 Hello and welcome to the Active Inference Lab. This is the Active Inference Livestream. Today we are in Active Inference Livestream 14.2 on January 26th, 2021. Today is gonna be a great conversation, so thanks everyone who's here live and thanks everybody who's watching live and in replay. Welcome to the Active Inference Lab. The Active Inference Lab as I share my screen is an experiment in online team communication, learning and practice related to Active Inference. You can find us on our website, at Twitter, Gmail, YouTube, or our public key base user name and or public key base team and shared username. This is a recorded and an archived Livestream. So please provide us with feedback so that we can improve on our work. All backgrounds and perspectives are welcome here. And as far as video etiquette for Livestream goes, check out the checklist and remember to fill out the survey at the end to give us feedback and then during the conversation we'll mute and we'll raise our hand so we can hear from everybody. To learn more about Active Inference Livestreams for 2021, you can go to this rb.gy link where you'll find a spreadsheet. Today we are on January 26th, 14.2 with our second participatory group discussion on the math is not the territory with author Mel Andrews. And next week as we head into February we'll go to a different paper. So check out this spreadsheet if you wanna read the papers ahead of time and get familiarized with them. Today in Active Inference Stream 14.2 we're going to go through our introductions and warmups and then we're gonna walk through 14.2 again being a followup on the math is not the territory navigating the free energy principle by Mel Andrews in October, 2020 and we're able just to kind of chill, talk, raise some questions that come to mind as we continue and then next week we're gonna be on a different wavelength always welcome to have participation. So introductions and warmups it's a nice group of seven today we can introduce ourselves and feel free to give any kind of introduction or check-in you'd like and then we'll pass it to somebody who hasn't spoken. So I'm Daniel, I'm in California and I will pass it to Ivan. Hello, my name is Ivan, I'm from Moscow, Russia and I pass it to Marco. Let's go to Alex first. Yeah, hi everyone, I'm Alex, I'm also in Moscow, Russia and I'm a researcher in systems management school and I pass it to Steven. Hello, I'm Steven, I'm based in Toronto, Canada. I'm doing a practice-based PhD at Canterbury Christ Church University and I'm gonna pass it over to Cherie. Sarah, if you want to, otherwise let's go to Marco. Hi, I'm Marco, based in Holland and I'm just an active enthusiast, I'll pass it to you. Hi there, I'm Mel Andrews, I'm doing a PhD in adjacent staff at University of Cincinnati. Is that it, is that all of us? Yes, cool, and then Sarah, anytime you wanna jump in is chill. So today, oh yeah, go ahead, Sarah. Hi, I'm Sarah, I'm a person interested in stuff and things, Mel's talk is awesome, here I am. What else can be said? So for the warm up questions, we'll just dwell on them, they're just openings and just opportunities for us to raise the kinds of things we've been thinking about today. And so what is the, what's something you're excited about today? And then also what did the article or last week's discussion make you curious about? So something you're bringing exciting today or just a question or even if it's adjacent or it's not a quote from the paper or anything like that, like who are we that has brought us today to this conversation with Marco? And then I believe it's Stephen, but Stephen and Mel, if you could add your name on Jetsy. But Marco and then Stephen. Yeah, well, last week we ended the discussion with at least three people, Mel, me and Alex Piper on the prospect of continuing discussion about the relation between a thermodynamic entry and information threat to your people. So very much excited to explore that because I think that was a very crucial part in the paper that deserve more unpacking. Cool, Stephen, and then anyone else? Yeah, I'm bringing an interest in dimensionality and how that relates in to like high dimensional spaces, low dimensional spaces and entropy. So that was something that this paper really kind of got me thinking about. So I think while it seems very abstract, I think it might have some profound practical uses. Okay, could you also speak a little bit more directly into your microphone? But thanks for that, so yes. Oh, I'm trying. Yes, yep. So definitely, yes, philosophy, paper, abstract, general, theoretical, a lot of adjectives that we hear with philosophy, but then this is almost about the philosophy of utility and about the reality of the models that we make. So it's kind of has a hand in both of those worlds, which is what makes it an interesting thing to think about, I'm also curious about that intersection of how do we generalize and abstract correctly in order that we can make more effective action. And I'll just put up the last two questions and anyone else is welcome to raise their hand. What is something that you liked or remembered about the paper? And then what is something that you're wondering about or would like to have resolved by the end of the paper, or by the end of the discussion today? So one thing I'll just start with is if the math, and Yvonne, you can use two, please, thank you. So if the math is not the territory, then what is our metaphor here? Like what is the territory or what is the math? If we're gonna go with the title of the paper that the math isn't the territory, what is our complete spatial metaphor if that's even helpful to walk down? But it's just, yeah, just a few thoughts. Anyone else wanna raise anything? I guess I have maybe a general question, which is for Mel, like almost like a one to 10 or how Mel maybe feels that FEP contributes to like epistemic unfolding, like the best possible tools to allow us to ask the most relevant questions, that's. Yeah, Mel? Yeah, I think, so I think I came at this with maybe a kind of naive and outdated view of the scientific method. And I really didn't like very general, very abstract models, and I really didn't like use of metaphor because I saw a lot of the kind of mistakes that were being made with very abstract models that were done with kind of analogical thinking. You get a lot of, you're like, I think I am. Okay, please do not, yeah, thanks. Make sure that everyone's audio, let's take a few minutes before the hour begins so that we're here on time, so that everything's prepared. But thanks a lot, the audience also appreciates it. Mel, continue. What happened? Oh, there's just audio on a loop. Cool, not me. No worries, yes, not you. Although I was also late. Yeah, so I tended to think of certain approaches in science as being kind of bad because I saw a lot of mistakes that were being made with them. So a lot of kind of reification where we start with an analogy like, oh, we've got selfish genes or oh, we've got, borrowing models from physics into biology and then confusing them, reifying them. But the mistakes are not the methods, right? There's a distinction between the mistakes that come along with certain methods and methods themselves. There are mistakes that come along with everything and with any kind of approach you're gonna take in science. And especially when it comes to sort of uncharted water and research, you really need to be more creative. You really need more generic models. You need to be borrowing models from one domain and importing it into the new domain. Kind of testing in that, right? So when you've got very simple systems and in like classical physics and chemistry, you can kind of build up direct models without doing much analogizing, right? But when it comes to dealing with something like the human brain or an ecosystem, things, you know, or even like social structures where we're just sort of out of our depth, you need a lot more looseness. And for that reason, I think when it comes to the kind of systems that we wanna understand in future science, things like the SEP are actually gonna be the way to go. Cool, well, this whole future science, I'm sure that's something we can return to. But this was a really great point that when we're exploring or constructing a new area, we do need to borrow ideas, vocabulary, methods, et cetera, because otherwise what is there to be discussing or converging on, but these come with baggage, preconception, and just incompleteness because they're not the Gidellian incompleteness, just their metaphors, they're not total one-to-one of everything. So how do we navigate borrowing from the past, but also moving into something that has different attributes? So Stephen, then anyone else? That's a very good point in terms of, you mentioned about this, like, does it belong to science? So we often think about, is something an engineering or applied science, or is it science? But then I suppose there's a question of when is something useful to have in a philosophical sense somewhere in between science and philosophy? So not completely philosophy, but maybe there is another category of foundational knowing that we kind of then use to build other things upon. So maybe there's something in that. Yes, it's almost like this question about whether the natural kinds are disjoint or not, and then, okay, all the stuff or all the approaches that we can take, there's science and not science. Is that a clean break or is there multiple kinds of non-science, even multiple kinds of science? So this is the demarcation problem and maybe the simplest demarcation would just be the bright line and line in the sand, but what if the demarcation problem is a little bit more nuanced and it has islands or it has other attributes that are not just a line across the gymnasium floor? So Mel, I'd be curious, how did you think about the demarcation problem early on in your research or when you initiated this and then where has that taken you on the demarcation problem? Which is, again, about what is science and what isn't science? Yeah, unfortunately, I think the longer I spend, my background is in cognitive science, I was doing, I had some more bio background and some psych in cognitive science. So this is the first time I've been, full time a philosopher. And I find that the longer that I immerse myself in philosophy, the more quinian I become, which is sort of to my chagrin, I didn't expect to see myself go in that direction. What does that mean, the quinian in this context? Yeah, naturalism, which means no backdoor to knowledge. The only access we have to knowledge is empirical. So there's like science and there's commentary on science. And there's this, I view the relationship as between science and philosophy as being kind of intrinsically connected. They need each other. The philosophy comes up with new questions that they don't know how to ask yet empirically. And then we figure out over time, ways to go about these questions empirically. And we come up with methodologies. And then the methodologies fail. They hit a standstill and then philosophers have to come and say, well, you've kind of arbitrarily narrowed yourself to the outset. We need to start thinking in a different way, right? And I don't think, I think I do, in contrast to what sort of quin... There's a kind of popular thought that goes along with naturalism nowadays, which is just that philosophy is this kind of finite stock of questions. And as we pull the questions out of the magic hat and had them off to empirical sciences, we run out of questions. And so, fortunately in the near future, philosophers will run out of questions to ask because the scientists will have solved all the questions and then we'll be out of luck, right? Which I think is just kind of hopelessly ridiculous. I mean, it's a really simplistic picture. And fortunately for me, I don't think we'll ever be out of a job. I don't think scientists will ever be out of a job. I don't think philosophers will ever be out of a job. I think there will always be questions for as long as we persist. But I do think that ultimately, I am starting to come around to the idea that there's just an empirical route to knowledge. And that means to me, where I'm at with that right now in kind of processing that and digesting that is that there may not be a sharp dividing line. There might not be a bright line distinction between science and not science. And what I really hate about this is I really want math to be entirely conventional. I want math to be invented, not discovered, constructed. And to be a naturalist about mathematics, you have to kind of accord it more reality and naturalness than I want to, but, you know, so that's where I'm at. That might have to, I might have to bite that bullet. What Steven was saying, so the two people, what you were saying about kind of like foundational things that come before science, this ties into what is being said by Inez de Poletero and what was said by Matéria Colombo and Corey Wright in a 2017 piece about first principles. So there's a piece, 2017 piece called First Principles in the Life Sciences, something in Mechanism. I think it's in BJPS by Matéria Colombo and Corey Wright about first principles and the FEP. It's about the FEP, but the first principles approach. In the live chat, put it in there. So, cool. People can check that out. Do you want to continue? Oh yeah. Or Sarah, do you want to go ahead after you've adjusted your chair, but go ahead? Yeah, I'm pulling a wire from under my desk. Yeah, I have a question for you, Mel. I'm as a person, a new student in philosophy of science. I'm bumping, and somebody who came from physics, I'm bumping up against this dynamic that I see between philosophers and science where science will come up with something and then there's this kind of ankle-biter dynamic where philosophers will be like, well, what about? Yeah, but what about? And I'm just like, eh, I mean, I want a more positive conception of what philosophy can be. And the only one so far that I see that's really clear, and I'm a new student, so it's not like, so I'm probably missing a lot, but like, I really like Chris Fields because he's more like, not afraid to be sci-fi, you know? Like, he's not afraid to have a speculation, which is originally what a lot of philosophers did. So I'm like, I just, I really want advice as like a student about what, how you feel that dynamic can be. Cool, we'll go Mel with a direct ask, and then Marco, Diane, Steven. Yeah, Chris Fields is awesome. He gave a talk recently to one of our working groups and that was fascinating. A lot of it was over my head. It's just introducing so many new concepts, but really fantastic. And he's joining in March, early March for Act M17. He'll be on. Oh, excellent. Yes. What are we reading? TBD, to be determined, but we'll find out very soon. Oh, fantastic. Yes. Awesome. Yeah, and there was something I read recently. I'm, there's a long story. I wouldn't, I don't want anyone thinking that I would just go out and read Heidegger because I wouldn't. What happened was there's a long story about tracking down Quine's typewriter and like writing back and forth with Quine's son and finding a Heidegger scholar who had just acquired the typewriter of Martin Heidegger who I used to correspond with as a kid. But I ended up reading some Heidegger in part of, as sort of part of writing a piece up with a Heidegger scholar. And just a couple of paragraphs, but there's a line about that there's this sort of, and this is, you know, this is like at the time of like, there'd be an historical, at the time of sort of the positive. And there was this line about logic and philosophy of science kind of like limping after the sciences and sweeping things up and putting cobbling things up and cleaning up after science in this very kind of hobbled, lame way. And it wasn't contributing anything of its own. It was sort of doing like logical reconstruction of scientific development, but nothing that actually really contributed to furthering the end of science. And Heidegger was declining that and saying that we need a generative logic, that we need to go back to the model of the relationship between philosophy and science of the time of like Plato and Aristotle and thinking ahead of science. And I do think, I very much think that's what we need. And I think to some extent, there's still very much like a live tradition in philosophy of science that is stuck in this kind of, we're just describing, we're just doing the bookkeeping, right? But at the same time, there's, I think, very much a live and active tradition of philosophy of science today that is really thinking ahead of science and coming up with new questions, problematizing things in a way that leads to real breakthroughs. Cool. Marco, Dan, Steve. Thanks, yeah. I really like the discussion so far. I would like to point to the notions of adaptive inference and the exploration, exploitation, and the problematarity to maybe approach this relation of science and philosophy. So, I mean, I also have my qualms with philosophy, like Sarah and also noted that I personally really hate the excessive, well, in my view, a test of trends to seem authoritative and absolutist and we fetishism of universal truths and stuff. I'm not fond of that. What I'm more fond of is the, it's almost, for me, it's an art, right? It's the art of taking something, a given set of assertions or propositions, whatever, and then generating more territory to explore from that, right? So, the way I see it is on the active inference, knowledge is generative, right? And so that generativity can be seen as leading to adaptive inference, so it generates new hypotheses, but it can also be seen as imagination. And that level, that step of generativity is then itself rooted in what's been generated upon a direct contact with empirical reality. And so, for me, it's like an intermediate kind of step. It's like a detour. You create all these hypotheses, you explore the maps you created yourself, and then you can return to the empirical grounding. And so that's kind of how I see philosophy. It's kind of a detour. And like Mel said, they're inextricably connected without an empirical basis. You have no grounds on which to imagine and explore and reconceive, right? But at the same time, you need to path back. Everything that's reconceptualized, explored, and imagined in these, say, philosophical modes of inquiry can then return to the scientific ground. So it's nice back and forth, at least that's how I see it. And like Mel said, there's no clear demarcation. I don't also do think there's a clear demarcation. It's simply, for me, the gradients of how well grounded it is in the empirical roots of these inferential processes. How much weight does it have? Where the other extreme of the spectrum would be that you generate a lot of understanding or sense or use to end your uncertainty by virtue of this reflexive engagement with your knowledge. I think Ryan Smith had a paper on model expansion and elaboration. So maybe that's also a nice connection to that sense. Cool. But yes, you used the term. Sorry, Marco, can you say again who that was? Sorry? Say again who that was that had an idea. You just read at the end. Yeah, I think Ryan Smith has something on model expansion and stuff. But I'm not sure. But also you can get the paper on abductive inference and curiosity and insights, which is actually the paper I co-authored with the Berlin 2017. But that's kind of what I read. It's always like abductive inference. The coolest form of inference. Cool. Well, yeah, inductive reasoning is when you go from the particular to the general. And then deductive is when you go from the general or the philosophical idea to the specific. And then abductive is when you go sort of laterally, like your abductor or something like that. It's off to the side. And so that's this imaginative or generative way that we're talking about as kind of threading between science with the exploit and philosophy with the explorer. Are we ever going to be trapped into this dialectic? Philosophy is broadening our horizon and science is narrowing them. And then it's almost like there's this abductive sideway that we can take where there's a fruitful discussion. And so that's kind of the niche and the intersection we're reimagining. So Steven and then anyone else who raises their hand. Steven Venmel. Yeah, I think that the, actually that's a very good point about abduction and in a way you have to say everything that we know started with abduction from a child because it's only with time that you start to set I often think with induction is also you've got a target and you try and narrow the gap. But you've got a target. So where's the target come from? How do you know what the target was? And you've also got that question then of, so I think this is, these questions about how they get applied are very, very relevant to bringing in this question of philosophy. Are you applying it to help scientists or are you applying the philosophy to help you decide to sit and how am I going to sit in church or how am I going to spiritually be in myself? So you've got this kind of, and I think this is where you're starting to see the same philosophical questions being addressed with the same framework as opposed to it being like, okay, that's for them. It's starting to break down some of the silos I think of how it hits the ground, and I think that's quite important. Yeah, Mel? Yeah, it's sort of to what Marco said, but also sort of to something you said in the beginning, which is, I guess Dan, you said this, which is, if the map is not the territory, what is the map, right? If the map is not the territory, what is the map? And I think that there's a kind of classical position in I guess both science and philosophy of science, maybe even also the popular conception of science, which is very prevalent in modern times, which is that there are like a finite number of atoms in the universe. And, you know, we're picking away at psychology and we're picking away at the social sciences and we're picking away at biology and chemistry and so on and so forth. But there is some fundamental level of reality, you know, whether that's atoms or, you know, whether that's the quantum level or whether that's string theory or what have you. But there's some ultimate level of reality and eventually, you know, we're going to figure out how the psychology reduces to the biology, how the social reduces to the psychology, how the biology reduces to the chemistry and changes, right? And then we'll be done. And I think that this is a very mistaken, outdated view of what reality is like, but more so what our kind of epistemic acquaintance with reality is like. I think it is, as Mark was saying, far more generative, right? Where we're instead of kind of constantly reducing the allowable frames of reference for understanding the world, we're actually constantly generating them. You know, we should be doing something like model selection, right? Some models, some theories, some accounts in the sciences do similar work to others and are not as good. And those get weeded out, right? But we're also constantly coming up with new approaches because any frame of reference, any model, any theory that we take is not purely a reflection of the world out there, but will also be a reflection of our kind of epistemic aims. Will also be a reflection of what we want to get out of that way of relating to the world. And so you can see how there's going to be a kind of infinity of possible ways of getting at the world because they all reflect our ultimately subjective aims as researchers. Yes, very nice point about the reductionism is misguided. Just the fantasy we're going to keep on reducing to smaller and smaller pieces. So let's think about the map. If someone says, I want a map of California and someone goes, oh, no, no, no, no. That's already been solved. It's made up of dirt. No, I don't want that. I want to be on a road trip or I want to be visiting my friends. So when we reintroduce action and pragmatism and utility, the argument against reductionism is very obvious, which is, okay, it may be that it's a substrate, but that's not actually how we're moving forward. And then it's interesting that there's a computer science algorithm called map reduce. Now, it's a different map in a different reduce, but it's almost like action and inference. We map, we do sense making, and then we reduce the scope. So we map and we sketch out in very broad strokes and then we have to go into that reductionist mode to go into a little bit more detail about what is actually on the road or in the ocean ahead, but we're mapping and reducing or acting and inferring with an eye towards the direction that we're already moving in, not just because we're going to get to the bottom of action or the bottom of inference and then just call it a day. Just like you said, it's an infinite game. We're going to keep on coming up with questions and keep on coming up with measurements for the dialogue between scientists and philosophers and ultimately just participants in the knowledge enterprise, whatever somebody's title or profession happens to be. So Marco, Dan, anyone else? Thanks, Daniel. That was very beautiful. The image you vote for me is like a directed cycles of contraction and expansion, which I believe is also very active in French. But anyways, so about the question about the maths and the language and everything, so my improvised pitch is the maths or the mathematical mode of active inference effectively described how the terrain maps itself. So if terrain is the thing itself or the actual reality of the brain or any ecosystem, then active inference is basically saying, well, all the constituents of this ecosystem or brain, for example, is constantly inferring each other. It's constantly engaging in mutual and reflexive inference and all these inferences will then leads to models of each other and effectively they all do this kind of subjective mapping, right? But all these subjective maps over scales and levels then also need to calibrate to each other. And so, yeah, if we have to play with words, I would say the maths describe how the terrain maps itself, which I think is, you know, quite nice. Also, I very strongly echo what Mel said. I was very briefly worried that she was actually arguing for reductionism, but I should go better, of course. Yes, so if it's appreciated, I really like thinking in terms of trees. So there's a book by Julio and Africa, sorry. But my favourite phrase is trees of the brain roots of the mind. For some reason it always stuck with me and basically it's because there's these generative trees that lead to roots at yet another level. And so, again, kind of the way the previous discussion. So the reductionist story is more about, you know, the kind of essentialism, properties and objects and kinds, which I'm not a fan of. But complementing that kind of view is proceduralism. And so I really, again, think that this generative perspective is inspired with proceduralism. It's how do these processes actually structuralize themselves? How do these flows and dynamics attain some kind of regular structure? That regular structure will be in the form of flows and generative relations. So it's like generative scaffolds. And I think that would be an interesting angle to approach ontology as well. Cool. Oh, sorry, one more point if I can. Is also one echo with Mel said of a pluralism. So I think looking at active infants, in my opinion, I try not to talk about knowledge because then you get the baggage of correspondence notions of truth. And I think that's distracting. If you just look at it from an active inferential perspective, actually all knowledge, the value of knowledge is going to be contextual. It's going to be contingent upon the actual agents, even though we can say always just the accuracy of the states in the world as believer, as maps. But these states don't have to conform to some objective truth. It's more about given the states that it has modeled, if it acts based on these states, how accurate will it be? And then there's a bit of a, I guess, a pragmatist account of truth, maybe? I don't know. But I guess what I'm saying is that active kind of implies also for epistemology that it's about the normative textuality and that we really should pursue pluralism in science, which are cool. Thanks, Marco. We'll go Dave, then Mel. So I can't hear you, Dave, but you're unmuted. So maybe try to reload or I'm not sure what's going on there. But Mel, go ahead. Or do you hear Dave? I don't hear him. Yeah. Okay. So yeah, Mel, go for it. But I know that you can hear us, Dave, but we're on a live stream, so we can't figure it out at this moment. So Mel, go for it. I Yeah, I just want to say something effective. I think that that's what I guess we're all sort of converging to describe it. It fits in very very well with the kind of Bayesian view at least the Bayesian view of science in the future. There are some kind of stagial Bayesian views of of the of the late 20th century, mid-late 20th century that I think we've surpassed. But I think if we view scientists as kind of subjective Bayesian agents coming up with like different kind of parallel models of the world and adding evidence to these and kind of retiring models that aren't useful in a way, I think this is this is very much what we're describing there. Yep, it's about these metaphors and we're talking about understanding the relationship of science and non-science. That's the demarcation project and we've heard a couple of different metaphors in this introductory section. Maybe Dave, if you want to reload and try again, but in this introductory section we talked about one was like a spectator model with a scientist doing the work and then there's the peanut gallery and the philosophers are either cleaning up behind them or are confusing in their trail and that was the following after science metaphor and we talked about how that was inadequate because there's actually a really fruitful bi-directional generative relationship and that reminds me a little bit of a live stream. So it's almost like in the context of science on the live stream the people who are on the panel are the scientists, the observable states being admitted, being viewable to the world the empirical states are like the state of the live stream and then we have all of these curious philosophers whatever their title or position is in the live chat and they're commenting asking questions and by doing so engaging in this real-time dialogue so we actually have this instantaneous system where we can talk about how the observable states are in feedback with what people are saying at that moment. So it's sort of like again it's just a metaphor, it's not that it carries every single dimension of the metaphor through but it's just really interesting that there's a discussion that can be held amongst many people in the live chat that's kind of like the philosophy literature and then the people who are in the live stream could just not be looking at the live chat and that would be like a scientist who just tunes out and chooses not to engage with that discourse or the people on the live stream might be obsessed with what is happening in the live chat and that could actually distract them and help them or prevent them from carrying out the aims of the stream. So it's a pretty fun way that our new technological affordances and the rapidity of them are helping us understand the relationships and the bi-directional dialogue between those who have, for example, technical expertise and those who are just learning like we've seen on these streams and in the model streams. So Dave, let's see if we can hear you and then yep. Yep, go for it. And then Sarah. Okay. Yeah, I'm glad you mentioned a road trip that reminds me of my puzzlement why people are talking about homeostasis rather than allostasis. It seems we've got the, you know, and this is a puzzle going back to Freud's proposal for a project for a scientific psychology. We want to get to Nirvana. We want to stop everything. We want to be perfectly happy in the eternal now, but then something happens and it's horrible. That's preposterous. That's not how any organism ever acts. Organisms want to go through a series of states. We want to be tired so that we can rest and we want to rest so we can run around and get tired and be a little hungry so that we can enjoy the food more. We want to follow trajectories. We want to take road trips. Was the notion of allostasis proven to be useless or to just become unpopular because homeostasis seems to fit with somebody's mathematical formalisms? I'll give a quick response on that and then we'll go Sarah Marco. So homeostasis is like the first pass absolute value. You say your temperature returns to a certain number so it's homeostatic. It's thermostatic. You know, it's just like a thermostat. And so at the first pass somebody who's looking at just the absolute value of temperature or a blood sugar might note that it returns to a certain point and then in a static conception of the universe rather than this process driven understanding, it's like right, if it's returning to this value of 10 then it's getting homeostatically attracted back to 10. But you're taking a look at the process and the derivatives, the rates of change and you're like wait it never stops. In fact it couldn't and it shouldn't stop. So why don't we just focus on the fact that it's doing this iterate road trip that yes it's returning to 10 a bunch of times but shouldn't we highlight the fact that it's on a circular road trip rather than just say yeah it's at 10. So it's really two sides of the same coin whether one chooses to highlight the instantaneous empirical observation or highlight the process and the approach that the system is actually taking to result in those observations. So just like many other things we're looking beyond the superficial to the generative and to the underlying models and that's what is revealing all of these patterns across systems. So Sarah and then Marco. I'm stuck on something math wise but I think it you know you could describe it as conceptual as well. You know when I look at as far as I've gotten anyway with active inference and FEP you know it's all predicated on Bayesianism and Bayesianism seems to have a kind of teal use in the sense that you do priors and you go forward but I don't know I kind of bump up against like well wait a minute but what about like you know built into FEP there's like a feedback type mechanism but it's kind of built on a substrate that is going forward and I find that kind of interesting like I wonder about the limitations around that and I'm not well enough diverse to like say anything more specific but another little side data point which hopefully isn't a distraction from the question is you know you see a lot now in descriptions of complexity they'll talk about well you need something that's of high dimensionality and that description is very much predicated on the methods that we use to get those dimensions you know it's predicated on the math techniques and I even when I think about this high dimensionality thing I keep coming back to this like well yeah but you've got this like you know F is a function of X Y Z you know like it's not there's no feedback in the actual heuristic well no there isn't a heuristic in the end result or in the in the end and you know like FEP like let's say it converges it doesn't quite converge like work up like it's or like the chain MCMC does but I don't know you get to something that's close by energy minimization but in the end you get to a thing where you're like where you're like something is a function of da da da and it never goes back around okay that's just I hope that's a useful question if anybody has any thoughts around it Marco and then if anyone wants to raise their hand to speak to that yeah that was very interesting wait oh yeah well Daniel already did most of it but I also want to note that it's also really what Sarah was talking about so there's this issue of having too much freedom and how you characterize a system it's not really well defined how you're going to define a state space especially when we're talking about complex systems that model the environment because then we have to not talk about simply physical observables but more the structures that they constitute so for example with homeostasis like Daniel said it's two sides of the same coin and it needs to be in reference to something else so you can say it's homeostasis in terms of temperature and you just try to stay there and then add to that next to it the process of anticipation which then gets you outstasis but another way to also look at it is for example you can also say instead of having one perfect state or value you can also talk about viable sets for example the domain or boundary in which the agents will survive or remain viable and so you can say okay that is the state being in the boundary being in a viable set then it reduces to one point again which again relates to what Sarah is talking about I think was a lot so about the function so for example you can have a market blanket that's enacting sort of functions and it's not built in the loop the loop is not built into the function but the function enacts a process perturbation, activities, flows whatever that expresses into the environment with the expectation that it expresses itself back to them and so that's left implicit I'm not sure if that makes sense also you said something about things moving forward and I'm not sure what you meant about it but it did feel like it was a very interesting angle you were pursuing so if you could elaborate it out it was just really simply like it was just you know you have this idea of prior so there's a kind of atelios going forward in the way that you minimize energy or whatever oh so well personally I feel like the forward metaphor then is the problem it's not necessarily forward it's optimizing or calibrating or tuning but I think maybe I'm just still understanding but it's a constant yeah you're right that wasn't the right I agree the moving forward is related to historicity and to developmental thinking so understanding that although there might be attributes like body temperature that are returned to so that's the homeostasis side of the coin in another sense there is no state that the organism returns to so it's about how will we understand as you pointed the atelios the end well there's this local end directedness as the prior is updated but then we're almost using this short-ended atelios because we don't know the ultimate end game and as you point out it doesn't necessarily have to converge it just has to be the most effective model in that ensemble in that niche at that time and that's who does well on that statistics test or in that you know ocean ecosystem so that's sort of this updating model and yeah Mel then I'll say something about the dimensionality yeah and I think the accident friend free energy framework and I don't know that this is speaking exactly to the question that you raised there but there's a way in which it kind of looks to kind of hierarchical architecture at the exclusion of parallel architecture we don't we don't often talk about like the jockeying of beliefs at the same level right is there any interaction between models that exist in parallel that kind of thing and I think that there are plenty of cognitive phenomena that get mixed that way or in biology right in organismal you know we miss stuff because the FEP doesn't kind of readily accommodate that as I have come to understand it in music I don't think it's very good at handling that and so that's one way in which you're sort of constraining how we conceive of kind of epistemic agents engaging with the world and there seems to be a sort of I guess you could say there seems to be a directionality to it that is maybe misleading that's the time hypercube FEP question with like the gradient descent and the moving forward through time what if it's kind of ironically circular and that we're experiencing it in the downhill direction the hill is just there and we're moving through it experientially in a way that is prospective and it becomes pathological and not adaptive when it's not prospective that would be like prioritizing retrospection without an eye towards anticipation and those systems fail to exist so we end up having this like distilled and extracted understanding of our past only partial aspects of it with an action oriented view towards an uncertain future so that's why FEP is such an interesting framework and as Mel just pointed out it's not like it's a lateral alternative to how the mitochondria works or how a protein folds it's not really a scientific idea at that level but again to take it to this initial claim of the paper the claim is that the FEP can be taken to belong properly to the domain of science so then we're talking about what is in the domain of science landia is there only one kind of model are there multiple kinds of models are those models hierarchical are they heterarchical how do they depend on one another who is involved in producing and communicating about those models these are some of the very interesting questions that are almost within the demarcation problem once you're on one side of the line or the other there's still more questions again highlighting the fact that it's going to be an infinite game with researchers and philosophers so Marco and then anyone else I apologize if I actually covered this I couldn't hear it properly but I just want to add that at least in my view the idea of tails for me is at least I don't like it because I have an allergy for anything that breaks up essentialism so Tilos kind of feels like something universalistic it's almost absolute great it's like what is the Tilos of this thing right but that breaks down where we're talking about entities or systems that are exactly defined in terms of their context right so each of these systems embedded in a big mutually inferring ecosystem they are at the mercy of their context right they don't see the world they only see the larger worlds through the directly adjacent systems with which they sat in an inferential relationship with right so for me the normative principle of FEP is the theological principle as in like I think Dan also notes this the Tilos kind of is constantly changing but there's nothing else to say it's not different than saying that there is a normative principle that is defined in terms of the relation of the system and its context and so that's what I find beautiful so one of my key phrases is contextuality because what's happening with these systems is they're enacting inferential context for me maybe nice parallel is because I don't know that the live stream parallel is a domain of discourse so what we're constantly doing in a live stream is we're cultivating or constructing and acting upon a domain of discourse we keep adding new elements to it, new perspective, new frames and then domain of discourse expands but it's the same thing for these inferential systems these market packets they are constantly changing their domain of discourse when they reconnect with other inferential adjacencies or interlocutors they change their domain of discourse because that domain is defined by the adjacent systems at least in my opinion this is again a bit speculative way from the Brown's truth but I'm curious to hear what you think. The patterns of discourse define the interface so it's not just that there's the essentialist nodes that are then like bumper cars on a you know in a chat room who are engaging in this thing called discourse it's like it's a co-construction just like you pointed out with the relational and the contextual or the ecological teleology the purpose is multi-scale and it's relational and then also teleology is just one of the quadrants in the 2x2 with Aristotle's 4WISE, the Timbergens 4WISE so there's the teleological cause but also there are the other quadrants so there's other aspects of why that we might be interested in and at any given moment the why that's interesting or meaningful is itself situational but there's also a pluralism there so it's kind of like we're weaving these similar ideas of relational thinking pluralistic thinking and then we're finding that we're kind of pluralistic across issues like we're okay with people having different perspectives in a live stream we want to have a plurality in the literature but then at the same time that's that expand cycle and then there has to be the reduce cycle because if we have pluralism that leads to somebody thinking that we should capsize our ship it's not that it's you have to do it in all possible universes this is like simply wrong it might be enough for someone just to say it's not my preference it's not my C vector that this ship get capsized so that's not a policy I want to engage in so it's kind of like this moderated pluralism and this pragmatically oriented dialogue with philosophy and science Stephen and then anyone else and that's this pragmatism and being able to move between scales the the idea and it talks in the paper now about thingness the idea of something being a thing even if it's not a living thing you've got that and I'm curious that didn't actually quite doesn't actually quote the the Markovian monism sort of tag in there as that and be curious if that that was a particular choice they didn't get locked in by that but this this this idea of dimensionality is if the free energy peak principle or is based on entropic change and entropic is is dimensionless to some extent it's just a scalable vegan vector sort of I don't know what you'd even call equality and then it goes into thingness then it goes into some sort of process theory which has to have dimensions at some point you know so you know once you in some ways once you go into active influence and there has to be dynamics across things you start to have dimensions or maybe certainly once you have something that's nested inside something else and that's that I just think that's really interesting I'd be curious what Mel's thoughts are about how the dimensionality drops in and out between thingness and say pure entropy and and just maybe why she didn't put in Markovian monism was that a choice for sort of reasons of not getting caught up in it or does she not agree with it or something like that could you just briefly introduce what Markovian monism is or why is it exciting or relevant to you okay so well Markovian monisms is the kind of overarching philosophical sort of paradigm where the the idea of something being in the world is because there's a differentiation between what's outside and what's inside and therefore there's a sense and the nature of how that inside and outside dynamic is structured even be it a stone in terms of the stones stays as being a stone so somehow the force is inside balance its relationship to the environment outside and therefore there isn't there isn't this dualism of the stone and purely outside the stone as a complete separation there's somehow emerging so everything is to my understanding monism is there's kind of an underlying kind of unifying kind of dynamic that we sort of builds on sort of ideas of David Bowen and the kind of those sort of principles that there's something in the background which is more implicit in our existence if I've said it right but that's my understanding okay it's definitely a big and deserving topic usually the monism would mean that there's basically one kind of thing dualism would be the idea that there's perhaps two kinds of things so the one type of thing in monism whether it's matter or whether it's mind is being manifested differentially these two views often converge because the monist will say yes but it looks like two things or the dualist will say yes but they're interconvertible or they associate with each other so it's a little bit of a weedy area so if there's any thoughts on that we can happily hear that otherwise let's maybe slowly walk through some of these slides and just highlight some of the very nice points of the work itself and then just think about how that kind of spring boards us into 2021 and beyond why not? Sound good? Yeah, all right cool yeah Mel, go first and then we'll continue on I could respond quickly so there's a paper a few years ago that presented one view on consciousness that Carl Pristin was an author on and I think the phrase that stuck was something about the Cartesian the theater it was very much a kind of classical dualist account and this more recent one, right, this is just this past year, right the Markovian monism one, yeah so it's another it's an updated view on it's an updated take on the consciousness problem on the mind body problem that Pristin is also an author on I think it's interesting it's an interesting they're doing interesting stuff with the formalism with the information geometry stuff I used to be very compelled by the the problem of consciousness the mind body problem I worked with that as an undergrad so that was a big thing for me, right, it was an evolution and an interesting consciousness I'm on an out phase of it I'm taking a couple years away from the mind body problem before I come back to it so I'm trying to like keep my blinders on and focus on on like one or two problems I want yep philosophy has that feature of you pull on one question and then all of a sudden you're asking how do I know and who am I, what are we it's because they're all very linked up and kind of like it could be the same with science you make a measurement about the gazelle or then you want to know about the cell but in philosophy it's almost hitting you all at once and there's less clarity about to structure those questions and that's actually part of what I think these conversations help clarify because in the example of the animal we know that there's a certain nesting or interactions between different types of biological entities and so we can coarse-grain and that allows us to make useful decisions like about wildlife management or something like that again with an end in mind and a team and a method in mind we can do useful science and so the question is with the method and the team and the question how can we do useful philosophy and then maybe even move beyond these titles overall and just have the group working on the project with the approach so Marco, then Mel I just briefly wanted to join in on the social consciousness because I'm not out of it yet, I'm still obsessed by it so I think the beauty of Marco's monism or dual aspect monism is because the most beautiful thing for me for active infants is the emphasis and the primacy of mutualism that nothing is without context, nothing is scenic context and that all their so-called identity, so-called property, so-called T-loss these are just heuristics we've accumulated over the years in ignorance of the relational nature of these systems and so the beauty is that these relations of relations of relations, these tangles of relations themselves become thing-like because they facilitate or allow for instantiation of the coherent activity I forgot what the paper is, I've been trying to find it I probably have to dig on Twitter but there's a very nice paper about how some neuropopulations that might for example aesthetically are about encoding cats, let's just put it very simply and they're just constantly chaotic until they get an input that is relevant or comes from the descent data about cats and then that chaotic high-dimensional noisy stuff collapses into a particular order and I think that's also interesting for you Stephen because you often think about spaces and this relates to a lot of well some people think it's too hyped but the work on manifolds and computational manifolds in neuroscience so the weird thing is that these really you know a lot of neurons and population and they're related to a relatively challenging imperative problem but then for some reason when you analyze data you can obtain a manifold you just say well the neuropopulation is just traveling on this low-dimensional manifold so that's very interesting the interpretation of that is another topic but since you asked about the dimensionality I wanted to add that I should go to Mark Mel? Yeah, what you were saying just reminded me I had a thought yesterday I was trying to nail down one thing and of course you know there's this right you pull at one thread and everything unravels I see that there can be a kind of systematic way of doing that I think philosophy is the it's not so chaotic it's right it's using the same kinds of conceptual pools just not with the formal rigor of mathematics but there's still kind of a system it's easy to it's sort of like you know a mean field approximation or solving a system of differential equations approximately right where you're you're taking you've got you know these 11 moving pieces and you're taking 10 of them and you're either sampling over them to get an average value or you're approximating them and you're kind of like freezing you've got all these dynamic moving pieces but you're freezing the values so to do this paper right there's no one definition of what a model is right in philosophy of science there are 50 people working on models and they have at least eight different like distinct conceptions of what a model is and what a model is rests on what similarity is what representation is what analogy is and then you dig into these and there are all these different conceptions of what similarity is and what representation is and what analogy is we don't know exactly what a theory is we don't know exactly what the scientific method is but to drill down on one thing you've got to kind of stipulate you've got to kind of like average over literature and find what it is that's a kind of commonality in what we think about a model and kind of stipulate well you know no one agrees on exactly what a model is but a model is approximately this thing and in order to figure out what this one model is doing which is sort of stipulate roughly what a model is and it it's quite similar I think to what we're doing in math and physics but just using language yes Mel I think you've taught me more in that like whatever three minutes than I learned from all my professors so far so thanks that's fun stuff I agree that was really awesome and again it's like philosophy science we're almost converging on this approach if not just converging on the conversation because when you were talking about how many definitions of models there were I thought about how many definitions of foraging there were for insects or definitions that go beyond insects and then I thought about well when we measured it in the field we said it has to be walking in a straight line with mandibles empty between this time of day and this time of day and there's no single point that we would draw the line in the sand is one meters it's always going to be like you said it's about looking at the distribution literature if it's outside of the bounds of literature you're going to have a lot of justifying to do to the reviewers that doesn't mean it's wrong it just means that you're going to be having to make special claims like you're going to have to write an extra subsection in your paper to justify why you did that but then if you use exactly what the literature did again whether it's right or wrong it doesn't ask to be justified and so people who are operating within that center that mean field approximation they kind of just point outside and they don't need to do an internal justification of their choices or of their approach and then you just said like you have to freeze other things so you did the scholarship to show there was this whole section on scientific modeling and then all you can really do is trace out the if thens you said if you are feeling this way about models then you would be consistent in thinking this and it's not that it's better to be consistent versus have cognitive dissonance it's just that you can do the scholarship and trace out the if thens of thought so just a fun way because there's probably people who are hopefully listening to this and curious about the conversation who are coming from the philosophical side or from a scientific or from an engineering background and so it's like there are similarities across all of our experiences like the need to iterate the need to freeze specific subparts of a model whether it's qualitative or quantitative and then work with teams and with literature corpus that aren't just giving a single resounding answer for what is a model or what is a foraging trip so really just important nomadic ideas that we're broaching here so it's just good to take stock of that so Marco and then we're going to just walk through a few of these slides unmute Marco yeah I was saying you again relieved me of a lot of stuff I wanted to say but yeah I also I could this what we're seeing also in philosophy and science is moving away from moving towards pluralism and I don't think it's a problem that there's no consensus on the single notion of models I mean a model is not a natural client it's a concept we're using because our again non-natural or abstract processes of inquiry necessitated the phrase model and so a model itself is contextual or context dependent so it's kind of a theory or context or norm but I also like the comments of now so we've inevitably accumulated all these conceptual tools and inevitably people will try to find system testing between them and I'm kind of wondering in a way a reflexive way if active intervention also can shed light or make interesting that phenomenon of we can sense it letting the fly out of the fly ball we've surrounded ourselves with so many tools bounded together with systematicity without actual real justification and so we're kind of trapped in that fly bottle because we haven't had any alternative so you know who knows active inference can fulfill we can sense dream so on the fly bottle in undergrad I worked in a fruit fly genetics lab and the joke was you'd have the fly in the bottle and you'd say fly in its natural environment because it couldn't live outside a lab it would get selected out so it was like no we're studying it in its natural environment now that was laboratory biologists who were just very aware of how selection works but it's such a deep insight into like yeah you are always studying it in its setting whatever it is and then you just broached upon this contextual and relational thinking and then also that modeling it's not just about the object or the essence or the natural kind model is a verb so it's like the dynamical and the relational approach is coming and imbuing all of the things that we're working on so Steven's working on performance Mel's working on the history of thought and about the discourse that's happening at a very high level we're all working on different attributes here and then there's something that we can align around like a strange attractor itself and that's that we should probably be thinking about context, relationships, dynamicism, multi-scale models across the systems and that's really the fun conversation that gets to emerge in these moments. Good, I'm glad that people got riled up by that so let's go Steven, then Dave, then Mel One thing that also comes to mind I think that's quite pertinent maybe in the broader context which everyone in the universities you see this interdisciplinarity transdisciplinarity participation mostly it's words that the universities are getting into because they it raises massive questions because how do you start to not just try and contest an idea within a field or you need to work between fields across fields you need to bring in context and multiple contexts and I think this question about how we actually start to understand knowing and at fuzzy levels of knowing which actually opens up potential to bring in people who are more marginalised because they have other ways of knowing and experiencing and they are better adapted probably to survive in some very difficult conditions compared to the research themselves so without making the genetic comparison but the whole ecosystem and bringing it all together so I just wanted to bring up that question of how all of this does relate to transdisciplinarity and the need to do that with the world in the state that it is OK, chill Dave and then Mel and then we're going to move through the slides Jock Panksup made a comment along the lines of your fly bottle he said he could not work with white rats these specially bred rats when he was researching the evolution and the socialisation of emotion because these are completely artificial creatures with no normal rat emotions you can't do those things to a real rat that rat runners do with white rats what he took was your regular old brown alley rats, the nasty horrible rats that everybody's afraid of raised them around people generations was really careful not to select out anything and they just got to loving people that could play with you and come over to be tickled and they giggle and have a wonderful time and he says these are real things these are actual animals that could survive in the wild I could let them go and go out and live the white rats would be just wiped out overnight OK, fair to say that yes it's not about the exterior there's no shades in any species Mel, on anything else you wanted to say there I actually forget what I was going to say but yes, chill it is very interesting and it's always the real environment and who's adaptive and who's not it's not something that could be speculated before the ecosystem emerges because most species go extinct and there's insects that had a strategy for 100 million years and then they went extinct and it was always situational questions though so we're going to pass over the abstract and also there's about half an hour left in case people want to put any questions in the chat so in the roadmap I think it's a good place to pause on our road trip analogy we can see some of these terms that we've brought up in the discussion reflected by sections in the paper and by the way if anyone has to leave that's always chill Mel, go ahead I have to leave thank you I saw the time drawing near Mel, thanks so much for participating in 14.1 and 2 it was just obviously great discussions and apologies for bashing you after you leave thanks for coming on and any other future discussion you're always welcome to contribute your perspective so we'll hope to see you soon fantastic, thank you cool so nice to continue to go around but really interesting stuff we did hear a lot of these things let's look at this table and maybe are there any terms that we haven't really brought up we've talked about for example Mel mentioned mean field I can see a few things already but maybe someone can raise their hand and pick a term that they don't see how it's fitting into the things we've already brought up because these discussions it's not the end all be all it's just a couple of hours so we're just sharing our perspective seeing at the first pass how these ideas are associated and then there's a paper and a literature and a research community to delve into if some connection is interesting so someone's like well with a connection between 6.1 and 4.3 was really exciting so Dave what would be something cool to go into yeah I first encountered the term free energy in economics where it means something just dramatically different it's a good thing in economics but the closest that I've seen to a use with the connotation that we've been using is as anxiety free floating anxiety paralyzing terror and I don't know what's wrong is that off the wall this is fun it's definitely an unconventional economics interpretation of free energy but it just shows how when using terms it's always important that we understand where we're coming from because in a way that's what this paper and also several other people papers have broached upon which is there's a thermodynamic aspect to free energy Gibbs free energy which is like after you ignore the activation energy so whether or not the kinetics of the reaction are such that it does go down after the reaction has occurred delta G change in free energy is like how exothermic the reaction was is what draws the reaction forward and so you have a candle it's like water in a dam isn't spilling over and then the delta G is negative so it wants to burn and then when the activation energy occurs it does burn so that's the thermodynamics free energy there's more to say there and that is actually addressed in Mel's section on statistical and thermodynamics the origins of this free energy term in this setting are from again statistical and thermodynamics but as that became more and more integrated with information theory what's called the epistemic turn there became more and more of a marriage between physical interpretations of free energy and more informational or statistical and even Bayesian and computational understandings of free energy so then we were at this point where free energy was being used pretty fluidly as a metaphor in machine learning by analogy to those chemical reactions but now in the statistical setting and that set the stage for the innovations that Friston specifically made which we can go into but that I hope was a little recap it sounds like the economy drew this free energy term and then just to close on that you said a generalized anxiety yes it's the differential between being too anxious and having too much surprise and being too rigid and fragile though converging to a perfect optimistic model it's not a bad thing to have a gradient descent on expected free energy because just the reality of the situation is that we're not going to come to that perfect model really in any meaningful respect it's just not how but these are there's other little pitfalls and landmines and things and these are all papers that are in a network like the paper that we're going to be reading in 15 about is the FEP realist or instrumentalist is it actually how things are or is it just a scientific instrument those debates are being skipped over in this section but that's a little bit how free energy got from statistics and statistical thermo to information theory Marco thanks just fair warning I'm going to be a bit spectacular here but I was like when people make nice metaphorical connections so they've yeah I would want to build upon your intuition so like mentioned we have thermodynamic free energy and no matter what these thermodynamic free energy can be used for work right but then we can ask okay is the work that's going to be done is it going to be useful or not for the agents and so this ongoing process now we're just talking about thermodynamic free energy this ongoing process of basically channeling or allocating or distributing the thermodynamic free energy in a way maximally useful for the agents is a challenge so for example it can try to challenge somewhere which will evoke a certain inferential challenge but when it fills you get negative after right so so so one part of you sends a challenge the other part can't resolve it so yeah I kind of feel like it's not that wrong in my opinion to see that because anxiety is very specific it's basically kind of like fear but it has no targets right it's generalized so you're afraid of spiders but you have social anxiety because the whole domain of social interaction is just fraught with uncertainty but without the availability of a specific reason why you're negatively uncertain or uncertain in a negative way that makes sense Marko I'm going to build on that because it was really helpful so you mentioned how the FEP has to be deflated because it's like generalized bad it's not the same as bad but it's something that is being optimized whether you think of it as minimizing or maximizing depending on whether there's a negative sign so it's the kind of thing that we're working on that's the key point and it's inference what we saw here is that we're going to draw physical analogy from the way that chemical reactions indubitably go downhill the candle doesn't unburn so we want to design governance structures information schemes economies that are like a candle and they just irreversibly according to our experience they just play out you don't have to specify the micro level detail just as an analogy so we want something that's going to go quote downhill in governance space the thing is we don't just want to simply do inference on it so in the machine learning area it'd be like okay what's the data set let's fit the estimator let's reduce free energy so that we can fit the estimator on the descriptive model like on a big K-means model how many clusters are there then that will be like an inference model we're actually doing free energy gradient descent on what policy selection that's where control theory and cybernetics comes in because we're not just doing descriptive analysis of parameters of large data sets we're actually doing prospective or anticipatory policy selection in light of a deep knowledge reflected by embodied priors in a niche so that is the action-oriented turn where we go from again just pure free energy minimization to do inference via tricks and heuristics like variational bays and Friston takes it to the next level by doing a few things not the least of which are incorporating action conceptually doing inference on policy selection which we've seen embodied in the models where like the pi is influencing the b so the way that the state estimate is changing over time the way that s and the b matrix are related is being modified by policy so policy is what free energy is plugging into so we're doing free energy descent on not just descriptive attributes of the world or any smaller large data set but we're doing a prospective free energy gradient descent on policy because that's about our world model and how the states the world change for some people I know I was a little repetitive but that's really some of the key history and that helps understand how statistics and thermodynamics and Bayesian inference which already is a big synthesis to make how all those areas and information theory and computation and everything that's kind of quantitative that we're talking about how did that get involved with ecological psychology control theory cybernetics and then even into this deeper relational way of thinking so that's a little bit of that and it's an intersection that will you know embroider over many times but it's really a great story and it's awesome that we all wrote it up in a way that isn't just back room lore but it's something we can be like yeah it's in 3.1.1 in Mel's paper so that's what's great about literature Steven. Yeah thanks Daniel just got to take a breath that was really that was quite a lot of useful information I agree what you're saying there and I suppose one thing that is good to bring it back to that Gibbs free energy is you know there are times when for instance when ice freezes and the actual free energy is such that it acts as a force it acts as a force so that the the lower energy molecules are not necessarily purely favoured and it takes a slightly different structure and you know you end up forming ice of which you know it goes to a state which is not necessarily energetically favourable you know so but that dynamic is still something that can happen over time with equilibrium so you've got this ability as you mentioned to sort of to select policies outside of equilibrium to keep going and to work with that but I think that what the free energy does and what this talks about that's a bit different to how a lot people talk about emergence and stuff is they normally work with these phase shifts and this kind of like if you put enough molecules in a box and get them to be constrained in this way and that it will emerge that you'll get the ice that the ice becomes water and that seems to be like the prevalent way of thinking about emergence sometimes in kind of economics and some other cases it's just the interactions under the temperature and pressure but what this gives a way to think about how it can be innovated upon and thought about in terms of policy selections happening at multiple scales and the challenge is it depends which scale you go at and you're talking to to try and explain it to someone so the cell versus the organ versus the organism but I think that that's something that this really helps to unpack because you can't deny that this is not just emergence as you put everything together and they'll interact and it involves some sort of sentient way of building up but starting from a non-intentional base and it's got both which I think is really cool Thanks Steven, I want to mention two points, so you talked about a phase change and so that's a reversible phase change with like melting and thawing so it's a little bit different than the candle you know melt, thaw, resolve itself into a dew, that kind of stuff so slightly different with the chemical reaction versus freezing, just from the chemical side but you mentioned it was not favorable for the water molecule to freeze from the liquid into the solid phase so go from freeze, from liquid water into solid water but actually it is it's that as the thermal energy reduces all of a sudden it clicks into a new differentially configured arrangement and so that's what the phase is is like when you're at absolute zero assume most substances are solid like they're in whatever their ground state configuration is, which is like a close packing, it's a crystal of some kind it's a solid and then there's some temperature T where that crystal melts where all of a sudden it's like the iron is vibrating it's a crystal, it's a crystal, it's a thousand degrees it's two thousand degrees, I don't know what the melting temperature is but at some point the iron is breaking out of its crystal and goes into the liquid phase and then there's still the temperature attraction of the different molecules in the liquid even the iron and then at some point it vaporizes when you're even hotter than that in plasma and so on so these phase changes are actually they're about the temperature and then if you want to go to that fallacy you're kind of mentioning it's like people think that you're going to get a quote phase change by adding more into a box or by stabilizing it but it's like no if you're at the wrong temperature you're never going to get ice, you need to do an enormous amount of pressure to get solid above zero, you have to crush with like millions of pounds of pressure but you can make water into ice at like five degrees Celsius you just have to put a massive amount of pressure and so the question is how do you do it with zero pressure that's by changing the temperature and that's where this multi-scale Yvonne in the chat 2282 Fahrenheit for iron thanks for that super clutch then the second point just quickly and anyone else then Sarah is which scale and you said the cell or the person in the society and to me that's making me think of Helen Longinows book on pluralism because she discusses how the different fields like the sociology the geneticist the molecular biologists they have stories about aggression and sexuality in the case of her book and depending on which method you're familiar with which affordances you have at hand which aspects of the world you're sensitized to you're going to tell a different story a molecular biologist is going to have a protein binding answer to for example a pandemic a sociologist is going to know about some social intervention and so there's all these different people and it's not just that one person's will do is like the right model someone else's is the right one that's kind of this question is which scale not for any one person per se to say but those are the questions that we need to gravitate around to make action that isn't unduly biased towards the molecular towards the social etc but rather something that works across those levels so really interesting stuff thanks for like always highlighting how it comes back to to bigger questions like that and then Marco yeah I mean maybe just to throw some more monkey wrench into this something about what you said really really reminded me of the phenomena of hysteresis and I never really thought about hysteresis this way before but I was like oh Jesus I mean it is true I think that hysteresis is an expression of hysterosity you know of history and so you know with FEP I'm not sure with active inference maybe but like I often ask myself how these various you know like you have an entity as you know by way of your model or natural kinds or whatever it is but you know and you hook in these entities you're kind of you know weaving or whatever these entities together by way of something and hysteresis has always been interesting to me because it's like that extra bit of dimensionality that phase if you want to call it that can hook into something one of its neighbors and so I often think about the like let's say with FEP or active inference you know like if there's only like one interface that can be used to hook to a neighbor you know maybe it's phase change or whatever that moves out to an adjacency so like how many different degrees can be used to connect as an interface to various neighboring agential type type thinking and I know that's like left field I don't but Daniel if anybody can take it somewhere you can or Marco I don't know Marco definitely take the first swing at it or anything else you want to talk about Thanks I'll do my best so let me just grab my thoughts before I lose them Sarah but what you just said I'm not sure if you find it relevant but one of my favorite papers is from Kuhnigsberger and and other people about routing strategy so I personally am very fond of the conceptual set of flows and paths and routes you can also see it's navigating a certain policy space so this paper a spectrum of routing strategies effectively is tackling an interesting question of when you have a given system how much non-local how much information in its adjacency in its context doesn't need to incorporate right so as it expands let's say it's domain of discourse or its inferential context as it expands there will be a non-linear growth in let's say the rank with the involved elements as it works and that's a cost but at the same time if you don't expand maybe you're just not able to resolve a certain inferential challenge and so this tradeoff of how local and how global or how expansive you want to go for a certain inferential context is I think related to what you're saying right so how many degrees can you afford when it comes to selecting how you go about selecting a policy if I understood your question but I linked it in the YouTube comments thank you and the other thing what we're talking before thermodynamics oh yeah I am still not mathematically competent enough but for those who are I linked also in the YouTube chat to an article by John Carlos Bayes who is a huge figure in mathematics at the moment and he wrote something about entropic forces and when I discovered this it was just mind-blowing and amazing because basically what he's doing is licensing talking about forces as governed by entropy so it's basically about whether the temperature is static or whether it is either thermal static or motion was it anyways but when the temperature is static then the forces are better described in terms of entropy which I wanted to emphasize because we had this discussion a bit with Steven's point earlier and I often like saying to people when they don't like normative principles there's always at the very least the ultimate normative principle which is the principle of least action the physics there's always that no matter what every system has to be governed by that so given that we won't deviate from that law we can say the brain is governed by the principle of this action but clearly it also is governed by a imperative to systematicity or mid-to-infant this constant co-regulation and so the question that I wanted to explore is kind of how does that link together how I like to quote it be phrased it's basically these kinds of systems are effectively cultivating nature to do the work for them how do you transform the dynamics governed by by the ultimate principle of this action how do you nudge that into also being systematically conforming not just physics but also your inferential imperatives does that make sense so this is a very subtle and really interesting point Marco thanks a lot let me try to trace a little of that so you were pointing out how bases work on entropy and some other aspects of information theory and the fundamentals of physics it's almost like integrating statics and dynamics into one more unified field and or you're muted but is that correct or where did that I'm trying to trace how this comes into the thermodynamics side because continue though I'd like to understand that a little better yeah I mean honestly I think we should have discussion maybe later it's actually a brief post I don't sure if he has a formal paper about it but in general Bayes has a lot of interesting work about entropy and biological system again you know way out of the element there to be for me but I would really be curious to see what more formally competent people think about these terms but the thing about entropy force is just a very simple block post it's not that complicated okay we'll read it, discuss it and maybe invite him or anyone else to be on a conversation so we have about 10, 20 a little bit of time left so let's think about some thoughts, questions we're taking forward with us but then we can kind of flip through a few more ideas Sarah to your idea of hysteresis and historicity which are both two kind of like similar sounding words but hysteresis is just the system where like when something happens it doesn't return to exactly the same state and historicity is just the idea that things have a developmental history like even if you reelect the same person they're older it's a different world etc so then you also mentioned the connectivity of agents and then that made me think about social agents and how the social networks have a hysteresis element to them like if there's a fracture it's not like you just zip it back up like a sweater it's like gonna be something that has to be recovered if it's recovered exactly or not exactly it has to be recovered in a different way so I kind of see where you're coming from with the hysteresis and the connectedness of agents now where does active inference play into this if it is related I think there's some elements because we're using a lot of these words so it's kind of like these networks of thinking about these topics I'm not sure where one would take that going from the idea that it's the networks and the relational insight that define function of the system and then hysteresis and thermo I'm thinking of a few things though Marco yeah well I mean I would love to hear thoughts about it so I think there's a bit of a tension because often the form is described in relation to market change which is explicitly about just being dependent on only the last day but I think what Sarah was kind of alluding to quite rightly so is when you have these not just one system but interconnected systems then the a past state of a particular system might effectively come back with the detour as it were and I think that's kind of what she's pointing at even though you could of course define the market chain for every point etc on a different level you can say that the past states for example that system expressed an influence into its larger context and then later another system that received that influence responded with again return to that initial yeah this is actually what I'm always obsessing about is and you can analogize it to complex numbers as well you know like the hysteresis adds that it pops out an extra dimension that something else can hook on to and that's what I'm always thinking about with respect to you know I don't know not even well versed enough to even say FEP but what I understand about active inference you know like the model seems to just creep in a forward direction but there's no side chains like there's no side hooks and so I'm always like but how does that how do you do I don't know like I'm and it may just be my limited understanding but I come from a place of thinking a lot about hysteresis and and stuff and it and I can't quite get those the math to answer my questions which elaborates on side hooks I think that yeah agree what would a side hook look like not a sky hook for those Dennett fans out there well what I mean by like let's just say you know like you have real numbers and then you have complex numbers and I think of the complex plane as like a side hook where you add an extra dimension basically where other things can interact with that number space by way of the complex side hook but it might be a too limited paradigm for me to like I'm throwing out these random ideas that may or may not work together but that I don't know it's it's let's say this is like totally sci-fi you know like you have members there's and they're in the but they're going in like voltage amplitude one direction but then they have different paths as they go forward versus reversed and and so that that phase that kind of phase relationship might offer a relationship with something else that's related to phase or related to you know whatever the physical dynamics might be around that so I don't know like when I think about active inference you know like they're just like well what's the happened in the past and what's going in the future and what are my prayers but there's nothing on the side that sorry it's as good as I can do I don't know if that's helpful you know you riled you riled everyone else else up so it's great so we're gonna just slowly we don't need to end exactly on the hour if anyone is gonna leave on the hour that's fine just maybe speak your piece as your closing piece but we'll stay on a few minutes after because yes this is a really nice topic so Marco Dave Steven yeah thanks I think I understand what you mean with sight hoods and I think maybe what you're bumping against is what I also find problematic so it's good that you raised this so the equations the formalism for free energy and for FEP and active inference they are stand alone they're like the most naive equations you can have right but the whole point is that when we apply them we have to fill in the things that were previously naive we have to for example only now only very recently we're going to multi agent systems with active inference but that was always implicit in the research as in from the first paper in 2009 in 2010 it was already implicit because it explicitly talked about how it would be able to scale how it would be like nested right it's all these different systems all doing transient minimization in relation to each other and so that that co-dependency right in the mutualistic relations those are the scientists I believe so for example if I'm not abusing maths here you know the sight hood for complex numbers is it's imaginary until you square it right so it's a potentiality until a certain operation occurs so in the same way when a particular system expresses an influence into its context it isn't affecting that system anymore until something happens when it returns right so so the fact that it's always situated in an extended environment that a particular system is very naive about actually that it can't know everything exactly that it has kind of trust or upload certain influences to its context is kind of the sight hood I would say does that make sense interesting once I look at the math I'll keep that in mind and it might start to that's really helpful don't look at the math that's the point right you'll hear all opinions the maths describe a general general principle for how they are by what they are governed but they are governed in terms of states and relations to other systems so I kind of confused because you've alluded to it many times before how they are all connected to each other right so that's not in the math oh wait I'm just going to continue with the order I think it's a little bit in the math though I think if I keep your this is how I think or how I learn is I keep these kinds of phrases and statements in mind and I look for them in the math and then it becomes clear so I think yeah and you've talked also your metaphors are great you know like you've talked also about context and you didn't say this but my brain did this like collapsing or like something becomes so it's really helpful to keep these things in mind as I'm battling my way through these formalisms so yeah I'm just going to ask Daniel because Daniel is way smarter than me I know there's a method of notion relevance here because of matrices you have a marginal the diagonal and then the other entries are for the conditional conditioned situations and that also reminded me of the sidehooks oh your physics we're going to return to sidehooks but let's go with Dave then Stephen okay hysteresis and sidehooks the first the most elaborate discussion I've gone through on hysteresis is in Rene Tom and Chris Zeman on morphogenesis structural stability and catastrophes in there the hysteresis are referred to hidden states or relatively inaccessible states so these amazing things like the financial collapse in 2007 and 2008 where you just absolutely fell off a cliff and none of the models said that would be possible or in say a psychotic break the most stable guy in the community just goes totally wild or good things too a really just conversion that turns the slaver into the guy who wrote amazing grace well now more recently Gerald Adelman emphasizes relatively inaccessible states a tremendous amount both in his work on immunology and in neuroscience the term he uses is degeneracy which he borrows I believe from analysis numerical analysis to mean something roughly like a function whose inverse is not a function on multiple ways of coming up with an activity which from the outside looks exactly like another activity but it's implemented in a completely different way and then Professor Friston's paper that was suggested reading a couple of weeks ago on where he's talking about as ascent climbing gradient ascent he says it seems as though what's going on inside the Markovian Blanket is it's learning about its environment but it can't observe the environment how can it be learning about something that it can never observe so I don't know where to go with that but it all sounds like it kind of would fit together and make a nice paper thank you Dave for mentioning that and lots of great points Steven and then I'll also go with the side hook and hysteresis discussion yeah I think this one thing this discussion really talks about is the relativity between what the organism Markov Blanket the generative model and the niche and the environment so the world out there can follow all the rules of physics and it can have a lot of side branches and it can go and behave in all these sorts of ways and which we don't know about some we can track through the entropy some we will know because the niche will hold it I my cup will still be on the table so some of the kind of the stuff about least action has a benefit of maintaining a trace more clearly in the environment so that there's an inch I think this is where it's quite interesting is how much the way it coming in it from both directions I'm starting to see what Carl Friston meant there in some ways is sometimes it's about the model on the outside and sometimes it's what's out there being shaped and humans we're doing a lot of that so when we shape something out there what is available because they talk about the adjacent possible a lot when they're doing work with people so is it that maybe as an animal you can infer much more diffuse, abductive kind of things if it's within your sensory range however needs to be within the adjacent possible whereas in the niche you can do things at distance you can fire something over the hill next week and you see where it landed you know which is kind of so anyway I thought that's quite interesting I think that really does speak to this two way movement and the fact that some pieces are more about it changes the name of the game for emergence for sure beyond pure physics into this non equilibrium and I'll put a paper because I found a paper that talked about non equilibrium chemistry and it was interesting it said that there's almost been no non equilibrium chemistry until like they only started doing it in the 80s and 90s and it's like it's all this at equilibrium type work where you go from one to the other and then of course all life is a non equilibrium so it's kind of interesting because you've still got both going on it's just happening in the niche Steven excellent point let me go on just in reverse order there's a few excellent things you said long range action through niche modification that's stigmergy the ant can't do long range instant communication but it can instantly put its abdomen on the ground and then three months later another nest mate could instantly detect it so it's at an instant with a behavioral focus but it's actually long range with respect to action of that one nest mate on another so then there was a discussion by Dave about how Friston was mentioning how does the organism learn or how does the system learn when it can't directly directly measure the states and that's everything we've been talking about with the observables the organism is actually getting the observables and using those to do state estimation on the hidden states of the world and then it turns out that the observables don't have enough instantaneous information to reliably identify what you care about which is the latent causes of the world so you have this imaginative generative model which includes that space of the adjacency you know the adjacent possible that's weighted by how likely it is and a bunch of other stuff that makes it more like a candle burning than just a marble on a hill so it's a little bit more nuanced because we're talking about something that's like a Bayesian computation at least as the scientist uses it and again that distinction between is this quote what's happening realism versus is this just an instrument that scientists are using in the world that's going to come up next week in 15 and so that's like really interesting alright so the side hook I thought if there's a idea with many sides then maybe it has many side hooks because like it's clearly by analogy to the complex number so I like this idea though of jumping out a dimension and that seemed like it was related to what Steven said with the adjacent possible imagination it's like if there's a wall in front of you and then you're only on the number line you're going to hit into the wall but then if you are logistically in a different dimension then you can go around it or you can figure out whatever policy it is so policy and imagination is like popping out to a different dimension where then you can enact a policy but if your imagination isn't connected to the manifold of reality your action selection it's like a fantasy it's like if I could fly I could just leave this building it's like yeah but E matrix is not that way so it can't happen so it has to be imagination with the affordances and then also just one last point on the statistics so the Markov chain that Marko I believe mentioned so like a Markov chain with a memory of one would mean a statistical process that's only dependent it only has a memory of one whatever the previous state was that is what influences it and then it's almost like as a statistical model so again whether we think of this as a scientist instrumentally with a model or the organism actually embodying this model let's just say that most things have the influence of one timescale but then once in a hundred times it has like a range of a thousand so there's two models you could make you could have a model with a Markov chain of a thousand so the different action sequences of like every single possible last thousand days and that model is very computationally complicated but if you had enough observations unrealistic you'd be able to fit that like to a T versus the more simple model which is just give me the memory of one so I can have just one bit of memory and then I'll be surprised when the other thing happens and so that is the compromise of the modeler of model accuracy minus model complexity is do you make this Markov blanket like simple like a heuristic or a bias and then be surprised by particulars or do you try to fit this ridiculous particularized model for every single case but then that loses all efficacy because then you're tied up so specifically with the very specifics of the model that you learned that's like overfitting so I know that's a ton of points but those were all like really awesome comments and I think we're getting to a lot of the stops on this little road trip but let's just head to the last section but anyone if you have any final thoughts or questions that was yeah basically the only other slide was this previous one with a little bit of and then just the usual questions but I'll just say thanks for participating and stay in communication with everyone and then let's pull back to the roadmap or to here and then each of the four of you thanks so much for these like just amazing conversations so each of you if you would like to just take a you know we'll go one last round of people want to so whoever wants to raise their hand otherwise we can go to this one I guess so this slide was kind of starting with the physics on the bottom with the Fokker plank and a flow so that was what Dave brought up and the way that this conceptualizes a flow like on a hill you can have like two orthogonal curls two orthogonal factors one is kind of curling around the hill at the same elevation that's like what you see on a camping map like the ISO contours and then the other one is like a ruler that's right on the hill like straight up and down and so not to go too much into physics but the ruler is like the delta G it's like the gradient descent like is the candle burning or is are we in a flat spot so that's the ruler is like are you skiing fast downhill or are you not skiing fast downhill and then the ISO contour is like is a steady state that's like when you're when you're doing a chemical reaction there's always a forward and a backwards reaction now the keq the equilibrium coefficient for the reactions completion might be 99.999% like a candle but actually the completion is really high because it's so exothermic but Steven from chemical manufacturing you know that reactions that are barely favorable you're going to end up with like 52% of the product that you want and 48% you don't want or something like that so that's that decomposition right here with a Helmholtz decomposition that there's the solenoidal curl and irrotational or divergence components so that's kind of a thermo-info connection so when people are talking about gradient descent in machine learning they're talking about like just like chemicals are allegedly physically skiing up and down these slopes also in machine learning you're doing informational skiing on these slopes and then that's where you get these Bayesian statistics and thermodynamic ideas the key insight and the integration with control theory happens here basically if you're going to be maintaining that far from equilibrium steady state better than all the other models like you in the niche then you need to have a generative model not just of state estimation because that's sort of like the low-hanging not really fruit even if it is low-hanging or not it's just not really fruit because action is what the agent needs to do to stay far from equilibrium whereas a machine learning model on our computer it doesn't need to behave well to survive because it's like not really the kind of selection pressures that we are but we need the action element so then this imperative to be self-evidencing as an information forager means that you get all of these cybernetic characteristics and that you see systems that act as if they're doing the optimal experimentation given their priors reflected in the generative model in their niche so that's where you get this sort of like forager information optimal information gain principle because that's what's needed to keep the action selection effective because they're far from equilibrium steady state that has to be subsiding despite these things that's what gets subsumed in free energy principle as discussed in Mel's paper and then on the top are corollary process theories so it's like okay whoa I guess that would have to be true for any candles I saw out there how did that energy get so organized so it had to be an active candle or it has to be part of the extended niche of us right there's no candles at the watchmaker analogy and then how does that happen and these are falsifiable process theories because theory falsifiable empirical unique explanations and predictions this is where we see the falsification happening and so we've discussed in this paper like some of the already kind of I guess less in favor process theories like the Gibbs filtering or Gibbs sampling which is done on computers but is not really tractable on biological architectures in the way that these more established or I don't know productive not falsified yet there's some good critiques of Bayesian brain in the paper in 15 as well as in the 2018 interview in alias with Friston but yeah and then what else because the whole idea is that like as free energy principle gets short up and connected not really to a bunch of other things there's going to be other process theories active inference is about some kinds of systems but who knows there's other ways we could think about it and so I don't know what does anyone think about that other than that we'll head into our final section Marco yeah I guess I can add hopefully pleasant perturbation so one interesting thing to raise here for this slide is it's conceivable that active inference might actually be more solid in its conceptual framework or structure than actually formalism for example you could maybe for example not choose KL divergence as a distance metric but other distance metrics because the KL divergence is I think also in itself idealization right in the end you're going to have very a lot of messy stuff in real systems such that even if you're going to model certain belief distributions in the action implementation it's going to be extremely unlikely that it perfectly follows this particular conception of distance to be minimized and so I guess I was urging for this question of what can come out from it that I personally just like to encourage people to do to be playfully exploring different variants of the free enterprise because personally I believe conceptual model taking it as a conceptual model is better than purely a formal model but that's personal well thanks for expressing your thoughts as an action suggestion Steven and then Sarah yeah thanks for that description Daniel actually I'm going to go back and be able to use that as a really nice reference point for lots of information I like this top piece and I think yeah the Bayesian brain was there as this kind of integrating idea and then they thought well let's have this predicted processing process and I think Marco was saying like some of the assumptions then for it to go to active inference you know the math is there to make it solid but we can think beyond that I would say active inference is really an active inference which goes another level that's not a math thing so if it's an active inference and then does there need to be like I say what else is there I actually quite interested in potentially looking at some ideas of some spatial layers like we had that thing about projective consciousness and some other layers like what are the what's the layers beyond that need to be there for these processes to be plausible and who knows that would be something that we'll talk about in a later date but I like this slide could actually be something that could be played with in the future as a learning tool because I think it's quite helpful great thanks so much Stephen Sarah oh the trail might be kind of cold now but you super had my attention on on yours you know you had your arrow moving around and you said you know this is the place where there's like a connection between thermal energy and free energy and I was like where because I mean you know math wise because I that's always just been the super separate world that's annoying and I can't really put it together so I want to find where that is and then go find it in the math that the word description in this really nice slide and then the math and try to figure that out great call great call I agree it's why one of our projects just to sort of like talk a little bit about the lab since everybody here is basically a participant in it it's one of the reasons why having the ontology and the structure of knowledge for the equations is really important because we want to be able to say this equation in this paper is where the variables in the you know the F from this paper in this other math area actually we're going to call it this letter here's this and that's where there's a point of contact with this and then is this one following from that one is that one following from this one those are the kinds of things that truly are in the frontier of what we can understand and then understanding how to bring directly from these analytical mathematical phrasings to the control theoretic in the policy selection that's a non-trivial it's not a velcro interface that's a serious maneuver that involves also moving through as pointed out equilibrium and steady-state and far from equilibrium non-equilibrium all these different things in this metaphor or is it not realm of info thermo so lots of exciting work to be done and traced good points and then also yes even just you pointed out it's active inference imagine if someone said in 2010 well active inference is simply one time step and it's XYZ it's like it was X and then now it's something else we've talked about and activism we've talked about affective inference now is the variable in affective inference equivalent to answering philosophical questions about affect no but neither was the variable affect in any other modeling framework which again makes us pull back and just look at that instrumentalism versus realism debate so just the realism would be this is actually telling us how the world is arranged and what it's doing versus just this is a framework that is being just like Mel's paper this is one of the main points is with respect to science it's in the realm of science but what is it in the realm of science well it's not one of these falsifiable theories up here it's something that's more like natural selection at that type of tier but it's like a theory generator and that's why if we look at the roadmap there's um 14 sections on kinds of models because this is like the system that this is being slotted into Mel could of or anyone potentially could go the other and say no it's not part of science it's part it's okay so you can debate about where it is with respect to the demarcation problem but since this is the paper we're in the journal club on we're saying given that that's the view being defended the literature has to be integrated in is not theology it's not poetry it's actually scientific modeling philosophy of science literature so then that will help us connect it to like okay but what does the variable in a linear regression mean is that realist versus instrumentalist interpretation nobody would say it's realist it's not there's not like a line in nature it's making one you know group of plants a different shade than another group of plants so that's sort of the process that this paper is engaged in is interfacing better so that we can understand what kind of thing the FEP is and there's a few exhortations like this is it true what is it true of how do we empirically know that it's true and Mel argues that that's a category mistake so it's like saying any other category mistake now think about these questions in relationship to the FEP being like so again great paper we'll just go to the closing thoughts now but just thanks everyone for being a part of 14.2 and all the other sections because this was really like two great discussions I'm looking forward to people can raise their hand if they want to make final comments but I'm really looking forward to the discussion with the authors Hippolyto and Van Es on 15 for both weeks and that will be interesting and we'll confront this realism instrumentalism discussion and then also the .0 videos will be a dialogue so hopefully more and more people will get involved with making the summaries of the videos and introducing them from different perspectives different lengths and each paper is unique in set of authors and set of people who are approaching it so it's been really awesome to see people engaging with some of these truly longstanding and fundamental questions in a new light and then also just co-creating with first just participation in live streams maybe even just by watching but then anybody who's watching live or in replay is always welcome to join or find out what we're doing in the lab it's just been a really good we grew a lot during 14 and so it's good times so any other final comments that the panel wants to make yeah I'm going to grow some neurons now peace Marco go for it yeah I just wanted to say ever since I first joined I can't find the survey you keep mentioning I have no idea where the survey is so maybe get to us and the viewers where to provide feedback sure well the survey for live participants is in event calendar so it's in the calendar invite and so yeah for those who are listening to the very bitter end the drags of the active stream basically the single source of truth is the calendar invite it has the slides that are commentable the paper the video URL and the feedback form and then the feedback form you know it's awesome if everybody fills it out but then the feedback form for everyone else is just email comment threads YouTube comments just leave feedback publicly or privately will always respect both public and private feedback and let you know how it influenced our trajectory so but for live participants we have a special structured form that we can always modify and evolve as well so that's a little insider baseball but always happy to talk about that because what it is we do is it not Stephen and Sarah on the discord and all the decision making maybe it is maybe this but it's been fun so thank you that was great very good explanation and I enjoyed today right like so I've got I've got to grow grow a spare brain and then I'll fill it up and put it back in my head just off source it put it in the paper all right good to talk to you all I'm going to terminate the live stream