 Hello, everyone. Welcome to Act in Flab, guest stream number 19.1. It's March 25th, 2022. And we are here with Marco Facken, and we are going to be discussing the recent paper, Extended Predictive Minds. Do Marco Blankets Matter? So if you're watching live, please feel free to put comments or questions in the live chat, and hopefully we can address them. And Marco really appreciate you joining, looking forward to hearing about this paper and your lines of research. So it would be awesome to hear an introduction and take it away. Okay. Hi, everyone. So I'm Marco Facken. I actually am a PhD candidate in Italy. And the paper basically is an attempt to understand what was going on in the active inference literature with the externalism, internalism debate from a philosophical perspective. So I studied philosophy for the most part. I'm not super mathematical savvy, and I have basically a background in embodied extended cognition. And I stumbled upon the papers by whole year on set and other, and I wanted to understand what was going on with the Marco Blanket talk. And so I wrote a paper about it, to understand what was going on, and to try to suggest that maybe that much emphasis on Marco Blanket is a bit misplaced. And yes, that's it for a general introduction for the paper, I think. And then here for any question or any doubt. Okay, cool. So maybe I'll just ask some questions and we'll just see where it goes. For somebody who was unfamiliar with extended cognitive processes or the state of the art, maybe outside of active inference, just what is this area? How are people looking at it outside of the active inference, FEP angle? And like, what is the contemporary state of that area? So, okay. So the extended cognition stuff, say officially is something that spawned out a paper by Clark and Chalmers in the early 2000s, basically. And their original idea was, as I understand it at least this, that sometimes at least you interact with the environment in a way such that part of cognition happens in the environment itself. So my go-to example is using pen and paper to do math. When you use pen and paper to do math, you write down partial results and that scrap of paper is working as a piece of your walking memory. It's just that it's not inside your brain is around you on your desk or whenever you're doing math. And so that was basically the original idea. And from that, there has been a reach elaboration with people like Richard Minari suggesting that actually, we should think of extended cognitive processes as processes that transform our cognitive capacities is not just more stuff, more memory, more symbols, but new stuff, new capacities, new symbols, new cognitive processes that take place in the interaction between the subject and the environment. So we have that line of research. And then you have a more, say, an active line of research that tends to stress the idea that basically all cognition takes place in the world, interacting with affordances or interacting with your surrounding environment. But I'm not super expert on that line of research on the extended minor, a bit more conservative. So the punchline would be this, that sometimes stuff in the external world really is in the most literal sense possible a cog of the machinery that does your thinking. That is the clearest and the most personal way in which I'm able to put a DAD of the extended mind. And of course, sorry, once you have this idea on the table, the question is, okay, but where does it stop? Is pen and paper is in, but I don't know, are our cell phones in? Are our books in here at my left? You might not be able to see it, but there is a big library with a lot of books I read. Are all those books part of my memory? Probably not. But you need some sort of boundary. And that's you for Mark of Black, of course. So I'll leave it like this. Well, I like this, how you said it was like a cog, and it's kind of like cognition. So it's right there in the word, even though that's a little bit of a wordplay. And also very interesting about the more is different, like we're not just augmenting the computer with a bigger hard drive or more RAM, it truly can be different. And so I think maybe we could return to this idea of like, what is the space of the adjacent possible for cognitive systems, especially if they compose in these nonlinear ways. But could you maybe walk us through some of the pieces of the paper, like what brought you to the introduction, what brought you to writing this paper on Markov blankets? And then what did you sort of weave together in the introduction? How did you clash them in the middle? And then where did you leave us at the end? So did you walk through to the paper is something not super sure I'll be able to do satisfactory at least. So the first step was to what motivated me was to try and figure out what was going on between Andy Clark, Jacob Hoey, Ramstad, and Markov blankets, of course. And what I wanted to argue basically was something like this, that even if you set a site, DDD works by Jelle Brunneberg and others or Menarium and the alerts that tend to give you this sort of message like we have a perfectly fine mathematical concept of Markov blankets, and then a strange concept of reasoning blankets that is a bit mathematical and a bit ontological, and we're not really sure what it is. Even if you leave this sort of worries aside, what I wanted to say is that the way in which Markov blankets are used for when it comes to discussing the extent of mind stuff is very limited to the specific application, but for the extent of mind the usage of Markov blankets is not really useful, does nothing of particular value and tends to give a big headache to the people that are not into active inference. So I would recommend not using them. The main argument, I think there is no main argument at all. I think I have three arguments that are, well, first, in order to find a Markov blankets, it seems that you must already know what is inside. If you think about a graph, there is no way in which I can give you a graph and just ask you, okay, please show me where is the Markov blanket in there. I need to tell you about which know I'm interested in. So you must already know what is inside, and when it comes to the mind, the cognitive machinery, knowing what is inside is knowing which are the codes, which will presumably into the brain plus maybe stuff in the environment, pen and paper and whatever. And if you already know that, which you must already know in order to find the relevant Markov blankets, that intuitively you already know whether the mind extends or not. And if you already know whether the mind extends or not, you know whether the mind extends or not, there is no new piece of information to be gained by thinking about whether mind extends or not in terms of Markov blankets. That is the first argument. The second one, which is a bit messy. Could I just pause there and make a comment on the first one? Okay, awesome. So though the Markov blanket, if only it were like the blankets on our bed and you could pick it up and it was a blanket, whether it was inside the box or around the box or whatever. That's not really how they are defined. It's really like a co-partitioning of internal and external states with respect to a blanket or a blanket with respect to some other internal and external states. So the blanket states are the ones that make two other sets of states internally external, conditionally independent. And so it's not that a node can be even in a purely statistical sense. So without getting into the whole, what do they physically map onto? A given node is not like a blanket state or an internal state. It's a blanket state with respect to some other states because it's part of this partitioning that's co-instantiated at the same moment. So it's an excellent point where if you know what your system of interests internal states are, then you can talk about what states blanket them with respect to some other states. But if the question is about the extended minds and what are the internal states of minds, then there's kind of this catch-22 cat-in-the-hat scenario where if you knew where to even focus, you would have already sort of addressed the question upfront, just to sort of unpack that and connect it to some of the other ways we've talked about it. Yeah, that's a great unpacking. The keyword you said and that I forgot to say is the system you're interested in. Why I wanted to focus on this right now because the extended mind stuff is supposed to be hyper realistic. The idea is not that, I mean, we can't read pen and paper as if they were a form of external memory. That is something you can do, but it's not the claim that interests me, Clark, Menari. The claim is that the interest me, Clark, Menari is that pen and paper really are with the capital A of R, we are a piece of memory in the external war. They are memory in the same sentence in which parts of your neural activation or memory system are memory. That is the irrelevant claim. So there is nothing observer dependent or interest dependent is really a hardcore metaphysics about how the word actually is independently about what you know about it. So that's the point that I should probably have me sooner. Well, you made it in the paper with the direction of identification runs from target variables to Markov blank. It's not the other way around. So yeah. So that's the first one. The second one is very muddy and also very dependent on the how the details are on the extended mind and Markov Blanket develop. But the point is basically this that from the very beginning, that would be probably who is 2016 paper, the self evidencing brain, the D message, look, there are many possible blankets, and we have somehow to choose one. That sort of message get got through the literature and D way in which we and others try to single out the superspect of blankets among the many possible was to look at self evidencing in the long run about keeping your keeping yourself close to your set point in the long run. And their idea as I understood it was to say that well, the machinery of the mind is identical to the machinery that keeps yourself near your set point in the long run. And that sounds persuasive from a free energy principle perspective. But it's not super persuasive if you're not very into the free energy principle, because there is a lot of stuff that keep us alive and kicking in the long run, our liver, our guts, our kidneys, and prima facie none of those organs does thinking in an intelligible sense. And so did you worry here that probably is never actually stated in the paper is that we're confusing cognizing in some broad and fuzzy sense with staying alive. I mean, I have nothing against the claim that living things tend to cognize or tend to behave intelligently. Plants do behave intelligently, even if very slowly says behave intelligently. But I'm not super sure about equating the two. And that would be probably my second point. Okay, let's stay on this because that's also a great point. So if not one in the same as staying alive, what other perspectives on cognition might there be? Like what is thinking other than staying alive? Or what is it that, for example, the neurons in the glia are doing that the cartilage in the knee is not doing? Prima facie processing information would be a different answer if you take DNA or at least censoring information. So that's the question I'm struggling with right now. And my two cents would be this, that we have a big problem because not because we don't have a definition of cognition, but because we have too many definition of cognition. We have too many research programs telling us what cognition is. So you have the classic cognitive science, what is cognition? Well, symbol manipulation. Then you go with the connection is what is cognition? Well, vector transformation, subsymbolic computation. Then you have the inactivists, what is cognition? Sense making, slash autonomy, geological psychologists, what is cognition? Well, interacting skillfully with affordances and so on and so forth. And my big problem is that no one here is clearly wrong. No one is obviously dead wrong and that thing clearly cannot be cognition. And so what I think I would suggest is a sort of pluralism about cognition. Cognition is not a single thing. It's not like dancing. What is dancing? Well, moving your arms. Yes, sometimes, but sometimes dancing might even be staying still in a certain piece of art. There might be some regimented staying still as a part of dancing. That would be the analogy. So something that gets done in a variety of ways, none of which seems to apply to kidneys, levers, clothes, and that sort of thing. If I could layer one more sort of and or there's the experience and the phenomenology of us as ourselves and as people and then there's sort of the behaviorism or at the very least animal behavioral sciences or just behavioral studies in general, which is like from the outside looking at a system. And so there's this gap between like us doing the thinking about cognition and the experience of that thinking about cognition and then the question about recognizing cognition in other people, in children, in nonhumans, etc. And so like I really like how you lay out like there's the symbol, the subsymbolic and active and then no one disagrees. No one is saying that bacteria make bad decisions on the whole. They're here, aren't they? And no one's saying like a tree doesn't have a metabolic coupling with a fungus and no one's saying that the fungus doesn't diffuse signals, but then the aspects of these systems that people focus their regime of attention on and where they choose to start their talks and what they decide to talk about and what experiments they design that frames the discussion about cognition. That's a wonderful way to put it. And it's sad because the paper about cognition has already been submitted if I hadn't submitted it already, I would have stolen the framing part because it's wonderfully sad. That said, so there are two things I'd like to say about that. So the first one is no one really disagrees. I'm not sure. I think that there is some genuine disagreement. So the hardcore ecological psychologist, I believe and feel free to correct me if there are any ecological psychologists really think that cognition is interacting with affordances food stock and that there is no symbol manipulation going on in the background. And I think classic cognitiveists, I think really think that cognizing justice manipulating symbols in the background. So there are genuine disagreements. What I'm not super sure about is whether those disagreements are solvable in the sense that you can provide a master argument or even master experiments from which you can just read off the ecological psychologist was wrong. Oh, the classic cognitive is right. I think that the best we can do is accept some form of plurality about what cognition is. And perhaps, but I'm not super sure we might have to accept that there may be things that are neither determinately cognitive, neither determinately not cognitive. I mean, plans do behave, do cope with their environment well. I'm not entirely comfortable about saying that there cognizers in the exact same sense in which I am a cognizer. Maybe it's just because their behavior is very slow. If compared to mine, maybe it's because I have certain core cultural prejudices, which I mean, they exist. And calling something a cognizer is influenced by your prejudices. You go back to the early analytic philosopher, pretty much no one, both animals think and now is unbelievable that someone for someone to have that position so. So there is that, but it's not super thought out. But the big message would probably be this put simply, there might be something, something of a gray area between thinking with symbols, surely cognitive and being a rock, really not cognitive, the sort of big gray area in which you have intelligence coping with the environment, coping with the environment well, adaptability, flexibility, but still not be really sure about whether there is some cognition with the capital C going on in there. And that's the first thing. Then you also mentioned probably about consciousness. I'm very scared of all this stuff. Okay, so another interesting piece as evidenced by the, think about what to say carefully, don't want to say that it's a lane we could veer off into, but that is a more purely philosophical discussion, whereas neurobehavioral studies and cognitive science frames itself as science. And in science, usually we think about like unique explanations, predictions, the ability to design and control and perturb systems. So is this science? What kind of science? What are we looking for here in a cognitive model? And how do we resolve or accommodate this tension between like unique explanations and predictions, which is the bedrock of one type of science with the call for pluralism that you just made? I think that the answer here is do not. In the sense that the call for pluralism is not a very, not a call. Cognitive science is splintered. You have everyone doing its own thing in its own lab. There is cooperation in the sense that, I mean, the various research groups are not entirely isolated. It is not the case that people working on neural neurons know nothing about connectionism or classical computationalism or whatever. There is some communication. But on the face of it, what I say as a philosopher looking at science is a splintering of the cognitive science enterprise in a variety of groups, each with its own peculiar research tradition. And I think that that is basically fine. If no one has a strong call or a strong argument or a strong model that does what Newton did for physics and unifies everything under a single banner, plurality is fine. So there is no call. It's more something like, well, that's reality. Let's try to like it. I'm on board with that. But with this banner, what shape would it be? Or how would we know it when we saw it? I have no clue. We probably just would just know it when we saw it in the sense that we would see a research tradition starting to absorb all others. We would probably see people trying to demonstrate us that there is a real and significant equivalence between our powerful research tradition, uniform research tradition, and all the others. And probably the winning research tradition, the one that really got cognition, really got thinking, would be able to make some novel prediction that no one else was able to make. And that would be probably a strong marker that's the way to go. If the implicit question is, do I think that the free energy principle is that sort of research tradition? I'm not sure. And my answer would probably be a qualified no. Because we do not have right now the DDT proof that is unifying as it wants to be. It makes big, unifying claims and it might be good in the sense that they may for circle operation at the very least have at least that pragmatic advantage. Whether it's succeeding unifying as it wants to unify is still an open question. And so the appropriate thing to do is to, for me at least, to wait and see. For others perhaps is to pick up your tools and try to unify and show that it really is your right thing or just argue against even negative arguments are useful. Nonaction is an action. That's true in world knowledge traditions and also in Bayesian statistics. Like it can be useful to wait to gather more information. I wanted to return to what you said earlier about you need some boundaries. And a lot of this discussion, of course, it's like extended. How extended? What is extending? And what are we looking for with a boundary model in the FEP and active world? Certainly we talk a lot about different kinds of blankets and Bayesian graphs and things like that. But those are not the only ways to think about boundaries. So just wanted to learn about what are some other ways of thinking about boundaries, cognitive boundaries of different kinds and what are you looking for in a boundary model? Not super sure about how to answer. So my understanding is that Markov blankets enter in the debate as a sort of tool to understand what the boundaries of the mind are. But there are no other big alternative if not for something said by Chalmers about action and perception, the intuitive boundaries of the mind. But what I think he had in mind when he said that was something like, look, our intuitive image of the mind is that of something that is freezed between perception and action. Something like we are intuitively pretty supposed to say that something like the classical cognitive sandwiches truck, we encounter world perception and action. And I'm not really sure whether that sort of boundary would be a boundary in the sense in which Markov blankets would like to be boundaries or are used to to play the role of a boundary in the sense of something that individuates a single system and tells it apart from us. The problem I think for the extended mind and the need for a boundary is a need fueled by the strong desire not to say that everything counts as a cog in the cognitive machinery. Why? Well, because if everything counts as a cognitive machinery, then the claim that pen and paper used to do math counts is not a big claim. It doesn't add much, right? In order to for a claim, well, X is part of the cognitive system to be meaningful to being packed full. There must be things that are not part of the cognitive system. Otherwise, I mean, you are just cleaning up. There is this which is exactly identical to any other thing. Not super important. In the paper, you you were threading together this question of short and long term features of cognition, like in the instantaneous action perception, cognitive behavioral timescale, and then the quote in the long run, where a lot of the arguments rest like it reduces surprise in the long run or on average in the long run. And then you talked about like jumping into it would force one to count inordinate amount of stuff as a cog in someone's thinking machinery, jumping into ponds of water while on fire or shooting at a tiger to avoid becoming the tiger's dinner, both contribute to one's prolonged occupation of one's phenotypic states. But neither ponds of water nor guns and bullets can be properly counted as cogs in the thinking machinery. And I can totally see how one type of extended mind perspective would say like, yes, the boundaries of the cognitive system are just so and they're so flexible that whatever it is in the situation that enables the cognition to happen is the cognitive system at which point the lines are extremely blurred with just the system as it is. It's like a pancognitivist perspective. And that ends up, at least from a Bayesian perspective, not being informative, because we no longer reduce our surprise by learning about anything. And we've sort of deflated cognition away. So where does that sit with you? Is that a goal? Like what are we looking for in delineating cognitive systems from each other? Do we prefer pancognitivism? Do we prefer anti pancognitivism? Or do we want something in between where like humans are special or linguistic entities are special? So I think that pancognitivism should be avoided. In general, if you have pan-Santhanism, you are on bad road, traveling down a bad road, because if everything counters X, there is nothing gained about it. I'm calling things Axis. No one is puzzled by things in general, I think. So that's the first point. Pancognitivism should be avoided. What I think is a desideratum many would buy is to have something balanced in between individualism and DDD. I think appealing intuition that sometimes interacting with stuff at least aids cognition in a very substantial way. So I think that DDD go to the desideratum would be to have a bubble around DD. Individual, pre-theoretical, I understood. Many would disagree. Many would be comfortable with group vote, with group minds. I think that is minority view, but maybe they're right. I mean, being right or wrong is not a matter of numbers, but still to frame DDBs, I wouldn't follow the minority view. So I would keep that option in the theoretical horizon, but not use it to orient research right now. So what I would do is to have sort of a bubble around the biological individual. And for the time scale, I think that the right thing to do again, just to frame things and get them started is to focus on short timescales. The hyper long evolutionary time scale is not something that captures our go-to intuitive sense of cognition. And when you start framing a delete, you start from your go-to intuitive native standpoint. So my symptom is for the short time scale, for the here and now. It's very funny, like you mentioned, it's not a vote. A small group of people could be correct, and a larger group could be incorrect, yet people often think about voting as group cognition, yet cognition isn't a vote. So that's kind of an interesting tension there. Let me ask about Section 5. So it's is vehicle externalism conditioned over Markov blankets possible? And you said, I want to argue that resorting to Markov blankets to settle the debate over vehicle externalism leads us to sidestep the whole debate in a very real sense, making vehicle internalism vacuously true. So just for like a little context, what is vehicle externalism and internalism? Like, what is the distinguishing features of these concepts? Why are people interested in this debate? And like, what does it matter for talking about cognitive systems? Okay, so externalism and internalism are things that pop up everywhere in philosophy. Here we go to with the vehicle variety, which is internalism or externalism about the cars and the thinking machinery about the material stuff that does the thinking, say. And you need to have a boundary in mind to say internal and external. And the go-to boundary is the skin, basically, the skin of the biological individual folks they understood. Internalism would be inside the skin. So cognition internalism or vehicle internalism is the idea that the machinery of thought is inside your skin and more precisely inside your cranium. Externalism is that at least some pieces of that machinery are outside, in pen and paper, cell phones, whatever you want. But notice, here we have an intuitive boundary, which is this game. If you go with Markov blankets, I think you end up called in a dilemma. Because you either start with some intuitive Markov blankets, say, at the skin, use it to define internal and external. And then you might discover there something external to that blanket counts as part of the system. Or you can do what, again, Jakob Highway effect did, which is, well, first I find the Markov blanket of that thing, the real one, the Markov blanket with the capital M, and then use that blanket to define what is internal and what is external. And if you do that clearly, everything that matters is internal by how you have defined internal, which is the one horn. On the other hand, you start with the standard intuitive distinction between internal and external, and then try to find a Markov blanket, which is fine. But then the Markov blanket does no longer distinguish between internal and external, or there is a tension, there are two senses of internal, and you need to disentangle the intuitive one and the Markov blankets relative to the real Markov blanket ones. And that is not what seems to be going on in the literature, right? So basically that is my problem, which is a purely philosophical problem, I think, is a purely philosophical one. But still, my concerns were philosophical, so it counts. Fair enough. Is that what you were referring to when you said, I doubt such a redefinition of internalism would buy the internalist something more than a purely verbal victory? Yes, that's exactly the point. I mean, the internal, the vehicle internalist one, at the end of the day, to say something like this, thinking happens in the brain, or close to it. If you go on the Markov blanket path, define internal in terms of a Markov blanket, and then discover that the Markov blanket contains the brain, but also the spinal cord, but also my hands, but also pen and paper, but also myself and so on and so forth. You can say, well, all that stuff is internal, so internalism is true. But I mean, that's a verbal victory. That is not what you wanted to say at the beginning, right? That is a form of externalism that gets passed as internalism because of how you have defined your terms. And notice that even if you go with that definition of internalism, then there would be still, say, super internalists that say, no, no, no, did the real internal thing that does the thinking is just the brain and externalists and internalists or lax internalists that say, no, no, no, no. It's the brain plus other internal stuff within that Markov blanket. So the problem wouldn't be closed. You would still have a clash about cerebralists and non cerebralists, if you prefer. People that thinks that it's just the brain or the brain plus the spinal cord and people that like external stuff doing part of the thinking. It makes me think of like a thought experiment, kind of like a tracer. So like a radioactive tracer might be injected in a blood vessel, and then you can see where it flows or like a metabolic label might be given to a cell, and then you can see how it gets metabolized. So the tracer, let's just say we're following like we're going to pluck one of the strings and trace the impact of like shining light on the retina. Initially, it follows like nervous system paths, but then let's say that light pattern induces the system to pick up that phone and reload a website. So now that tracer is spreading everywhere. It's calling servers in different parts of the world and it's like communicating through satellites. And it's like that perturbation on the retina as space and time sort of zoom out. It's like more and more gets wrapped into that. And so like what I'm hearing you say is if we just say, well, wherever the tracer goes, that's internalism. Internalism is the total cognitive apparatus, including the satellite and the server in the other country. Or one could like retreat back to the epithelial boundary and just say like it is extended once they are everything outside of the skin is external. So it's sort of like, again, no one denies that somebody could see something on the retina that causes a server far away to do something. But where is our implicit or explicit boundary? And will we just stay with the epithelium, which gives us a boundary, but then over different time scales, clearly different parts of that total apparatus matter. Like over one seconds, maybe the ocular motor circuiting and the some brain region is involved with processing meaning. But then over hundreds of years, there might be a more valid discussion about how it's actually interactions amongst people and slower reading and writing of like books or of other resources that actually contribute to like the semantic discussion and evolution. So it's like infuriating sometimes because again, no one's saying that what's on the table isn't that way. It's just like what are we really looking for in a cognitive theory? And what are we gaining or what kind of value are we adding with these different framings? Yeah. So again, to kick things off, I would stay close to the basic individual. I would put the go to boundary and the skin and then wait and see how things evolve from that. There is an important sense. However, in which the privilege of the scan, putting the boundary there at the scan inadequate in the sense that really, if you look at us from a longer perspective, perhaps as a species, you notice that much on the of defense we do are extremely dependent on how our environment is, our toes and our niche totally understood. It's just that due to perhaps evolutionary reasons, perhaps cultural reasons, we tend to attach minds to single bodies, say, and we have an expectation, one body, one mind. Once that expectation is in place, you can discover that the mind extends and that pen and paper and your cell phone and so on and so forth really help you think and really should be considered part of the machinery that does your thinking. If you don't have that expectation in place, it's very hard to understand what the extended mind could be for you because there is no boundary relative to which there is an extension, right? You can't go farther, farther than what? Exactly. You just would have the mind cognition as it is in the real world and once you have that, it makes little sense to say, oh, it's extended. It might be extended relative to the individual, to the biological body, but if you don't have expectation that the two things go hand in hand, you don't have the violation of expectation when you discover that they don't go hand in hand. You don't have the sense of discovery. It's just nothing strange happening that nothing to be amazed about, nothing requiring inquiry. So that's a nice point, but as a matter of fact, we have the expectation that minds are attached to individuals. So there is a sense of wonder in talking about the extended mind, something that strikes us. Of course, a lot to say there and connect back to what you mentioned about the cultural priors, but sort of to hinge on the cultural priors and go in a slightly different way. How does this relate to computationalism or cognition and computers, not just us using computers to have like a human-to-human convo, but what does computationalism add into this discussion or how would you relate what you discussed in the paper to computational theories of cognition? I'm not super sure. So I think that cognition is computational, but I'm not super sure. But if asked without any reflection, I would say yes, of course, cognition is computation. So I think I count as a computation. I see no big problem in extended computations in the sense that if you think about computation as symbol manipulation or manipulation rule-based or rule-sensitive manipulation of certain primitives that might be symbols or sub-symbols or whatever, clearly you can have rule-based manipulation of stuff outside your body. There is an historical point which everyone I think knows that is, well, originally computers were people doing the math with pen and paper before the Second World War, before a digital computer came about. So there is nothing intrinsically strange about extended computation and computation taking place outside your body. That means that it is true that most, I would say, computational models focus on the brain and that most of cognitive scientists think that the really interesting computation happens in between the two years. Then you have, I mean, Rodney Brooks, which surely is one of the inspirator of and prime movers of embodied cognition, anti-representational cognitive science, and that really has inspired extended cognition or at least try another extended cognition. His models were entirely computational, were finite states automaton for the most part. So what I would like to say is that you don't necessarily have a clash in between computation and extended embodied cognitive science. As for computational models of wider agent environment systems, I know there are. On the top of my fingers, or on the top of my head, I don't have any examples, but I can gather them if I were to remember by my artist, but there are such models. And I think that there should be no in principle problems of making a free energy model about agent environment system or even extended cognition. Indeed, I think that Fristen, Clark, Kierschoff, and someone else wrote a paper about that with a quick minimum model. Not sure about the title, but I think it's published on manual language if someone wants to check. This kind of, it cracked me up with the Freudian slip, with the pluralist slip. You said at the top of my fingers, the top of my head, and it's like, aren't those the different worlds though? And so it's just very interesting. And it again comes back to the water we swim in, the cultural priors, and biases that help us shape whether an idea is outside of the Overton window, like rocks can't think, versus you're a heretic if you don't think that rocks think. As we sort of move towards a little fractal resolution, of course, this will probably be a discussion for many years to come. So you provided a quote from Kierschoff and Kierschoff 2019. I'm not going to read the whole quote, but it's visible on the stream. And basically, they just mentioned that the Markov blanket formalism as applied to systems that approximate Bayesian inference, or we could even take it one more instrumental level, to systems that we approximate with Bayesian inference, serves as an attractive statistical framework, in their case for demarcating the boundaries of the mind, but other situations as well. And then you mentioned like, perhaps this is the correct way to think about the role Markov blanket should play in the debate over vehicle externalism, maybe asking, where can we draw a Markov blanket around the thinking machinery yields more satisfactory results than trying to find the mark of the cognitive. So this is really like a really nice point. And like what kinds of questions help us move forward as we do seek like productive, but also useful and accessible ways to think about these topics? What sort of question was forward? Yeah, I have no idea. Probably the only sensible answer is ask question and try to solve them somewhere. Any retrospective, you will be able to identify the ones that really push you forward from the ones that were not well posed. So probably the only sensible question to which the only sensible answer is wait and see. As for the DDD Kirchhofen-Givestan quotes, I'm not super a big problem I had with Markov blankets and extended mine is this is to me a place not super clear whether when Markov blankets are used, the Markov blanket is something out there in the world, like some mathematical entity instantiated somehow in the world, or whether there are some sort of accurate representation of what is going on, or whether they are just a mathematical tool to make sense of what is going on. It seems to me that those three things are used together when it comes to discussing the boundaries of the mind, but there are three different things, I mean there are three different ideas about what Markov blankets are and by adhering to one coherently you have a very different outlook and a very different metaphysics say of the free energy principle. If Markov blankets are things out there in the real world, then it seems that you must take the free energy story at face value, you must be a hardcore realist about the free energy principle. If they are just a tool, you can take the free energy principle to be just perhaps a sort of mathematical machinery you can deploy to build interesting models of phenomena you want to study, you can try just as you can use standard computational theory to build models of things, you can use the free energy principle or you can use dynamical system theory, you have various tools you can deploy, but the point is that if you go super instrumentalist, oh that's just a model and I do what I want with it, my only guide is not truth but pragmatic discoveries and making science grow. If you take that standard free energy principle, you're forced not to make strong metaphysical claims about the boundaries of the minds and how things are. You can make strong and interesting claims about how things are modeled but not about how things are. On the other hand, if you go super realist, you can make strong claims about how things are but you also have a very straight outlook on reality I think with these sort of fractal boundaries within boundaries that are reciprocal and so on and so forth. My two cents would be that to follow the pragmatic non-realist usage of the free energy principle, to take it as a model building tool, but don't trust me, there are my two cents and I'm not super savvy about instrumentalism and model building. Definitely this question or tension of instrumentalism and realism kind of map and territory in Mel Andrew's 21 paper and several other times we've seen this discussion arise of course is active inference just a statistical model that we can apply to action cognition perception just like we could apply a linear model to height and weight but that doesn't make height and weight a linear aggression so does applying the active inference structure to neurobehavioral data that cannot be taken as positive evidence that that system for example is doing active inference so that's definitely a point well taken and the pragmatic turn which is associated in cognitive science with a movement towards action orientation and then there's like this meta pragmatic turn in cognitive science which is like better theories are more useful and that actually is in contrast with what we were speaking about earlier with science as providing explanations predictions because the pragmatism actually highlights the ability to design and control it's like well if I used the epicycles to explain predict design control it's useful enough and that in a pragmatic world would be sufficient but then somebody is always coming out of left field or right field or whatever with the questions about how things actually are and say well but they're not actually moving in that way yeah I perhaps do not see any strong clash between pragmatism and urea reason what you definitely have here are two say tendencies one to be hyperplotonic and say look reality is this way there is nothing in your data that can prove me wrong which even respectable times for certain class of entities like mathematical entities and the opposite tendency to say I don't really care about how things are I just care about prediction and control and making good science and predicting stuff and making discoveries I think that a sensible observation is that typically theories and models that tends to do make discovery and give you good prediction have a degree of accuracy tend to capture something about how things are I mean you cannot proceed arbitrarily in model building and do whatever you like you have some sort of resistance and intuitively I'd say that resistance is reality and so even going fully pragmatism prediction and control only you get something about reality maybe what is lacking is the sort of platonic ideal reality that is super independent about what you know I mean that is something you will encounter rarely and only when you do philosophy so if it gets left behind might not be the biggest tragedy of the human intellectual history that's ended in this way um so one thought is that in active inference the way that we think about action selection policy selection is as related to providing pragmatic value which like brings you into alignment with your preferences and then epistemic value which reduces your uncertainty and it's almost like the utility is loaded on the pragmatic side but of course just narrowly pursuing the pragmatic value ends up leading into a blind corner and so it's good to balance both and to have like diversity of thought within a person within a scientific community we don't need to be a hundred percent convinced or only on one side of the line as individuals or in a collaboration and then to connect that to one other point that you made in the paper so it's on not sure if there's page numbers but it's um there should be page numbers but there aren't but there aren't so I think it's just online version and so without yeah page numbers yeah exactly um so you were talking about benchmarks that adjudicate whether something counts as a constituent of the thinking machinery and so that's like something we've been talking about so then uh you wrote traditionally this benchmark is provided by the mark of the cognitive that one endorses again relate to what we've been discussing that is by what one implicitly or explicitly takes to be necessary or sufficient to make something a genuine contributor to thinking in the broadest sense so then I'm thinking about this paper so it's a single author paper however you mentioned like in a footnote like this observation and the example are due to Chris DeLega you mentioned in the acknowledgments some people's names their citations which is of course in science how we communicate like attribution of ideas and then there's even like a little discussion with a reviewer and so it's like there's all these different contributions actually to this cognitive artifact this epistemic artifact and so um it's like the like you're you're supporting the biological individual epithelial boundary single author paper all these external contributors like they can get mentioned but they're not authors and then um like if it were going to be written by the radical pancognivist like why not make everyone an author just let's have everybody as a contributor and so just sort of like a funny little microcosm there no no one not even the hyper radical would put everyone as an author because no one likes the academical swiss now I tend to privilege the individual boundary but I mean I am pro extension at the end of this I think I mentioned it in the paper that I am in favor of cognitive extension and perhaps because I want cognitive extension to be real I favor some conservative boundary which is is lead from past is perhaps more second analysis about my choices and my preferences but yes I mean I'm on board with GTR that you can have cognitive processes that are that run through multiple brains I'm not super sure what how common they are because even unless you do some real extreme brainstorm typically the input coming from others is very limited or not very sizable but I mean there are I would say generally multi-brain cognitive processes like a conversation yeah like very deep conversations in which at the end of the day you agree you do bring about some well-defined artifact or output just free floating conversation I wouldn't say that is a a genuinely extended cognitive process but if you want to know there is a paper by Jelle Bruyneberg and Regina Fabre in which they take all the extended mind stuff with mind wandering and that is fascinating because mind wandering is clearly a cognitive process is clearly thought with the capital of tier I wouldn't say but isn't not something that as a definite output there is no set endpoint to mind wandering right and yet they claim they might it might extend and if you take this sort of no set output no set endpoint perspective then maybe even converse causal conversation may qualify as intrapersonal cognitive processes but I'm not super sure about how one could develop that idea precisely it it kind of makes me think about in a very transposed analogical way like to integrated information theory like if two people are speaking at each other but they're both having the conversation that they would have with a wall then it's like they're speaking past each other it's not like an emergent conversation whereas if people are engaged in like active listening and like actually integrating some novel stimuli then coming up with something they wouldn't have said to a wall then that is like where that that third space arises where things can be productive but then also that's very interesting that they might not be and one other note on the mind wandering is um as our technology in the niche changes with some neural recording opportunity say just let your mind wander and all of a sudden through biofeedback or the enabling of new cognitive linkages that mind wandering actually could leave a trace on the environment in a way where if it's just brain bound it wouldn't which I think returns us to that earlier question just like that one thing I want to address about like composition of cognitive systems so whether we're internalist or externalist or whatever how do we think about this space of the adjacent possible for cognitive systems like what can we think of next what can we think of next or how could we think differently if there were some structural changes to our environment like the pen until it existed or the writing implement until it existed wasn't there and there was pre-adaptations in the evolutionary basis for utilizing it etc but when it was there it changed the game in a quantum way so how do we think about those kinds of transformations of cognition I'm super not an expert about this topic and I'm a bit ashamed by this frankly but I and I'll need to catch up but I think that this probably strongest case study you could make for transformation is with reading and doing math we have of course basic visual capacities but reading seems to be a purely cultural thing something that you need culture to learn and something that deeply transforms your cognitive abilities so just the ability to target your own thought crystallized in a written form gives you a level of meta reflection that would be hard to maintain with your brain alone or the ability to crystallize your thoughts in a scrap of paper leave it there to ferment and then go back at it one week later two week later three week later with a fresh mind this sort of thing is not something you can do without externalizing your thoughts you can just remember your passports and objectively look at them as you objectively look at your passport written in paper and the other thing of course is math we have some basic familiarity with numbers that is biologically endowed but pretty much from arithmetic on it is all bootstrapped by culture and of course here you have the biggest transformation possible with mathematical objects modeling and all sort of things so those would probably be the case studies to try and have a sense of what ddd space of a decent possible mind might be and this is pretty backward looking I'm trying really hard right now to think something more forward-looking something more positive something more hands-on let's build a new mind but I don't have any idea right now about that we'll like look back to look forward it's an it's an interesting idea and theme I you know of course wish you look on the forward-looking but when you talked about asking questions it made me write down and subsequently reminded myself later either the question will have been asked already doesn't mean that it's been satisfactorily answered to anyone or to you but either the question has been asked in its rhetorical structure and then it'd be awesome to point people towards a knowledge resource or it's a novel question and novel questions are awesome and then looking back it may be possible to know which of those novel questions led to a useful line of discussion or application but to wait before the question and say well I only want to ask the best question it's like in a way we don't know since it's so many steps down the road from the question at least one or two till we can evaluate the impact of the question and so like as a hack or a heuristic having rigorous scholarship so that when the questions have been asked we're not doing truly redundant work but then also like recognizing and raising up new questions helps expand the frontier and the fringe and some fraction of those may develop into useful avenues and then later on people might say well it was obvious somebody should have connected the sequential database with the encryption and with an economic system but those things existed long ago and it's like looking back that some of the synthesis conceptually or technologically um post-hoc make more sense but again waiting and not acting for it to make sense a priori it's like it it doesn't seem like just a 2022 thing that that's not gonna work that seems to be like in principle a very challenging way to expect effective action yeah i'm not super sure i got the point so uh the idea would be this that you don't have any sort of precognition about where a question will lead you you cannot evaluate questions beforehand before asking them to see oh this person right here a long way to go is a good one and this other waste of time right is that the point yeah even if there was no um yeah we shouldn't uh prejudge questions we could easily sort questions into those that have already been asked and those that have not been asked and so to those that have been asked we can use scholarship to those that have been asked we can like recognize novelty um so even we don't need to finesse that too much and just to kind of connect that however vaguely to your idea of like looking back but then the challenge of looking forward in cognitive sciences sometimes yeah it's surely a challenge but i think it's per the big challenge for me because i'm illiquid to look forward in cognitive science if you probably ask people that do build cognitive architectures and cognitive models for a living uh i think they would be much more forward looking than they i mean um you have something you'd like to try some twist on the architecture you'd like to put and that is i think forward looking right well what is your forward looking like over whatever time horizon or capacity set that makes sense like what kinds of questions or experiments or research areas are you like excited to build on from this paper i'm not super sure where this paper is headed i mean i'll continue probably to work on extended minds i still think i'll be working on predictive processing in a sort of very vapid fashion probably i'll try to merge d2 in a more positive way probably but i'm not super sure probably in computational terms but this is super vague it's super what i want to do with my postdoc if i had one nice it's not a concrete proposal right now hypothetical postdocs are the best exactly is a list of wishes what i wish to do well i wish to do yeah probably that's that would be one thing and another thing i'd like to do is to try and say open up and specify the grip metaphor you have Andy Clark and those other people telling you what predictive processing actually free energy principle gives you is a grip on the affordances that matter is not super clear what the grip is apart from not stumbling into things and moving around appropriately the epistemic sort of contact with the world is left i think analyze but i have no big idea on how to analyze it it makes me think about how words related to our physical cognition and describing our interior landscape like get a grip you know can you see what i'm thinking about or let me sketch it out for you comprehension means to grasp with a hand so like there's so many and of course the whole optimal grip active inference as getting grip on things from a physical perspective like as the hand goes for the cup of coffee but then okay now we transpose that into the cognitive space and we in a conversational way do talk about grip and then here's all of these formalisms that it can be hard to get a grip on but are they about grip or like what would that actually mean that's very interesting question yeah i mean there is a way to impact the grip idea in terms of beige and inference but that is a bit too internalistic for my likens the idea that perception is a form of inference that you are inferring every time everything it works in the model domain the model is inferential whether you think itself is inferential i'm not super sure and i'm not sold on the idea that the inferential talk really captures what goes on in most instances of cognition uh i mean most of our thought is not described inferentially when i let my mind go and start mind wandering not inferring things at random i'm mind wandering when i draw an inference i draw an inference typically with high concentration to one specific topic and usually did the panel paper so the inferential talk might not be optimal but still i feel the need to do to give some sort of epistemic connotation to the grip is not just not stumbling into things and moving around quickly and swiftly there is some sort of comprehension and intelligence and mindfulness into that that going with a pure behavior good behavior good behavioral patterns vocabulary might not just miss but it might even make you open to the charge of being a behaviorist of being a philosophical zombie without any inner life and that sort of things we need some mentalistic lexicon some epistemic they charge lexicon which is not full-blown decard and a meter challenge would be to develop it it makes me think about like an ocean and the river and then where they meet and it's brackish your cognition and pure cognitive uh the rest cognitive and then there's the physicalism and um we have the ability to like be uncertain actually and to evaluate and to oscillate between different perspectives or different regimes of attention on these different ideas like more highlighting the physical boundaries of a system more highlighting the cognitive more highlighting the statistical causal which is where the Markov blanket concept comes from from pearl and Bayesian causal graphs and those causal graphs the nodes and the edges may or may not link to anything that is a physical thing or a cognitive thing so it's a very um open and interesting area um is there anything else that you like wanted to just bring up talk about ask no i don't think so i think it's now time to thank you for the the space the opportunity and the wonderful talk uh to thank the the audience which i not seen because i can only see you apparently from my screen but hi audience i hope you enjoyed it i hope you didn't find me too rambling if they're listening this far then i don't think either of us are too rambling for them oh maybe they're just doing exercise driving around and once i'm standing back around be back in France awesome well yeah Marco thanks for the paper and the contribution and for joining for a guest stream you're always welcome to continue the discussion and also thank you for all of those who listened and paid attention for this conversation great okay see you later see you later bye everyone