 Hello, everyone. Welcome to the Active Inference Livestream. It is Active Inference Livestream 14.1 on January 19th, 2021. Welcome to the Active Inference Lab, everyone. We are an experiment in online team communication, learning, and practice related to active inference. I'm going to share my screen for the participants. You can find us at our website on Twitter, email, YouTube, or our public keybase team or shared username. This is a recorded and archived livestream, so please provide us with feedback so that we can improve on our work. And whether in the live chat or the comments, just feel free to share whatever you're thinking. All backgrounds and perspectives are welcome here. And as far as video etiquette for livestreams, we'll mute if there's noise in our background, raise our hands so that we can hear from everybody and share and use respectful speech behavior. So for the Active Inference Livestreams in 2021, we're going to be having these regular Tuesday, 7 a.m. Pacific times. And then we're also going to be having these special sessions like the model stream and several other kinds of events that we're really excited to be planning. And go to this big red link and you'll find a spreadsheet. And in that spreadsheet, you'll find the details on past and present and upcoming Active Inference Livestreams. Here we are in 14.1, discussing the math is not the territory with author Mel Andrews. Next week, we're going to be discussing the same paper for a follow-up paper. And then we're moving on to a different paper. But for this week and for next week, we're going to be in this awesome paper and this awesome group of discussions. Today in Active Stream 14, we're going to have introductions and warm-ups. Then in the sections of 14.1, we're going to be going through the paper, the math is not the territory, navigating the free energy principle by Mel Andrews in October 2020. And we have some slides with aims and claims, abstract, roadmap, et cetera. But this is pretty much going to be whatever people want to discuss today and whatever questions we want to address in next week as well. So if you have any questions, just save them and submit them however you can and get in touch if you want to join live. Okay, awesome. So let's start with the introductions and we'll just give a short introduction or check-in and then pass to somebody who hasn't spoken yet and especially if it's your first time, we would love to hear what your background or interest in active inference is. So I'm Daniel. I'm a postdoctoral researcher in California and I will pass first to Mel. Hi there. I'm Mel Andrews and I think for better or for worse, my doctoral work is going to end up being largely about free energy principle and active inference. So that's why I'm here. We'll pass it to somebody who can then introduce themselves. Marco. Hi there, thank you Mel. Hi, my name is Marco. I'm from Holland. My background in psychology and cognitive neuroscience and German master's also worked a bit on its engine principle. My interest in active inference is well, in most of things, I just believe that it really has a potential as alluded to and written about by Mel in the paper which doesn't today. In general, just a lot of potential to understand a lot of aspects of the world in society that we so far haven't been able to really grasp with alternative models. So happy to be here. Then pass it to Ivan. Hi, my name is Ivan. I'm from Moscow, Russia. I'm a researcher in system school. I'm an engineer, a system engineer. And I pass it to Shannon. Hey guys, I'm Shannon. And I'm working in cognitive cognition and brain stimulation. And I'm also interested in social cognition and how we interact like in a musical ensemble and how the Fringe principle applies to both scales of those questions. And I pass it to Alex. Alex, we can't hear you. Distance X, Alex. But you don't appear muted. Let's go to Alex. Yep, now we hear you. Go ahead. Yeah, hi everybody. I'm Alex. And I'm in Moscow, Russia. And I'm a researcher in systems management school and trying to find a way to find integration between active inference and system engineering approaches. And I pass it to Alex. Ivan. Okay, thanks. Thanks, Alex. I'm Alex. I'm interested in... I'm a philosopher, I guess, interested in free energy principle and active inference and sort of fundamental questions about mental representation. And lately I've been more interested in applied work and sort of combining, eventually maybe combining these approaches with other approaches to machine learning and active inference as like a technology. But I will pass it on to Dave. Dave, you muted. So we have Dave, and then we have also... There we go. Tim, Steven, and Phil. Dave, I'm retired from information technology. I live in the Philippines. My background is in cybernetics and machine translation. I'm teaching myself neuropsychoanalysis and I was so pleased to come to Mel's paper because I've been trying to figure out what is this thing after all? It's like a metaphor among metaphors among metaphors and maybe you can tell us toward the end of the presentation whether that has anything to do with what you're saying in your paper. And I can't even see who else has not spoken. Philip, there we go. Hi, I'm Philip. I'm studying theoretical biology in Prague and I'm interested in computational neuroscience and I joined because I only read Mel's paper and I found it interesting but I don't know anything about it. Perfect. Let's go to Tim. Hi, I'm Tim. I'm in Toronto, Canada. This is my first time joining in the sessions. I'm really eager to learn about this stuff so it's pretty fascinating. My background is in science and engineering, specifically I guess electronic systems and software and that kind of thing. And I'll pass to Steven. Steven or fellow Jitser. Okay, and Sarah, have you gone? Hi, I'm Sarah Davis. I'm in Berlin and right now I'm a master's student in philosophy of science but before that I was an artist and the through line I guess that's interesting to me is the relationship between information and materials and media. So I'm interested in like reservoir computing and how and when you can separate those things in what is life, like all these funny things that are circling around that. So active inferences, just yet another useful thing that I want to understand better because it seems to connect to a lot of other disciplines. Starting to know each other through the current seeking circle or big current seeking fund. Cool, it'll be interesting to hear about that soon. And then I see Blue or yes. Hi, good morning. I'm Blue Knight. I'm an independent research consultant. Nice. Is there anyone who has not introduced himself? Otherwise we're just going to keep rolling. All right, great. So let's go to our warm up questions and I'll just put up the first two which are what is something you're excited about today and what is something that you liked or remembered about the paper and so people can just raise their hands and we'll go in order. And until I see a hand, I think something I liked or remembered about the paper was that there was no figures. So from the point of view of usually reading papers with figures that was very notable but then also that there was clarity and structure despite figures which are often used as a, not quite an aid to structure a paper but it's how we often think about in empirical research like, oh, what is figure one going to show? But there was clarity in this article without the usage of figures. So Marco and then anyone else who wants to continue? Yeah, I fully agree that it is unique to clear. So I think kind of lacking behind on the literature but for me, Mel's paper is the best paper so far when it comes to contextualizing French and Spanish and like the title implies really mapping it out and textualizing in the big complicated epistemic landscapes that this all connects to. So huge props to Mel for that. I think the only other critical piece I've read is from Gio Colombo which was more unwieldy. And I think the kind of importance of mapping really shows in this paper because it opens up a lot of directions to think about the FEP which stands in stark contrast to a lot of other papers are read which kind of take it as authoritative and then just move on instead of actually trying to interpret it. And so Mel does a great job at guiding the reader in understanding the nuances and the questions that would arise when you see those nuances. So excited to hear about people's thoughts today. Cool, very nice. Tim, and then anyone else who raises their hand? I thought that I was noted to myself just, you know as I was reading how clear the writing was I thought the writing was really clear and expressive and the explication really well ordered. And there was kind of like, I guess, a humility. I remember somewhere in this content I saw the idea of the list of epistemic values I think it was, you call them. And humility was one in the list and that kind of comes through in that paper the way that there's no sort of diving for like a discovery or a conclusion it's really just laying it out and giving it to you to leaving the questions open which was really nice. Yep. And also lots of nice philosophy type things like the idea of defending an idea. And so it's not common to see in an empirical paper like we're going to defend this. It's a lot more like a battlefield of ideas in philosophy and in science it's a little bit of a different structure. So Sarah and then anyone else? To like lather on the praise of like to Mel's paper but yeah the writing was really good and actually when I first read it I feel like I understand a lot more than when I first read it and so now I kind of feel like I'm happy to be here because I need to read it a second time I've learned a lot more. But yeah the exploration of kind of the beginning of the exploration of all these different models it was saw a lot of potential. Cool and that's kind of part of the idea with the dot one and a dot two is sometimes you read the paper and you don't talk to someone about it you have thoughts one two three and then you have a first conversation about it or you listen to one video and all of a sudden you have thoughts one through ten and then you have the discussion and all of a sudden you're you know up to 20. So that's part of the idea of taking the time to really read through these papers. Does anyone else want to add anything? I'll throw up the third warm up question which is what is something you're wondering about or would like to have resolved by the end of today's discussion? So Alex, Kiefer, then anyone else? Yeah so I think having this resolved by the end of today is probably a tall order but I'm interested in the question that's sort of explicitly raised toward the end of the paper about whether the FEP is a very abstract sort of model of or in some sense maybe representation of nature or a pure formalism and like that's an interesting line so I think it's an interesting question. Great question, Sarah, then Marco. Yeah I was, to me it all just seems like Mel, fix it by the end of the hour. But yeah I was curious to know like if you have any insight into ways that the model could never be applied you know like just trying to situate the model in all these worlds that it has potential for. Cool, any other intro thoughts there? Marco or anyone else? Yeah I have a lot of questions for Mel I think but I think what maybe I hope so possible to resolve by the end of this stream is there's only a very brief note about it but you also said that it's also compatible with processionalism. Something about processionalism I mean I didn't see anything else about it so I'm not a philosopher so natural kinds for me always was a bit weird iffy concept for me but I was wondering if you could resolve maybe the problems of mapping of natural kinds of stuff but for processionalism because it seems that active interest is much more processionalist kind of framework not really compatible with the essentialist objects, properties framework of ontology. So yeah I'll leave it at that. Cool let's definitely return to that and define it and get clarity on it. Shannon then anyone else? Yeah so I'm really interested in what role the free and intro principle or active inference plays in empirical hypothesis testing and I think this paper did a really good job of explaining that the mathematical models can be very useful to generate questions to generate hypotheses or just to further investigate some empirical phenomenon and I think it would be good this week and next week to just sort of talk about what it means when you say I have a free energy principle characterization of system. Very nice agreed and yes we have several fun hours of discussion ahead so I'm just going to flip to this summary slide and ask Mel when you're conveying this work to different audiences how do you communicate it or what would you like to start with just how do you give your 30,000 foot overview perhaps to different audiences or perspective yourself in the past yourself in the future however you want to do it. Yeah I guess I saw a lot of confusion about what the FEP was and what it was doing and it was before I could really play with it I had to I had to get clear on what it was and make everyone else clear on what it was and so I tried to tackle kind of like the most fundamental questions in this paper about what is the FEP and what it's doing and in some ways I think a lot of those remain kind of unanswered by this paper but no longer unexplored like I put it all out on paper so we can talk about it now whereas before there are these things like is the FEP like is this a scientific model or scientific theory or is it does it not belong to the domain of science that was a big one no one had asked that then well Alex's question that's one and I didn't come the first sense is saying I think it's the domain of science but then I'm writing a paper where we're going to come down kind of differently on that I mean it's not written it's not 100% written but we come down a little differently on the domain of science and then in terms of Alex's question that was one of my main driving questions the notion of is the FEP really abstract representation of natural systems, natural processes or is it not a representation of anything in nature at all and that to me is like totally not answered yet and there's almost like a Aquinium that's almost to me it's like getting to the heart of like naturalism as an approach and philosophy Carl Friston's line this whole time has been while there's a dissolving line between just math and physics proper and that drives me crazy but but I'm coming around to it a little bit that wasn't a good summary of my paper that was a good summary of what I'm excited about exploring further Fun, yep well just a few things in there that might fly above or below those who have different listening regimes of attention you said before I could play with it I had to be clear on it and also help other people be clear so that just captured this learn by doing learn by playing and also learn by serving and teaching because those are some of the key pieces and without being playful and understanding that it's a deep well to dive into and then also that there's a community that's also curious those are some of the key aspects and then also it almost goes without saying that this is a really exciting philosophy paper and it's an ongoing philosophical question so people are maybe familiar with a measurement is made on a protein or a new species is found in some forest and it gets nature science and people are really excited about a measurement that's made about nature and then the philosophical questions potentially because they're fundamental or there's no simple sound bite or there's no figure there's no graphical abstract you got to read the pros or got to think through it the philosophical questions are often seen as not contemporary or not exciting or not cutting edge but we're seeing in real time with free energy principle and active inference the way that maybe we could even call them top down philosophical priors on what the FEP is are shaping the realization of what emerges from the bottom up for example it's really from on high and FEP is part of science and it's part of a scientific program that's going to be very different several years down the line than if people said this should be thrown over there with the fiction books and you know with the biography of somebody else so it's pretty interesting just have these top down and bottom up ideas are playing out in the discussion about what this framework is so pretty interesting and anyone can give a hand or give a thought or a question the other piece I just wrote down until anyone raises their hand was this unanswered but not unexplored which is taking us back to the map metaphor and it's the math is not the territory of course a little bit of a pun on the map is not the territory and what is the map what is the territory and where are there dragons where do we have just the coastline on the interior or where are the underlying features of the world stable or changing and then where is our map stable or changing in an attempt to understand and just like the navigation there's probably pure navigators out there but also there's those who want to move cargo so how are we going to get it done with this map and this territory so we'll go to Alex Kieffer, Sarah, then Mel yeah so I guess I'm just following up from some of what Mel just said so I don't want to take this too far in the direction I'm thinking about because it's pretty general so to me this paper raises questions that are fundamental like whether or not you're interested in the FEP and then it's partly a matter of seeing how these two things fit together but like so Mel mentioned quine so one problem that I've always had with the line that the FEP is not falsifiable or that it's like distinct and kind of interactive inference as a process theory is just like I guess confirmation wholism so so I just assume that if this FEP thing is informing our empirical theories at all then there must be some semantic logical connections of some kind between the empirical theories and the FEP so and if you're a confirmation wholist do you think that you know sensory evidence what we observe kind of back propagates if you will to like all of the beliefs our network then I think the FEP should be vulnerable to that but again that's not really about the FEP so I just want to sort of raise that issue more explicitly and I don't know just before we go to Sarah could you just define a little bit better what is the confirmation wholism who would think differently on this issue and what would be the implications for active inference well so I don't know what people think about this issue these days actually I'm just this is sort of like I guess it's attributed to Quine and do do him I'm sorry if I'm misprouncing his name but the idea is just that you that sort of scientific theory does it like it meets sensory evidence as a corporate body so like the idea is that if you observe something that's in conflict with your total theory there's no I don't know way a priori of telling which of the claims in your theory you need to revise right so like if there's I mean you could for example you could say well there must have been an error there must have been a malfunction in the measuring device and so you could revise that part of your theory or you could say no this really does falsify my more fundamental assumptions and so that's kind of what I'm getting at and I'm not sure if people hold the alternative view these days that like no there are certain statements that are meaningful but immune to falsification but that's the kind of question raising. Excellent Sarah, Mel, Marco I think actually just to not overwhelm Mel my question could fit in anywhere I'd rather just give Mel a chance to let's go back to Mel then Sarah yeah the web of belief thing yeah the do him point hypothesis this idea that we've got an integrated web of beliefs and you can't test any atomic proposition in affiliation from you can't you can't protect anything in your web of belief from revision with by new evidence which I think it's almost like very compatible with with this kind of Bayesian frame view right there's you get this you get this spreading its propagation of of evidence through the network all the way up and down I like that idea but where did I want to respond to this oh yeah someone said someone said active inference is a process theory and I just want to like set the record straight on that the the reviewer one on my manager I just I just got reviews back on this not published yet just got reviews back and they pointed out that it it's probably misleading to refer to accident princess a process theory there have been process theories on the basis of active inference and variational message passing developed so when you see the like the there's a paper fristings comic review Joe Pizzolo's on their 2017 it's called active inference a process that that is a process series made from active inference that's that's set in variational message passing but active inference itself is not a process through so much as a corollary of the FEP applied to to activity in an environment and then process through all that have been made so I should probably I'm going to go back in and like be more careful with the words I use cool Marco then I'll go ahead I'm not sure if you've met me with a process theory but I actually also avoided the phrase process theory because I think it's what I was trying to say with that comment earlier is love the treatment of natural kinds and mapping to real states or affairs of the world seems to me more grounded in a more object property oriented philosophy of mythology right but for me it sounds more that the ontological assumptions or commitments in process are far more compatible with active inference which I wanted to address it later I'm not sure now is the right time but there's this issue of not demarcation for is FEP AI scientific but also the demarcation of market blankets right so as you pointed out rightly there's this dilemma of you know where do you put the threshold for where the boundary of the market blanket is but for me that fuzziness is not an issue if you would be more specialist is then the claim with the necessity would be more there is a boundary but that we cannot affirm or confirm where it exactly is what the you know perfect absolute pressure is is not an issue itself simply that there is a there is certain a boundary in of states would be sufficient I think because the processional dynamic nature of these systems would mean that you can't say there's a static bond simply that there are boundaries between subsystems but that was not my question my comments for now was basically one thing I think is also interesting is seeing FEP AI as a language so that's the canonical kind of typical notion of math is the language of physics discipline in which that's written in maths and I kind of feel that this is the same here so Daniel give very beautiful metaphors of navigation and mapping and to me it seems that this whole project not just here but the whole project of FEP and active inference it's more of a vehicle of tool kids of languages and allows you to translate back and forth in disciplines translate into mathematical abstraction and as you explore in the mathematical abstract world you can translate back so it's like if you can't see the patterns in empirical data translate back to abstraction find patterns there see if you can translate them back and then you can test them and so it's like a scientific narrative with a detour but if you find it interesting would love to discuss it more later but yeah it is interesting and also the navigation brings us back to cybernetics and the word cybernetics comes from navigation and so it's almost like if you can't navigate on the water surface and then you go you abstract beyond some sort of limitation and so it's kind of funny because the mapping metaphor again it is a physical one it's related to cybernetics but then also there's this idea of a map in mathematics mapping is a function and so there's this abstraction to abstraction mapping and then there's also that as a map onto the physical world because we're embodied agents so if Mel or anyone else wants to raise their hand otherwise I think there's a few other threads to highlight there go ahead Mel yeah I just want to flag something Marco said quickly the Markov blanket isn't actually all that it's not as a boundary the Markov blanket is not actually all that diffuse and dynamical it's actually harness to like particular particles and so when we think about the boundaries of an organism or the boundaries of a cell wall or something you're constantly you are dispersing the cell is dispersing the material of which the cell, the organ, the tissue the human being the society is made up these are diffused and they're dissipating there's a constant turnover in the material composition of these things and the Markov blanket doesn't have a lot right the Markov blanket is wrote to particular particles so to speak in your system so it's there's a kind of a desire in the original like Birkhoff stuff on dynamical systems you get 1927 he kind of proposes an ocean of wandering and there's kind of a desire Carl has a desire to extend the Markov blanket so as to be able to accommodate wandering sets where there is a turn where the material make up goes away and yet the boundary persists you get turnover it doesn't do that yet it's not nearly as dynamical as you'd want it to be the wandering sets and the attracting sets again with a navigation metaphor the attracting set could be like a pool and all the boats around it are kind of being drawn into this attracting set if they move away then they're drawn back in and then the wandering set is like a flotilla of ships maybe some are joining and some are leaving but they're maintaining a coherence as they dynamically are moving through the ocean or as their subunits are changing because there's this dissipation that was brought up and then the living systems that are far from equilibrium are the ones that are apparently successfully resisting that dissolution and so what is it that those living systems are doing and that's the what is life question that's Schrodinger 1944 to Maxwell 2018 that pipeline so it's really interesting questions Marco then any other raise hands thanks for the clarification maybe I'm misunderstanding something but so I guess for me maybe it's my personal interpretation of it all but for me the primacy is actually less about formalism as for example Julia Perl initially formulated as but what for me is interesting is more the conceptual interpretation of there's an intermediating kind of interface there's a unidirectional flow that's in its dynamics allow for this protection of states that shouldn't go beyond certain boundary for example and in that sense oh wait and one more thing is that maybe that which is to be bounded or protected doesn't have to be a particular state as such maybe it's like a relation for example you're protecting the property of certain nodes of certain states of certain densities that they are adaptively viable for other systems that they might be in relation but I think in general I think what this all shows is that there's a lot of freedom at least in my opinion to the picture that follows from FAP AI I'm still kind of unclear why it's not dynamic because the mathematical nodes can be variably mapped to physical nodes in my opinion because I'm not sure what you mean with particles it's something like we do the boundaries of particles but more in terms of sensory states or that through which the Bayesian influence is mediated but I would love to hear more about this if anyone else raises their hand I would just kind of almost take stock there's a few threads happening here there's some philosophy of science questions what is a principle what is a theory what is falsification all these types of questions were brought up from the more philosophical side of the literature and then there's this a little bit related discussion about Markov the definition of empirical measurements and how cleanly we sort them into these different types of natural kinds if we want to use the philosophical word or maybe just a functional whole to be less essentialist about it there's a few different threads happening here so it's just good to keep track of these pieces and Steven I saw your hand raised or anyone else I suppose one thing that I've been quite interested in is dimensionality in the sense that entropy is a dimensionless sort of property so you get this interesting question so the free energy principle in a way is mostly dimensionless or it's some sort of trend and then when that comes into active influence it doesn't get fully dimensionalised it doesn't it's like lots of different active influence processes each one can contribute to dimensionality as like multi sort of threads the actual signal unlike when you mention like say in cybernetics where you have something that is guiding you the signal itself doesn't have the signal doesn't contain per say the data the entropic noise it's through influence of the dynamics within the noise that the generative model can sort of then process that and so it sort of raises this question about when things go from one level to another in an active influence in these states is that transference between states also dimensionless in that sense or where did dimensions start to come in as we go up the blankets or maybe as things get aggregated Interesting question and that reminds me of some other questions that I've been raised about the dimensionality of active influence like is each sensory modality enshrouded in its own blanket or is it sort of a tuple that gets passed a single organismal blanket that represents multiple sensory possibilities so interesting questions anyone else want to raise their hand or we can just start to slowly walk through the paper and just any thoughts that people have will raise them so we checked out the abstract in the dot zero video here's the roadmap so this is a pretty thorough roadmap and it's going to be a lot of the sections and a lot of the terms that come into play so Mel I have a question what was your intention with writing it you said that you wanted to clarify it and learn as well how did that translate into this structure and then maybe where are you headed off this roadmap this was one journey what is your next journey a friend brought my attention to the 300 principle and I think full of 2017 which would have been like my penultimate semester in undergrad and I was taking a course on like mathematical psychology so if you one of my references is right to the reference to this normative model normative versus descriptive process versus that reference is Ardonk and Luce mathematics which is he's one of the primary figures in maths like a lot of cognitive science has subsumed maths like but the idea of having these formal mathematical models of cognition so that's what I was enmeshed in so that's what I was thinking about most at that time I was like oh a new mathematical model of psychology perfect and I I I was very skeptical at first and I dove in and I it it brought me under there was a lot there and I really struggled so at first I really wanted to do that too it was immediately apparent to me that there's this big complicated thing that very few people understand and there's a lot of widespread misunderstanding of what it is and what it's doing so I tried to do that at first I came up with like a little a little dossier on like the basics of the SEP in like some sort of reformulation as an undergrad and I put that up but it didn't solve kind of a lot of the bigger I guess like philosophy of science questions about it what it is and what it's doing so this is like a second stab and from here I so I was thinking of the FEP one one way it helped me to think about it is as a model and I don't I guess I don't use model as strict natural kinds or like strict the scientific kinds I don't I don't think that there's like a a list of necessary conditions for being a model or like an essence to being a model FEP is either you know definitively a model or not I think I think it helped me quite a lot to think of the FEP as a model it reduced my prediction error if you will to think of it as as a very abstract formal generic target with models but but you know and I think in a lot of ways thinking of it as a modeling language or a modeling framework is also really fruitful I think that's important there's not there's not a philosophy of literature on modeling frameworks for modeling languages right but but I think that it it's probably more of a modeling framework or modeling language than it is a model per se and and next so I'm next I'm going to try to work on paper with with it in as kind of exploring what it is to be a principal right like what's the principal because there isn't a field side to run on principles really right what is the principal that's next I also there's been a lot of comparison of the FEP to natural selection to Darwin's theory of natural selection and when you ask people about that and this is this is sort of the the core like pro FEP camp the core like we're working on the FEP camp and often draws that comparison and I think that this so what they mean to be doing there is saying well it's not directly falsifiable to Alex's point I think I think everything is sort of falsifiable every everything in science is falsifiable and so much as it if it if it doesn't get traction it falls out of use if it's not useful then we don't use it right and then it it ceases to be kind of part of our collective consciousness as researchers and I think that if the FEP fails to be useful you know it falls by the wayside that gets confined to dust and the history so there is that in that sense I think there definitely is that kind of like confirmation wholism there right if it fails to produce useful lower order theories then it falls out of use so so I think that in trying to draw a comparison between natural selection and the free energy principle people want to say well the FEP like natural selection is not directly falsifiable in the way that natural selection isn't directly falsifiable and it's also supremely useful it has this widespread kind of both empirical use and you know grand explanatory power and I think that this is sort of ultimately conflating multiple notions of natural selection so multiple kind of discrete formulations of natural selection one of these due to Popper and one of these one of them sort of due to Darwin and then it gets formulated in like Lewinson and Godfrey Smith and stuff like that and one of them that's basically due to Popper and Rick Hansen as a mistake it's an error and here's what I'm thinking so I want to do that and I also think there's kind of a meta narrative there because I've been working on models and reification in models so when we what's happening with the FEP we can think of as reification conceptual reification or pernicious reification where you get confusing the map for the territory basically Perfect, thanks for that response and also welcome Scott we will go to Alex Kiefer, then Marco then Tim Okay so I guess there are a number of things that I want to say so as far as right so I think what Mel just brought up reminded me that there's a there's a one way of I guess I've discussed this with Mel and others so I'm apologies if this isn't in the paper I'm not quite sure I need to look at it again but basically one way in which this thing could fail to be falsifiable is if it's not so much a claim as like an approach or like a sort of stance that leads you to make certain claims so to put that if you wanted to put a negative light on that you could say it's a moving target for the FEP at all I think you know one thing that's going on here is that Carl Friston is a fairly creative scientist obviously sort of trying to do something new and so he's this is a framework Mel mentioned thinking of this as a modeling language it's a framework that's still under construction to some extent and so you could say it's well it's not falsifiable because it's not a particular mathematical description it's a creator of descriptions like that so let's stop there for now yep and just to draw out one quick thought there it's the principle has been metaphor the metaphor is between the principle and the language so could anyone falsify English could anyone falsify the idea of English grammar or is English a moving target is it something that can't be falsified but you can make claims under the umbrella or using the English grammar approach that the claim could be useful or not or it could be useful for one person or not for another person so once you get into the specific process theories the specific claims then you're talking empirical data or you're talking about the perspective of individual researchers but then when you pull back to the principle level and Mel it sounds really interesting about the work on fleshing out what a principle is it's a little bit more like you can't falsify Python but there's programs that don't work okay so Marco Tim and then Dave thanks so first of all I apologize to the fans of the philosophers because I'm kind of cautiously tried into pragmatism so again I'm not a philosopher but I'm kind of interested in the angles of pragmatism on these matters so reiterating again the theme of seeing the French principle as more of a language and a framework we also I think obtain an interesting maybe mode of education that actually Mel just gave us which is if a certain approach or framework or system of doing science is selected by the niche that is a scientific enterprise then that is maybe an interesting form of the opposite of falsification right so if it doesn't get adopted then that maybe is kind of falsification but then an interesting implication here is following the popularity and a strange like of FEP in so many different circles is I think a sign that that the scientists or the researchers who have utilized it have experienced and embodied signal of usefulness and then following that comes a question of okay then where does it come from and then for me what's really really interesting is then FEP is the perfect kind of approach to answering or approaching a question right so upon adopting or upon being exposed and internalizing the FEP and active influence what changes within so now we get also two kinds of approaches to language not just you can communicate with other people but also how do the different models you embody communicate with each other so the unifying capacity of Fianzhi principle is not just in the abstraction in the papers but also for people themselves at least for me and totally also others that is I think one of the more unique aspects of active influence it really evokes the sense of unification that allows you to use a single framework and lens for a variety phenomena and theories and to me that is maybe the biggest argument for it being scientific for the merits it evokes our genders in scientific enterprises including that situation so just people and researchers so yeah so we'd love to hear more from you especially philosophers on pragmatists and perspectives sorry I've been binging a bit of Hasok Chan recently so maybe that's why I'm asking this thank you Marco Tim then Dave then anyone else who raises their hand did you say pragmatist just now yes that was pragmatism yeah a good segue to what I wanted to mention I guess because I was when you mentioned reification that resonated with thoughts as I read the paper about Whitehead's fallacy of misplaced concreteness you know the idea that this not in the processes that exists that's being revealed is actually a process and the things that we see in the objects of the world are really just sort of a reification of the processes and you mentioned natural selection that was another thing that came up but not in the way I don't think you were mentioning it now the way I thought of it in the reading the paper was basically the objects as we call them are simply processes that we can take a perspective on that survived they're there because they survived they sustained themselves as you like to call it self-organized and all that right yeah so anyway yeah thanks for tying that Dave and then Alex Kiefer yeah Mel you don't seem to give the exact title of the 1995 loose paper what was that I've got his 95 book but he doesn't seem to hit is it just the introduction to that um the taro indow collection up Mel you're muted but look it up and then post it in the youtube chat or comments not the jitsie chat because it will make a noise so post it on the youtube and then anybody who wants to post any of these papers they're mentioning really the place to do it is the youtube comments or send it to me and I'll put it in the youtube video description so that everybody can access it Alex Kiefer then anyone else who raises their hand uh yeah I guess I was just going to create some more problems for the simple discussion here um in one okay so so privatisms on the table and ratification um so um I guess similar quantian type reasons I'm not sure how to think about these things I think what Tim brought up about um perception forgive me if I get it wrong but perception being something like a ratification of an ongoing process I think is a really good point so like um I mean in a there's a sense in which I'm a fan of ratification because it's like yeah maybe that's just what we're doing when we you know we take some we have a language uh maybe it's an internal neuro language or something maybe it's an external one and we take it to be about the world and that's our lens on the world so without ratification in some sense I don't know um how we we would have any any kind of perception of things but I'm sure that there's a way to distinguish bad kinds of ratification for good kinds um and I was going to say something else but it's I completely forget so yeah good point it's kind of like stability and plasticity remember those two top level descriptors of the cybernetic paper that we heard from in the last couple of weeks and it's kind of like they're not intrinsically good or bad because you could be too plastic or too stable and succeed or not and then beyond whether something exists or not humans have this kind of second layer where it's like oh yeah that exists but I don't like it so it's perpetuating itself but then we just don't prefer it and so that's just our agents level perspective and so it's not just enough to be reified there's something else happening there so Marco then Sarah yeah oh the build on Alex a great point so one of my favorite notions is Rekonstinian ladders or in Buddhism they also talk about the simile of the rat um and I think this is really what it's about right so we're kind of like Alex said we kind of stuck with reification that's just part and parcel how we engage with the world because the alternative was believing in nothing and then if you literally believe nothing then how the hell you're going to engage with the world right so all you can do is have as good as possible reification or at least worse reifications and again for me this is this processionalist kind of approach or at least perspective that we are just engaging in finding better reifications such that that particular reification or such that that particular embodiment or incorporation of a model leads to better generative dynamics and then to better active influence and to use active influence as the framework here um yeah so I accept cool Sarah then now um I don't know it this is possibly going to be an embarrassing moment for me but I feel like that's my job um I don't really understand like as I was kind of stepping through FEP the math of it there's a there's a point where I'm like eh it seems like Bayesianism has got some like it doesn't seem like even it the foundations of what FEP is a lot of things are based on I mean this idea of belief and priors and and that's something you know something is absolutely true I guess I just wonder Mel if you you know if you ran up against that or like if there's a deeper background that I'm probably not aware of in philosophy about the limits of Bayesianism um or these kinds of models so that's I don't it's really a big question but um it's the best I could do Oh sorry yeah it is a great question go ahead yeah it's an excellent question so there's a um yeah I think I think when I was first grappling with Bayesianism there's a way in which um there's sort of like a Laplacian demon smuggled in like there's a there's a view from nowhere smuggled in there's a a perfect knower smuggled in um and that would be the case for the FEP but we've we've um we've worked out a way for it to be just a a cognizer that's like self supervising right it's knowledge is limited it's access to the true world out there is intrinsically limited and yet it's able to self correct and self supervise and if you want um I think sort of the best recent explanation it's like why this is the case and how it's able to do that is um yeah compose paper and I want to say a synthesizer and I want to say something self supervision and the I guess I have to I guess I have to drop this link as well let me go do that yeah that's what's so fun is the discussion thank you yeah the discussion we raise so many fun links and so many connections and then we just make sure to post them so that everybody can asynchronously or synchronously be in the game so Marco then anyone else no raised hand yeah uh oh yeah sorry um I was going to say oh yeah and I really love that but I I want to propose saying instead of a perfect knower is Laplace indeed knows everything but for me Bayesianism is really more about if you would have to make an analogy it's a perfect pragmatic learner so in essence what it it's the north principle kind of by design says this is the perfect way of knowing if you also add a lot of assumptions for the particularly eight but but uh yeah I guess a bit half jokingly you could say it's a distance demon you know the perfect pragmatic learner um yeah I just I just ran up against this just in my everyday life I contradict myself in so many ways I can't even keep track you know like one one part of me thinks this the other part of thinks that and they're literally and so I yeah I just I just started realizing my own limitations with respect to knowledge and how knowledge would even be modeled but that's part of the design I think that's something I often like to point out is it's not about having no conflicts it's more about given that conflicts will arise how to best cope with them so the adaptive engagements with the world is also reflexive so your own struggles your own challenges your own seeming contradictions are also opportunities to grow and that's again only possible with this demon that is pragmatically learning all the time I'm curious about this that's a bit more optimistic yeah I'm curious about this perfect demon because I would say that a Bayesian updating agent could make a update that helps it survive in its niche or not it could over learn or under learn and just simply fail to exist in the future and so natural selection again is the hand that just sweeps off the table the Bayesian learners that aren't existing and I would challenge you know there's many thoughts on Bayesianism and a lot of debates but alternative being what frequentism or some other vaguely unspecified mathematical framework that somebody doesn't want to put a name on to and so I think it's an interesting question how does mathematics and how do formalisms come into play with philosophy at all what would any equation have to say about philosophy but then once you're in that world of trying to make models that have some element of formalism whether it's natural language or whether it's another language then I think you start wandering over towards the idea of a learning or updating agent and not needing to take the baggage that it's a perfect learning agent just that it is an adaptive agent of some kind Sarah and then anyone else who raises their hand yeah this gets for me though this gets into a kind of a I don't know the clever philosophy word but you know this is also why I asked a question in a prior week about the Markov blanket because it's like I picture like a monster truck rally where two FEP models are battling it out basically in my own brain or whatever and so it's like where is the so in that paradigm where you have these kinds of objectifications you know of one FEP with another FEP like where but then you have questions around like the boundary between the two and there's this kind of it there's like there's a thingness about all of this that kind of gets me in like an infinite regress so that's yeah fun Alex keeper than anyone else yeah so I mean I think this this issue we're talking about now is really interesting I yeah I was wondering recently whether Bayesianism is a is a prior and what that does when you start thinking about it I think this is totally relevant to the theme of overall paper because of course it is but directly relevant because like it we're talking about things that get to be so general theories or frameworks or whatever that are so general that they start to raise the really weird questions but one thing I was going to say is just that like I think maybe this is implicit or explicit in what people said already but like the fact that we're dealing with variational inference well for okay first of all Bayesianism even if it's Bayes theorem itself is supposed to like descend from God or whatever it's like actually building subjectivity into science more right explicitly than the alternatives I think people mentioned contrasting to frequentism but then we're taking a further step away from objectivity and saying yeah and each creature sort of learns its own approximation to that so I think I don't think you can eliminate claims to objectivity entirely I don't and so this maybe maybe this is the best we can do or at least a step in the right direction very interesting point Marco then anyone else Marco then Mel yeah I fully agree so objectivity can't be eradicated and I think that's kind of the beauty of how like Alex said this act of inference stuff is taking the rejection of pure objectivity even further it's really approaching or making at least the building the foundations for an objective science of subjectivity right and and to to add to touch back on the of jokey idea of the best demon like Alex said right so they create their own particular approximations they have to have first instantiated particular prize but more importantly you also noted on I'm sorry you also noted on infinite regress so so that's another beauty for me there's no problem of infinite regress because traditionally with a monkey line stuff and controllers or agents or will there's this yeah but what is controlling that and there's this movement towards smaller and smaller but that's not the case with this scenario because the dependency is not like a vehicle dependent on the driver but it's more system dependent on each other and I think that's kind of the beauty where the agents in the perfect DNA scenario it would be more like minimal dependence on the agents faultiness and more about how well the niche of the context prepares that agent for the niches to be incongru in their life and for those you know with a penchant for Buddhism it's also very beautiful how it relates to this idea of the mutual arising or co-origination it's just in essence like Alex said it's about subjectivity the co-dependency of how densities or how states and identities how densities kind of adapt each other and what is worth Mel then Tim then anyone else I guess I want to push Mark or maybe to bring up a question he raised when you were giving us the sort of like breakdown of the paper which I thought when he raised I was like wow that's a really difficult question that I can't answer I think that would push a very interesting discussion if only because it's a question that I don't know how to answer and it's a question of it was something about well aren't physical descriptions just kind of formal descriptions at some level anyways so Mark do you remember do you know what I'm talking about no well you raised a question about about this line that I drew between a mere mathematical, formal statistical description of something at a physical description proper and you sort of challenged that such a line which is Kristin's line but somehow I love Carl but somehow the way he has always raised that it feels to me like a cop out it feels to me like some sort of evasion but when you pushed me on that I was like you know yeah you're right maybe there isn't a hard line between and this sees into Alex's question as well it's all continuous this question of well is math natural like is math an arbitrary formal system where we even invented all the rules and it just has nothing to do with contingent nature or is it continuous is it just descriptions of nature at a very abstract level is there a hard line between information theory well you know going full James in it is there a disappearing line between between inference technique between techniques for Bayesian inference as a you know just as a statistical technique and thermodynamics you know statistical mechanics those are the questions that I'm like oh man I can't I can't that that seems to me to be the most pressing question behind all of this to me I was just kind of the homunculus infinite regress thing versus the dependent arising sort of co-emergence of the agent I think that contrast is really something neat that I don't know how to explore it but they seem like opposite ends of you know two poles of the same kinds of consideration and I feel like I'm repping whitehead here today but it also touches on what he talks about when he talks about the bifurcation of nature that he decries right so I mean so when we talk about taking a perspective saying that object emerges to my perception first of all because it survived as a self-recreating thing over time you know and I did too we're both the same thing you know what I mean like it's not like I'm apart from the world the reason it's in my world is because I'm in its world and nature's not bifurcated into the mind and the physical anyway awesome point Tim because we often say in active inference that the organism comes to embody the statistical regularities of its ecological niche and so people look out there at the ecology around them and they can say okay that's natural but then the organism is also natural and wouldn't the generative model of its niche be just like its hand be just another phenotype that's generative and therefore natural as well so Alex Kiefer then blue then Marco um I might just I might actually pass for now I just wanted to drill down more on this idea of like mathematics what you know pure mathematical formalisms and how they might represent but I think there's a lot of interesting stuff on the table right now I'd rather hear about it at the moment perfect blue then Marco so maybe it'll I'll wrap it back for Alex because I kind of want to just touch on this dependent arising right or like the mutual arising um you know the world exists because it exists in my mind um and also the idea of whether or not mathematics is natural right like is this like something that we invented or something that we discovered but I would like to make the argument that really like as humans we are also natural right so so if it can be discovered in more than one place at more than one time by more than one human as many strong theories and science are they bubble up together I think that there is some kind of natural arising or naturalness to um these kinds of studies thanks blue Marco then anyone else thanks um so I think I know what you meant with that comment Mel I think but let me allow to push back on myself so I think if we're we only have just descriptions um and the only kind of extra qualifiers of properties of these descriptions as such that's relevant and grounded is their etiology or their genealogy where do they come from right so if we're already talking about descriptions we already have to be talking about systems that can generate descriptions and so I don't think there's a hard line in terms of an essence right that's intrinsic property but I think the hard line is more about their path their genealogy how did that description come to be um and then I think we go back to language again and I think this thanks to um Liam Bright shout out to him who introduced me to Karnat and he Karnat talked about valid physical languages um and I think that's kind of the beauty that the FEP is maybe not perfect for every state of affairs or space affairs but I do think it's true that that you can transfer a lot of what we now at least see as valid physical descriptions to the language FEP but more importantly to me the beauty is a generative aspect so if you transfer something into um uh translate something into active language of active inference then you also transfer the generative aspects to it so if you transfer then to active inference and then you explore in the space that's generated there uh and you see a pattern then that hypothesis obtained or derived in an abstract space can be also translated back and so it's the validity of that back and forth translational path that I think uh is more important to the question of is it a physical description right is it is it physically rooted rootable if you have these valid languages I don't know if that makes sense also small point uh very small points is um I've always also taken issue with Carl saying that we embody statistical regularities about the world not that it's not true but I think it's it's it's only a partial view for me what's more relevant is that there are statistical regularities due to the fact that the world is systematic and these regularities allow this kind of tethering right so it's for me it's more about embodying a bridge with the world and what you then get at least my motivation is then that leaves a more openness to also give room for the unique aspect of humanity as in humans are weird we're like really weird I don't know why it's weird exactly but we're really weird and that weirdness has less to do with statistical regularities and more about what is evoked or generated or cultivated upon embodying this statistical regularities and upon being faced with the challenges of dealing with the world um and so I guess we have statistical regularities of the world and kind of weird irregularities of ourselves if that's yes we swim in those regularities of the niche you know if the fish were just the statistical regularities of water there wouldn't be a fish there so it has to be something a little bit bigger now another cool thing is we've been talking about this similarity between the physical ecology and then our social ecology which is also natural so we can think of a physical example of Stigmergy like the ants that are digging out their nest and they're modifying their physical environment which then changes how their physical behavior is implemented but with this conversation and in the literature especially there's a Stigmergy at the level of information the informational niche which has different perspectives within it different agents that have different access to different kinds of information and those kinds of communications in the informational niche are natural and the generative models that we have are natural as well so there's just so many fun ways that we can think about all right well now that we're niche constructing in this info niche what kinds of things do we want to put in this construction that we're working on together so just like the ant engineer might say well we could kind of have a corollary to this tunnel that branches off this way maybe there's something useful over here oh we falsified that corollary tunnel it's not useful it retracts so in this social world and this informational world how do we take that natural perspective and also make it useful for the colony scott and then anyone else who raised their hand just wanted to say with regard to the fish I think some of you heard the example I've read research a number of years ago where they took a semi-rigid piece of plastic and put it in a stream and then they varied and increased the speed of the stream and ultimately the piece of plastic started undulating with the periodicity that resembled the fish and so what they realized is the fish doesn't swim through water it actually passes through the water so it was really being over each time fish are formed by the water now again not the entirety of the fish it also formed by other relationships with other organisms and media but it's similar to the way that insects bats and birds independently developed wings because they were flying in the medium of air but I think Mel has a point yep let's go yeah we'll go Mel, then Shannon, then Marco I put there's a like a lovely little video of a dead salmon in a stream and it's it's behaving in this way that really seems very lifelike but dead and I think I once posted this with the caption like is this morphological computation and that made a lot of people there was like people were very up in arms over whether or not that was morphological computation Interesting question and synapses and neurons being physical what isn't morphological computation under certain perspectives Shannon, then Marco than anyone else That was a brilliant picture of what I'm thinking in my head right now to piggyback on what both Mel and Scott just said so the dead salmon or the piece of plastic you know is passing through the water and you can very clearly see how the environment or the medium that it's in is forming its behavior even though it's dead or it's plastic and as you get into more weird creatures creatures with culture like humans even fish that interact in groups and communicate with other fish in a cultural way then you end up adding different layers of medium so you don't just have water you have now the other fish are in your medium they're part of this social organism and then with humans now the other humans are making another social medium that you're interacting in so you have more in the terminology like nested levels of environments that you're embodying Thanks for that point Shannon, Marco, then Scott than anyone else I'm not sure if it's appreciated but I'll continue a theme of linking active entrance to some spirituality slash eastern traditions is so the notion of the Dao which stands for the way I think it's also for me that always evokes this resonance with active inference because there's this emphasis on this the way the Dao of how the universe works specifically that is unnameable and I think that that is also related to this issue of reification earlier that there's more systematicity there's rules to how the world works but fundamentally it's unnameable the moment you name it it's not the real way and meaning the intermediate parts the one of least effort or as often as described no effort away this too is exactly resonant with that picture of the life the dead fish in the stream so yeah I think that's just in general there's this theme that I find also very exciting how much active inference actually helped me appreciate a lot of philosophies in tradition not just eastern but a lot of let's say forgotten wisdoms and forgotten insights from not just religion and not just spirituality but also arts right we often see this false dichotomy of arts as something frivolous in science truth but a lot of insights scientific insights were actually prefigured by a lot of artists in conceptual form and so in my experience active inference where this general lens or perspective allows us to re-appreciate that which has become too decoupled from our modern postmodern over rational of analytic and a way of engaging with the world thank you Marco it's almost like dare I say it the math is not the territory so we'll go to Scott, Blue, Sarah and then Alex Kiefer so a couple other just observations along the lines of what Shannon was alluding to I think in Foucault philosopher Foucault not the pendulum he talked about pastoral control which was the idea that the persona of a person is formed by their external norms and rules and laws and culture that they're exposed to so the same kind of thing like a plastic toy in a metal mold being formed so that same kind of notion and one of the things I think is interesting about active inference for me is that those differentials what is the motivation for closing differentials and the motivation is de-risking and leverage grossly stated I think so the idea is if you have a differential between you and the environment you want to protect yourself from the harm from that differential or exploit the opportunities of that differential that differential is either temperature differential in the case of a heat engine or an information differential in the case of a von Neumann or Shannon extension of the second law of thermodynamics so the same differentials you have in Carnot's equation for heat and cold you need the differentials and information to have market exchange and I think to have Beijing exchange so what's the status of a Beijing process is to have no differentials between the input and the expectation so along those lines what it seems like is we have these identity stacks this is what I think Shannon was alluding to not Claude Shannon but our Shannon here and both perhaps but you take these combinations of things in the environment that make you who you are we collect up these externalities and internalize them for our expectations and then there's a whole set of them and then we go out and identify those that are anomalous for our expectation so on another call earlier there's someone who talked about the difference narrative and story and how your internal narrative and your story is being told you have to line up well to the extent this is a differential that motivates activity of interaction and markets have been described as places where you find price discovery and solution discovery at large scales so the Beijing process seems like a way of incrementally identifying the externalities that are anomalous for your expectations and then being able to internalize them is kind of that Beijing process I think so anyway something the reason that is so appealing to me is that I've been arguing for years that markets and countries and companies are biological because they're iterations and artifacts of structures of biological beings which is us they happen to take on an abstraction and an informational embodiment but they represent those same processes as biological systems in terms of those differentials and trying to render externalities or exploit externality differences so anyway this is feeling very there's a strange attractor here it feels like that's pulling the conversation in the direction of that generative systems that I define the last point I define life as being auto-catalytic entropy secreting systems so these auto-catalytic they keep going and entropy secreting the disorder but the problem is that the neighboring system has to absorb that disorder and so how do you manage disorder among systems is again when Shen and not Claude Shen was talking about the different things the different characteristics that's the neighbors good neighbors when they bang into each other in terms of their creating anomalies for each other's expectations manage the anomalies instead of just winning over each other anyway just a couple of things thank you Scott blue, Sarah, Alex, Marco so I'm going to skip over what Scott just said and go back to what Marco had previously said about the connections between active inference and eastern religion so I'm a student of Buddhism for many years and I've been wondering do we in fact have free will like if we're considering karma and also considering active inference so this is something that I find this parallel like karmically we're connected we've got collective karma with groups with systems with the universe in this philosophy so and then we have our own karma like our previous actions and have collectively formed us and so when we're faced with a choice like I wonder do we in fact have free will because all we can really do is like if we're just the product of our karma we only have one choice like it's the choice that we're going to make but obviously that's the only choice that we could have made because of all of our previous karma right so and in the same way in active inference in thinking about active inference as like a computational from a computational aspect if this is really representative of you know the brain and how we think and how we operate and how we make decisions so thinking about active inference there's really only one decision that we could make right like given the inputs the sensory inputs like we have this generative model like given our model and given the inputs of the system and our selected policy there's really only one choice that can be made like as a result of all of this like collective computation going forward anyway it's just something I've been thinking about for a while blue it's awesome and I just checked on search engine karma as multi-scale Bayesian prior the title is not taken yet so anybody who wants to write that one go for it Sarah Alex Keeper then Marco um yeah just more random into the mix I guess but um you know I was I was reading about like early Babylonian I don't even know if it was philosophers but somebody you know said oh those guys those guys weren't scientists because XYZ and um of course I disagreed but I was just thinking about just generally the um evolution of science and math and there just seems to be this constant like bifurcation bifurcation and kind of contrasting this ever more complex ecosystem to the fact that we're um really diminishing our we're going through species extinction in the actual material world but in the model world we're going through this proliferation and that just like totally trips me out like what does it all mean in terms of um connecting those two things I often think about the informational versus material and more in the loss of um I'm more in the loss of analog actually like this is a big thing for me and I don't quite know where to go with it but um it at some point when things became digitized we really did a couple um behavior from material in a way that I think is really significant but I can't quite get my head around where it matters and where to where to cut at the joint I love that phrase um and how to how to explore that so if anybody has advice about that also um but yeah thanks Sarah it's like Marshall McLuhan's work and other media theorists on the innovation of the written word as one kind of uncoupling and you're hinting at even beyond the written word there's a digital uncoupling and how that influences us and especially just you captured it so well there's a diminishing biological diversity in some ways as we modify our environments but at the same time at least over historical time periods we're proliferating there's more songs there's more scientific theories there's all these types of things but what happens if we reduce our uncertainty about the biological world till it's just something that can't support that kind of theoretical proliferation. Alex Keeper then Marco then Scott. Oh man really interesting stuff going on here so I mean I originally raised my hand just because I wanted to explicitly say that I do appreciate Marcos by connecting this stuff to Buddhism and things like that that's something that like I hope to write about once I've established myself as a serious philosopher of science type person enough that then I can say things about that and no one will react badly because I think it's awesome but and this issue you're bringing up now with like we're kind of enriching the model space at the expense of the physical world is really deep and important I don't know what to say about it but it's there and yeah that's the thing that I was that I wanted to bring up I figured since we're all talking nicely and getting along about this these big you know interesting large scale themes that maybe what we should do is instead argue about like some detailed annoying thing so I just wanted to push this idea that so the reason that I don't see the FEP as innocent of representation is just this might be sort of because of my provincial view of what representation is but I figure if you think of a mathematical model as like a structure right and then we can sure you have different windows in that structure depending on like the actual mathematical language you use to write symbols down but maybe this presupposes mathematical realism in some sense but like you've got a structure if you've got any system of equations and then on my view that just represents whatever is isomorphic to that structure and it could be like a idealized physical system I don't know you could argue about applications and where applications come in but I guess I don't see like how you can escape representation so see what that does if I say that yep and the infamous representation wars earlier on active streams Marco then Scott then anyone else who wants to raise their hand thanks I tried to address a lot of points and hopefully so I didn't catch everything which is that Scott but I fully agree that there's this huge hugely important issue of dealing with externalities right so I have some issue with saying that we simply throw disorder out in the world I think that's maybe a consequence of taking the notion of exuding entropy a bit too oversimplified entropy isn't strictly disorder because as it's been noted it's all about actually almost pretty much yet every information theoretic measure is subjective it needs to be relative to something often forgotten but re-emphasize invasionism and now also machine learning because they're discovering that it's a huge issue anyways so the point is that first of all I want to also let you know that amongst research and active influence is indeed something that's being addressed right there's for me one of the most exciting applications of active influence is kind of a philosophy of society or a philosophy of how we engage with each other in the world and how I like to say and Bryce Huber also took that word I think of co-regulation we regulate each other so I wouldn't say that we intrinsically exude disorder in the colloquial sense it's more about the basic notion of active influence you perceive something you do something the question is to what extent is that problematic do the changes that you enact in the world lead to undesirable consequences that's the only question really but indeed we can't really address the question systematically in my opinion without a framework such as active influence and until then we will be myopic until then we will take these externalities for granted and have to wait and we're at the mercy of our discovery when it becomes too problematic such as what we saw with industrial revolution great great progress in the terms that you care about then way later you find out the consequences the catastrophic consequences of the externalities at the point then natural revolution ignored but you know we're working on it and a small side point to add to that is having said that our exuding of entropy is not problematic itself comes also with a re-emphasis on how the entropy we exude are actually really great think of children you know those children are still learning they've taken all this stuff they don't understand and they just express so much chaos but we love it it stimulates us it apparently nurtures us you know people get happy when they see children playing partly because I think when people are doing this chaotic what a chaotic weird messy behavior in the real world we're able to kind of understand what's behind that and I think that's quite a beautiful way to look at it because then you can also see that what maybe messy random behavior for some person might perceive as something stimulating nurturing nourishing so there's a nice kind of ecology going on there and as for blue about three wills I hate the notion free will mostly use philosophy because it seems it almost always seems like free from determinism and I think that's already a trap that you cannot fall into instead it's more about autonomy the behavior that you engage in the actions that you make to what extent are they driven by factors that are your own factors that are more internal rather than external and then of course you would get the problem of oh but we don't have a self instrument journal sure but it's okay then then that is simply because that which is internal is also cultivated as an external but it doesn't take away the fact I think that we can say when there's a journey of process to what extent is that due to idiosyncratic factors which I think is beautiful because a lot of these idiosyncratic factors are cultivated in parts their engagement with other people with other cultural artifacts and so our cultivation of culture, of narratives of interactions with each other our practices, our rituals these are all participating to cultivate this collective kind of autonomy to what extent can we as a collective shape the world niche construct the world in a way that's more adaptive to us as an agent on multiple skills so also I actually don't really like the notion of discrete choices but more a fluid navigation magic as a conscious you have a conscious being space that which is conscious you're more fluidly navigating the density and it express itself maybe as a choice but I think that too is the kind of reification we talked about earlier so not to go too long on that yeah and also really like the notion of this uncoupling so for me the danger is kind of to use a word there's this body catalytic power to some narrative to some practices which shifts the fitness of these practices to a more overly pragmatic mode of behavior one other way I want to describe this is basically a lot of our digital technologies create a training set that is too decoupled from the actual test set that we will encounter in the world which you're seeing the kind of rise of let me know the notion of meat space and service space when you live in service space you're being attuned to a world that isn't conforming to the meat space and I think exactly that is a danger and I fully agree with that even though I'm much very much interested in I do think this is a problem that hasn't been fully addressed is dangerous and the fact that we can't with all the smart people here that we can't formulate it properly or characterize properly is another indication that this deserves more attention yeah so thanks for all the Scott and then anyone else who raised their hand so a small fun point and then a big fun point so the small fun point is Marco you're talking about children throwing off wildness and entropy and that's recreational disorder right because you know that there's a thing there it there's two points on that one one hot sauce is another one of those right because it's uncomfortable but people like it right and so on that there's actually a physical there was article in nature this is a small fun point article in nature magazine that in your senses of taste there are five senses right there's actually two channels good and bad the good channel is sweet which is carbohydrates and umami which is proteins the bad channel is bitter and sour which are poisons typically an unripe fruit and so salt doesn't have its own channel salt goes from the good channel until a level of blood salinity and then switches to the bad channel so the reason for mentioning that is in food you use bitter and sour as highlights as recreational disorder right and so there and there because they're not indicative usually of consumables in nature so that was just an interesting side light on that it's a similar to this notion in the dentist there's this research now that if you have a very painful dental procedure you should actually finish up the procedure with a little extra pain but not but a soreness type pain not a difficult pain because the mind remembers the last pain experienced so the the article said do you want a little pain with that and the idea was extra irritant and the ADA said they can't because of the do no harm they can't administer extra pain as a matter of policy but the idea was to hijack the sensation the memory of the bad experience with the lesser pain which is kind of interesting so that with those small fun points let me get to the bigger point which I think is goes back to what Sarah was alluding to so this is that idea of symbolic and physical beings okay this something I've been playing with for a little while takes a short period to describe and so let me try that so what if put aside all your notions of causation for a minute I'm going to go back to something that Seth Lloyd raised for me a guy at MIT who wrote Programming the Universe and he asserted in his book that all interactions since the Big Bang have increased exponentially all interactions in the universe so two particles hit each other then they hit four particles at 16 particles whatever and then that iterates out to us okay so Moore's law you have this exponential increase in transistor density on chips that has led the fifth order effect of that to an exponential increase in human interactions okay because the chips are more ubiquitous they can digitize more interactions we're having this discussion now was not possible before okay so the more interactions are happening exponentially let's go with that for a second suspend disbelief and just imagine that all interactions are increasing exponentially okay what happened with humans is physical space did not offer enough vectors of engagement to contain interactions increasing exponentially and so we had to go to symbolic space so there were a couple of other I'll get back to that in a second so think of clonal reproduction versus sexual reproduction clonal reproduction you resemble your parents and for niche space and again this is not intelligent design but stay with me for niche space clones are not sufficiently varied so that if the niche space is changing rapidly the clones won't have as much survivability mutation random mutations can introduce that noise that can allow the accommodation of that change in physical space I think what's happening with humans is that we are becoming information beings increasingly and I say that to people like yeah yeah yeah and then I say hey you know that 401k that you have if anybody has any retirement savings the I say to them do you think that's a pile of groceries or a house waiting for you to retire that is data that is nothing the system goes you got nothing it's totally symbolic so one of the if and to the extent that interactions have in and their artifacts are increasing exponentially throughout time throughout deep time it was inevitable that we move to information space because the the phase space available for solutions to our problems was insufficient in physical space we needed more dimensionality in order to address things and that's what language does for this it opens up dimensions it opens up abstractions it opens up things called war it opens up things called stock markets it opens up things called insurance law governance all these things that allow for us to be human and the thing that separates us from other organisms to some extent is that language is where the mind exists because that's the symbolic element of who we become as information being says a lot there but I think one of the things that's been I've been playing with just try to understand what this might be is what does it mean to be converted increasingly to information beings and not when I still have a physical presence but this came up first and last point here this came up in the identity work I did years ago when people were saying oh you can control me on the internet I said sure you can a government a sovereign can put you to death they have the monopoly legitimized violence or they can silence you which is what we've seen happen now and so the physical body if I die right now while I'm talking which some people might like with this long thing I'm saying here if I died right now immediately my legal agency to do things on the internet would change and it's an entanglement it's almost like a quantum entanglement because it would happen instantaneously I cannot have agency if I'm dead so anything pre-programmed in my symbolic self could keep running but the physical world then doesn't respect you as having a will anymore so there is an entanglement so as long as governments continue to control the physical body and have the monopoly of legitimated violence they can control the digital body because putting you to death de-animates the digital body Scott and we have Dave then Sarah then Stephen and we'll steer it back to the paper and also get towards the last thoughts that each person wants to share so again Dave Sarah Stephen and then anyone else okay the it's make a lot of chaos you look at especially one on one games that kids do they are largely dominance game one kid is going to tend to monopolize being the dominant figure the cowboy the sheriff for a while and then switch back now if they hang with that role for more than about 60% of the time eventually it gets actually almost sadistic and it becomes too aggressive the little and things get out of hand the dominated kid cries runs off the smart mommy just kind of gives a little pat and lets the kids go on their way the overprotective mom though almost takes over the bullying and makes it a terrible thing which means the kids can't go back and achieve restoration so when the child who has been paranoid doesn't have the reassurance that oh well okay the world does eventually become okay and the kid who's been in the sadistic infantile mode doesn't get the reassurance well even if I was evil mommy for a while I'm not evil mommy anymore now I'm good mommy so another reason to let your kids whack on each other and get dirty okay thank you Sarah Steven and then anyone else oh I'm Scott's comment kind of reminded me of what I unconsciously was bridging with respect to information Scott said something verbatim I don't remember but basically that something about language and for me you know language languages as with ecosystems as with species like languages are going extinct and so languages are really interesting bridge to this you know information space and it seems that that fact of languages going extinct seems to contradict something you said Scott but again because this is I feel like Mel's show I want Mel to like hook this into whatever she's working on if that makes sense for her Steven, Tim, then Mel or anyone else Steven muted? I was looking for a paper that I was trying to find on something but it's okay what I was going to say is that there's a lot of stuff coming up here and one thing I like about this paper is it sort of starts from the free energy principle and really pays attention to that and the active influence and I think that there is something important because we get into this conversation about information and stuff is like the free energy principle is a colliery for active influence of these stages and different scales and then once that scaling is happening there's also a free energy principle that sort of scales up anyway the natural nature of how these societies try to work with these models and there's a paper that was I think in the physics review of life but it sort of talked about how we naturally try and reduce free energy over time and I think this is particularly what's happened since capitalism in the last 500 years or 200 years particularly is we want to reduce or there's a tendency in different ways to try and reduce the amount of free energy out there and we can do it more and more because we can construct and our focus is constructing these larger scale societies, social networks ecosystems, a super organism as Mel talks about it and we're kind of we're sort of constructing this niche to such an extent that we just take that for granted and these perspectives on things and the rarefication is almost de facto but there's also probably more tradition in history and it's probably also true if you think like what you do when you're out in the middle of the ocean you can't control the ocean so now you have to try and adapt wherever the winds are doing based on what you've got so this that sort of comes back to that idea of the organism as an active influence process to try and minimize free energy and just to stay attuned and alive in a niche is something that's sort of very foundational understanding in physics and chemistry still alludes to and then we have this ability to construct through action and this is where we can start to feel that these actions that we do like we're talking about that we are information but this is the danger of the rarefication of information and we at some extent because we've built on top of things we kind of construct things in our niche which makes it and it is kind of an informational hold in our niche but we're still a being that has to dynamically engage with that niche and that information doesn't travel unlike traditional inactivism that information doesn't come in as like that's a tree it comes in again in amongst all this noisy statistical variations and we infer from that so there's this question about that danger that information starts to get rarefied and becomes seen as something being transmitted down the channel but that's only something that effectively emerges once some processing is done so I think that's what came up to me Thank you Steven so in our last couple of minutes we will go to Tim, Marco and then anyone else so let's just keep it brief and summary and remember that next week in 14.2 we're going to be on the same paper so no need to cram it in in the last minute maybe raise some questions or ask somebody about their perspective so that we can mull on it in the next week and then come back next week with the links ready and with the ideas ready with the questions ready and read the paper another time so Tim, Marco and then anyone else yeah just briefly the child's play maybe it's obvious and it's like what everybody already knows about things but it struck me when you were talking about it that that's kind of like active inference on hyperdrive right because the child's play is this you know generating these fanciful hypothesis you know like the imagination is just going and then that puts them in a position to interact with the world and maximize surprise as they run all over the place and trip over things right so it's like it's learning on hyperdrive and then the only other thing I want to say is wrapping up was as I was going through reading the paper and all those attempts to answer the question what is the free energy principle in terms of the category there was one kind of startling reductive idea that came to mind it seemed really silly but I'll throw it out there could it just be like an idealization like frictionless you know because it's compared to the principle of least action right but if you look at the actual free energy principle through the layers of statistical mechanics like all those layers going into that kind of translation the free energy principle almost just seems like since we can't really seal that through all that cloudiness of statistical and probability and such and it seems like just an idealization a way to say we can do something with this even though it's this distant thing that's gone through this probability machine but we can do things with it because we have this idealization that we call the free energy principle so it's just another another option for what it is maybe it seems silly I don't know yep Mal then Marko yeah so interestingly we've had in philosophy of science we've had a series of how science works for a long time and we've gone through many kinds of popular accounts mainstream accounts of how science works and these can traditionally emphasize series and accuracy you want an accurate true representation of what's going on in nature but and falsification this sort of thing the kind of en-rogue view right now the the idea du jour is really in a lot of ways focusing on models, focusing on idealization, focusing on pluralism there are a lot of great books on this Michael Weisberg's Simulation and Similarity Fantastic Book great introduction to models Steven Downs has a new like tiny little introduction that just gives you a breakdown of all the literature and models and Angela Petocznik in my department has a fantastic book called Idealization in the Aims of Science and then I what's funny to me is I wrote this paper I wrote this paper and then the doctor is not the territory and then a few weeks later someone a professor in my department announced that they were running a course for the spring term on this new book called When Maps Become the World which is actually a fantastic book I'm really enjoying this book and it's touching on all of these things that you guys have been talking about like different modes of representing and do these things represent or not and how the kind of feedback between our maps of the world and like cultural differences and how we represent things and where the world is going these kinds of kind of big picture questions and it's a really lovely read the idea with models right is that what they do is chiefly idealize right if you look at like the WIMSAT the WIMSAT account of what's going on in modeling right the idea is that it's not meant to be a maximally accurate depiction of some natural system or some natural process it's not its fidelity to the world that makes the model useful but in fact it's the idealization it's what gets coarse grained block box away it's about latching on to some kind of key feature of the system of interest and that's always mediated by what our questions are what our interests are in these natural systems of research so it's kind of value-laden in a way it's the starting the information that is inessential to the question at hand so idealization is at the core of what a model is awesome response so any very rapid final words this is the true lightning round Marco, Scott, Alex Kiefer oh dear I'm not good at lightning runs okay very brief so I agree about the importance of friction but I do think it's already you get it for free you get it for free with active inference because you have two situations where I think you will get friction one is pathologically stuck when you're overly rigid because there's no alternatives of sufficient credence so you can't move out of it so instead of saying there's a presence of friction it's more like there's an absence of lubricants or something and the other side you buy a base factor can gets too high because everything is kind of similar so you get stuck in that and I think for me those are two forms of friction I just very briefly want to wrap up that big theme of how we scaffold all these things upon society which is attract and the many many dangers that lie there where indeed you would have I think an autocatalytic pitfall of trying to cope with the complexity of the world but falling into heuristics and when these heuristics are used to minimize your experience consequences of excessive free energy and then adopt that heuristic then it will spread in this kind of self-selecting manner as in the people exposed to these notions such as for example QAnon and conspiracies they heuristically connect everything together and so you seemingly experience coherence as long as you don't stick too far or commit to proper aesthetic virtues and indeed this will become worse and worse unless we have an alternative to better ground at least in my opinion and I do want to push back a little bit on this mostly true claim that models are just idealization but I think what's really really interesting is not just models as presented in papers models as defined in maths but take into account the human enterprise that is science because everyone has to adopt this model not just program it in some code way more interesting is what happens when you adopt or incorporate or internalize an idealized model and for me that's the expressiveness factor upon incorporating it you will de-idealize it you will fill it up you will make it your own particular and that expressiveness in my opinion is what makes it interesting and what Mel has really laid the groundwork for in my opinion we should have talked more about the paper it's dot two it's not about answering every question in the dot one or even the dot two it's about staying excited about including as many people in the conversation hearing from everybody's perspective if you're curious and listening joining the Keybase actinflab.public is a great way to get engaged with action not just inference so again single kind of closing thoughts but less personal review paper my Ted talk but more what are we going to move forward to into 14.2 Scott Alex Kiefer Stephen Sleid the painter Kandinsky said that violent societies yield abstract art and one of the things I was wondered is is the reverse also true is abstraction a form of violence and so that's something I posited to you that 2008 financial break was abstraction of mortgage instruments that caused a violence of people who were kicked out of their homes the math the math of the mortgage was not the territory of the lived experience and the need for housing and so that is where the disconnect happened so something to be aware of in terms of abstraction and the possibility of it being a form of violence and the monopoly of legitimated violence being a definition of the sovereign from mills you might think about the monopoly of legitimated abstraction being a form of sovereignty thanks Scott Alex Kiefer Stephen and then Sarah so one thing I'm still interested in talking about potentially next week is I guess this probably is like a microcosm of some of the stuff we've talked about but sort of the Janesian perspective in thermodynamics in physics and how this relates to all this because I think there is potentially something interesting so one thing I like about this paper I will be brief here one thing I like is that it goes through the history of the formalism of the FEP and where it came from and I think in part that's meant to cast doubt on any direct connection between the FEP and physics but at the same time the Janesian perspective in physics is something the FEP inherits from and that sort of already maybe gets epistemic stuff into physics at a lower level so I want to talk about the relationship between entropy in the thermodynamic sense and entropy in the Shannon sense thank you we're going to go to Mel with the first author privilege and then again anyone can drop off if they need to go but let's bring it to a close together otherwise Mel, Steven, Sarah yeah actually I would really love to talk about that point that you raised and my other reviewer I know who it is they find it my other reviewer Andrew Corcoran actually a great guy raised that there's this new paper by Gottwald and Brown published December 3rd 2020 that is in plus computational biology and it's really getting into Janesian so it's really getting into free energy it's really getting into the Janesian perspective it's like the first time anyone has really dealt into Janes and the FEP so I'd love to discuss that but I haven't read it yet and I'd love to discuss that thank you Mel Julian Janes, which Janes is that? no ET Janes so maximum entropy principle I can drop this link in the comments as well representation and how models represent if they do excellent Steven and I'll go back to just my regular camera Steven go ahead and then anyone else yeah I think that I would like to hear more on that Janes the entropy the maximum entropy approach the whole entropy aquatic thing is really interesting I think this paper is really tapping into that and I suppose one thing I'd like to say is we talk about models we're using models to get the dynamics and there's this slight trap because well that makes sense but you have to take a perspective and verify it and it's like this modern world we just do it but the key thing that's also present with the whole flows of information is that your models can reveal other ways of knowing which are not perspectival and are not themselves so the interesting thing about dropping something out of an active inference model isn't the model, it's the dynamic shifts that the graphs show when the model is running so that and that can tap into indigenous ways of knowing because the thing with indigenous people I was saying earlier is that they're trying to attune to the world not say what the world is and model it and that maybe is what we need to do more to actually survive on this planet as it's an ecological collapse because we might need to sense what is it that we need to dynamically embody that isn't entirely modelable so anyway we could talk about that, that's enough for me Sarah did you have something Marco any last thoughts I just want to echo what's been said by Mel and Alex I also think that one of the most interesting things especially for the next session is about the relation to thermodynamic and information theoretic entropy especially because there's one sentence in Mel's that I just want to point out so it says the elements of the framework do not map on to any known features of real world systems at least not with any more granularity or specificity than the causal dynamics of subsystems but I find that a very strange sentence because it seems that the granularity or specificity of their causal dynamics would be sufficient to act as a model and I think the really really interesting but Mel gives a good starting ground for is asking more about the relation between the thermodynamic and information theoretic entropy where I think we should maybe move more or emphasize the perspective of interpreting entropy as distribution, how flat, how homogeneous, how distributed is it as a tendency to kind of flat out to spread out in the thermodynamic sense and then I think with that conceptual metaphor it's not really a metaphor it's just true but with that concept in mind I think it's much easier to have a conversation about the relationship between information theoretic and thermodynamic entropy that being said looking forward to 45 foot 2 Thanks so much Mel any last thoughts as the first and last author otherwise I'll close it out Phenomenal, thank you this was great Mel thanks so much for not just the paper but for engaging with us you know asynchronously on video chat all of the participants there was so much we brought up today and it was just awesome to hear what everyone had to share so if you're listening live or in replay you're part of the community of practice and we want to just have you participate however so if you're curious about something post a question if you want to know how you can participate just reach out to us through any number of these mechanisms other than that on January 26th we will be right back here for 14.2 and we'll deal with some of these questions everybody will come with a few new digested thoughts couple of polished questions and we will go from there so thanks again everybody and we'll see you next week