be all, all those ways could be ways of changing.
They
Michael Levin: could, but, but some of those [00:52:00] ways would probably be, that's it, that's it. Some of those ways will be more effective than others. Yeah. And there, and there are many people, I mean, look at an alternative, uh, uh, alternative medicine, right? There are many people who, who say we will, we will cure your cancer through positive attitude and through meditation.
Am I saying that can't happen? No, it may be, it can happen. Am I saying it's, it's a kind of an effective thing you can count on? Not at the moment.
Andrea Hiott: So. Right, because there's so many variables of. We don't know how to do it. Each agent would not start at the same place in terms
Michael Levin: of. And yeah, and so at some point we will find out which is the best, most reliable way to do it.
Andrea Hiott: Last question, I, I think a lot about transportation because of course, like for me, this, the way that we move through the world and these companions, this technology that we build changes our ability to way make, which for me is our ability to navigate our cognitive light cones or capacity changes through that movement.
And so sometimes when I'm reading your work, I think about regeneration. Um, in this, through this, from this perspective, for example, like electric motorcycles, [00:53:00] when you're going downhill, you're regenerating, right? You're actually like refilling your battery just by, by nothing, by just letting the machine, the vehicle go with the inertia.
So this word inertia, I wonder like if you've thought about that or if there's any. If you can see any kind of way in which that's also kind of what we do, or you might be doing when you're changing the pattern.
Michael Levin: I think absolutely. If you think about it as in the simplest way of, of resistance to, uh, to change and resistance to stimuli, yeah, every system, if you have no inertia, you're not going to live very long as a, as a complex system, because every, every input is going to destabilize you into some other region.
If you're going to be a coherent. thing, you have to have inertia in all these different spaces to have some degree of stability. Um, if you have too much inertia, then, then also you're not going to be, you know, so there's some kind of a sweet spot in different systems for how much resistance you're going to put up to, to various [00:54:00] external changes.
Andrea Hiott: Are you trying to change the inertia in a sense when you're trying to, uh,
Michael Levin: yeah. I mean, so, so one way to think about that, uh, for example, would be, yeah, uh, yes. Uh, there are, there are ways to, if, if you have a mod, we haven't published any of this yet, but, but if you have a model of cellular decision making.
That includes components of what things do they ignore and what things do they pay attention to the inertia or is by virtue of ignoring a whole bunch of inputs. And so one way to change that to lower that inertia might be to help cells to change their mode of attention so that they're now paying attention to other input.
So, yeah, that's potentially we haven't done that in any existing papers, but that's certainly
Andrea Hiott: something that we're working on. That would change the competencies
Michael Levin: too, right? It can, it might reveal novel competencies, right? So you might have, you might have, so you might be super competent doing X, Y, Z, but if you're, if you, if you're not paying attention to the trigger, then, you know, as observers, we're never going to
Andrea Hiott: see it.
Patterns in inertia are [00:55:00] not dissimilar. Inertia is a kind of a electrical pattern in a sense, or could be thought of in
Michael Levin: that way. Uh, it could, I, I, I think of it more as, uh, The, the, the degree of willingness to, to reconsider your pattern, right? Yeah.
Andrea Hiott: Oh, that's great. That gets to a whole other thing about awareness.
We don't have time now, but awareness and how attention and awareness change
yeah. Yeah. Yeah. I'll let you. Okay. Thank you
Michael Levin: so much. Yeah. Thank you. Yeah. Great. Great. Great to see you again. Thanks for the conversation.