 Hello, everyone. Welcome to the Active Inference Lab to the Active Inference Live Stream. Today it is Active Inference Live Stream 15.1 and it is February 2, 2021. Welcome to the Active Inference Lab, everyone. We are an experiment in online team communication, learning, and practice related to active inference. You can find us on our website, Twitter, Discord, Gmail, YouTube, or Keybase public team. This is a recorded and an archived live stream, so please provide us with feedback so that we can improve on our work. All backgrounds and perspectives are welcome here. And as far as video etiquette for live streams, we'll mute if there's noise in our background and raise our hands so we can hear from everybody. Today we are in 15.1 on February 2, and really excited and thankful to have both authors of this paper today and also joining us for next week. So that's going to be really exciting. Check out this spreadsheet if you want to see the most updated event calendar for Active Inference Live Stream. But here we are in 15 with a rapid introduction. Today we're going to have an introduction to each other, especially some of us who are new on the stream. And then we're going to be kind of walking through the paper, having ideas, having fun conversations, and also writing down ideas that we want to discuss more in 15.2, which will be like a follow-up discussion on this. So if you're participating live or watching later, then just write down ideas and questions you're having and get in touch with us if you want to participate. So here we are in the introductions. Today we'll go around, introduce ourselves, and then just run it so we'll have a fun conversation. I'm Daniel Friedman. I'm a postdoctoral researcher in California. And yeah, I'm just really excited to be having this conversation. So I will pass it first to our authors, Enes and Thomas. We can't hear you, Enes, but you're not muted. So reload and let's go to Thomas. Thomas, can you unmute and introduce yourself? And then Enes, just like control our reload. And then you will probably work with the audio. But go ahead, Thomas. Yeah. Hi, everyone. My name is Thomas from S. I'm currently a PhD student in Antwerp at the University of Antwerp. I'm working primarily on inactivism and free energy principle. Thanks for having me. Thank you. We're going to look forward to hearing a lot more from you. And Enes, can we hear you now? Okay. Can you hear me now? Perfect. Great. All right. All right. So I'm a PhD student at the Berlin School of Mind and Brain. And yeah, and my topic of research is on dynamical systems and e-cognition in general. Great. Then we can just pass it to whoever hasn't spoken. So go to Shannon. Hey, I'm Shannon. I'm a PhD student at the University of California in Merced. And I work on sensory motor neuroscience and the dynamics of crab behavior. I'll pass it to Sasha. Sasha disappeared. Let's go to Steven. Select then Scott David. Oh, hello. I'm Steven. I'm in Toronto. I work a lot with participatory theater and community development. I'm doing a practice-based PhD at the moment through Canterbury Christ Church University. And I will pass it over to Alex. Hi, everyone. I'm Alex. I'm in Moscow, Russia. And I'm a researcher in systems management school and still trying to find the ways to join active inference and systems thinking frameworks. And I pass it to Sasha. Hi, I'm Sasha. I'll try this again. I'm a graduate student in neuroscience and studying early brain development and based in California. And I will pass it to Blue. Good morning. I am Blue Knight. I am an independent research consultant based out of New Mexico. I don't know who hasn't gotten it. Maybe I'll pass it to Marco. Hi, I'm Marco. Marco Lin. I'm based in Holland. I'm basically just a free energy active inference enthusiast. Happy to be here. I'll pass it to Scott. Hi, folks. My name is Scott David. I'm the director of the Information Risk Research Initiative at the University of Washington Applied Physics Lab. My background is a lawyer. And what we're doing is engineering systems that are socio-technical systems. So it's parameterization for technical and social elements of active inference-based systems. Thanks. Well, thanks, everyone, so much for coming on live. This is really going to be a great discussion and also a really prescient paper. So great to hear everyone's views on it. In these warm-up questions, we can just begin wherever people have been heading into today's discussion from. So I'll just list them and then people can raise their hand and we'll hear from everybody who wants to speak. What is something you're excited about today? What is something that you liked or remembered about the paper? What motivated you to come to this discussion, to read the paper, think through these topics? And then also what's something that you're wondering about or would like to have resolved by the end of today's or next week's discussion? So until anyone raises their hand, Ines and or Thomas, I just wanted to ask how do you start by communicating this research or what led you to this question? I'm going to kick out the ghost, Ines. So should I take this, Ines, or do you want to...? You're muted right now. Sorry. No, Thomas, just go ahead. Yeah, so I think the way I got to the idea of the paper is that I'm primarily coming at it from a radical inactivist perspective where there's a lot of antagonism towards representations, models and whatnot. In the meantime, I got very excited about the free energy principle and I've been sort of trying to sort out where that all fits, what the models mean, what everything means in everything that's being said. And so Ines and I were talking about it and we started talking about representations and models in the free energy principle and we figured out that it really depends on how you interpret the models and what exactly you take the mathematics to mean. And based on that, we just started listing out the different options you could have and what they mean and what we personally thought about those, what we thought were good aspects, less nice aspects. And through all of that, we sort of came up with instrumentalism as the only path forward. Awesome. I'm sure we're going to have a lot of great second thoughts and conversations on that. So anyone can raise their hand, but either Ines or Thomas, just to lead into those who might not be familiar with long-running debates and the philosophy of science or multiple possible ways to interpret formalisms, how do you convey this instrumentalism, realism question and motivate its importance? Shall I take it, Thomas? Sure. Okay, so one thing that we thought was really interesting to think about is that the majority of the accounts within the free energy principle or process theories, they are very much realist. So that's what Thomas and I started questioning and we think that it's interesting because there's been other people also joining that particular consensus that we also have. So we thought that this would be something that we needed to look at. And the question just to give a little bit of context on where we are in the literature is that the majority of the approaches were on the realist camp, even those that were coming from inactivism. And that's why we found things to be a little bit strange because the first question is whether one is licensed to take a realist account from the fact that we can model the brain or we can model the neuronal activity or we can model cognitive activity. So that's the one question that we wanted to think about and look at whether we are licensed to take a realist stand and if we are, what is the argument, what is the evidence to think that this machinery that we use in our scientific modeling can also exist somewhere in neuronal activity or cognitive activity. So that was one thing. Then the other thing was that it started to be sort of like a wave that was taking the free energy to be compatible with inactivism. And we started thinking about and questioning this because it might be the case that they may not be compatible after all because, as we know, on the side of inactivism, there's like a claim against representations, for example. Then on the predictive accounts of cognition on the other side, you sort of need representations to get your theory going because you need to have predictions, you need to have prediction errors coming, so you need to rely on representational machinery. So that's what we thought that it's important for us to question and think about whether it is actually the case that these two things can be made compatible. So on the one side, the predictive accounts of cognition, which take representations in a very literal account. And on the other side, the inactivism, which for the majority of the people on an inactivism side, cognition doesn't come down or it's not all the way through representational. So this is sort of like where we are navigating the literature here. It's sort of like between the very representational predictive accounts of cognition and the realism that that's a standard they kind of need to take. And on the other side, the inactivism, which there's sort of like a tendency recently to be made like sort of compatible under the same roof, so to speak. Great. Thank you both for contextualizing and everybody can raise their hand so that we can make sure to hear from everybody's excitement here. But yes, the bigger question is, does making a model have a realist interpretation? If you make a model of the brain as a steam engine, is it really a steam engine? No. But then if you make a model of it as a free energy principle, corollary process theory, is it really that? That's kind of the discussion. And there's a lot on the line, which we're going to tease out. So Shannon and then anyone else who raises their hand. Yeah, so sort of jumping to the end of the paper already. You say like the model and scientific practice is a representation of the nervous system to the extent that it holds explanatory capacity. And I think like a lot of neuroscientists that I talked to, if you start bringing up, you know, the representation or the representation debate, that's like, ah, that's something that philosophers worry about. You know, when we say representation, we just kind of need a word to stand in for our conversation so we can talk about like patterned activity of neurons. And we don't care that it could stand separate or not from the situation or the environment, but it's useful. It's a useful word to use so that way we can have language and talk about the system or the pattern of activity that we're interested in that's associated with some behavior. And I just think that sounds right, like how you conclude it. And maybe I didn't have a question. Maybe this was a moment where I had to comment more of a question. So apologies, but if that sparks some conversation, that would be great. It's something that stayed with you and will change how you think about even other kinds of models. Next time you see a linear aggression, maybe you're going to wonder if it's being used in a realist or an instrumentalist context. So thanks for sharing it, Shannon. Steven, and then anyone else? One thing I think is quite interesting is that just like Shannon was saying about, you know, often a word gets used, but then it's kind of fuzzy and then it kind of gets like hand-waved away around why it's maybe not that clear. I think this also maybe the same challenge comes up with pragmatism because pragmatism is really useful and it gets used in a lot of contexts. But now, because everything's slightly pragmatic in the sense that it's all about action through the free energy principle, it becomes so calling instrumentalism sort of helps in a way get a bit of clarity when it's something which is about theory. So I wonder, I'd be interested to know about how the pragmatism wasn't used as the word and used a more sort of clear unpacking, I suppose. Yeah, great question. So anybody, authors or otherwise, what is the relationship between pragmatism and utility and then instrumentalism? So we're using the instruments ostensibly for pragmatic ends, but the authors chose to frame the discussion in terms of realism versus instrumentalism instead of pragmatism versus say anti-pragmatism or some other, you know, could be pragmatism versus realism, perhaps. So any authors or other hands? I don't mind saying just a few words and then perhaps I'll end it over to Thomas, but I just wanted to, the one way that I think that it's useful for me to think this is the use of representation. So in what sense are we using representations? Because Shannon was saying she was raising a very good point. It is very useful to refer to these things that this activity that's going on in the brain as representations. But I think that in what sense are we using representations as important? Because if it's in a sort of a metaphor, then obviously there is no problem because it is actually very useful. We are building all of these, we are trying to make sense of all of these data that we gather from the brain, and that's in some sense building a representation, because what is a representation? Representation is a sentence, is a picture, is a map. These are the kinds of representations. Why is this problematic? It's because these are things that we as human beings do. We do that in our daily lives and we do that in our labs. So we built these representations to make sense of the observations that we do in our labs and in our daily lives. So then we can obviously use this word, this concept, but it doesn't have the same meaning, or it does in the realist sense. So I think it comes down to how serious are we in the use of this word? Are we using this in a literal sense, in the sense of sentences, maps, pictures that we as human beings have and build and develop, and the brain also has them, or are we using it in a metaphorical way? Which is, if it's in a metaphorical way, because it's useful to use the concept, then it should be fine. It's a metaphor. Okay, and that's it. Really interesting, a metaphor is really an instrument. When somebody says that this movie was, you know, was like, it's like an ocean of ideas or it was like an ocean, a simile, they're using that sentence representation instrumentally to convey a point, to, you know, communicate something that is internal to the thinker. And then that is not meant to be interpreted realistically. But in science, where there's a strong realist kick amongst some people, they think it isn't science unless it's not going to be the realist, most mechanistic thing. Now, whether it's mechanism all the way down or what the lowest level mechanism is, that's sort of where those realist ideas bottom out. But without going that way, it's just, it is across fields and across different media. We ask whether we're meaning something really or instrumentally. So any other thoughts? What was some like a, oh yeah, Scott, people can raise their hand on Chitzi because I'm not always looking at the actual video screen. So thanks, Scott, and then anyone else. I was just wondering if it's interesting in terms of the encoding and the decoding notion. So if you have different isms on either side of your encoding and decoding, that seems like where you have the problem potentially. So if someone's encoding a communication as a metaphor, but it's taken not as a metaphor, but as something stronger and is decoded that way, then there's a miscommunication. One of the things that I've been trying to fuss with and it feels like this is that place, the painter Kandinsky said that violent societies yield abstract art. And I always wondered whether that's reversible, whether abstraction is a form of violence. And here, that feels like that might be relevant because if I abstract something with one ism and then it's in my encoding of my message and then it's decoded by someone else who's decoding it with a different ism, then that abstraction has led to a violence. In fact, it might lead to actual harm to that other person because they may misinterpret my message based on that schism of isms. I had to go there. But anyway, does that something, that challenge of communication across boundaries feels like it's invoked here and what might be some strategies for doing something about that? Thanks Scott for the great question, Thomas. And then we just continue down the stack so people can just throw up their hand raised whenever they have a thought and we'll get to everybody. Yeah, I think these three comments really together cut at the heart of the matter. I think that indeed from what I have heard myself, neuroscientists pretty much often know what they mean with representation and they don't take it to mean anything much more robust than a sort of correlation between particular patterns of activity and particular things that we see happening. And it's more robust than that, but not anyway what the philosophers take it to mean. But the issue is then exactly in communication where it feels like among neuroscientists it's a bit of an insider's knowledge that representation is sort of a metaphor. Like the computer metaphor as a whole, it's a metaphor and you shouldn't take it in a realist way, as we would say. But when students read a handbook, this is not always that obvious. And I've also talked to neuroscientists that really believed that the rats had a literal map in their brain like we have a sort of JPEG on our computer because they were talking about representational maps in the brain. And if you know a little bit about philosophy in this area, philosophers take these neuroscientists talk about representations to mean just about anything that they like. So I think that the issue is indeed one if you want to talk about that way of encoding and decoding that they encode a metaphor. And some people that are in the know decode a metaphor, but many people don't. And I've noticed that exactly with the free energy principle where it feels like it gets very muddy with what is metaphor and what isn't and what people take to be metaphor and what isn't. In that sense, my main project here is to try and create clarity in what it could possibly mean. And even if people completely disagree with the position, I think at the very least I would like just people to be more clear about what they really mean. Thanks, Thomas. And really one of the goals of these discussions that we're having. So we're going to go to Shannon, Stephen and us. Yeah, this is actually really great. Thomas and Scott, I hadn't thought about it this way before. So no one in the general public cares about the representation wars. That might not be true, but like nobody cares. But whenever science from a neuroscientist gets picked up by a journalist who then makes the article with the headline that says your brain has models inside of it that it's running and you should learn music because it will make you smarter but learning coding won't make you smarter. Which was like a headline recently. But that translation process that Scott was talking about that encoding and decoding really gets stuck and it doesn't get decoded when it gets written about in sort of like more mainstream news outlets. Like maybe someone's medium blog post that they're writing and they're a really dedicated grad student who's grappling with these ideas all the time and they're very clear in their writing about what scientists mean when they're talking about representation or a model. What this means for the actual brain itself and then what this means for you as a person just reading about it. But that doesn't always come across in like the headline on CNN or like the Facebook feed that's scrolling through my adult parents' computer screen. And so being clear you have to go through this step of using it as a useful metaphor in neuroscience getting clear what the metaphors could be if they're real or just instrumental readings in the philosophy camp and then like the one more step is bringing the science communicators in and getting them to parse this huge debate down into something that's able to be distributed to folks who aren't enmeshed in this every day. Thanks Shannon, so Stephen and then Sasha. And this clash is definitely a useful discussion because I think radical inactivism and free energy are both trying to come in and say it's not all in the head. So just like you're saying the challenge you've got is any and as soon as people talk about neuroscience and that it's like there's just an assumption in our culture that it's all in the head. It's not even so that metaphor is like a signpost and then everything else becomes and is it a film playing in my head and it's like oh wait a sec it's the mind-body environment with a dynamic system and it's about action and they look at you with this like wait so it's like it slips into that so I think this is a really good good point because metaphors when you embody them are kind of like you inactively understand what it's like to be inside them but then when they just become like a signpost for what to do they lose all of that kind of architecture so I think this is quite valuable. Yep the metaphors and the language Steven and related to the science communication point it's oh well this is hardwired in or the DNA codes for this protein it uses the language that's implicit in a very specific understanding and that metaphor gets overinterpreted and then you start well where's the RAM in the brain you know where's the GPU and then you start building out this metaphor before you know it you're way down the rabbit hole with something that was basically in terms of this paper taking a realist reading on an instrumentalist project and again we just think about other kinds of modeling if you did a steam engine model the brain it's not a steam engine if you do a linear aggression it's not a linear aggression so if you do an active inference model it's not an active inference brain but that is very nuanced to communicate because of the way that our language is so Ines and Sasha then anyone else yeah I think these are very good remarks I just wanted to I think I have to disagree just a tiny bit just to make a point about I think that there are real issues and there are real dangers in conceiving of in the ways that we conceive things and especially in the ways that we use our language and our concepts and particularly looking at the concept of representation but there are others that I can think of the concept of hierarchy for example that could be a metaphor could be used in a realist way there's new papers on this which are very relevant and very interesting but what I want to say is that in the particular case of the representation I think there are real dangers in using it's not just how the science is communicated or oh maybe this is not a problem for me to deal with let's just the philosophers handle that that's not real for my particular lab it's not going to be an issue I think it's going to be an issue because I think the way that you conceptualize however the processes are going on in the brain is going to dictate the way that you design and you interpret the results so I'm thinking for example there's a bit in the paper where we make a distinction between two kinds of modeling techniques that we can use and they have two very different goals ambitions, worries and they end up in very different conclusions and so one of them takes up as for free that there is information in the brain so this kind of modeling is the kind of modeling that uses structural causal models and what they aim for is to think about the relations between the nodes these are nodes that are highly connected and the way that we design and understand or interpret the data coming from those models is by the presence or absence of information now this means that we are taking information for free information for existing in the brain so what kind of information are we talking about is it the kind of information that I also have in my model because if it is then we're talking about semantic information and I mean it may be the case that the brain actually works with the very kind of information that I have in my model but we need to prove that that's an assumption and this is how we are building our models and I think that's one worry now the other worry is that in building these models that's the inactivist worry which Stephen was just mentioning which is if we say that well the cognition comes down to these representations happening in the brain then so much for the rest of the cognition that is in the body and the body being in the environment so that these all is inactivist sort of like tradition and I just wanted to add that cool thank you we have Sasha, Shannon, Alex yeah thank you for explaining that so well and bringing this to the forefront this is an issue to be grappled with I think from the neuroscience background yes it's just relegated to philosophers to deal with these bigger questions but leading to this going back to this idea of encoding and decoding it really does matter what is meant by the phrase and then how it's decoded and I think in the world of like molecular neuroscience and just recording signals from the brain I really think that we're measuring something real about the system whether or not that's what's actually happening that's just one measurement of the overall behavior and to just lean into instrumentalism it's a useful metaphor for what we're seeing but obviously it people don't have the training or the decoding of the philosophy that's active on what that means and so I've never thought about it until this paper that this could be a real issue even in my own work so I think it's really important to get into the nuance of what we mean when we use these metaphors thanks Sasha Shannon Alex well okay yeah go ahead Shannon yeah so that's really great I didn't mean to say that like only science communicators and philosophers are should worry about this question but I think so like something that this reminded me of in sort of my field so I look at music perception in the brain and I kind of have a question of does the brain use internal models to predict timing mechanisms and this metaphor that you brought up that how it affects the assumptions that you make and how you design or analyze or interpret your experiment I think it also makes a case for pluralism like if you're realist then pluralist about realist approaches if you're making metaphors then pluralist about the type of metaphor that could be useful so we thought or a lot of neuroscientists will think there's a clock mechanism in the brain or a pacemaker that's just like tracking rhythmic activity but maybe there's this move now that there's more of a Bayesian mechanism that's building up a timing model that's now predicting when or what the next event will be that occurs and one of those might be more explanatory useful and the other one will fall away but I don't know if we can actually ever prove that there is a Bayesian mechanism in the brain or that there is a clock model in the brain I think as far as we can get is that taking this approach is more explanatory useful than taking the other approach who and it made me think from my background as an evolutionary biologist about the question what is a gene and that's a philosophical question that they don't really discuss too much in genetics departments but of course the way that you consider what a gene is it's going to set up what you measure who's included on the research project how you write about it, how you analyze it I mean it's the primary determinant so whether what's happening with the electrical signals in the brain is part of a real model that is being built out or whether that is something like we're just measuring the temperature but then we're modeling something totally separate and then also just quickly the pluralism it's easier to be pluralistic with metaphors because it's like we can always write new poems we can always have new metaphors we can always build and do yes and with the metaphors but realism it's a little harder to be pluralist because ostensibly it's not multiple things at once or potentially we can't all be living our own truth for what the brain is but we of course can have our own perspectives so actually it totally opens the door to pluralism in a really qualified way where we can come together talk about what's useful rather than debate about who's right about how things are so really nice framing so Alex and then Ines thanks for me as I came from engineering world to learning active inference instrumentalism approach was an only way starting to understand what is happening because when I start finding and look at models in this framework it was a way for my mind to get basics and trying to go deeper but as for this paper for me it's great because it gives great promotion for instrumentalism approach and I just think if there is an ocean for radical instrumentalism if so can we use it and how it should be described and again for me honestly I don't care much about philosophy debates or even neuroscience debates I'm watching on a world around in terms of systems and trying to understand with systems basically as black boxes without understanding how this deep rabbit holes inside working so that's why it's interesting for me as for reading this paper and with discussions too thanks Alex so Ines and then Thomas yeah I just wanted to add something to what you were just mentioning Daniel about the pluralist approach I think that yeah you said that something that sounded very right to me which is one can only take this pluralist or it's at least easier to take the pluralist route if we are on the instrumentalist side so for example with the example that we have in the paper with these two kinds of different modeling approaches that we can take like dynamical causal modeling or structural causal modeling if we take an instrumentalist approach then we can say that these two different sorts of modeling techniques they can work together where for example what you get from the structural causal modeling which is you get like a sort of a nice topological description of the connections arranged in the brain as we get a nice map of that which can sort of combine with dynamical causal modeling where you get for example the dynamics in terms of states and how the activity occurs you can combine this in a pluralist way as long as you take this much more instrumentalist approach and you don't take it to be the case that what you're getting from these models is what is actually the case so I just wanted to add that Yes, we have one instrument that helps us find the structure and the topology the connectedness we have another instrument that captures or helps us do inference on a slightly different attribute and it's not a philosophical debate whether it's actually just dynamics or actually just topology it's actually us investigating a system we care about and then that's what can bring people to the table so Thomas and then Scott David then anyone else Yeah, I think that's really quite right I've spoken about this paper with other people as well and some were worried that instrumentalism was placing too strong of a limit on our scientific endeavours that if we say oh well we can only ever be instrumentalist then they said why do science at all if we can't really get at the heart of the matter if we can't actually add some sort of finish line say okay well now we have just an accurate description of what cognition is and how organisms navigate the environment and I think that's a fair worry I wonder if perhaps it's just a bit too ambitious but you know who knows what we can do in 100 years and I think I do think in that perspective it is important to see that the paper doesn't really get into that grander philosophy of science debate about model instrumentalism across the board right so it could very well be that at some point we'll get a model that really well describes everything that really happens in in an organism maybe the main point here as I take it anyway Ines might disagree of course is that at the very least in the free energy principle you don't get that entailment of because we can model it the organism models as well but that's not it could be that in the end we can take a realist position on some other model in the future nice so we're not committed to yet this instrumentalist across the board we can be realist elsewhere and we can be realist in the future but you've laid out a research agenda and a logical structure for asking whether you know have we met some of these claims that were brought up in your paper on another topic or for free energy in the future on that note to quickly get back to what Alex was saying it doesn't necessarily mean that we have to commit to radical instrumentalism we don't have to be that radical although that might be called engineering by some so maybe it already exists so Scott and then Ines yeah and along those lines so I mentioned before that I was an attorney so rhetoric is persuasive speech and it is what we attorneys produce and so it's interesting when you talk about instrumentalism and realism in terms of rhetoric so I used to say to my kids all the time that the only time you see reality is when you see paradox and any other time you're living inside a model and so that means that when people talk about resolving paradox what they're talking about is whose model wins but I always told my kids you should manage paradox not try to resolve it because that is reality so let me take that what I'm suggesting is metaphor may be reality in certain contexts because the management of the paradox happens through rhetorical speech the incommensurables are brought together with rhetoric to create a story to make it a thing and so what I'm wondering is I was thinking of the analogy of well tempered clavier when Bach was tempering musical notes there's all sorts of notes out there and then the scale was decided upon and now it's a scale and it's objective and it's mathematically demonstrable why it's a good scale but it's not the only scale and so that kind of idea of managing paradox through I guess it's instrumentalism because you're allowing for multiple approaches what I'm wondering is is there a preference for one or the other in in trying to motivate change in groups and then the second part of that so that's one and then the other part is I'm beginning to understand that the I believe that the mind doesn't exist in the brain it exists in the language and the brain is an antenna that's tuned into that so from that perspective can anything be other than instrumentalist so you made a claim about what the brain is so that was a realist claim about it but I yes great questions Scott so we got Annes blue okay I just wanted to add something to what Thomas was saying actually whether we lose something when we take the instrumentalist part and I think that my concern is that I'm not sure if we can even have the realist part and I'm not sure if that's an option so the reason being and I'm going to come back to the examples that we have in the paper because I think these are like practical useful examples concrete examples that we can think about amongst ourselves even if we come from very different backgrounds but there are some scientists here some neuroscientists here and I think that's useful to you know come back to what we're really talking about with examples so with the two different kinds of modeling or let's say just two major ones that we refer to in the paper where you have one kind of modeling that is trying to capture a topological description of how things are and then you have these other kind of modeling where you're trying to explain why things behave in the way that they do and that's sort of like the more dynamical systems one so in a sense in the dynamical systems one it's explanatory because you're trying to explain why things behave in the way that they do and the crucial point here that we also point out in the paper is that when you explain to answer the question why you need to have a theory you need to have a hypothesis so these means that these models in dynamical causal modeling they are hypothesis dependent so you are looking for an explanation as opposed to a description in for example structural causal modeling attracting functional connectivity for example so what I want to say is that in these two particular ways of modeling things you already find two different goals and two different strategies one is descriptive the other one is explanatory now my question is whether modeling this in this way we lose any explanatory traction I don't think we do I think that this depends on what we are doing and how we are doing because if our research is going to be driven by a hypothesis then you're going to get this explanatory traction and let me just give you a very quick example so let's say that you are modeling a pendulum right you're modeling a pendulum and you're trying to explain so there are two things you can do with this you can describe the pendulum and you can also try to explain come up with an explanation for why the pendulum behaves the way that it does right and once you do the explanations obviously it's very obvious for us because it's a very simple system the explanation is going to be well that's the laws of physics obviously right but what you've just done is you just explained what a pendulum is doing why the pendulum is behaving like that by using these two different ways of modeling the pendulum why it behaves in the way that it do and how it behaves in the way that it do you have your why by using the laws of physics and you cannot say you cannot at any point say that the pendulum is representing the model and yet you explain why the pendulum is behaving the way that it is so I just want to say that I'm not so sure that we should take the realist or that we need to take the realist part just on the account that perhaps you lose explanatory traction well we are explaining the natural world by modeling the natural world so I just wanted to I'm going to retrace that gem of an argument actually realism is looking for descriptions explanations about how things really are instrumentalism is testing hypotheses and then the key twist in the discussion in the argument is our hypotheses are about explanations of the system so we actually get the benefits really without the baggage of realism because we're hypothesis driven investigators as scientists searching for explanations through hypotheses so by centering the hypothesis quest we can be instrumentalists and get the realist fruit but if we act like we're realists like we're searching for an explanation or a description of a system we get lost, we're not hypothesis driven and we're trapped with all this baggage of realism so very clear and really thanks for sharing it we got blue, Thomas, Stephen so I hate to like jump off of that like wonderful explanation by Ines like regarding realism but I have to go back to what Scott David just said about the mind and the brain because it's something that I think about a lot and that I am pretty deeply involved in but language cannot be where the mind exists and I'm not going to say that I think that the mind exists in the brain because I don't but I think that the mind also cannot exist in language because that just it totally you know negates the experience of someone like Helen Keller who didn't acquire language until she was well into her late childhood so before that did she have no mind? I don't think that's fair but also like in any kind of trauma experience people don't have language for the trauma so were you mindless at that time or do you have no mind to conceptualize the trauma so there are many experiences that happen on a very primal and very emotional level that cannot be explained or represented with language so I think that the mind can't exist there and then maybe to tie this back into the realist the realist versus instrumentalist like is language real so I mean I'm one of these people that really questions reality and what is real like what can we take to be a realistic representation and so language to me is an instrument so I mean it's something that it's not even, it's a human concept so it doesn't exist outside of our perspective reality which I mean to me like it totally pushes the realist debate off the table because like what even is real and I have a neuroscience background but I know like even in neuroscience there are very wide schisms like people that think that reality is a concrete objective thing but I'm like you know reality exists here and like that's it right so I'm on that other side of the wall anyway that's it so sorry no questions just some thoughts. Thank you Thomas, Steven and then anyone else? I think my initial point was just to react to in as her beautiful explanation there and I really think that should have been in the paper because it's actually a really really good way to set it up yeah that's really good as for the where the mind is and what's real I think I'd love to chat on about that for probably hours I think it's very interesting and I think the inactivist perspective is a very nice middle passage between there's an objective reality that we are approaching or it's all in the mind however I think really getting into that will probably take us a little bit too far away from active inference so I'll leave it at that but yeah the you know the conversations on the freeway all these fun keywords and side discussions the off ramp so thanks for sharing it Steven then Marco yeah I mean the culture does come back to how we frame looking at reality so we tend to look at reality either with a perspective and not or to get that perspective we use these tools as we're a very kind of like tool based kind of culture now but I like in the paper it does say it says in that first abstract it remains disputed whether it's statistical models are scientific tools but then it says to describe non-equilibrium steady state systems so in a way non-equilibrium steady state systems as kind of like a foundational ground that could be to some extent real it's the question of the models that are used to describe how they work that isn't real you know so I think that's also like this doesn't like there's still room for the idea that the brain at some level even though it's quite abstract is a non-equilibrium steady state system it's just that the models the statistical models shouldn't also be the I don't know the way that that system's doing its job so to speak so I think that's kind of and I suppose what that also then brings back is how do we get at that and this is where I'm quite interested in indigenous ways of knowing because well if you don't come on from a perspectival room and if most of what if you come from other like metaphors is one level but other sort of swarming kind of non-non-literal ways of knowing you know non-verbal maybe that's another avenue that opens up it says okay maybe we have to come at it through other ways of modeling anyway thanks Steven Marco then anyone else hi thanks so first of all just fair disclosure I haven't fully read it yet but it's an interesting discussion and Scott mentioned paradox and I'll take that up I actually suspect that the paradoxical interpretation might actually be the correct interpretation so we have a nice table of possible interpretations for the FEP a two by two table but what if it's all of them right so the FEP especially in the elaborations and active inference is already kind of paradox because you already have these trade-offs between accuracy, complexity or the epistemic and the pragmatic aspects so in kind of a weird way the whole discussion in the fifth topic around the different interpretations seems to me to parallel the way that the let's say first deflationary the statistical densities in the brain actually fulfill different roles depending on the actual relation with the world right when I'm actually just imagining about the world then the representation relation between the densities and the targets are not actually with respect to the world they're actually respect to each other in my brain but nevertheless we still situate in the real world at the very least if we assume an objective reality as in a reality that is not dependent upon our interpretation until we act upon it then at least we can say that by virtue of FEP at least what it suggests that the way that these models sorry the way that these statistical structures and densities are shaped are shaping by something that is real so in that sense I personally sorry I have a lot of allergies for some stuff but I'm very much against this whole notion that the representational content or representations have to be adjudicated by some accuracy or correctness condition and I think it's exactly FEP the challenge is that why the hell would that be reasonable that a representation has to be correct rather it is shaped by some sort of correctness but if it would be stuck in that we would never get somewhere as an organism it's not about being true or right or correct all the time it's about being to ratchet yourself in virtue of this adaptive imperative I'm not sure where I'm going I still have to partly observe everything but maybe what do you think of instead of choosing one of the tables cells one of these possibilities more embracing what is in my opinion part of the FEP and saying that it's actually paradox of it's all of them thanks Marco and I also agree the FEP helps us by thinking about how the observations get to us whatever side of the blanket or the interface we're on we can take those observations with deathly seriousness instrumentally it's not that the actual movement of the water molecules is going to kill you but there are temperatures that kill you and so we can take our measurements seriously and be instrumentalist from our side of the blanket and say yeah you know there might be a real system that's hypothesized as instrumentalist about what is out there but if we're waiting on our side of the blanket until we have resolved what is quite literally on the other side of the blanket then it's going to be arguing about what's inside of a box that can't be opened and then policy failure is going to be the outcome so it's a really interesting point about how we're kind of using these FEP scaffold to actually navigate this paradox so really cool there so Ines, Scott, Thomas Okay, I just wanted to add to what Marco was saying I think it was a wonderful remark and I just wanted to that also allows me to draw or highlight a point that we made in the paper that was also very fundamental and this is just that the FEP in itself does not have to entail this table we can talk about the FEP talking about this table and that's a point that we try to make in the paper so this table only comes up as something that is relevant to us once we start talking about process theories or the corollary active inference because the FEP in itself comes from a variational free energy which is a measure of the variational free energy to attain model evidence in variational base so we don't have to talk about this table to talk about the FEP and we don't even have to talk about active inference or process theories that's when we start talking about how is it that the system is minimizing free energy that's when we come up with this process theories to try to explain how is it that the system minimizes this free energy right and that's when we start bumping into these conceptualizations that our explanations or our models that aim to explain the process by which the systems minimize free energy also exist in break right so we don't have to take any realist stand or any instrumentalist stand until we start talking about this process theories and I think that sometimes I think that it happens a lot that in the literature these things come conflated that the free energy principle necessarily entails active inference or necessarily entails or is compatible with process theories like prediction error minimization or predictive coding or predictive process and these are actually not there's no entailment between them as we saw and as we tried to point out in the paper if I could ask a quick follow up on that actually so let's say we're in the FEP world then there's all these corollary process theories that are falsifiable we've talked about that last week and in several other weeks what is it though that leads someone to accept or reject or be attracted or repulsed by the FEP so within the FEP we know how to get from here to there and you're right there's no strict entailment it's actually a totally scientific empirical question but how do we get to that FEP understanding that leads us to the downstream process theories I think that there are different goals and at least for me it's useful for me to think about it like that so what we have what we gain we're talking about FEP is that it gives us much more sort of like grasp or gets us closer to what things really are biologically speaking so what we get is we get these bounds between like life and death by the form of entropy so it is this key concept which is entropy that you do not get in process theories because precisely process theories are in a different game they're in the game of explaining the processes by which cognition is an activity is conducted so it's the processes and the FEP is on a different business which is the business of looking into stages that's why it comes much more related to dynamical systems theory so that's a point that we try to make clear in the paper that it seems to us that comes very often conflated in the literature that there is an entailment I mean active inference is a corollary of the free energy principle so it is very important precisely to tell us how it is that systems could be minimizing free energy but there is no necessarily entailment even though if you want to tell a much more detailed story then active inference becomes very useful thank you Scott, Thomas, anyone else who raises their hand? Thomas I think you had your hand up first is this something you want to jump in on? nope, you're okay you're muted Thomas yeah, my fault unless you have something that's not really related to the current conversation you can go first because if it's related, if it's related Thomas go for it otherwise go for it I noted some comments to a lot of the things people said regarding the question of what we get from accepting the free energy principle I'm honestly not very sure I'm still trying to figure it out and I'm hoping we can get some very neat statistical models of behavioral dynamics and that we can somehow get a better grip on the patterns that we see coming back to what Marco said if I understood the story correctly by saying that why not all of the options simultaneously in some ways the model is a very neat description in another way the brains model captures something is based on something real in the world I think what happens there and this is something that I've seen in the literature as well I don't remember who wrote about that but I've seen some people saying it's the best of both worlds because we get both representationalism and inactivism and I think the issue with that is that it assumes that realism is a valid position it assumes that we can talk about such a thing as a model in the brain in a non-metaphorical sense it assumes that there is a model in the brain which does or does not belong to something real and I think in that sense the way that you often see people talking inactivism and the free energy principle together is by assuming realism and ignoring the inactive criticisms of what realism entails and there was a second point about whether we need to see representations involving correctness conditions I think that's a very fair point but my response to that would be that sort of comes back to our talk about encoding decoding and language and being clear in general we for the sake of our paper decided what we call representations involves correctness conditions so if you have a different notion of representation that does not involve correctness conditions that's a totally different story and maybe those types of representations could be totally fine if all you mean with representation is a co-variation relation between two different systems I'm not going to dispute whether that exists or not that seems fine but the problem is that if you are that liberal with what representations mean everyone's going to use them in different ways and you're going to get that we all use them but nobody really wants to talk about what they really mean and that's pretty famously at least among philosophers described by Bill Ramsey in 2007 book representations we considered I think it's called so the reason to say representations have correctness conditions is mostly just clear on the terminology so just a signal of this is what we mean when we say representation and if your notion of representation does not have correctness conditions that's something else to talk about and that's another discussion whether those are valid or not then Thanks Thomas and in an earlier discussion with Maxwell Ramsted and others we had a conversation about how part of representations is that like they have to be able to be wrong or they have to be able to be off base otherwise they can't be on base so yes lots of other interesting work Scott if you would like then Steven, Marco just said something going back to the earlier conversation it was when we were talking about the basically the ism schism and the ability to I'm staying with that and discern what was going on in the system it reminded me of the holographic principle and I started reading a paper another paper last night on why holography and it was just kind of an interesting notion that the Markov blankets and the holographic principle bear a similarity in there's an opacity to them and so the internal states and the perception of those internal states from outside may be based on different isms and that's something that may not ever be bridgeable I guess and it may not matter we're going to talk to Chris Fields about holographs in March 2021 so coming up very soon thanks for this great point Scott, Steven and Marco yeah just to build on the this idea that you go from free energy principle and then we go to this process theory so it's almost like at some point you've got to put dimensions on what's going on like there's an inside there's an outside there's so with active inference there's the idea that the stuff's going to go via the Markov blanket and there's a way to know and that way of knowing it's still to some extent dimensionless because there's no direct way to transmit the dimension information it's only through this entropy and I know entropy is mentioned in the paper so I'm kind of interested in this idea of when you go from say the entropy in this kind of general sort of non-dimensional sense it's just a way of knowing and then you've got this kind of high you talk about this idea of a low entropy states or high entry states but you've got this kind of this is sort of the jump between the two in a way where there's like how because with active inference it's the change in entropy which gives the way through the back door to encode knowledge so because it's through the change in entropy in relation to action in relation to sense or data that you sort of somehow build up a dimensional understanding even if that understanding is only in the mind maybe not at the organism level so I suppose I think that maybe I know there's a new paper you've brought out on skillful performance that's just come out so maybe this sort of opening up these other questions around entropy and dimensionality and when you actually go from modeling changes to how do these changes relate to information on knowledge flowing or being generated from the blanket through to the generative model and the world Yep, Steven, thanks a great point you're getting at this question which is how does the dimensionally compressed information from the world like a panorama that just presents itself get enriched through our experience and internal model into something that has what you're talking about dimensionality and then it's a good question to ask how it relates to the thermostatistical lineage with entropy and some of these very pervasive top level concepts so really well brought out Marco Thanks I was just wondering about Thomas' comments on my comments if you could elaborate on first of all the criticism from inactivism as he mentioned but I would love to hear more about that especially in relation to Max Lorenzo and other paper on inactivism tale of two densities also about the correctness condition I understand that you have to choose a definition of representation and again I do understand that you have to respect the lineage in philosophy and what has been established as authoritative or conventional but I do believe that there is a certain need I think to also revamp philosophy because I feel like there's always this kind of weird dynamic of like the two clubs and the school playground with philosophy and science one is constantly trying to radical revolutionize the other but philosophy seems to be so stubborn in not allowing science to revolutionize them right they keep trying to the philosophers keep trying to to yank the rigid prize out of the shell but vice versa somehow there seems to be no path we're not allowed to do that so for example why not let's this field of science slash philosophy yank that prior of representation having to have correctness conditions out of that shell because for example you can also just kind of play with it not the correctness condition but saying that in the limit it should settle into correctness relation with the target as if it's kind of fixed point or attractive right you can also say it's not a condition it's not a limit but it's a factor that contributes to that thing that we call representation so I think I would love to hear a discussion about that opening up the whole role rather than this essentially property the role of a notion of correctness in relation to representations as situated in the world and in turn I also want to again emphasize even though that you're right in the difference between the corollaries, the many corollaries potential corollaries of FEP versus FEP itself where FEP constrains a particular way we create process theories and there's another thing that I think is not often enough talked about which is the kind of implied picture of sentient systems for example from FEP and corollaries where we're not often enough talk about the inter-relational system the way that these systems in sentient agents and complex systems relate to each other so it's more specifically for humans or us we also self-model and when we're self-modeling we need to model all the different structures we possess and know what role they play in the big ecosystems that we are and so in the same way we can say that in our self-modeling we also intuit or sense that some of those subsystems have a very instrumental role for example you could say executive or retentional mechanisms of the prefrontal cortex is effectively instrumental and in a way realist about itself but the fact that it has to organize or regulate itself such that non, let's say primary physical observables needs to remain a certain bounds that seems to me instrumental because it doesn't matter if there really is some quantity or some notion of stability of feeling good or feeling safe right? it's just settled in that way but at the same time if it does latch onto something real cool and maybe you can elaborate upon it sorry yeah it's all good no you bring up okay so point one was related to inactive critiques point two was about correctness conditions moving beyond them, point three is sentient systems so if we want a question that's answered let's just try to trace out the thread so that we can actually return to it because they're great points yeah it's great points Marco is that good or is there more there's just what the nuance of third point is more about in the process of self-modeling we include the notion or implicitly into it's with model notion of do these things have instrumental role or more of epistemics that these theories are more about modeling processes of modeling rather than modeling, essences and properties yes the relational all the way down type thinking so thanks Marco three sides of a very nuanced discussion and an ongoing questions so Thomas then anyone else who wants to raise your hand yeah thanks for that so I'm not sure I completely follow everything so if I get it wrong just interrupt me and let me know so I think you mentioned regulation like bodily regulation as a form of self-modeling so I think that accepting that bodily regulation is a form of self-modeling is assuming the realist position there is a clear fact that you see biological organisms self-regulate whether they do that by exploiting an internal model is another question entirely and that's exactly sort of the distinction between instrumentalism and realism that the paper tries to tease apart so it's almost that when you mentioned the modeling the process of modeling is the instrumentalist side on Marco's way of phrasing the third part and then the question about essences is the realist question so the movement towards instrumentalism is related to this relational insight so I think the issue is I didn't exactly get the relational insight that's the part that was a little fuzzy for me and I wasn't sure exactly what he was getting to yeah Marco where do you see that relational just because it's something that comes up that it does sort of cross different idea or debates so what does that mean to you or how does it play out here yeah so thanks so often when we we see literature on active influence or fantasy principle there's always a centerpiece of a market centerpiece of assertive system slash agents but very rarely is it about figure one I'm showing figure one just if you want to refer to this exactly so it's always this but it's never this with another one that's like this it's very rarely there's the market blankets market blankets of life paper that does go into it but very little and start the shocking in these kinds of philosophy of soft papers is there actually talk about how market blankets interact with each other and the reflexive of the mutual influence because in that process of mutual influence how do you actually do that model the other models that you possess or are it's not just in terms of properties as such it's also about the role they have the bigger inferential ecosystem but I guess maybe that requires a leap of faith in active influence first if you already assume that the dynamics of government by inferential imperatives then you also have a small step to believing that in modeling each other they also need to model each other's inferential contributions where we go again back to the definition of 300 principle formulation in terms of epistemic and pragmatic or the active complex right maybe one mode of engaging with the worlds or incorporating one particular set of systems of your total constitution will lead to more of instrumentalist approach right when you're just instrumental engaging with something not assuming you have the correct view of something when you have a new game or some weird object you don't know what the hell it is you're not going to engage with it based on a realist assumption you're going to engage with it based on idea of I'm going to figure out how to manipulate I'm going to find all the instrumental points of latching onto it right and so no matter what you cannot escape the fuzziness or the vagueness with the fluidity of these so seeming dichotomies in my opinion and that's the relational aspect right that in terms of modeling and its many facets you need to understand the relations of the different systems as yourself model especially if you're extremely complex like humans playfulness human as a curious investigator playful insight relational thinking and a very something that kind of is personal and almost distinct from philosophy and or science is the personal which is we're really studying this and we're people who have certain valences towards different modes so pretty interesting Thomas in us yeah so I think I think that's a very interesting question and whether we approach something from an instrumental as a realist perspective but I think still that all I mean I might be way off base here but from what I hear it sounds like it's all still within the realm of realism and only when you take us to be modeling at all can you say about can you start talking about whether that modeling is instrumentalist or realist so the instrumentalist position that we defend in the paper is one in which we only model when we literally do so in the world when you're using your computer ram to start up a model or when you're drawing a diagram when you're literally making models in any other sort of activity that you're doing you're not modeling you're doing the activity so in picking up a cup I am not modeling the cup when I'm not in any sense models are not relevant there I'm picking up a cup and and then I think the other point you made a little earlier about the interaction of science and philosophy and that it seems that philosophers are always trying to revolutionize science and science is not allowed to revolutionize philosophy I think that's a fun point as a philosopher but I think that I think that with the popularity of predictive processing and active inference in philosophy you see where science really is revolutionizing philosophy as well and I think especially in philosophy of cognitive science there's always going to be a bi-directional interplay I wouldn't have been sitting here if it were not for the influence that science is having on philosophy of course I'm delving into all of this and then when you're talking specifically about the notion of representations if scientific endeavors could make it extremely plausible to draw away the correctness conditions for representations then maybe that is something we need to accept but the issue is that I think anyway just at first glance that representations are it's a word we use also outside of science right a picture that I draw as a representation of whatever it is a fancy drawing and there are certain features of that picture that I drew that are particularly relevant in making that picture a representation of whatever it is I drew now when you want to use that same word in science you might say well none of that is important and we're going to drop some of the very important features of the picture and use it very differently in science jargon that's fine but it's a little difficult to deal with when we already have a notion of representation and what representations are in just the regular dealings in the world but you could technically say sure we'll call it something else something that we'll have to all agree on and then be clear about what that means or not thank you Thomas so Ines and then anyone else so we have like half an hour or so lots of thoughts prepare them, get ready we can flip around to different slides Ines then Mark up yeah I just wanted to first I want to say the great points Thomas that's precisely what I'm going to spend or try to explain a little bit on there's a lot that I'd like to say but let me just go back to the self-modeling point as the bodily regulation so I'm really struggling to understand what these means and the reason for that is because we need to motivate why would the body have to model itself so typically one starts this modeling sort of like endeavor when we do not direct access to things so we wonder about what could be causes so I wonder why I'm really struggling to understand what this self-modeling is because in terms of body bodily regulation because what is it that the body needs to model really because we start modeling we want to model things so we don't have access to a true posterior when we do not have the causes so we need to find the posterior but the self-modeling seems a little bit counter-intuitive because it's the self that is right there so the body is right there so it sounds very counter-intuitive to me to think that that should need any kind of internal modeling in the body so the other point that I wanted to react to was Marcos point because Marcos seems to be a little bit worried about the lack if I understood it correctly the lack of having a program or a framework that would allow us to get things an explanation there would be much more relational all the way down and how would we link all of this together and then when we came to this slide I think so I just wanted to say a few words on that I wanted to say that the cool thing about Mako Blankets as we know is that we can apply them regardless of the scale we can apply them to any scale and we can nest them and I think that this is something that Mako could be a little bit worried about and perhaps this is not going to be the solution for what he was mentioning but maybe it will get us in the direction so the idea is that with Mako Blankets we actually understand the relations between different aspects and levels of neurophysiology as being together related in a multi-scale kind of system so basically the idea is that you can use a Mako Blanket to zoom into the level of description that you want to investigate but without losing sight that this particular level is already embedded in a multi-scale level system so then this allows us to zoom in up and down for example within the brain and that's a paper that I have that's going to be out soon so it allows us to zoom in to zoom up and down in the brain but it also allows us to zoom out to the organism and then get out to the organism in the environment so that's a nice thing that we can do with the Mako Blankets where obviously the internal states and the active states are going to correspond to the level of description that we want to look at and then obviously we are going to look at the embeddedness of this system within its environment because its environment is going to be the external states and the sensory states so that's the cool thing that we can do with Mako Blankets is that it allows us to zoom in out and up and down in a multi-scaled system Thanks Ines great points and I think for this zooming in and out seems very relevant to a lot of our work and a lot of what's on the cutting edge so maybe we can have like a workshop or a panel or something like that or we'll see it later but this is really interesting stuff so again maybe 20-30 minutes so if anyone has live chat comments then anyone else who raises their hand then Scott Thanks I just want to go back to models and multiple levels skills and stuff so Thomas says something about you're not modeling something that you're just doing and I agree again like I said my claim is that it varies over the situation context so again that's why I argue we need to zoom out to taking the relation so if you take the paper on active infinite curiosity with the games right so first you're exploring you're just doing the game allowing the statistical patterns as are impinged upon you kind of accumulate shape and adjust each other etc and indeed at that point you're just doing at the immersive and co-variational relations etc but at a certain point there's some kind of transition you have a moment of insight the model sorry let's say the picture more naive because in a sense of the game arises into your awareness as a complete picture and at that point you're able to at least hypothetically write down the model that you think that the game is governed by and so you said something like we're only modeling when we're doing the computer thing and claim that this is going to be a reversible model of the environment but the thing is what's the chicken and egg right if you're doing that game you're at the mercy of the patterns and then you obtain this internal understanding of how the model looks that allows you to create a computational model that means that the point where the model is instantiated is actually in the agent itself right and so that's why I talk about the relational thing and the end result regulation when you as a system are able to model yourself then you have different orders of that model the moment that it's just intuition just you're guided by the variational patterns just letting that be latched onto your action and policy selection yeah it's not really maybe just maybe it's not the model then but the moment that at this level above you're able to say that this is a model of that and this is the systemic relation pragmatic relation then I think we should be able to say that's proper and model proper right and another thing so something Thomas said about when science allows us to throw away the correctness condition I think we have that in psychology and social psychology a lot of people can develop an understanding or a model or a station for worlds that is actually not governed by correct conditions if you I don't want to get political but you know you see in certain political areas where people seem to be governed not by truth conditions but more about pragmatic conditions right but which beliefs if adopted or which ideologies if adopteds will I be better off in the world right that is not a correctness condition that that is taking instrumental this or pragmatic primacy for the sake of your fundamental imperative a lot of ways to sort of preservation etc and so this is a point about an indirect axis so yeah we're one self but it's most level multi-skill so there remains inferential distance between the locus of certain areas of yourself and other locus of yourself so the hypothalamus is crucial in self-regulation in terms of guiding the endocrine system but obviously it's distant from the actual systems that it is supposed to be in relation to right so there needs to be an acknowledgement of the inferential relations and inferential distances by which the self as a whole learns to regulate itself yeah I'll just leave it at that great thank you so in the last little bit let's think about tying threads up and phrasing things what we want to know it's great and exploring what we want to do next week so we're raising questions to think about and for next week as we tie up some of these really awesome threads so Stephen Scott Ines Thomas nice so well I just wanted to sort of add on to this or sort of maybe this is something that you can come back to this idea of representation I suppose it is ultimately seen as something that's in the brain and a model is also sometimes this is maybe where we get a problem is we kind of have this folk psychology that models are in the brain as well so if I think if we were to say models are in our niche so if we say we take this figure one the model is something we create in our external states in our niche that we work with and I have tools and we manipulate out there and we engage with our active states sensory states internal states to work with it can get rid of it helps to sort of get around some of these challenges and I think because I think that that is the nature I think this is really useful thinking about correctness and accuracy of a model is when can that need for directness and accuracy be loosened somewhat because it might be that you don't have to model the body doesn't have to model its active states or model its sensory states or model those exactly but its somehow using something like them and there is some sort of bigger model which we can try to use as a tool to understand what that aggregates to so I think this is actually maybe opening up quite a rich thread Thanks Stephen Scott, Enes, Thomas, Shannon I am raised thank you Scott Enes, Thomas, Shannon So I just I think that it's important the question that Marco was raising so I just wanted to sort of highlight that and just like sort of gesture to it for the next session so I think that Marco was mentioning this important self-regulation aspect and I wonder and this is like a very embodied kind of thing it's not available to our conscious levels so we don't know what's happening so it's not available to consciousness that idea that it needs to be to entail some kind of like modeling aspects because for us to modeling aspects it's because there is not a direct access to the body and I think that one follow-up question to that would be to think about what self-modeling would mean on a conscious level do we ever do that do we need to self-model on a conscious level because if we do then perhaps there is like a sort of like a barrier between the self and what I think is weird so I would really like to discuss that with everyone and the other question that I think that would be nice to discuss next week is to actually get down to the needy greedy to the detailed technical details that would allow us potentially to think these things are what Thomas and I had in mind in the paper which is whether from the fact that we can be trained to use for example a vibrational base to attain likelihood of model evidence which is like what we do patient inference and modeling that kind of thing and also from the fact that we as human beings in our daily life we can think and engage in sort of like a productive reasoning to wonder and make inferences about what we think is the most likely case of things around us from the fact that we as scientists can do that and we as regular people in daily life can do that does it follow that the nervous system itself or the body itself does that and engage in those kind of like tools so it doesn't be mine, yeah nice, I'm already capturing them on a slide so we're definitely gonna get to them Ines, Thomas, Shannon, Marco oh sorry, yeah Thomas, Shannon, Marco yeah, thanks I have a lot in answer to what Marco was saying so I'll try to slim that down as a very quick probably not to convince you but just as a description of how I would see things if you're playing a game and at some point you've figured out the patterns of the game and you're able to make a model of it I would say where the model is instantiated is when you draw the model out that drawing you made the description of the patterns is the model and when you're producing that drawing that you can share with other people that other people have access to you are modeling any other modeling models are never in the brain, models are never in the head models are in the world in our interaction with the world that only exist because of our social cultural heritage in which we have learned to use models sort of what Ines was getting to as well and then when you say when you're talking about some people being led not by correctness conditions I think that's very fair and I think none of us are generally led by correctness conditions of what we do except for signs when we give it our best shot we're just doing stuff but the most important thing there is not whether that means then that we need to change representation for it to mean something not involving correctness conditions but maybe that means that the notion of representation isn't really applicable here and instead of changing the words we use we just use a word that is more appropriate to what people are really latching on to and I think that will be just communicatively maybe an easier strategy instead of continuously changing the words to fit our current understanding just sort of come up with a different word that fits better. Some neologisms and also our growing terms list Shannon and Mark. Thanks for all of that discussion I'm wondering if in search of terms that would better fit our description is the term expectation more or less vague than saying representation or a model and what I'm thinking of is when you're standing your body has a certain expectation that you'll remain standing in a room you expect like you at a conscious level maybe expect to stay standing but also your cerebellum and muscles and your trunk are expecting to be in a certain relation to each other because the ground's not moving so they're just maintaining some standard standing posture maybe a little bit of balance side to side if you feel like you're standing still on a train and then the train starts moving when your expectation was to stand up straight now you have a deviation from that expectation your cerebellum, your motor cortex all of the muscles in your trunk and your legs react or predict that they should have been standing upright predict that they should have been in a certain relation to each other to stand upright and they either update that prediction or they act to change the world so if acting to change the world and all of your muscles will tense up in a different way to keep yourself standing upright if they're just accepting the prediction that you shouldn't be standing already you'll fall over and I just wonder is it helpful to say that your body has a model of standing or like Ines says your body just is in direct access to itself so even if there's an expectation of standing upright there's still like maybe it doesn't help to call it a model but there could be correctness conditions in that case if you called it a model and like if you're so far out if you're so far out of allostasis then you're in an incorrect correctness condition for standing and so to respond to that perturbation is to bring it back to that correct condition and I don't know I guess does expectation help or is that just even more vague than saying that you have your body has a model of itself Fun, thanks Shannon Marco and Es Yeah actually what Shannon says is very nice because I think again that re-emphasizes the importance of of distinguishing the kind of operational constituents of how models play a role in agents so what she was describing seems to me like the interaction of how the afferent and efferent expectations and data have to interact and that indeed I agree that doesn't have to be called a model I would just see that as the element or the operation of inferential interactions and kind of a zero with order or a proto model maybe but again it's a couple of that when you have the order above that where I think we have to say that there is a model because then there's also the modeling of how these inferential constituents relate to the world that has to be modeled too for example sometimes we interact with the world in weird ways and we notice it and we have to actively reflect about how we were guided by our actions I mean I don't know which philosophy it was but philosophy about self-examination examining your beliefs assumption if that's a modeling I don't know what it is and it has raised some very nice points so the rule of conscience too about conscious of regulation and I personally believe I am one of those people who does attribute a function to consciousness or at least attribute a functional drive that leads to the emergence of consciousness where by the sheer fact that it is the kind of the apex of higher and higher levels of modeling similarly how multi-sensory integration is the consequence of this in the very same way it seems stands to reason especially if you look at works like Graciano's attention schema modeling that's the very drive that leads to multi-sensory integration consciousness is the need to model that which is below that I'm sure there's an active interest in that too I forgot in response to what it was but also wrote down somewhere that it's relevant to take also Mel's paper in this discussion about models because I think it's very beautiful there in contrast to this discussion is the diversity of nuances regarding what a model can be and what they are useful right although that would probably stand at odds with a lot of an activist but you can draw a lot of parallels between the different roles and values that model plane have for scientists and the various ways that models as embodied could be useful right for example some are good for conceptual exploration some are more truth oriented some are more to kind of integrate different things and I think these are all valid ways to look at models and they survive simply by virtue again by virtue of what free energy principle entails is that the success of these models is what leads them to persist right and the success in terms of their pragmatic and epistemic contributions to minimizing or reducing variation free energy right and so it's like oh wait yeah Alex Johnson has a recent paper right well one of them had the recent paper on it's not about it's about what must be true if you've gotten access to paper yes yeah yeah so given that we have for analysis models what must be true given that we have them right so the answer would be they got there because apparently they contributed to a reduction of the variation free energy right so atoms I gotta say I think it's kind of one of those disagreed things because I feel a sense a bit of axiomatic commitments to model simply cannot be embodied although I would love to continue the discussion well I also think that adapting a language is key and again taking the norm of principle free energy it's the philosophers who disagree most people use it like the way that I've been describing right folk psychology and scientists use representation in less of a seemingly anal status rigorous notion of correctness condition it is simply a part of a bigger picture of how their value should be assessed so I think in general it's an example of niche construction we are embedded in a world of language where linguistic bonds so adapting a language instead of throwing it away seems to me the far more reasonable approach especially if you want to avoid confusion so I'll leave it at that well it's always interesting to re-listen to one's live stream because while we're speaking we think a certain way it's a different mode and an S you really have a finesse with constructing philosophical arguments on the fly as many other people embody and the way that that is done performatively Stephen I'm sure also you know resonance but like the way it's done for the first time performatively is very different than hearing it in a different state of mind it's like it lines up in the moment but then a different moment and then it's like because that's so it's really interesting points like are we self-modeling when you're improvising are you self-modeling or you doing are you just controlling your larynx where is consciousness fitting into all of that those are really great questions in S and then any like closing thoughts we could go anyone else who wants to speak in S and then feel free to drop off whenever you want okay I just wanted to sort of like address Marcus and then address Shannon points very good points so there was one thing that Mark said that strike me very very interesting because you're saying about that we need to so we need to choose the models models that are more you know suitable and I think that by saying that obviously taking a realist a realist perspective and thinking that there are actually those models in the body or in the brain so that's sort of like what Thomas and I challenged so I just want to sort of like give my two cents on that particular thought and say that I don't think the success in choosing models but it's more like the success is in choosing the most desired states that are going to get you what you want which is like you know avoidance of dissipation entropy and certainty and all of that so states are preferable models don't exist so that's kind of like my take on that and I just wanted to also address what Shannon was saying which was really interesting because I thought that it seemed to me that Shannon was trying to you know build up like you know okay let's just really get down to hands-on and say like how do we really apply these things that we are discussing here into like for example my lab and I really like that one thing that I think that is interesting that happens that I see happening a lot with my conversations is that we don't even see how we are already sort of like in completely and fully enculturated in a way that we speak about the body and the brain and cognition in general with so many assumptions that these things exist because it's just the way that we've been enculturated so as Shannon was oh I think I wonder if maybe its expectation is a better word right and then like it struck me because I was thinking about it okay so the body has got this expectation but from what from what so is that like because for the body to have a certain sort of expectation I would probably think that maybe the body would need to be like sort of like separated from the brain so it would have like some kind of like expectation or some kind of like inbuilt system where the body becomes like sort of a umunculus and also has its own expectation so detached I was like a little bit like I'm not sure of what to do with that so yeah so that would be like a point on whether the body has got this expectation and and how I think that another way to think that would be to think that the whole organism is evolving towards something right there's a sort of like at least like one goal that is shared by all living systems right of course interactions are very idiosyncratic we really context dependent but there's like one goal that is like sort of like in general times shared by all living systems and that they don't want to die so it is evolving towards sort of like avoiding this dissipation so whatever the state the body is at it's gotta be dealing with uncertainty right either so then Shannon was giving the example of like oh it's standing so is it there's an expectation of like a certain expectation in regard to being standing I think it's more of like the state that the organism itself is which is either certain or uncertain it's stable or it's not stable it's fulfilling sort of like a state that is gonna lead me to the next state so I think that would be like a sort of better way to describe the behavior that is entailed in these cognitive activities that we do because I'm not sure how to think like the body itself expecting something but maybe I'm just I missed something here Phil, thanks so much for that. Shannon, Marco and then anyone else who wants to give a last thought Great thanks for that Ines where you sort of ended where you have like most organisms have an expectation of being in a more stable state than a more unstable state depending on the context and depending on the environment and everything but the point you made about enculturation like how does the body have an expectations and I guess when I said the body I kind of was implicitly including the brain so if when your body is actually upright your brain and your muscular system have an expectation of being stable in that upright position and I think enculturation we could also just talk about like the development of certain structures or expectations or situativeness or something so the same way like enculturation gives us an expectation for a certain form of language our body just by existing in the world and being in standing states in lots of different contexts develops an expectation of what a stable standing state would be like but I think where you ended sort of on this like constant evolution sort of away from death or away from perturbation or towards stability I think that covers it and that was helpful Thank you, Shannon, Marco and then any other comments Thanks Regarding anticipation in the body, some examples I'm not sure if this translate is normally Dutch but I don't know if you've heard of it orc, orc, orc soup is eaten with a spoon but most people will say fork especially when you're kids in kindergarten so that is more of a linguistic example because it relies on notion of rhyme which you feel to it and the prediction error is again contingent on the model or the linguistic models you have but also just in general for example with friends with great relations or imagine you're at a bar, someone really drunk comes to you as is your really big guy with scary tattoos and there's no words need to be exchanged but if he comes really close to you perhaps you tightly in the shoulder there's gonna be expectations and it's felt in the body and a lot of these expectations especially imagine you're a racist and someone colored comes to you so then you will feel a physical response of anticipating a certain action or something but because it's contingent of uncertain racist belief which is culture there clearly needs to be a certain interrelation between the models with the abstractions that you've gathered through culture and the physically or viscerally felt anticipation of an action and so I think in a lot of ways what I'm trying to say is that a lot of the physically embodied anticipations clearly are contingent upon abstract or model or representations of something because if not we wouldn't have these visceral bodies so I think a lot of my personal problems with activism is often they look at the manifestations, the tip of the iceberg the marks, the labels but not the necessary path that it has to have taken to get to that manifestation and you said also about you know we don't need models we just get our preferred states but then you get the problem of how do you get to your preferred states it's not that easy I want a million dollars and a happy family but you know how do I get there and so it's going to be contingent on a lot of complicated detours and I simply don't believe that you can escape the notion of a model if you let go of the whole correct this condition but I don't think you can escape at least a model in the broadest sense as Mel's paper I think nicely alludes to especially with these examples of getting to preferred states given the complexity of our world and given the cultured reality of that experience thanks Mark so I would love to talk about that next time awesome well on discord or however we can continue some text brewing and thinking and it's awesome because it's just like a snapshot into our learning and thought on all these very important questions so Steven, Scott, Ines Thanks well I suppose one thing Ines mentioned there and this is this is quite rich is maybe how does expectations can expectations always be to some extent they can't buy minimizing uncertainty it's like a backdoor way to get at it you know so it's kind of but I do think there's something quite interesting about another way of thinking about representations beyond the model and beyond some internal brain state and that's where but there's something called mental space psychology which is quite interesting about how we might be constructing designing understandings of a space now I'm not going to mention that now but that just something that comes up is maybe there's other mechanisms for modeling that we work with which is in between just the body moving and it being in the brain but is everything to this active state sensory states relationship so has to be capitulated at some level using our sensor and using our actions so there has to so anything that's represented in a way we think it is somehow being capitulated in this active influence process it's not being just happened offline so to speak without any kind of embodiment Thanks Steven and the good news is we only have to mention the keyword once for the AI so Scott and then in this one thing I was thinking about is you know we talked about this as a way method of analysis of existing systems biological systems etc and it's also the thing that's it I've been doing historically is creating synthetic Markov blankets I think I alluded that to that before a contract is essentially a synthetic Markov blanket that creates a being pursuant to the attributes in the contract and so you're taking two different systems and linking them harmonically coupling them through a set of expectations that are documented in the contract and so one of the things that's kind of interesting to me is that idea of realism versus instrumentalism I think it goes away in the context of synthetic systems because you're essentially instrumenting a system from whole cloth saying hey we're going to create this entity and it's going to have these attributes like a corporation doesn't exist except to the corporate law and the attributes that are assigned to it by shared meaning but it's all fiction but it's a very broadly shared fiction and so it's interesting that notion of what we use as descriptors for existing living forms and those things that we set up to create niches in a sense for the new synthetic forms and that's something that I find intriguing with this model from the paper is that when it's taken to the synthesis of new forms it changes, I think it changes that analysis because it's not a question of discovery but of creativity in terms of the setting up of the attributes in the Markov blanket, thanks Thanks Scott, if I could rephrase that it'd be like the realist is focused on those essences of the world they're going to try to describe what's already out there that's like big databases of our current legal systems but once we're in instrument landia we're talking about designing apparently arbitrary fictions and then manifesting them and enacting them and so there, the rate limiting step once you're taking the instrumentalist turn is literally the innovation and the narrative buildup and how much collaboration you can achieve exactly without getting into the realist allowing people to have a different take on what is quote really happening so really interesting and it may allow for the design of scale free interventions in that case because you're not just relying on what happened naturally but rather you can create some predictive possibilities by the way you design one level to allow for the engagement on a scale free way thanks yes and then all the last comment okay so I know that we are already a little bit over time and I do have another meeting so I'm just going to try to be like super concise and I just want to say like one final remark regarding regarding up the realism instrumentalism so I just want to say that imagine that I have a model I just have built the model to understand how make sense of brain data for example that I collected from the scan I can have this is a philosophical attitude I can have either one or the other I can go and say no my model but that uses all these tools all these mathematical and conceptual tools is actually how the brain is working and generating that data is not only on my model it's actually how it is or I can say that no this is actually my tools that we as a scientific community have built to make sense of the data that we actually have access to and try to generate the activity that we don't have access to so we need to infer it so just one comment on that particular distinction and then another thing that I wanted to get to is that Marco was saying that I can't see how we don't have models how do we get from one state to the other and that kind of thing I just want to say that in that regard that we don't have models we do have representations on the agent level we don't think we use abductive reason we use logic we have concepts we do make representations we represent things we can do that on the agent level so scientists do it we in our daily lives do that and that is totally fine no one is disputing that the dispute is precisely on whether the underlying mechanisms of what allows for us to do that such as the body, the environment of a system actually themselves also have that capacity so that's the question about models models on the agent level yes on the underlying mechanisms of what allows us to model that's a different story that I'm not convinced about and then about how do we get like from one state to the other I think that we choose we think we use again reason, abductive reason we have representation we use all of these intellectual machinery that we can and have because we are in cultivated systems and living beings with language and we choose how we get from one state to the next and then that can be all mapped out with dynamical causal models if we want to I just wanted to finish with that thanks so much thanks everybody for this really fun discussion I'm glad that we opened so many great threads in point one I'm going to provide a technical wish anyone can drop off if they want I'm going to provide a technical wish and then a concluding thought so the one, technical one is it'd be fun to have an automated system for philosophical classification of statements even if only within the domain specific FEP so a statement like the brain has to maximize the A matrix is a real the brain is a a realist take even if the person has some other perspective but in that sentence grammatically versus we use this version of SPM to model this, it's an instrumentalist sentence so classifying sentences and then even just kind of like a time series like shading our live stream by the way that people are speaking with an underlying generative model and then are they speaking from a position of implicitly like are people speaking from a position of you know in some other way things that are like admitted states that could be inferred from linguistic patterns and then the concluding thought was Scott David talked about paradox he's talked about it before, Stephen brought up that humans are like a tool culture everybody, not just digital culture humans are like a handy work tool people and so Thomas said what do we lose with instrumentalism and so that's really just to conclude by saying that's what's on the table is do we just cut our ties with reality right for realism and whatever that is lost along the way there or do we cut our ties with reality and go fully instrumentalist where does that leave us adrift if we have a really good ship where do we steer it because now we're going to be in more direct contact with our preferences with our governance structures because we're going to be really clear that science isn't like advancing some noble truth agenda per se it's something a lot more cns thank you it's something that's more on the utilitarians the technological it's an engineering development if the FEP is within the realm of philosophy and engineering but not academia for example it's not too foreign of a world that could exist which would be different for an idea than other kinds of ideas and then it's the heart of the matter and is the heart of the matter how things are or is it how things are related to each other so that's which story we're in at that top level is the heart of the matter that instrumentalism sidestepping the question you're losing everything because the heart of the matter is how things essentially are well instrumentalism is only sidestepping that target with the heart of things are if you think that's essence but if you think that the heart of how things are is relational then maybe you can be a realist about instrumentalism you say it's really I'm really making this scientific model it's really the code that I put on the paper this is really the conversation we're having and all these sort of things like that so just it was really an awesome discussion much appreciated all of you active info labbers and we hope to see other people participate in the future if you're listening this far you probably could come on one of these conversations hopefully so thanks for participating everyone and the forum is in your calendar event for those who participated live everyone else just stay in touch but see you next time for 15.2