[00:00:03] Speaker A: When you focus on the precarity of living systems and how all of their parts depend on each other, you get a story about where timing is really important in the sense that you can't build part of it and then build another part and then build another part, which you can do with a machine.
And for that reason, you get a very a good reason to think that the idea that you could build an organism is almost impossible. You can only start with a protocell and then wait billions of years.
This is what I find very interesting at the moment.
All of these fluctuations in synaptic structures and how they turn over on these rapid timescales. And I think that is starting to look at the brain as this inherently spontaneous, constantly falling apart, constantly rebuilding itself kind of system.
So it's not that we as individuals try to represent the world, and then once we've represented it and thought about it, we talk to other people about it.
It's rather that these individuals are trying to survive in the world, and then we have to interact with others whose ways of surviving become interdependent with our own and also in conflict with our own. And in having to negotiate that, you start to build up the idea of something beyond yourself and a reality beyond your own needs.
[00:01:29] Speaker B: This is brain insp.
Powered by the transmitter. Hey, I am Paul Middlebrooks and that was Kathryn Nave.
Kate is a Leverholm Trust Early Career Fellow at the University of Edinburgh and the author of the book A Drive to the Free Energy Principle and the Meaning of Life.
In the book, Kate dives deep into the free energy principle and active inference, which are popular approaches to studying brains, minds and organisms in general, and which are being used in artificial intelligence.
Ultimately, Kate finds that these approaches come up short as explanatory frameworks for life and autonomy and intelligence.
Instead, Kate and many others advocate a framework that goes by various names. Kate calls it constraint closure. People say closure of constraints, organizational closure. Anyway, this is a concept from philosophy and theoretical biology that people like Alvaro Moreno and Matteo Mosio have put forth in their 2015 book Biological Autonomy. The core ideas are also found in various forms from people like Robert Rosen, Stuart Kaufman, Alicia Guerrero, Terence Deakin and others.
So we discuss what constraint closure is, why Kate thinks that it's a solid foundation to build on, and what, if anything, it means for cognitive science and the brain sciences to embrace this constraint closure View a link to the book and to Kate's information in the show
[email protected] podcast242 and I highly recommend the book even if you're looking just for a primer on the free energy principle and active inference as we discussed, Kate's journalism experience has helped her become a wonderful communicator of these notoriously difficult concepts. As always, thanks to the transmitter and Patreon supporters, to get full episodes and other bells and whistles support Brain inspired on patreon. Go to braininspired.co to do that, here's Kate.
Okay, here we go. I'm going to begin with a quote from your book, A Drive to Survive, the Free Energy Principle and the Meaning of Life.
Okay, so this comes at the end of the introductory chapter.
What makes organisms special is that the apparently invariant structures that constrain and enable these precarious flows of energy processes are also themselves reciprocally dependent on those flows of energy in turn.
Well, that's a mouthful. And that's what hopefully we'll try to unpack a little bit here today.
Excellent book, by the way. I'm sure you've heard that this multiple times or many times over.
So I've heard many reactions to your book go something like, oh, okay, like the free energy principle. She finally nailed it. This is finally. I can finally move on and not worry about the free energy principle being the answer. Fep. It's not the answer, right? For people who have been interested in it, that's, that's what a lot of people seem to take away from it. And that is like a sort of a destructive, constructive destruction that you do in the book, but you're also concerned with a positive account. And that's kind of what I want to focus on. And to get there, we'll, we'll need to go into free energy principle and predictive processing.
But the other thing I want to discuss, which is sort of my eternal question, is what is the question, right? How to even formulate a good question in the domain of what we're going to talk about.
So I want to pick your brain on exactly what your question or questions are, right? And then what you think the right hints are that are pointing in the right direction to better formulate a question, perhaps. Or maybe you have the right question already.
Um, and, and basically how far do we have to go? Right?
So anyway, that was, that was a lot to begin with. But let's, so, so let's start with maybe just your, your background. I think it's kind of interesting because you, you spent some time, and I guess you're still, still, still doing some journalism work. So you spent some time writing for Wired and, and other outlets and stuff. So what I kind of want to ask. And then you've written this excellent book, which is, like, very articulate and very well written and easy to digest. So I appreciate it very much. It's an excellent book.
But I'm wondering how your journalism experience has sort of informed or affected your academic work, whether it's thought process or just production. Right.
[00:06:16] Speaker A: Yeah. I think for me, they're actually very, very closely connected, particularly the kind of journalism that I did. As you mentioned, working for Wired, that was always.
That was always the place I wanted to work for the reason that it is like a deeply interdisciplinary. It seems weird to describe a magazine as interdisciplinary, but what has always characterized Wired was an interest in, like, art, architecture, design alongside technology and science and how these sort of blend together.
And I think that is what has always interested me.
And then going into the reason, I sort of went into philosophies, I did. I've only ever studied philosophy. And, um. And I kind of missed. I was talking to all of these experts and I kind of missed getting to be an expert.
[00:07:03] Speaker B: Um, and from that, from your journalistic perspective, you were like, oh, I'm missing out. Okay.
[00:07:09] Speaker A: Yeah. Like, I think, you know, you talk to these people who have worked on something for, like, 10 years, and you're like, I'm just constantly flitting around. You want a bit more depth.
Which is why I sort of went back to philosophy.
But I see. I think there's lots of ways people see the role of a philosopher, but I think an important model is sort of somebody who synthesizes and communicates between different groups, because people are not always good at explaining themselves across these kind of disciplinary gaps.
And that's the way I'm most sort of comfortable, I think, viewing my role as a philosopher, because there's very much the sort of old model of a philosopher as this kind of, I don't know, distant authority who sort of deduces things from pure reason. And I find that quite alienating.
But if you think of it as in a more journalistic model, that's something that I like, and I also like. I think there's. It's important that I think you can do a lot of philosophical work just through a clear explanation.
You know, it doesn't have to be solely, here's my argument and why you should believe me. But more, let's take a look at your argument. Let's pull it apart, let's see how it works. Let's simplify it if we can. And I think that is, yeah, you can get Quite far in conceptual work with that kind of approach, which is what you learn as a journalist really.
[00:08:26] Speaker B: Oh, okay. Well, maybe I should have gone into journalism. I mean, I've told this story before, but like I, when I was a research technician in a neuroscience lab early on, before I went to graduate school, I was able to moonlight in like a neurophysiology course. And I remember at one point he asked if there were any questions or something, or asked for comments on something, the teacher, what, what he had been presenting. And I, it was like a. More of a open up to like bigger questions, whatever. And I said, I think we're asking the.
I, I think we're asking the wrong questions or, or maybe I posed it as a question, like are we asking the wrong questions? And he was like, yeah, maybe. So what would be the right question? And I was like, I don't know, that's, you know, that's why I'm here. Right. Because I even to articulate a good question, you've almost done all the work that, that needs to be done in that case. And that, that's kind of what you were explaining just now. But you mentioned that the classic notion of a philosopher is someone who's sort of deducing theological structure. Some argument.
But before that you mentioned about how you're most comfortable in like the synthesis role. And I think that's wonderful to hear because that's so rare that people synthesize ideas and it's so difficult to do. Right. When I think of a philosopher, I think of like sort of a specialized topic and then you go deep on that one topic. Kind of like in the sciences you get like really specialized. Right. Like people in neuroscience, 99 out of a hundred of us maybe would like, if I mentioned biological autonomy, they'd be like, I don't know, what are you talking about? And why would I want to care about that? Right. So, so you're really, that's your comfort zone is synthesis in multiple ideas. That seems very difficult.
[00:10:12] Speaker A: I wouldn't say it's necessarily comfortable.
Yeah, maybe comfort is the. Maybe when I said comfort zone, that's the wrong word. I think it's certainly is quite a difficult zone to be in a lot of ways because you always know less, I think. Yeah, you always know less than the person you're talking to because I'm always talking to people about what they know. You know, you're always talking to somebody on their terms in a way.
So you have to be comfort. You have to try and get used to, I think Being sort of. I always think of yourself as like the idiot in the room asking the stupid questions. And that is what you do as a journalist. Right, but. Yeah, exactly. You know, this. Right. Like this is that you have to ask those. Be the person who asks those questions, questions that other people want to ask but wouldn't ask because they're supposed to be an expert in this particular.
[00:10:57] Speaker B: Right, right, right. There's a lot of freedom in letting go and realizing you probably are. You are the idiot in the room. Right.
[00:11:03] Speaker A: And you have an excuse to be. Right. Because I'm not. You know, any room I go into, I go, I'm not a neuroscientist or if, you know, I'm talking to philosophers about meta ethics. Well, I'm not a meta ethicist.
It's getting more difficult, I think, as things go on. People expect you to be more of an expert.
[00:11:17] Speaker B: Yes, I'm familiar.
[00:11:19] Speaker A: Yeah. So that becomes harder. But I. I enjoy that more. I like asking questions, but, yeah, it is. It is difficult. You never really get that certainty of having expertise in any one thing. So I think you do lose that, and that is quite hard.
But. But you.
[00:11:37] Speaker B: Yeah, a friend described it to me. So there's like T shape, right. Where you're. Well, so there's. What is it? Comb shape? There's comb shape, there's T shaped. And I don't remember what the different metaphors are, but, like, you know, as learning neuroscience, I was becoming very, very specialized, and I was always really resistant to that because I want to know. I want a broad and general knowledge. Right. But if you have only a broad and general knowledge, you still know nothing. Right. So now what? I think the best thing to do is to have a broad and general knowledge and expertise in all domains, which is impossible. So I'm not sure what the right path is, but you have become an expert in things like predictive processing, free energy principle, et cetera, and along the biological autonomy lines. So, yeah, I mean, you are the person to. To come to. Right. Which is why you're here right now. You wrote a book on the topics.
[00:12:27] Speaker A: I never feel like an expert when I talk about it, though. I think this is what I like about writing, is it's very easy to be expert in writing when you have time. And then when you have to talk about it, it's like somebody could ask you a question about something you've literally written a book about, and you'd be like, I don't actually know. I have to read.
[00:12:41] Speaker B: I know. Yeah, yeah, it's the same thing with scientific papers. Like, did I.
There's a question about your second. The second section in your methods, you know, rather second paragraph in your methods section. So. Yeah, well, let me read my own paper and figure it out. Right, yeah. Okay. Well, so that's an interesting thing. So you went to do philosophy then, with the aim, like of going back to journalism or what is the. Did you have a goal?
[00:13:11] Speaker A: Yeah, so I knew that there's not a career in philosophy is not really a very plausible life choice.
So my.
[00:13:20] Speaker B: What do you mean? All your colleagues are doing it in
[00:13:23] Speaker A: the sense that, you know, I particularly, as you go through the PhD, you hang around with lots of other incredibly impressive people and you see, you know, people ahead of you, they don't get jobs, they leave.
And I was aware of that from the outside to begin with, that there are.
But yeah, what I wanted to do was to get to a point where I had enough sort of philosophy credibility that I could do popular writing on philosophy. And that's something I haven't done as much of as I would like to. It's hard to find time for it. But that was always the motivation for me was to do popular philosophy.
[00:13:58] Speaker B: That's not what a drive to survive is. What do you consider a drive to survive? Specialist or what?
[00:14:03] Speaker A: Yeah, I think it's definitely specialist. I think, I hope and from the people who've read has been read by a reasonable number of people who are not academics themselves.
So even though it is pretty technical, it's accessible. Yeah, I mean, I hope so. I mean, it's hard because the French principle is not an inherently that accessible a thing. But I mean, my granddad's reading it, which is amazing.
So that's something.
But yeah, I think I sort of. I think. I mean, the model for me was. So my supervisor was Andy Clark, and he is in many ways one of my models for what good philosophy can be. And a lot of his books, I think they're not.
Some of his books are like the Experience Machine is more popular, but a lot of his books are this sort of boundary line, I think, between a super academic text and something that non academics would also read.
And that's what I would probably like to aim towards with the stuff I do.
[00:15:04] Speaker B: Yeah, that's kind of my goal with the podcast too. And I constantly question whether it's the right path. Right. Because it necessarily means that you're not going to reach a super broad audience. But also I am uninterested in complete layman science communications stuff. It's just not interesting to me because I want to learn stuff and so it has to be some weird in between state that I don't know quite well what, what that. Right. Is. But so, so your book was like perfect for me. Right. So for people like me, your, your book is, is just perfect. I'm sure you've heard like great, great feedback on, on the book.
[00:15:41] Speaker A: People have been nice. Yeah, I haven't had it, I haven't had any, any like sort of angry confrontations about it yet or anything.
[00:15:47] Speaker B: So the, the, the free energy principle pitchforks haven't come out?
[00:15:51] Speaker A: Not yet. But I avoid, you know, I, I know where to go and why not.
[00:15:55] Speaker B: Yeah.
Well, okay, so. So you have that background, right. And, but, but. And your advisor was Andy Clark. And so what I want to ask you is like how you got interested in these topics in the first place. Because this is like a weird time. Sometimes I ask myself, well, why am I interested in this? Right. So, but, but it's a weird time because AI is successful and you have all these like CEOs and well educated intelligent people worrying about LLMs becoming conscious. And if we just scale up and this all seems like so crazy to me. So that's one important reason why studying like. Okay, well, just as an antidote to that, studying something like how do you get from metabolism to full blown intelligence? Which is a lot of, I think what you're interested in, intelligence and mind. Right. That consciousness won't just pop out of a computer because the mind and the brain, the organisms are not machines, which I know that you're sympathetic to. Right. So.
So with that in mind, like how did you get into the predictive processing, Andy Clark kind of stuff and then maybe just a brief summary of like that, that traveling that you've done since then.
[00:17:12] Speaker A: Yeah, it's a bit of a funny path.
The reason I got into predictive processing was just because I was in Edinburgh and everyone was interested in predictive processing at this point.
Sort of like I did the. They have a mind language and embodied cognition masters here. And that was very focused on predictive processing. But it did chime with me because I've always been, I've traditionally been very interested in I suppose two things like epistemology and visual experience.
And for visual, visual phenomenology, what was very appealing about predictive processing is it gave you a story of visual experience as being very structured and action oriented.
[00:17:57] Speaker B: So can you just very briefly overview what that story is? I realize like sometimes you got to back up and yeah, yeah, with predictive
[00:18:03] Speaker A: processing, I think what comes through is a story of, I guess most principally the idea that the brain is oriented around prediction error minimization.
But one of the things that the sort of broader predictive processing story adds to that is that prediction error minimization is hierarchical. So you're trying to reduce expected prediction error over all of these different timescales.
And that that is also relative to what kind of actions you want to take in the world.
[00:18:37] Speaker B: So, but originally you were just saying like you were interested in like the visual experience. Right. So originally, I mean a lot of neuroscience was based on vision and still is, but. But in this case, like you, you would be looking around and the idea of predictive processing is like you're in a zoo and so you expect, you predict you're near the lion cages, you predict you will see a lion, and then when your eyes move there, you actually see an elephant and then there's this big error and then that cascades up the hierarchy in your brain and then your brain is like, oh, that is a elephant or whatever I said instead of a lion. Something like that's a very bastardized version of it.
[00:19:14] Speaker A: Yeah. And I mean, so the thing that I found quite compelling was how that helped you talk about lots of visual illusions because I always enjoyed that stuff. All of the ways in which you can show that your experience as a world isn't quite what you would think it would be.
That in some ways it's more impoverished than you would think. In some ways it's richer than you would think.
So I was very interested in how these kind of illusions could be incorporated, could be addressed through a predictive processing story.
In particular, I got quite interested in this is 10 years ago now. So. But in an inattentional blindness type phenomena where you should see things, you think you would obviously see something happening in front of you, but you don't.
Which I think pushes you towards a recognition that your perception of the world is very high level. So what I ended up focusing on was this sort of cluster of ideas around, like intermediate level explanations of visual content where your experience is all. It's not the fine grained details, it's not sort of abstract thoughts, it's at that mid level of abstraction, which I think is sort of right. But the way that's often pictured is there's just sort of this fixed intermediate level of experience of things like tables and chairs, but not, you know, photons or, and their philosophies.
And what predictive processing gives you is a story about why those contents could shift in terms of how granular they are.
So that, yes, your experience is relatively high level, but depending on what particular actions you're engaged in and what particular things you're going to be sensitive in for those actions, you could expect the kind of granular structure of your experience to change.
So if you're reading text, you're going to be more sensitive to the specific characters of individual letters. If you're shorting a pile of documents, you're going to be more sensitive to the shape of the pages.
And I think that just does capture a lot of both phenomenological report and experimental work on what kind of visual sensitivities we have in different contexts.
So it gives you that, that sort of flexibility of where visual content is that I quite liked.
[00:21:31] Speaker B: Well, so you just mentioned the word phenomenological, right. And I know that you are.
Oh, what's the word? I know that part of your process is thinking about phenomenology, right. And how it comes in. But like predictive processing itself, when I think of it, I think of like clean computationalism has nothing to do with phenomenology. So that was important to you, that it meshed well with actual subjective experience?
[00:21:59] Speaker A: Yeah, definitely. I think people's, you know, there's lots of complexities over how to bring first person experience into how we think about
[00:22:11] Speaker B: these things,
[00:22:13] Speaker A: but it sort of hard to deny to me that it is a rich source of understanding the brain in the sense that you don't just have. So I did a whole lot of psychophysics experiments at one point in my career to try and understand visual experience, not because I wanted to. And that gave me a real sense of what this could and could not tell you and how much you can learn from sustained discussion over one person's experience as opposed to averaging across tons and tons of people's experience. And that both of those are going to be important ways of thinking about what's going on in those conditions.
[00:22:50] Speaker B: Well, this automatic, I mean, this immediately makes me think, okay, the nice thing about averaging over a bunch of people is that it is very amenable to scientific questioning. Then, right? Where, and that's the big roadblock with phenomenological studies is that where how do you scientize this? Right? It's all like narrative and stuff and, and you know, you talk about this a little bit in the book, but also in more recent paper that I'll link to as well.
And maybe just in discussions, I'm mixing it all kind of up in my Head. But so my goal, right, is to sort of. What we're going to be discussing more is to bring it into the scientific realm. But there is talk maybe that's not even possible, like, to formalize this sort of thing. So maybe it's not possible to formalize phenomenological accounts. Right. And so how do we mesh that with, you know, doing. Actually doing science? So there's this tension there that I don't. That I want to resolve, and I'm not sure that it's possible.
[00:23:49] Speaker A: Yeah, I mean, I'm not fully sure how these kind of tensions are resolved either.
The main way I think about it is just.
The simplest way I think about it is like not having. I guess what you might think of is like, undue deference to formalization.
The idea that once you've done something and it's been formalized, a sufficient level of formalization becomes somehow kind of unquestionable and invariant.
And I'm certainly not against the idea that you can formally describe contingent aspects of phenomenology in, like, a way that happens to work in a particular circumstance.
The problem for me is the idea that those formalizations could become sort of definitional.
So, I mean, this is the thing I think you do get in the trajectory of phenomenology and kind of people like Merleau Ponty and the way he understands the relationship between phenomenology and formalization and empirical science, where you have this kind of continuous sort of circularity. I mean, for Merleau Ponty, experience is still primary, still has this primacy.
But that doesn't mean that you can't revise your understanding of your own experience through experimental methods or through even formal models of it.
So there is always this kind of back and forth. And I think that's the version of phenomenology that's most appealing to me.
[00:25:11] Speaker B: But yeah, yeah, okay, well, I sidetracked us there with phenomenology, so maybe we'll come back to it because it. Because like, bioanactivism, what you call bioanactivism, right? Yeah.
Kind of runs into those same sorts of problems when you start looking at the theory and philosophy behind it. But. So you were doing. Learning about predictive processing and getting into that. And I don't know if you want to kind of quickly take us up to. What I want to ask you now or eventually is like, well, how did you get into biological autonomy work? Right. The Moreno and Mosio stuff that. That you come back to as a potential solution in the book to the question and what is the question? What is the question? And, and I don't. And, and continue that narrative of being into predictive processing and where it took you.
[00:26:00] Speaker A: For me, I guess the question there is quite a strong underlying thread or question that I see as kind of having gone all the way through even when I was maybe an undergrad, which is making.
Making sense of. Making sense of error in a way that's not about error in relation to some kind of external truth.
Because you get this kind of tension, I think, where you have in. In all sorts of areas of discussion about things like cognition and agency and also in ethics, which is something that I have a sort of side interest in between trying to get away from a view where there's like a standard that's out there that we need to meet and it's independent of us and a view in which we don't need that external standard. We're just these kind of interactive processes where I mean, I guess you see it sometimes in inactive accounts where there's this concern that there's no account of how you can get things wrong. For example, if perception is ongoing coupling with the world, how do you get illusion? How do you get misperception?
And that pushes you back to a representationless story. But then if the thing. If the standard that you need to conform with is as external to you as the mind independent world, then how can you ever know that you're right? So you either have this external standard but no way of meeting it, or everything is internal and there's no possibility of error.
And what I'm interested in is how you can get a sense of something being an error for you is the kind of thing that you are, but not in the subjective sense that it's just up to you whether or not you choose it to be an error or not. Which is the significance of autonomy as a concept. But that's the sort of philosophical significance of autonomy that doesn't really come up when people talk about autonomy in like self driving cars, for example.
[00:27:52] Speaker B: Yeah, it's a different kind of thing. It's just a different.
The term means a different thing. Mark Bickhardt tries to address this misrepresentation issue that you're talking about with this interactivism, which I know you know because you cite it in the book and stuff.
Okay, well, so, so that. So then did predictive processing not make the cut for you? Then how did.
Was that it? Because that was kind of your original interest, right? And you're studying predictive processing and you think. Did you think Eventually. Oh, it doesn't account for my question or something like that.
[00:28:27] Speaker A: Pretty much, yeah. I think what drew me to it and what drew a lot of people to it, I think is the idea that it did give you that story where predict fiction era was this kind of.
You had these predictions you were trying to make and you could get them wrong or right, and that gave you this internal sense of like, I guess the kind of story Jakob Hovey tells in his recent book. I think he gives the sort of best defense of this kind of story where there's a basic normativity towards some form of regulation that your identity depends on the preservation of this model of the world.
And that because everybody has their own model, it's not that, you know, we all have to converge upon the same model of the world, but we have to find a way of interacting that can preserve our sort of our self model.
And that seems at first glance to give you a kind of normative story where it's.
It's still grounded in your own existence rather than some sort of external standard.
[00:29:28] Speaker B: But this is a social thing. He's expanding it to the. So you're saying like. And interacting with others.
[00:29:33] Speaker A: Oh, so no, not necessarily.
Okay. More that.
How to describe it.
So that on this kind of story to exist is. There are certain features that will define me that have to stay the same. And for me to continue existing is to keep those things stable. If they're no longer preserved, I no longer exist.
[00:29:54] Speaker B: I see.
[00:29:55] Speaker A: And so your.
Your sort of individuating features are things that make you. You could be different from mine. So what's good for you is going to be different from what's good for me.
But there's still a fact of the matter about what defines me versus you such that you could get that wrong or I could get it wrong. I could think I need to keep my body temperature at a certain level and be wrong about that and then die.
So it's a very.
It's one way in which I think Priddy Depressing and the free energy principle have tried to get a normative story that meets the kind of thing that various inactive accounts would be looking for.
[00:30:33] Speaker B: I'm not sure if it makes sense to sort of.
I fear I'm burying the lead. Although I'll say this in the introduction because I'm about to go into sort of the natural path of like, okay, well then. And then you came across free energy principle and what's wrong with that? You know. But I want to start from a more sort of Positive account. So maybe what I'll ask you is like at what point did you discover or did you find like the biological autonomy, closure of constraints, organizational constraints, did you, at what point did you find that as like sort of a solution and, or a potential solution and maybe we can work backwards from there?
[00:31:13] Speaker A: Yeah, I think that makes sense. So the thing I was looking for that specifically brought me to biological autonomy was a way of talking about self production that was more general than cellular metabolism, but not so general that it could be seen as a property of all sorts of non living processes.
So cell production in terms of cellular metabolism is how the idea sort of initially gets introduced in active cognitive science and autopoiesis theory with Maturana and Varela, where they try and capture the features that make a cell count as a self producing system.
And it gets crazy vague in the sense that people understand those essential features in lots of different ways. Maturana and Varela understood it in lots of different ways in their own work.
But there's two. I think you can divide those two ways of understanding it up into ways that focus on the specific sort of molecular properties of a cell where it has to be like molecular synthesis and a membrane and it's very, very restrictive.
Or people who take that specific cellular process to just be one possible realization of any sort of circular causality. So any sort of system where its output affects its input and it keeps going, and that could be an oscillator. Right? That's very, very trivial.
And neither of those works. Because if you want to talk about this kind of self production as being a general basis for agency and cognition, then you need to way to talk about it in multicellular organisms, in organisms with nervous systems.
But if you do that by just defining it in terms of sort of a circular feedback loop, a closed loop of circular processes, then you get something that doesn't capture this sort of importance of like mind life continuity and the idea that agency is an inherently biological phenomenon because you get something much weaker and that that direction is what leads to the free energy principle and these sorts of formulations of self production that become to my mind much too trivial.
So what I like about biological autonomy is it is the specific account of biological autonomy given by like Moreno, Mossier and Montaville is it tries to capture the sort of, the more thermodynamic and material specifics of cell metabolism in a way that doesn't have to be specifically about catalysis and like chemical synthesis.
[00:33:46] Speaker B: So can you describe your version of. I don't Know, I never know what to say. Closure of constraints, organizational closure. There's sort of a cluster of terms that are kind of interchangeably used. What do you prefer? And we can stick to it throughout the discussion and then just describe it overall.
[00:34:01] Speaker A: Yeah, I say constraint clos. Cause it's sometimes called the organizational account. But I just don't think that captures
[00:34:08] Speaker B: what the important parts of it organization seems. Sounds static almost. Right, so constraint closure, like the constraint. I don't know if. I don't know what's best, but. Okay, yeah, yeah, but I can see organization as aesthetic. So constraint closure. All right, we'll go with that.
So, so then broadly, kind of like, what is the idea there?
And yeah, yeah, first we'll start with that.
[00:34:30] Speaker A: Yeah, so the reason this is also the reason the constraint part is important to me.
As I said, the definition of this in sort of circular closure, that's sometimes described in terms of processes.
So in like a standard kind of description of it, a network of precarious processes that mutually depend on one another.
And you could have something like one of the counterexamples that's often given is say, a robot that seeks the sun to charge its solar panels, that then powers the process of seeking the sun to charge its solar panels. That would be a circular set of processes that all depend on each other.
And what that means is for that kind of system, you can deprive it of the energy required to keep that process going, and all of the structure will remain there, such that when you reintroduce the energy, again, that process will just start up.
So that's a much weaker notion of precarity than you get in something like a cell, where it's not just the processes that depend on one another, but the structure itself. And so you need to make this distinction between structure and process, which isn't immediately obvious.
The way the constraint closure story does that is by understanding the kind of causal relationships you get in the cell as being ones where the structure restricts the degrees of freedom of some process that would happen anyway. And then by restricting those degrees of freedom, you make something new happen, which is a very standard notion of causation. If you think of a piston reducing the degrees of freedom of an expanding gas to move the piston upwards, or, you know, a channel restricting the flow of water downhill to make it have enough pressure to turn a water wheel, those are just all cases in which constrict predicting something would happen anyway enables new things to happen, which is something Olivia Herrero talked about a lot.
So that's why you get the constraint process distinction.
And that's very natural for cellular causation in the sense that a lot of cell causation is based on catalysis, which is literally constraining something to make something happen at a rate that it wouldn't happen otherwise.
Also, ATP synthesis, so the synthesis of energy supplies in the cell is based on a constraint in the sense that you build up a gradient across a membrane to power something, to power a series of reactions that creates energy.
So constraint is just a very natural framing for what goes on in the cell.
But it's also much more general than catalysis because it doesn't have to be literally one molecule catalyzing a chemical reaction.
Right, right.
[00:37:17] Speaker B: But the, but catalysis. Catalysis fits into that general constraint closure framework.
And so, yeah, so when, when I came across that work, I too, and you know, this is reminiscent of Robert Rosen, et cetera. Like, I mean, it's, it's not a new thing. It's just formulated in a new way with some improvements, you know, which sounds easy, but it's very hard. But so I, so I read that I get excited and I, I think, all right, so how can I apply this to scale it up, to build it up to start? Because you talk about these things, it's always at the cell level or some processes in the cell. And it's sort of easy to sit back and abstract and say, yeah, I could see how that could work. And there aren't that many constraints to name them all. And it's somewhat potentially tenable to come up with a story. And people like Hoffmeyer, I believe, has an of, you know, how a cell does this, like in the constraint closure sort of way, and is starting to push that forward in theory and in actual science. But what I want to do is scale it up to, like, how do I think about multicellular organisms? How do I think. Can I think about our mind and our brain functioning within this domain? And automatically there's an explosion of like, okay, well, now I have to name 6 billion constraints and how they all fit together. So it seems like it's just impossibly daunting task. Right? It seems like we're at the very, very, very beginning. I think in, I think I've called it pre scientific almost with regards to, like, how this could be used to explain cognition. And I realize I'm sort of jumping the gun because these are the things that I'm thinking about and excited about.
But yeah, so, okay, so you explained the constraint Closure. I mean, are we at the very, very beginning, like, do you see this as a clear path forward?
How do you see this? Are you optimistic?
[00:39:18] Speaker A: Certainly not a clear path.
I think it's a position we're at where it's very nice to be a philosopher in the sense that you can make these.
And I think, and also a position where I think the role for philosophy, for like conceptual analysis actually becomes, I think, where you can actually hopefully really see that that has an important role to play. Because often, and what I mean by that is I think you can make arguments based on this notion of constraint closure that don't depend on empirical discoveries such as how it connects to adaptability and evolvability, that you don't. And lots of people don't agree with me on this, but you don't have to go out and check that this works. You can say that because of the way a machine is constructed, it can't have the kind of properties that a constraint closed system has. And that's not a hypothesis, that's a analysis of the kind of inherent features of these sorts of systems. It's a conceptual analysis. And that is something we can start to do at this point.
So that is, it's nice to be in a space where these kind of arguments are useful and obviously they might be wrong, but they aren't like empirical claims when it comes to extending this story beyond single cells. I think there's so much work to do and I think as you say, it is like crazy difficult.
There's a couple.
[00:40:50] Speaker B: Yeah, we're just talking about the conceptual engineering part of it right now still. Right?
[00:40:54] Speaker A: Yeah, yeah. I mean, it's almost like sometimes I have a bit of sympathy with these kind of mysterian positions where it's like it's in principle going to be incomprehensible to us. And I'm not necessarily. I think there is always going to be an element of incomprehensibility.
And there are, and there's. There's kind of argument, positive arguments you get for why that would be in the sense that, that when you have a system that's not defined by invariant features in the way that a homeostatic machine is, that limits the kind of predictions you can make about it. So there is a real positive argument for a limitation on what we can definitively say and predict about this sort of system.
But when it comes to the more specific question of how does it scale up to multicellular organisms and neural organisms,
[00:41:49] Speaker B: I can see, like if that's even the Right, sorry. Yeah, yeah. If that's even the right way. Scaling up is not the right way to phrase it even. But yeah. Sorry to interrupt.
[00:41:57] Speaker A: Well, how does it transform?
Because I like to think of these informations in autonomy and one thing that I think would be a really useful framing was like these kind of major transitions type stories where you have major transitions in individuality. So you have systems that are.
You have single cells that have their own autonomy and systems that come together and have a kind of collective autonomy and how those kind of trade offs are regulated. And you can look at that in systems that are simpler than the human nervous system, but where you still have the same kind of issue of two systems that have their own autonomous process have to somehow give up some of that to gain more flexibility in another scale of organization.
I think that's the way I would think about any kind of scaling up type story.
[00:42:42] Speaker B: Is this like a phase transition, to borrow a phrase from physics or something?
[00:42:48] Speaker A: Yeah, I mean one of the things that people sometimes there's a quote in the, in the free energy principle literature actually where people talk about phase transitions as amounting to the death of the organism. And the whole point is we need a story about that's clearly not true. But what we need is a story of how a.
What is it that persists through that transition that allows us to say this is a adaptation rather than a death.
[00:43:13] Speaker B: Yeah. And then you need to also. Okay, so, so I've used the word ontology before in thinking about these things. So. So then you would not need like an ontology of the transitions and like characterize the types of transitions and what it was before and what it was after that transition point.
[00:43:31] Speaker A: Yeah, I think because, I mean, one of the other kind of strengths that a constraint closer type story is going to be that there are so many different ways that you can realize something that counts.
And that is why I think you can't make sort of general statements about every single system will have this particular property or this particular property because you can realize this at all sorts of different scales. You can have multiple levels where cells have their own kind of constraint closed organization that's part of a larger multicellular body.
But you can tell stories about those kind of.
Because the account is so open ended, you can have transformations within it.
Yeah, it gives you.
It's not very constraining in some ways, which I think some people dislike about it. It has that level of abstraction.
[00:44:26] Speaker B: But yeah, I mean, do you even see like. Okay, you said you were sympathetic to the mysterian positions. Right. It's all like, too complicated.
Do you. Are you optimistic that we. That by the time I'm retired or whatever, you know, which is probably not that long enough, but that I'll be able to formulate a scientific question and to form a hypothesis within the cognitive sciences that has some connective thread that I can even, via narration, trace back to biological autonomy at the metabolic level?
[00:45:04] Speaker A: Yeah, yeah. I don't. Whoa, how in the sense.
[00:45:07] Speaker B: Wait, this is like. This is like a yes, in the next five years answer that everyone in AI gives.
[00:45:12] Speaker A: Well, I mean, what I'm optimistic about is I think there is a lot of work that I find super interesting on the kind of metabolic situation of the brain, for want of a better word, the way in which the brain isn't just the system that kind of responds to inputs in a.
In a passive way. And it's not just that the brain is spontaneously active, but it's. Which, you know, it's not just a spontaneous activity story, but also this spontaneous remodeling type story where all of these different levels of neural structure are spontaneously being turned over and regenerated, not necessarily at stable levels. So you have a lot of work on just. This is what I find very interesting at the moment.
All of these fluctuations in synaptic structures and how they turn over on these rapid timescales. And I think that is getting people who are much smarter than me in neuroscience interested in how that process is governed and starting to look at the brain as this inherently spontaneous, constantly falling apart, constantly rebuilding itself kind of system.
And that's new because that's, I guess, past 10, 20 years, because of the fact that new measurement techniques are making it possible to observe that kind of turnover.
And so I'm quite excited about that because I think that's getting people who aren't like me, these kind of abstract philosophers, interested in these kinds of questions.
So I think that's somewhere where there will be sort of interesting scientific questions that can be formulated, like how does this structure actually depend upon the processes that it enables? And how does that help us understand what these kind of neural structures are doing in a way that's not necessarily representationalist, but not just a sort of regulatory homeostatic story either?
[00:47:00] Speaker B: Well, yeah. So can you contrast that view that you're excited about with sort of the dominant view in neuroscience about.
[00:47:08] Speaker A: Yeah, I mean, I suppose I never know quite what the dominant view is because it depends on. But like, certainly in like, cognitive science, which is what I know more of the debate has and which includes neuroscience, but I guess neuroscience, there's so many different types of neuroscientists.
But the main way the dialectic I think is set up is you have brain as a computer versus brain as a control system.
So brain is a computer. We're interested in what the brain is doing when it's not interacting with the world, how it's modeling the world, these kind of detached internal processing.
And then brain is a control system.
You're interested in using dynamical systems theory to study the brain and looking at couplings and all of this kind of thing.
And to me that discussion misses this third option that I'm interested in, where it's not a dynamical system or a computer, it's a self producing system.
And so the hope is that there's going to be a move towards this third way of thinking about the brain. So there's a recent BBS paper on metabolic constraints in cognitive modeling, which is probably one example of this, where it's saying when we're building our models of what the brain is doing, we need to take energy demands and energy governance into more consideration. And I think that would be a step in this kind of direction.
[00:48:31] Speaker B: But who's the author? Authors?
You can send it to me later and I'll save it and flip it.
[00:48:37] Speaker A: But I got the surnames. I can't remember.
[00:48:39] Speaker B: Oh, okay. Because, yeah, okay, okay. So this is like, but the problem is like a computationalist approach and maybe less so a control approach. Because with control you have all these like thorny, like nested hierarchical struct structures that you have to talk to each other. But if you just take a strictly computational approach, it's easy in the, it's the easier of the three, I would say, because it's so detached, it's like flat, horizontal, like you just got to figure out the computation and then you're done. Right. And so that in itself is kind of straightforward and it's also liberating in that. Oh good, I don't need to worry about something as mundane as metabolism.
So. And the idea that we need to tie all of the grandeur of our super expert general minds. Right. Intelligence.
Now to talk about metabolism. What are you talking about? Right, so it seems like a downer. Right. And so it's maybe like hard to get people excited about this, but I'm at the point where like, okay, if that is possible. It is really exciting. But, but I, I think that you're. I would be hard pressed to excite many of my colleagues for the same thing.
[00:50:03] Speaker A: Yeah, I mean, another, another angle on it that I think gives a way of thinking about. Why should I be excited about metabolism if I'm interested in cognition is a. There's a different angle from the one I've talked about, which is just the basal cognition stuff, where you do get to see much more like it's back to single cells.
But you do at least get to see in that story how closely action perception loops are coupled to metabolic processes.
And I think a good example of where that could be transitional to neuroscience is the kind of stuff that Roman Barrett talks about where using like a paramecium as a model of a neuron. So you have a whole organism that has neuron like features such as firing action potentials, has a sort of sensory motor, coordination type story.
And so that's a different way of simplifying your problem because that's the advantage of computation is it simplifies things.
But you can simplify things in a different way where instead of abstracting, you literally reduce it down to a basic unit.
Again, the question is how do we go from single cells to networks?
[00:51:15] Speaker B: But then again, well, that's sort of a bummer for people who are like, you know, because I had romaine on recently, right. And some of the feedback I've gotten was like, well, yeah, like he sort of gave up on the brain because like he threw your hands in the air. I guess I'll study a single cell. And he admits as much. But however, it's like super exciting to him because it's finally like something that he feels like he can make progress on. Right. So you're faced with this overwhelming complexity of the brain and you really want to be able to sink your teeth into something and actually make progress on it. I think a lot of us feel like that we can't make progress on the big questions yet. Right. So. So sorry I interrupted.
[00:51:54] Speaker A: But yeah, no, I, I totally get that. I think it's.
I suppose it's just not something that I would. I like things where I can't see where the progress is because again, that's the kind of thing that maybe motivates someone who's not a scientist. It's confusing and messy and that's exciting.
But one thing that maybe is also a thing that I'm becoming a little bit interested in as well. There is where. There is another connection in terms of one of the things that we understand more and more about single cells is how they're regulated by their genome and the properties of these genetic regulatory networks.
And there We've moved. Well, biology has moved from a sort of code script type view of the genome to a very, very different view where you do conceptualize it as this network of mutual inhibition and excitation and it has all of these properties.
A particularly important property that I think is interesting is like this kind of degeneracy where you can have very, very different genetic regulatory networks, produce the same output.
And that's very similar to what you get in neural networks. But people have looked at how that, how that complex network in a cell is connected to what cells do and the function of a cell and the metabolic situation of a cell. And I think there are lessons there that you can apply to neural networks as well.
So that's a different.
I guess there's sort of three different angles of interesting story I can see, but that's one of them particularly focused on learning and evolvability. Like what enables biological systems to develop and evolve in the way that, which they do and how randomness and degeneracy power that in these kind of organisms I think could have interesting connections for the brain. I can say more about that. It's a bit of a tangent, but it's up to you.
[00:53:52] Speaker B: Yeah, yeah, do, do. Because, well, I was going to say like this, I don't know how different that is and I was going to ask you to clarify how different that is from the first of your three things where you reconceptualize the brain, right. As this like massive sort of turning itself over and always poised at this precarious state. Right.
So how is that different from.
[00:54:15] Speaker A: So I think one thing that the turnover in the brain story gets you is a way of identifying relevant functions and relevant levels of abstraction.
So if I'm interested in like knowing what, what is this particular, what is the level of abstraction that I need to look at the brain from to understand what it's supposed to be doing.
So one of the things an autonomy or self production type story gives you is the main thing it needs to do is rebuild itself.
So I look at. So I might think this particular variability is just noise. But when I look at the whole system and how that that contributes the processes that support the rebuilding of this system, I can see why that variability isn't noise, its function.
That's what I think you might get there. But what you get from the looking at how genetic regulatory networks work is.
So the thing I'm interested in at the moment is these ideas of neutral networks where you have systems that you can have all sorts of Variability with the same output.
And that what that does is it allows you to have populations that look the same for all intents and purposes, until something happens and that neutrality is broken. And that reveals this hidden variation that allows some of that population to pre adapt.
So you have cases where like a classic case is these fish stacks that live on the surface. And if you put them in cave water, you find that although all those fish would normally develop with fully formed eyes, some of them have implicitly developed these hidden mutations that support the loss of an eye.
So that something about that space for neutral variation and the inherent instability that causes you to move through that space allows you to develop these responses in advance of them being needed.
And I think you could probably tell a similar story about what goes on in brains as well, where they do tend to drift and vary in these ways that are consistent with preserving a stable behavioral output.
And that creates these puzzles about why are these structures changing?
Are they coding for something, Are they representing something, Are they secretly tracking something?
One story you can say is no, they're just inherently exploratory. So they have this tendency to fill out all of the different possibilities that are consistent with some constraint, such that when something changes, some of these populations, whether it's neurons or people, will be pre adapted to be ready for that change.
So that's a kind of neural level story that is already quite well developed at the level of gene regulatory networks. I think is one interesting possible that feels like at least a little bit more tractable as a thing to explore, if that makes sense.
[00:57:16] Speaker B: Yeah, for sure. I mean that's like a positive account of like what. So you know, traditionally neuroscientists like to get, like to get rid of noise because that's why you average over multiple trials, et cetera, et cetera, because you're actually seeing the true signal through the rest of the stuff, which is noise. But on this sort of, you know, account of, well, maybe the noise is actually useful in this neutral network kind of approach. The noise could be what we think of as noise. Well, maybe that's like an exploration of the possible state spaces that when perturbed, when the system is perturbed in a certain way, then you can engage that one like state space and that's like necessary maybe what we think of as the noise or these are actually important for the self generating autopoietic like nature of the, of the system.
But maybe it's also noise. Maybe 40% of it is noise and 30% of it is exploratory. You Know, at any given time, right. So you have, even within those fun stories which are, you know, could, you could write it up as exciting, you saw these boundary conditions of like. Well, you can't say that, like, unless you, you measure everything and run the simulations till the end of the universe. Like, you. How do you figure out the boundary conditions? Right. So I don't know. It's, it's, it's. There are daunting things. And now I feel like I'm just being negative, saying, well, you know, it's still going to be hard. Still going to be hard, Kate?
[00:58:47] Speaker A: Yeah, I mean, completely. And also a part of this is that you don't know in advance, like, which of those distributed possibilities is the one that's going to become useful because it's defined by. The only way in which they're selected is by being neutral, by not having any particular effect.
[00:59:03] Speaker B: Well, this is like where it's a relational thing with, with the environment and being embodied, et cetera. Right.
[00:59:08] Speaker A: And so, like, the only way to sort of tell in advance which particular variation is going to be the one that's going to produce this useful novel adaptation in the future. You can't do that just by looking at the organism. You'd have to look at the entire interaction with its whole environment. And then that does give you a story where there is a limit to what is going to be predictable.
[00:59:28] Speaker B: Yeah.
Oh, I think that might be kind of a key, right. Because like, so, so science, you need to predict things, to call it science, essentially, you need to be able to make predictions. But one of the most beautiful inherent things about biological organisms is that they're inherently unpredictable.
So if you did predict it, you would be losing something of what we think is special about it. What. That's quite the paradox, though. But that means I can't ever do science on the thing, right?
[00:59:57] Speaker A: No, I mean, you can do science on bits of them, right? You just can't do a science.
But I think that, I don't know, that's it. It's a personality thing. But that to me is very appealing, that there are limits, that they, that they, they'll always be this possibility for surprise.
To me, that's a positive thing.
And like, I think that should, I don't know, to me, that's as much as you should want, right? You should want to be able to do science on bits and pieces of the world and tame parts of it and like, locally tame parts of it. But the idea of taming the whole thing is incredibly depressing, really depressing.
[01:00:30] Speaker B: Yeah, yeah, it's a deterministic sort of outlook. Right.
Well, I mean, not, not to bring up AI like too soon or whatever, but I mean, do you see, is this where we should talk about how AI does not, you know, is this a bright line between biological agency and autonomy and anything that we could create in LLMs currently? But, but AI neural networks in general?
[01:00:56] Speaker A: I mean. Yeah, for me it is.
I think probably one of the reasons we confuse them is there's a, you know, from your individual, from your own perspective, it's hard to tell if something is genuinely predictable or unpredictable due to your cognitive limitations.
And because these systems look so unpredictable relative to our ability to comprehend them, we think of them as genuinely surprising and creative.
And then this is where you get into like this space of conceptual arguments again.
But it's not hard to tell a story about why that is just a cognitive limitation because they are bounded systems. They have very fixed constraints that they can of the possibilities that they can go within.
So you can list in advance the different ways in which this particular model can respond to this problem, and then you can change that higher order fixed function, but that's you changing it. The system itself can't change its objective function.
And that's why, even though they seem unpredictable to us, they don't have the kind of unpredictability that organisms have.
And like, I think for some people, it doesn't matter if it's unpredictable. To me, it doesn't matter why.
For me, the genuine, in principle, unpredictability gives you very important differences and reasons to care about that kind of system in a way that you just don't care as much about. Unpredictable.
[01:02:18] Speaker B: To me, yeah, I think that word care goes a long way toward what I have been experiencing. So in my own head, I don't care about an AI system. And I'll have conversations with, you know, colleagues or people who are working in AI and who just sort of assume we're going to have AI welfare, you know, soon. And that sounds crazy to me, but that hammers home the point that, you know, okay, we need to actually clarify why it's okay to turn this system off, why it doesn't feel anything, you know, and it feels weird to have to do this in a somewhat convincing way because it's so intuitive to me.
But I suppose we do have to do it in a convincing way.
So you're not worried about AI being conscious?
[01:03:03] Speaker A: No, I mean, it's one of those intuitions that just doesn't hit me like, as you say, it's sort of. It's a very strange thing to talk across because it just affect. The idea, just affects people so differently.
I mean, I can see. The only way I understand it is I can totally see that there are signs of a conscious intelligence in the outputs of an AI, because there are. Because it's trained on the outputs of a conscious intelligence. Like, there is so much that you could not have without living systems, but that's language. Like, language contains that structure.
So when you think of that as
[01:03:38] Speaker B: the result, lo and behold, yeah, we're like, impressed that it's doing what we told it to do.
[01:03:43] Speaker A: Yeah. And I find that, like, I think if you have a view of language maybe where you don't think of it as being as structured as it is, then you're like, how can it get all of this information from individual words that refer to objects?
But when you already think of language as like a very structured, meaningful space, then you're just copying what is hidden in that structured space.
Yeah. Although. Something I wanted to say.
[01:04:08] Speaker B: Yeah, yeah. No, go ahead. I don't. I'm interrupting you too much already.
[01:04:13] Speaker A: Oh, well, I was just gonna say, I think the other aspect to. Why the thing that stands out most to me is an argument that you could make to somebody who's not concerned about.
I think consciousness is such a frustrating space to have the discussion in because consciousness is such a mess.
And I think at least life is less of a mess. I actually think life is like, what is it to be alive is becoming relatively tractable in a way that what is it to be conscious isn't.
And one of the things you also get there is when you focus on the precarity of living systems and how all of their parts depend on each other, you get a story about where timing is really important in the sense that you can't build part of it and then build another part and then build another part, which you can do with a machine.
And for that reason, you get a very. A good reason to think that the idea that you could build an organism is almost impossible. You can only start with a protocell and then wait billions of years because getting one of those processes synchronized up is pretty much impossible.
Even getting the basic processes of a single cell synced up is pretty much impossible.
So there's something inherently unreplicable about a living organism that isn't about an artificial intelligence machine. Like, I don't know, a model on a computer chip.
And I don't know, that's one thing to me, that at least that's, I think that's the argument for me, that's the clearest that I can give to somebody who cares about an artificial intelligence and doesn't, and thinks it's as important as a living system is there is a real difference in what you can replace and what you can't replace here.
[01:05:59] Speaker B: Yeah, I mean, like many neuroscientists, I got into neuroscience because of my interest in consciousness and trying to understand how a brain could generate consciousness, et cetera. And now I'm finding myself, you know, questioning. Well, I don't. I feel like society like overvalues intelligence and undervalues life.
Is maybe sort of my current conclusion about this. Like for someone to like be worried about AI welfare and, and then, and then not express in the same breath like, oh, I don't know, starving humans or something like that, you know, like it's, there's so obvious like a difference to me that it's like I find myself in a very confused state.
And part of this is how definitions change over time, you know, like what we consider what is intelligence and consciousness, etc. But I guess what I am getting at and one of the reasons why I'm interested in this bio in activist account starting from these closure, constraint closures, metabolic processes. Can you, can I see like a thread, you know, can I eventually explain is there explanatory purchase using that thread to mind, right, to consciousness and mind. Because I see current AI and it's like a full stop for me. Like, well, that is awesome. It's an awesome tool like everyone says. But it just has no bearing on how I think about mine because of things like timing and precariousness and things that I actually value. But I'm not 100% certain that I can tell the story from that, you know, where meaning is supposed to come from because we have to stay alive. Right. And that, that's the promissory note of this approach is like we'll find meaning there and maybe eventually mind. Right? So yeah, I didn't even ask you a question there. I was just rambling on, I suppose. But I guess the question is do you see a path like toward, you know, because you mentioned phenomenology earlier and I know you're interested in like the phenomenological approach from Husserl on, you know, and those.
And before is this approach and part of like this. The subtitle of your book is like Free Energy Principle and the Meaning of Life. Right. So do you see us having eventually a good explanation of Meaning and then mind and phenomenology based on this approach.
[01:08:28] Speaker A: Yeah, it's. I mean, I think the first thing to say is it's going to be a very different concept of mind and meaning than a non phenomenological one.
And that's possibly where a lot of the sense in which. How on earth does this cellular thing you're interested in here relate to this thing of mind and meaning and intelligence? Because I don't think it does relate or it doesn't go direct to the kind of mind, meaning and intelligence that is sort of standard, I guess, in analytic philosophy, and at least I'd love to.
[01:09:00] Speaker B: What is that?
[01:09:01] Speaker A: Is that like detached or detached representational, that kind of story? I think you get that back.
But in a phenomenological story, that's a social phenomenon.
That's something that you get when you have.
That's a product of intersubjectivity and things that have their basic intentionality and have to negotiate across it.
It's not the basic. It's almost a reverse direction of explanatory priority, I guess. So it's not that we as individuals try to represent the world and then once we've represented it and thought about it, we talk to other people about it.
It's ra. We as individuals are trying to survive in the world and then we have to interact with others whose ways of surviving become interdependent with our own and also in conflict with our own. And in having to negotiate that, you start to build up the idea of something beyond yourself and a reality beyond your own needs that looks more like representations, but that becomes that sort of detached, intelligent centered viewpoint, becomes this very much later stage achievement that you don't get from basic biological autonomy at all. It becomes like this sort of social intersubjective phenomenon. So I think you can get there, but you get to a very different version of that picture.
[01:10:12] Speaker B: I'm okay with that. You're okay with that?
[01:10:14] Speaker A: Yeah. Possibly. That's. Yeah.
But what.
[01:10:17] Speaker B: What is. Okay, so, you know, we've been talking a lot about how like the agency talk and biological autonomy talk that is part of this constraint closure. Your story, it's so much of it is at the single cell level. And we've been talking about how these are processes that are definitional of life and. Or necessary for life or whatever to stay alive. Basically. Where do brains come in? Why do we even need them in that story? Why should I continue doing neuroscience? Because I can explain a lot without brains.
Where do you see brains coming in?
Yeah.
[01:10:55] Speaker A: And I. This is why I don't have a great answer yet.
I think it does.
The story that I would be interested. The kind of way I see the story going is having systems that are sort of. Once you have systems that are oblique obligately collective, so you can have. Have single celled organisms that can come together and act as a unit and then divide. So you know, like a slime mold can split off and go two ways around an obstacle.
I can't. So I am stuck together as one thing.
When you have that kind of system, then it needs a sort of.
There's a kind of higher level unity that is required that isn't for something like a slime mold where all of the parts have to be coordinated to work together at every point point.
And that. That requires more sophisticated. I don't like to think of it as like regulatory apparatus because I don't like to think of the brain as being a regulatory system on top of the metabolic system.
[01:12:00] Speaker B: I think you just said coordination. Is that a better word?
[01:12:06] Speaker A: You need a system, one of these precarious systems such that when parts of it break down, they automatically trigger compensatory responses. I think so it's not that there's something on top that manages it, but that the whole system is such that these breakdowns are connected to each other in ways that they trigger responses elsewhere. Which is. Comes back to the kind of degeneracy story that I'm interested in.
And I think you get like this massively degenerate possibility space with a brain where you have tons and tons of different ways in which the same thing could be achieved. And that gives you a kind of adaptability for how to keep the whole together.
Because you can keep that whole system together in so many different ways that look the same but have a wide variety of different internal configurations by which it's achieved. I don't know. That's very vague.
[01:12:54] Speaker B: Yeah, well, I mean, I have it in my head. I'm not sure what kind of job it did to the public. Right. But yeah, I mean again, it goes back to like, well, this is like a very daunting endeavor. And maybe it is something, I guess I'm wondering what am I going to be satisfied with. Right. So like what you just said in some sense is satisfying. But while you're saying it, I'm thinking, well, how could I perturb it in a way that I could make any predictions about, like what? Which little part in the space of possible compensatory reactions where the system's going to go like. And so maybe I don't need, maybe eventually I need to let go and thinking about it in those ways, right? And what I want is more like a law, like description, right, where there are these overarching principles that I can be comfortable with. It reminds me of like Conrad Kerning and I'm forgetting the other authors right now, but in studying like, you know, these massive neural networks, saying, okay, well, so maybe we're not going to be able to like probe the units and understand the units. And maybe our understanding needs to be at the level of what the algorithm is, what the architecture is and like the learning rules, you know, or whatever. And that should be enough, right? Because. But it, that feels unsatisfactory. So I guess my, my desire is to come to something like what you were saying, but have it feel satisfactory. And I don't know if that's a great margin.
[01:14:20] Speaker A: I mean, maybe one of. I, maybe I see it slightly differently than from this, offering a different story at which you have understanding because I not somebody who thinks that we shouldn't use computational explanations or dynamical explanations.
So it's less this is going to be an alternate, this is going to be an alternative way of getting the same understanding and more that this gets clearer on the limitations of that particular model and when it works, when it doesn't and why it works.
So you can still describe, and I think the noise, when we talked about noise before, that's a good, that's one good example of how this can be helpful, I think in the sense that when you're trying to you model the particular, regular dynamics of this particular process and you're identifying what counts as regular, you could find patterns in the noise or you could not find patterns in the noise. That's sort of up to you, depending on how you describe it.
But if you have a story where you understand, if you're at least thinking about that process in terms of which bits of it are relevant, are the bits that the system that produces that pattern of activity depends on, on, then that gives you at least a way of thinking about that signal noise question. That's not. I choose to focus on this signal or I choose to focus on that, but more that's the reason that I think this is genuinely noise, because it's not needed. You know, if that noise was stripped out, the system would still be able to. If that variability was stripped out, the system would still be able to sustain itself, whereas this variability actually underpins the very process of it sustaining itself. And I think that is helping.
[01:15:56] Speaker B: But yeah, yeah, it's a matter of finding. As a scientist, I'm tasked with finding how to measure which variability that is and what, you know. Now I need an ontology of different variabilities.
[01:16:09] Speaker A: I mean, one thing I should say as well is I think a lot of the time maybe that's not a problem because we're sort of implicitly relying on a vague notion of what a living system is.
So it's not like people are always like, like I have literally no way of telling what is useful variability and what is just noise. I think we're relying on these sort of base, like these implicit ideas of what that system needs because we do understand it as a living system.
It's just making more precise what that judgment is based in.
And you know, there are examples of things like motor variability where in some cases being too variable in your movement is bad for you. In other cases it can like, protect your joints. I think that's a nice example where you see how this sort of story can speak to that question. Once you understand joints as being these precarious things that need variability to prevent them from decaying, not sorry, to prevent them from being damaged, that's maybe one of the simplest places I can see this story formalizing a judgment you'd already be making.
But what is that based in?
[01:17:13] Speaker B: I suppose, yeah, I think maybe one of the interesting turns here is that. But we're going from these living processes, right?
Neuroscience continues to neuroscience writ large, right? Air quotes with whatever neuroscience is. I mean, the great victory is, okay, we can ignore the processes that are based on living. Right? That's not the interesting thing. The cognition is the computations and the algorithms and, and what you're asking us to do. I know you're not asking, you know, but what this kind of view is asking neuroscience to do is, is to actually turn and hug the living processes. That's actually where the action is, you know, and that's so foreign. But I'm in a neuroscience lab and I'm recording brain activity and relating it to behaviors.
And the question everyone asks is, well, okay, well, so what's the mechanism of like how the brain does that behavior? And there's that word mechanism. So I'm in a weird spot where, like, I'm trying to divorce myself from that kind of speak. But you know, because I see, because it's very computationalism, functionalism based and brain is a machine, organisms are machine based. And so I want to divorce myself from that vernacular and even that conception.
But it's so hard to even formulate the question scientifically like from my recordings and the behavior and all of the data that I have to even start to begin to tell that story within the constraint closure world is difficult. And so that's like a big ask.
And maybe I'm just in the wrong place in the wrong career, right?
[01:18:59] Speaker A: I mean, maybe then the sort of approach I would want to argue for is less challenging and less, less difficult than it sounds. Because I think what I'm more interested in is a story of why you can talk about things as mechanisms.
So they're not machines in the same way that pendulums are machines. But once you understand how that mechanical like phenomenon is sustained, then you can talk about it as a machine like phenomenon. It's just underpinned and it's not defined mechanically in the way a pendulum is. It's contingent that it's mechanical. It depends upon the living process.
And insofar as the living, the process of life is this kind of stable thing that you understand well enough, sometimes you can ignore it because you it's implicitly in the background of how you think about that mechanism.
So I think a part of this is not saying we shouldn't talk about the mechanism for this or the mechanism for that. If we're careful by what we mean with mechanism, it's understanding that when we're making those judgments about mechanism, we're basing that on this implicit understanding of a living system that is not there in a pendulum or in an artificial intelligence.
And there's a nice quote just from Hegel because I like that where he talks about teleology is the truth of mechanism. So that whenever we're talking about mechanisms, we're always relying on an implicit sense of purpose. And that's what the story of living systems gives you.
And you can background that sometimes, but just the danger of backgrounding it too much is you forget that your whole account depends on it. And then you look at an artificial system. And so this is the same thing because you've forgotten what's in the background.
[01:20:39] Speaker B: Ah, okay.
So eventually we'll be so implicit in the background of our texts, of our papers is going to be this story. And so that eases the burden of retelling the story every time because we can go on carrying on and using terms like mechanisms, but we have to understand them in a different way. That's the idea. Yeah.
[01:20:57] Speaker A: And when you have, and it's nice because it means that when you butt up against a problem where somebody, why this scientist says the Mechanism is this. This scientist says the mechanism is that you're not stuck with completing intuitions. You can go back and look at what that decision is based upon and see which one actually integrates better with this background assumption that's underpinning all of our talking mechanism.
[01:21:19] Speaker B: We're in a. Potentially it is. And we're at an opportunity to reframe right now if we took that opportunity, because we're getting so much data, there's so much compute, we can do so much more. You know, there's this turn toward more naturalistic kinds of experiments where we're embracing the variability, recording all of the variability. Right. And so there's this potential opportunity, not maybe for a paradigm shift in the Kuhnian sense, but for a reframing of the questions toward these things. And I kind of of the opposite of the alternative is to carry on as usual. And I can't figure out what we're doing. Like, there is sort of a turn happening, but it's still in the same kinds of language and probably with the same kind of background assumptions. So I'm kind of curious also what the field is going to look like in like 10 years. Right.
So I worry. I don't want to miss the opportunity. If reframing is the right way to do it, I don't want to miss that.
[01:22:18] Speaker A: So, yeah, I guess it's interesting to hear from you how you see it.
But that's. I think that does.
[01:22:25] Speaker B: But I'm already on your side. Like, I'm already on. Like, we're on the same team basically. Right.
[01:22:29] Speaker A: I think maybe the thing is I don't want to be. I think there's a tendency of people coming in and going from the outside. This field is doing things this way and they should be doing it this way. It's like, right, you have another field. Like, who are you to say that? You get that a lot with, like, debates and evolutionary theory.
[01:22:42] Speaker B: Right.
[01:22:43] Speaker A: Like, traditional selectionists think this way and like, no, they didn't. And it gets people's backup.
So it's interesting that you see it that way. And I definitely do see what I think of as like an almost a sort of taming approach where you take some finding and you go, oh, yes, this is a homeostatic feedback mechanism or something. And it's like, sure, you can describe it that way, but that's probably not the best way to describe it. Like, lots of people who are doing work that is really close to how I would. Would want to describe things, but not quite because I think the way I would want to describe things does involve giving up a degree of predictability and control and accepting that the system, these mechanical models that you will make or these dynamical systems models that you'll make are themselves vulnerable. And you don't have the same guarantees that you do with ordinary physics type modeling.
And it's interesting seeing people talk about all these results that like push against that and then reconceptualizing them in a way that doesn't really isn't quite as threatening.
[01:23:51] Speaker B: I worry or think, I don't know, judge, that maybe the main culprit, from my own judgment is this reification of the model concepts that we use in, in neuroscience. Especially for some reason neuroscience seems to be vulnerable to it, where once we start using a model, like a computational model, then then the computation starts to become the real thing and, and we forget that we're actually modeling it. We're, we're finding out the real thing. The map is not the territory. But I think that error occurs over and over and over in neuroscience, maybe because it's easy to do, right? Because you want to be able to say like, if I solve the figure out what algorithm it's running, I've solved it and it's running the algorithm. You don't want to be able to, you don't want to say like, well, the space of possible algorithms, if you want to call them algorithms, there's an invariance there. And that's actually, you know, what you need to account for. It's not like a single algorithm. There's, you know, an invariant structure of the algorithm. So I mean, I mean, do you agree with that? That could be like the main culprit is just the old fallacy of misplaced concreteness from Whitehead.
[01:25:11] Speaker A: Yeah, I think that is the culprit. The interesting thing is why that's particularly problematic for models of living systems in particular.
So even, you know, even for a pendulum or any other physical system, a model is going to be slightly wrong. It's going to abstract away from certain features of it, or models are wrong, some are useful.
But I think sometimes people say that line and then as though we've accepted that all models are wrong. So that's okay. We don't need to worry about the different ways in which they can be wrong and the particular way in which they're wrong. That something like the fallacy, misplaced concreteness capture quite well in living systems is, that is, and Whitehead's a good person for this is precisely their processual nature that they don't have that kind of fixity. And that's a different kind of wrongness from the sort of abstraction you get when you abstract away from friction and a pendulum.
[01:26:09] Speaker B: Oh, I see. Okay, well, let's say.
Okay, let's say neuroscience embraces this turn, or a portion of it does. Right. There's a lot of talk about perspectivalism and pluralism and the benefits of that in sciences. So instead of saying like, okay, well, neuroscience is doing it wrong and we need to think of that in a different way, could this just be just another slice ways to approach it and we can all live together and be pluralistic and happy?
[01:26:38] Speaker A: I think to a degree.
I think the only way in which we couldn't live together is if everyone doesn't agree on the provisionality of their particular description.
So when somebody says, look, book, the brain is literally the fundamental algorithm of cognition is Bayesian inference. And that's what cognition is, solved it, and therefore artificial systems can do it, then that's not going to be a lived together kind of story. But if it's like from my perspective describing it this way works, but I understand that that abstraction is based upon this kind of. More that's one possible abstraction of a process that is not abstract, that is inherently not an abstract thing, then yeah, you can argue over which abstraction is better and these become more.
They're not like totalizing, sort of.
It's not a totalizing setup in which one abstraction has to be better than the other. You can still argue, but it's about the particular situation and what you're trying to do, which is very.
[01:27:41] Speaker B: Yeah, so we've sort of successfully talked about some potential paths forward, which I don't know why I'm slightly surprised about this, because we haven't even talked about free energy principle and why you find it to be less than the solution. So I'm not sure if you want to spend much time talking about that, but maybe, yeah, I mean, if you don't mind even the broadest general summary. So we don't have to talk about the ins and outs of what the free energy principle is because people don't understand that anyway. Like even people who use it and study it. But there's a lot of, of excitement about active inference and the free energy principle. And I feel bad because I keep getting asked like, you need to have some active inference people on.
And now I'm having someone on who's critical of active inference before I have the people who espouse it on.
So you spend a lot of time in your book. And it's probably good that we didn't go through a bunch of the reasons why you find free energy principle falling short, although very useful. Right. In predictive processing, et cetera. So I'll leave people to get into the nuts and bolts, like, I'll point people to the book for that. But in a broad sense, with the background that we've been discussing about these issues and how to move forward, what is it about free energy principle that you find lacking and active inference?
[01:29:09] Speaker A: Yeah, I think what the free energy principle looks like here is these general issues that we've talked about are really nicely crystallized in the free energy principle.
So it's a very.
It's defended in different ways, but at least at one point it was sort of like the clearest and most sort of strident statement of exactly what I would be arguing against in terms of the idea that life is about maintaining stability, It's a regulatory process. It can be described in terms of minimizing error and.
And it doesn't. You can abstract away from these kind of metabolic, energetic questions.
So in one way, it's like, I think the role of the French principle, at least in my life, is like a really clear crystallization of a lot of ideas that people generally have in the background and making them an explicit claim as a theory of everything, which is actually great because if you take people's background assumptions and turn them into a theory of everything, then you can criticize those background assumptions in a way that you can't when they're implicit. So it's very useful in that respect.
[01:30:15] Speaker B: The other way, it's like, thank you so much for doing your science so well that I can criticize it. Yeah, yeah.
[01:30:20] Speaker A: It's like very explicit at various points. And I really like that there's a lot of quite definitive quotes where people state exactly what they mean by it. I mean, so people complain that the theory, the way in which it's developed, changes a lot, and it absolutely does. But at least at each point in its development, I can find a clear statement of what. That.
What people were saying at that point.
And you also get the.
This sort of retroactive approach I talked about where you say, oh, whatever happens, we can always retroactively describe that in terms of some invariant principle, which is the sort of more recent development.
[01:30:58] Speaker B: And again, sorry, does that mean like making a just so story about how some phenomenon was. We can tie that to minimizing free energy or something?
[01:31:07] Speaker A: Yeah, it's essentially just so story because if you have A flexible enough mathematical formalism, then you can describe anything under it. But what's the point?
I mean, so not what's the point, but you know, like it doesn't tell you in itself anything new. It allows you to describe it in more tractable ways, perhaps, but.
So that's what the free energy principle sort of has represented for me. And where it is useful is that I like to focus on things like unprecedability and transformability of learning systems.
But there is clearly a very important regulatory aspect to life as well. Like certain things need to be kept stable enough for you to be able to transform.
And if you are trying to describe process of maintaining stability, the fact that you can use an inferential formalism to do that, which is what active inference does, it gives you a way of redescribing control in terms of inference, then that's super handy for connecting all of these high level inferential processes to more basic control processes.
So I think there's lots of interesting and cool stuff that is done under the sort of active inference formalism that I have no issue with at all.
It's just that I don't think it's a foundational sort of story of what cognition is.
[01:32:31] Speaker B: We talked about consciousness and mind and the prospects of relating this bio and activist account to consciousness and mind. And you've discussed even in our discussion here that it's the invariants that need to be paid attention to. We don't want to average away what we consider noise, et cetera, but it strikes me as like, how do you think about the fact that I might be in a different mood based on whether I'm hungry, but generally I can still do mathematics or something, Whether I'm hungry or I'm ill, I might not be able to do it as well, but I also still feel like the same person, right? My identity, there's a unitary flowing of my personality, my personhood, my subjective experience.
And it seems. Seems odd to have that unified subjective experience in light of the vastly var. All of the variance that's taking place at the biological level that we want to use to build up to tell this story. Right. So is there a paradox there?
[01:33:48] Speaker A: Not so much a paradox, but like, I think a point that this kind of story about what it is to be a particular biological system needs to integrate both of those sides of it in terms of like how, like how this inherent transformability of living systems actually often does tend to not be used.
So they are constantly capable of transforming and varying in all of these ways. But most of the time they don't. Right. Like, most of the time we are pretty stable.
And I think the story there gets into the.
I think the interesting thing is that single celled organisms can often be much more transformable than multicellular ones.
So because once you have a multicellular system, you have all of these systems regulating each other, because if you change, that's going to cause problems for me.
Whereas in single celled organisms, you might be like changing your genetic material all over the place.
You can have a lot more flexibility because you're not constrained by the whole
[01:34:55] Speaker B: that you're part of and your partners aren't depending on you. You're just trying to get along in the world.
[01:34:59] Speaker A: Yeah, yeah, yeah. But once you have this reciprocal interdependence that actually enforces a kind of stability. So all of these parts are inherently volatile and liable to change, but put lots of volatile, liable to chains, parts together and they start to stabilize each other.
And there's a nice sort of.
So you end up in a very similar place. You end up in a place where you have a mechanism that largely keeps itself stable, but it's built upon the suppression of intrinsic volatility.
[01:35:29] Speaker B: Oh, so is mind like this dynamic singularity of unification of like total constraint, mutual interdependency. I'm just throwing out word salad now.
[01:35:39] Speaker A: Yeah.
But I mean, like, there's really nice work from like some people in the. To Mayo Montville, one of the constraint closure people with like Ana Soto, Carlos Son and Sheen and Giuseppe Longo, where they talk about approaching multicellular life from this perspective and how that gives you a different lens on things like oncogenesis and cancer as being not a failure of an individual cell, but a breakdown of the collective constraints of the sort of whole system. And there's like good empirical. There's some good empirical evidence for that sort of story in the sense that you can. If you put a cancerous cell in healthy tissue, in certain cases, it can suppress the cancer, and if you damage tissue, it can be an initiator of cancer. So you can see it as this tissue level phenomenon is what they're arguing for.
That's a whole extraordinarily complicated debate that I don't have anywhere enough knowledge to have a position on. But I think it's. So it's like one different way of seeing how this framing does actually change the way you think about something when you think of stability as a collective achievement rather than something that you begin with.
[01:36:52] Speaker B: So I can blame my depression on my family, then, right? Is that what you're, that's what you're saying?
[01:36:58] Speaker A: I mean, there is like a connection to these sort of social stories, right? Like, I think of it as multicellularity, but it totally does, does extend up to, like, as organisms in a social system, we also constrain each other in ways that can be beneficial or not for the individual.
[01:37:16] Speaker B: Okay. I realize we're coming up on time here and I don't know what we did and what we didn't do, but. But I, I do want to ask you, you, you've hinted a little bit at what you're, what you're interested in these days. So what have we not discussed that, that you're working on now that you're excited about about?
[01:37:32] Speaker A: The other thing that I'm excited about is like a very different chain of thought. So the two things I'm doing now, one is this looking at the importance of neutral networks in evolvability and how they are there at all levels of biological organization.
Because that gives you a story for why.
What I want is a story for why tolerating as much variability as possible level should be the aim of a biological system rather than suppressing variability.
Because the more variability that you can tolerate, the more poised you are to adapt to a new situation.
[01:38:07] Speaker B: Is that related to capacity?
[01:38:10] Speaker A: Sorry?
[01:38:11] Speaker B: I think sometimes I think that the job of the brain is to just increase the capacity of possible state space, essentially the capacity of the state space. Is that a related notion?
[01:38:22] Speaker A: Yeah, I mean, that's definitely one way I think about this, right. Once you have that sort of architecture, you have all of this spare space for variation that is sustainable by you as a system, but not directed to any particular goal or problem.
[01:38:39] Speaker B: But it's horribly inefficient if that's what you did, is maximized your variation ability. So there has to be some sort of evolutionarily bottlenecking of the efficiency of it. Right.
[01:38:51] Speaker A: It has to be a constraint. But I think it doesn't have to be maximally efficient.
[01:38:54] Speaker B: So I think maximally. But you can't just run wild, right?
[01:38:58] Speaker A: Yeah, but I think when people think of efficiency and they think of biological systems as sort of optimal, then it looks like all of this variability that isn't doing anything useful is a failure of efficiency.
And whereas this framing gets you a story where it's not a failure, it's not a failure of efficiency. Exactly. Because efficiency, there's no. So we're not optimizing systems efficient for what?
[01:39:23] Speaker B: Right? So you could say you need this much capacity for variation to be the most efficient survivor given your niche and your society, blah blah, blah, blah. There's always a story you could tell about that.
[01:39:37] Speaker A: There's always a story you can tell. And I think the efficiency story doesn't capture what this variability is for.
It's more that it's. It's not that it's efficient, it's that it's tolerable. It's that you can continue to be viable and be this variable.
So you have this intrinsic tendency to vary and proliferate and spread out and be different.
And that's what subsequently can be co opted in a particular adaptation. But it's not driven by that. It's a sort of physical tendency of life to vary.
But because life has that physical tendency, that's what enables it to evolve in the way that it does.
So it's a. It's trying to look at variability through that lens.
And then the other project is stuff with.
I spoke at the beginning about my interest in autonomy and sort of this idea of error being relative to the individual but not subjective.
So looking at accounts and metaethics of autonomy and connecting them to this biological autonomy story, which is quite a far away thread, but I think think there are a lot of overlaps there.
[01:40:41] Speaker B: What is wilding?
[01:40:43] Speaker A: Wilding? Oh, in terms of like rewilding the web.
[01:40:46] Speaker B: Yeah. What is that?
[01:40:47] Speaker A: Yeah, that's my other project. I forgot about that.
So the other thing I'm. Yeah, the other project is which connect. This is where the variable, the interest in tolerating maximal variability comes from in the sense that there's lots of people arguing that it's a problem, that the Internet has become a monoculture in the sense that you have a few centralized organizations who control everything. So Amazon for cloud compute is crazy and Google for search and so on.
And that's very. Not in the original vision of the Internet where it was supposed to be this distributed system where you could have all these free individuals just harmoniously organizing without top down control.
Which I think is a really nice case study in why decentralization isn't automatically diversification.
So this project is kind of bringing together lots of different people. But what I'm interested in there is trying to figure out what does drive diversification in biological systems and why that seems to be absent in artificial ones and then applying that to questions like what would it even mean to rewild the Internet?
Oh, it's very blue sky, but yeah,
[01:41:59] Speaker B: yeah, very blue sky. I just had Yogi Yeager on. I haven't released the episode yet. But one of the things we were going to discuss, we were going to have William Wimsatt on with him because Yogi has written sort of a summary of Wimsatt's positions. Because William Wimsatt wrote this really long book Re Engineering Philosophy for Limited Beings. But it's notoriously dense and really long winding sentences one after another. It's almost. It reminds me of like Terence Deakin almost in.
In his. What is that anyway, his style of writing where it's like really difficult to follow through. So. So it's kind of ironic. So, you know, Yogi wrote this summary paper of a lot of his ideas which was very much needed saying I guess the title of it re Engineering Whimsat.
Yeah. But it made me. So reading works like yours, going back to your journalism work and how that has sort of honed your skills at communication and writing, like it's such a. A pleasure to read work like yours. It's really hard to do what, what you have done in the book. And so again, I just, I really appreciate it. I hope a lot of people check the book out and. Yeah. Anyway, thank you for, for being here with me today and nice, nice chatting with you.
[01:43:20] Speaker A: You. Yeah, nice to meet with you too. I'm sorry I'm not as clear in conversation as in writing.
[01:43:27] Speaker B: Yeah, yeah, no, no, it was, it was great. All right, I'm going to stop this real quick.
Brain Inspired is powered by the Transmitter, an online publication that aims to deliver useful information, insights and tools to build bridges across neuroscience and advance research. Visit thetransmitter.org to explore the latest neuroscience news and perspectives written by journalists and scientists. If you value Brain Inspired, support it through Patreon. To access full length episodes, join our Discord community and even influence who I invite to the podcast. Go to BrainInspired Co to learn more. The music you hear is a little slow jazzy blues performed by by my friend Kyle Donovan. Thank you for your support. See you next time.
