 Hello. Welcome to the Acton Flab Livestream number 28.1. It's September 7th, 2021. Welcome to the Active Inference Lab. We are a participatory online lab that is communicating, learning, and practicing applied active inference. You can find us at the links on this page. This is a recorded and an archived livestream, so please provide us with feedback so we can improve our work. All backgrounds and perspectives are welcome here, and we'll follow video etiquette for livestreams. Today, on September 7th and next Tuesday, we're going to be discussing this paper on computational phenomenology. Then we haven't set the paper for 29 yet, and then we have a few papers set for October. Today, the goal is just to learn and discuss this very cool paper towards a computational phenomenology of mental action, modeling, meta-awareness, and attentional control with the parametric active inference. And we're really appreciative that Lars, the first author, is here to discuss with us. So we're just going to start wherever we start and go from there. I guess I think each of us here probably brought some questions and anyone who has a question in the live chat can ask it. So let's just introduce ourselves and end with an author and go from there. So I'm Daniel. I'm a postdoc in California, and I'll pass it to Dean. Morning, I'm Dean. I'm retired. I'm in Calgary. I'm in Canada, and the work that I do ties in nicely with this idea of being able to bring some precision to how we learn. So I will see who can I pass it to. Dave? Dave? You might be on text-to-speech mode, which is fine. Right. Welcome, Dave. And we'll pass to Lars. So Lars, welcome to this discussion and just thanks again for joining. We're happy to hear any introduction. And then we'll go from there. Yeah, sure. Well, thanks. Excuse me. Thank you. Thanks for having me. I'm kind of excited to be here. Sounds like a fun couple of hours ahead of us. Just sending regards from the other co-authors as well. Obviously, like a lot of these papers, this has emerged in collaboration with other people and they couldn't make it here today. So, but I'm glad I'm able to be here. So my name is Lars. I'm based at, my affiliation is with the University of Rio in France during my PhD there. And my focus is explicitly on meditation, actually. And that's what I'm interested in. It's what brought me into academia. I had a first stint in academia in physics a while ago and kind of left it and went into corporate stuff for a little while and then was drawn back in by disinterested in meditation. Sorry, there's a bit of background noise. I don't know if you can hear it. But yeah, I guess I'll leave it at that. The kind of genesis here was from that angle through meditation and ending up in Carl's lab and asking questions around attention and meta-awareness through that lens of practice. And that's where this kind of paper emerged from. So we'll get into that in a little bit more detail. Cool. And where did active inference come into play? Was it from something you were learning about from that practice perspective, like you mentioned, or a theory-driven perspective? Or did a collaborator bring it to the table? It was a collaborator. So I sort of came into computational neuroscience looking explicitly to contribute to the field of meditation research. And so I'd reached out to a lab in Lyon, Antoine Lutz's lab, who's one of the co-authors, and joined the lab there. And they had an ongoing relationship with Carl Friston's lab because Jeremy Matou, who's one of the co-authors here, did his postdoc over at UCL. And so they had a bit of an idea for an attention project using active inference. And I kind of came in, not really knowing very much about it, but EGIS Learn and then was sent off to London with a pretty open blank check, as it were, of like, see if you can figure out how to apply this to meditation and attention. And we had some directions that we were thinking about, but then what ended up emerging was this hierarchical sort of deep parametric stuff. Interesting how it kind of came to you. Yeah, it really was right place, right time. They were looking for somebody with a physics background with an interest in meditation and computational neuroscience. And I sent them an email and that was pretty much the extent of the interview process and off I went. Cool. Well, welcome, Steven. Dean or Steven, if you have a question, you can raise your hand or, of course, anyone in the chat. Maybe, Dean, did you want to talk about a crosswalk or something like that? Yeah, I want to do that later, but I don't mind asking. Yeah. It's fantastic to get the background about the meditation piece because when I, before I picked up the paper, what I was wondering about was how precision plays a role in learning and especially when you're moving from novel contexts and trying to familiarize yourself with something. And so for me, my curiosity was around the idea that when you have the simple model and then you move to these more nested paradigms, you're kind of moving from the unplanned because you can't plan your perception to the plan, the idea of being able to think about your thinking. And so that transition was what really fascinated me. So now that you brought up the idea of meditation, I wonder how meditation goes from the unplanned to the plan. So you've opened up a Pandora's box for me because now my original question has kind of now been taken down this tangent. So I'm curious. Right. Well, I mean, I can speak to that a little bit. The idea for me going in here was to see, I mean, just intuitively, as soon as I got familiar with active inference, there was a resonance there in terms of how the maths seems to echo a lot of the kinds of ways that people talk about practice in general. There's this sort of balance between, you know, attention and the, that our perception is this combination of prior and observation. You know, that idea and like how these two things are balanced and getting into the optimal space there is very, very close to, you know, if you start thinking about mindfulness, obviously is a way of paying attention and increasing the impact of the sensory observations. So then also things like don't know mind, beginner's mind, this kind of fresh kind of stance that is very common in Zen, but in other areas as well. And so already there's this kind of like, oh, there's something, there's something here. And so I arrived at Carl's lab and what I started out with was just getting familiar with the attention literature, because that's at least, you know, the good way in for meditation, because at least in the very beginning, that's what it's all about. And in some ways it's about that right to the end. So the question was, you know, how do we talk about paying attention within the already and I was really fortunate because I was kind of coming into the space. When all the scaffolding and the maths I needed to start talking about this in a more practice related way or metacognitive way was already there. All it really needed was somebody to come in and ask a question that was for me phenomenologically driven, which was in meditation, you are paying attention for sure, right? You are deliberately paying attention to some object, you know, traditionally the breath. But really the aim of the game is or an additional part of that, which is really, really important is being able to monitor where your attention is. So in other words, being able to pay attention to attention. And so that's where that kind of shift happened, right? And you mentioned like thinking about thinking and it's like, well, okay, so if we know some of the maths about, you know, what it means to pay attention. What it would mean to pay attention to attention and what would it mean to be able to have control over the attentional states? And how would this, how would we make that an allowable thing within within the framework? Yeah, so exactly the deep affective framework kind of set the set the scene for that. And Casper Casper tells work sort of gave the gave the hint and that was on a different precision in the model and here we're just talking about likelihood precision. And then that kind of created this this opportunity then right that then becomes almost generalizable across and that's when we can get to it a bit later. But really what I'm more excited about in this whole work and paper really is what we start to hint at in the discussion at the end pointing to the fact that we could generalize this across precision throughout the model that could then be interpreted in as different forms of mental action and what that might mean for understanding. Yeah, I mean, it gets broad at that point. Awesome. Thank you for the summary and welcome blue at all at feline. So Stephen with the raised hand and then any live chat or anyone else with a raised hand. Yeah, welcome. Yeah, I was I'm really curious. I like the you're saying about the generalizability or the potential for generalizability and you're sort of extending beyond your phenomenology to phenomenology in general, I suppose is an aim. I'm curious how how you see action potentially coming in with meditation, because you talk about sort of reflecting on mental states, like breathing is an action I suppose but in many cases, they use meditation because you're reducing the amount of action that's happening in the amount of visual stimulus because often the eyes are shut, you know, bodies contained and and now I see that, you know, okay, so how does this then come back into an action state and I know that there's a kind of in between like, I know vip vipisana type meditation on the body senses or like you mentioned the breath but I'm curious how you see that kind of an active piece starting to come in through this. No, thanks for the question and it's a really central one, because I mean, I guess to give a little bit more history to the this paper either is this is also building on a little bit of some of the intuitions of previous work by Antoine Jeremy and another collaborator Giuseppe Pagoni, and they wrote a paper called on the epistemic value of inaction, which speaks to this this point. I think to in the beginning off from the from the outset, we can kind of separate, you know, mental action and overt action and a meditation, you know, you're putting yourself in this situation where you deliberately don't have that much overt action right you're sitting still. And yet you're doing things you're definitely doing things with your mind, even sort of do nothing meditation is still a kind of doing in a sense. Well, it gets a little bit sticky. But the, for me, to answer your question, the, the cessation of overt action puts you in this highly epistemic state because now the sort of self evidencing side of the equation, you're putting yourself in a situation where that is harder to access so that you, you no longer can fulfill your expectations through, you know, action acting in the world, right. So like, if you were to sit down and start getting agitated or bored, rather than the sort of fulfilling the expectation there that you shouldn't be bored by going and doing and doing something, you have to stay stay bored. And there's a certain discomfort that can arise from that and then the role of mindful sort of acceptance in that moment becomes central. But then what it allows then for is that that your model has to update right if the model can't be sort of can't find evidence forward, well, then you got to you got to update it. And there you get to this notion of vipassana, right, which is like inside clear, seeing inside, right, is that you're you're learning something through through that that wasn't available to you the moment, the moment before. And, like, apologies to everybody out there who's like maybe very into the meditation that like this is very sort of hand wavy way I'm talking about it, especially kind of like matching it over to the over to active inference of various sort of loose way but that was kind of the, the, the gut intuition anyway coming into it and remains remains the the intuition and I think there's this really for me anyway and where I'm kind of seeing my my career with this going just is the very beginning of an exploration of how to how to articulate meditations, different practices, what they how they might cash out in terms of precision dynamics different forms of learning different forms of optimization, and what that might mean, you know, for us so I think that's why I wanted to start with attention because it's pretty controversial, as it were, and just start building up the building blocks and be able to start making more and more interesting inferences about what it might mean to be doing other kinds of practice. Awesome answer. Thank you. So blue or just go for blue. Hi, so I don't know if you had a chance to watch the discussion or like the preliminary discussion we did yesterday but something that came up was in the idea so Daniel can you flip to the circle about like tension and tension. This one. Yeah, that one. So something that came up there. When we were discussing this figure was like the discrete stops between like distracted awareness of distraction redirection of attention and focused like, you know, there's kind of like an intuitive feeling that the distracted awareness of distraction and redirection of attention back to focus is like discrete but like there's this transition that's kind of muddy going from focus to distracted and something that I thought about was like you can become so intensely focused on something and like the example that I'll give you because you're late it's like going down a scientific rabbit hole, like we're chasing reference after reference after reference. And so you're still on task and you're still correctly paying attention to what it is you're paying attention to. But like, you're so focused that you're finally off task right because you just start getting interested in other things and and so it's kind of a gradual transition from focus like you're still focused but you're you're like focused on the wrong thing. And so I just wasn't really like where in the bottle do you think that kind of like focus until you're so focused you're not paying attention like where does that like come into this or how do you think tomorrow something like that. Yeah, thanks for the question so this model that we're seeing here is of course a bit of a caricature, right? It's an intuitive sort of description of what's happening because you're right there's degrees of distraction. And you know meditation we talk about you actually can classify these you know there's when you sit down to try and pay attention to the breath. There are times when you've totally forgotten that you were even meant to be meditating and you were like gone right you're thinking about something completely different and then the timer goes off and you become aware that you were completely gone but then there's other things that are much more subtle right this like maybe you notice some sound in the room or you know some sensations in the body but still somewhere in your awareness the breath it remains present but maybe less and more in a background foreground kind of situation. And so like are you distracted then or not and to what extent and all these things and kind of everything in between right. So in terms of how that would cash out here. The first thing to kind of point out to is that the the precision parameter that we're talking about is a continuous parameter. And so it doesn't have to be relating to discrete states and we could have been free to granular like make more granular the attention all states was not just distracted or focused and sort of have grades of focus would have been possible But then the other thing that isn't modeled here is what are you distracted by here distraction was just not paying attention to the thing. But in reality distraction is your attention being grabbed by some other stimulus and for some particular reason. And that is not shown in this where that's something that we're hoping to start to work on in the in the future. Part of the reason why we didn't go down that direction here in detail is because there's still a I think a really central part that is is difficult to tease out still at the moment which is a intentional resource. You know in some ways you could you could run this simulation and given the preferences on wanting to be paying attention. If you put that preference on a bunch of different observations. There's nothing stopping the agent just paying attention to all of it right maximally and having really high precision observations across the board. Now that doesn't seem to be what we're able to be there seems to be like a a finite something finite about it. And so when you're paying attention to one thing it's at the exclusion of something else and maybe part of practice and part of you know what you're getting in meditation is that that you know is softening a little bit your ability to sort of pay attention to one thing or have more allocation of resources there in some way that might be learned over time but that's not that's not shown shown here which is why you don't get that. What you're talking about that kind of intuitive organic feeling of what it means to be what it really needs to be distracted. Awesome. I have a question and then blue. So we've talked a lot about how spatial goals are set and even how some symbolic goals are set like sentence structure but how are cognitive goals set. So are they reducible to sensory preferences. If so just the preference for you know some salt and sugar that's not going to really reduce your options in the city. So how do we set those higher level in these increasingly like abstract and ephemeral cognitive goals. Yeah. No thank you for that Daniel. This is the question I'm working on right now actually on a follow up paper where I'm trying to take this model and specifically model focus attention meditation and in that and in practice in general the the question of motivation and goal is really central and where where in the model are you actually putting like how are you driving the attention behavior or meditation behavior. The way the way that I did it here was to say that in this kind of setup that the agent is given an instruction. Hey I want you to pay attention to this stimuli and that that then translates to them wanting to see yourself in an in the attentional focused state. And so the way that the dynamics here are driven is by having a preference on that attentional state and observations of that state so the kind of like metacognitive observations of seeing yourself paying attention here. Yeah. Thank you. So oh to there we put a preference on those ones exactly. And just as a side note there you might think like but to those observations exist really are they since your observations what kinds of observations are these. And I think that's a valid valid question is something we talked about for a while but my intuition here is that you know if I were to ask you the question. What are you paying attention to right now. Right. And you're going to be able to give me an answer from having observed something. And so there's some kind of an observation there that you're able to make that you can use to what attentional state that you're in. So anyway that's where the preferences is is there. You know I prefer to observe myself paying attention. Exactly. So then the question is what is it that drives you know like a practice and I've been inspired by some of the work by Merza and colleagues on sort of seen classification or seen as perceived perception where there there seems to be this attentional driving of attentional policies given another kind of preference which is to resolve a scene which is to clearly perceive a scene. And that's actually the way that I'm I'm starting to think about at least in a practice context in like a meditation context because in reality when you're meditating it's not so much that you're trying to pay attention to the breath although we talk about it like that. It's more like you're trying to be continuously aware of the breath. So you have some kind of a preference on perception or being able to be sort of being able to resolve a hidden state and that that is then what drives the kind of dynamics behind the scenes. That's pretty interesting. Thank you for the response. Lou. Sure that kind of slides in nicely to what I was going to ask. So as a follow up to my question earlier and also to Daniel's question Daniel can you flip to the mind wandering slide. Yes. Thank you. So I had a question here. And these are two different models. The one on the left specifically is from the Mind Life Institute by a PhD who works with you know meditators right and so and then the one on the right. I'm not sure if that's an explicit meditation task. But I just thought there was an interesting context here. So this paper on the right which is the neural model of mind wandering. So you have like this paradigm where you're on task which is like in the meditation context or I mean maybe it's not focused on the breath but I just I think about meditation as you know being present in the in the present right being in your body aware of your hands feet breath etc. Yeah so so there's this model where you're on task which is like focused on what you should be focused on in present and then mind wandering they talk about as the pursuit of internal goals which I thought was interesting but I think about that in terms of like thinking about your shopping list or all the things that you're not supposed to be thinking about while you're meditating just going off and pursuing your own internal goal oriented thoughts and then have a separate like category for being actually off task which is like where you're not thinking about anything you should be thinking about was just thinking about like random like whatever nonsense. And so I just wonder like do you or could you think about making those kinds of distinctions in this kind of like and it goes from from being focused on the right thing to focus on the wrong thing like where does this off task come into play I'm just curious if you have any thoughts on that. I'm not exactly sure how they talk about off task but what I could just talk about here is the distinction between to me the the distinction here comes down to the meta awareness piece right where on task is pretty understandable off task you can imagine just doing something else but aware of what you're thinking about. Of that fact you know like I've just decided to change and start focusing on something else but there's sort of a metacognitive awareness of the fact that you are doing something else. Whereas mind wondering is this slightly or at least the way that we're talking about it here different situation where you're distracted or you're paying attention by when I say distracted is being paying attention to something you hadn't intended to pay attention to but also not aware of that fact. Right you haven't updated the fact that you're that that that's happening and that's the crucial piece here that we're trying to regulate right with practice or whatever in an intentional in an intentional task. And for that distinction I think I think active inference kind of you know it almost gives you that for free in a sense to kind of show you where where that phase would be and that's what we were kind of showing in the results is that that period of. Updating of where your attentional state is when you sort of think it's one but it's actually another. So you think you're paying attention or you're somewhere you don't really know where you what you're doing but your pay your distracted is then how we would sort of express that or cash that out computationally. In terms of active inference. So is so I just wonder is like off task or my wondering like I had never thought to split those two apart until I saw this this paper and maybe check it out if you have time at some time but but really like I think about this mind wandering is this like epistemic exploration is that like. I mean is it's an exploitation when you're on task and you're focused on what you should be and then is it you know we think about active inference in terms of optimizing exploration exploitation. Is there room in a model for that epistemic exploration or is it always like a bad thing. Is it always a bad thing. You know I don't have a I don't have a good answer for you to be honest this is what we kind of want to go towards when I was saying you know what are you distracted by and what's the why behind. You know the default loan network basically like what is what is the value of that on the one hand. I could imagine that yeah there's a there's sort of like an epistemic wandering through state space and future possibilities and just sort of monitoring of potential negative outcomes in the in the future. Would for which there's obviously value. And I don't think it's necessarily a bad thing I think when when I you know in a practice context anyway it's like that. That action can also lead to discomfort now and anxiety now because you're sort of simulating future bad events and there's actually a great paper with Casper wrote a good paper on that. Deep. I read it. Yeah the simulated future events. Well and so often meditating like freeze up space in your mind. I mean I don't know if you've had the experience but like where you're not thinking about some other problem but like you'll be meditating and like whoa have this like brilliant flash of brilliance or like key moments of insight. Right. So there's just like insight but it's maybe not insight into like some internal like deep spiritual discovery but it might be like insight into some external problem that you're having. So there's this there's that and so I wonder like that might be the difference between off task and mind wandering. Like if you're off task it's like maybe a non beneficial or like there's no excuse for it's just like doing blah blah blah whatever and then mind wandering might be like exploring your neural space. And then like a recombination phase. I think about that. So anyway it's just something to think about going forward. Interesting. Interesting. I think the notion of insight and the kind of why practice leads to that is really exciting place to start thinking about like what is it about paying attention in a particular way that leads to these moments of. Aha. Right. Thanks. I'm going to read a question from Dave. Dave wrote a practice I've been following based in Naropa Institute Shimata practice but not much stress there is to attend only intermittently say permit mind wandering during the out breath but attend during the in breath. There are two effects. Obsesional thoughts are disattended without intending to dis attend to any content. The motion of this attention of the transition from this attentive to attentive occurs many times the usual return gently to resting the attention on the breath happened several times a minute rather than several times per hour. Does anyone else do this or discuss it maybe to warn against it close parentheses hyphen to your knowledge. I like that. I didn't know I haven't heard of a practice like that before but in the way that you're describing I could see why that would be really really powerful and really useful because one of the pitfalls. I can tell how this conversation is going to go more and more into meditation. I'm up for that game but one of the pitfalls of course in practice is all the subtle ways in which resistance can come up towards our own minds. And if we are paying attention and we start to see distractions as things that are stopping us from paying attention and that we don't want. Then it creates this tension and is actually counter to what we want to be going for and I could see how I deliberately letting mind wondering happening and coming back and sort of cuts that tendency to have resistance to thoughts which actually just ends up perpetuating them or giving them more potency in the mind. Really the aim is to pay attention to the breath and it's funny because when we say pay attention we think like it's this like focus you know thing. In reality 90% of it is relaxation and letting go and letting go of the resistance to distractions distractions come and then we come back back to the breath and that's what kind of does does the work so I would love to. I'm going to look into that a little bit more detail that style of style of practice would be interesting to see when we have a more fleshed out version of this like how different approaches to practice might lead to different kind of learning outcomes at different rates etc. That topic of distraction it's like if somebody were a captain and they cited something very subtle that would be a distraction that'd be on task. So then there's this aspect of the multiple levels like there's the eye circling and then there's higher higher and modeled in this paper greeting Scott at slower and slower kind of clicks of the model. And so it's just interesting that yeah what the distraction implicitly says that it's not on task because if it was even something that was unexpected that drew the regime of attention that way if it were relevant for the task. Then it would still be considered not a distraction part of the relevant stimuli for the task but the exact same stimulus might be off task in a different context like five minutes later. Yeah yeah completely I mean if somebody is trying to talk to you and you know share something with you you only paying attention to the sensations in your hands is not very helpful. But I think where what we're kind of flirting with here as well is is this notion of task and goal and on task and off task. And I think in a I mean in a meditation context that's a really you know difficult difficult thing to define. There's a big difference between somebody who comes into practice because they want to you know sleep a little bit better or not be so stressed or have a little bit you know more ease in a relationship to somebody who comes in. And says you know I'm in it for really understanding myself and who I am and what I am and for you know questions deeper questions of like awakening and. You know expanding awareness and these kinds of things and those intentions and motivations are you know these task priors that you're setting over this whole thing and are driving the way that you practice and you know traditionally. What is sort of said again and again and again is that defining that is pretty important in terms of what you're going to get out of the practice because to pay attention to the breath. To just pay attention to the breath without really understanding why and where it's leading to leading is is one thing whereas having it within a context of you know a wider program and goal is is a completely other other situation. Now modeling those in active inference is interesting right I would say that the latter sorry about the background noise here but the latter example there of you know thinking a bit more broadly is having some kind of a preference on. Your own happiness right like your own model fitness ultimately you know if we kind of go on the affective inference kind of view you know Casper at all that like affect in some ways related to how well you're doing and how and how well your model is is predicting. And so to have that as the as the intention rather than just paying attention to the breath but seeing how paying attention to the breath can can lead to that ultimately is a different kind of goal goal statement than then they sort of narrow attentional task itself. Thank you Lars Dean and then Scott. Yeah, I think Lars I think you're I think. No I'm on okay I think you're talking I think your topic lens self to wanting to gallop ahead because it gets people so excited about this idea of moving from subliminal to liminal to kind of a limital like precision around edges right so that transition gets people excited because now they say oh I can think of a ways that I can use this in whatever it is and I'm trying to. Figure out so at at some point that transition pulling that transition apart but not breaking it is is going to have to we're going to have to sit down and really parse that with you because it feels to me like we're doing a. Are we 28 or 29 we're doing the point to right now because everybody wants to really get this thing going because it because it's. It's kind of exciting stuff right like you've given us this this backpack of stuff and we're like opening it up and it's all this candy and we're. But I'm still really kind of interested how that how that goes from subliminal which cannot be planned and cannot be thought out because we're talking about act sensory and activation to thinking about thinking because when I when I see the diagrams themselves. They're there but it doesn't really help me understand that any better because I still think that's a really really tough bridge to build. Yeah you know what I'm saying I know it's there but it's still hard for me to get my head around. I think I think I know what you mean you know what you're referring to which is like where does the where does the structure come from in the first place is everything. Yeah well we know where it comes from in the first place and we know where it ends up it's the transition in between it's those Casper goddess to the third layer. Now we've gone to a deep transitional we generated a context we're thinking about our thinking it's metacognition. It just appeared it just emerged so how does the how does the math. Work on that this I think I think what you're getting at there I mean there's kind of two ways that you could think about. A model like this is like either you say well the structure is in place and what's changing through learning is. The. The precision on observations essentially like you become aware of it it becomes something that can factor into your perceptual belief updating. That's one way around or you can say that there's something about human development and maybe the evolutionary process that sort of leads to this or structure growth and learning. And then the question is well how does that come about. The first question is a little bit easier to answer than the second question because obviously structured learning with an active inference is still an ongoing. Area of you know investigation research and how exactly that comes about questions of neuroplasticity and. Is this is quite it's quite difficult to grapple with and then how why do particular practices and you know different paths in life leads certain things and others but then. To take the other stance or the other way to talk about this which is I think what you were saying correctly if I'm wrong but like. It's assumed that the structure is here that we have some kind of a deep hierarchical you know parametric structure that is giving rise to to metacognition. How do we go from being not aware at all of the contents of our mind to being aware of the contents of our mind. Right. Yeah well it's kind of contradiction right because the way it's set up is it's a bridge to everywhere. Not a bridge to somewhere and then somehow in that bridge to everywhere metaphor we select out a particular and we give it our focus or our attention. And so I'm not saying that the that the math doesn't work I think the anchor pieces that at the basic level nobody's going to argue that we have a Markov like it. And we're going to we're going to statistically move closer to what we think is going to happen next we're going to update. It's the it's the last piece which is we're now thinking about our thinking that just seems to pop out and make itself available. You know because we've now built up a set of priors but we I don't I'd like to see how the math explains that. That's what I'm because it's still it's still a distributed set until it collapses to what we focus or attend on. And so it seems like a contradiction that and maybe that's what it is maybe that's what the math is trying to explain to us. I'm still not sure if I've completely grasped the the point of interest here. Well how that how the transition works. Again I'm sure you'll be able to explain it but what I'm seeing is a transition to first starting out being very open and then the mind is able to buckle down for lack of a better expression and get its attention on something starts out open and then it moves to something closer. What I'd like to see is how the math explains that. That makes sense. I think so I think so. Dean maybe one way to saying it would be just looking at figure five like the lower levels are certainly outside of our control and the higher levels are moving towards phenomenology. Whereas the lower level in traditional neuroscience would be seen more as like sensory processing and yeah it's indeed similar kinds of models. So how do we kind of bridge not just the science humanities gap but like the parameter experience gap. And then what is that sort of marsh between them and is it really the case that just by nesting this model that we know and love we ended up getting the or a structure of cognition is that the or a structure. Does that exist as part of a broader family of related models. Is it just about finding patterns of nodes that explain variance in multiple components of phenomenology and physiology. That's the question. Well I for me it all comes down to precision within the model. Then that's kind of how I been thinking about this is that you know at the sensory level that first level there are multiple places that you can start to. Identify patterns you know school patterns and then start to layer on top of that a confidence or precision over that which you know we have evidence for you know how that's mediated in the brain for various various places. But the precision then it sells itself becomes this thing that you can layer on top of your entire sensory experience and gives you this kind of internal flexibility as to how you're how you're interacting interacting with the world. And then the modeling of how those precisions are being driven you know without your awareness is then what what those states that that be the be matrix at the second level would be would be encoding as you're starting to learn. Okay so this you know my attention gets grabbed by this or that or like these are the kinds of and here we're talking explicitly about attention but there's other precisions around that you can start to understand how they are evolving. And then also learning where you're where you're reach of control is like how much of how much of this can I control and influence. And what does that mean and what are the impacts of that. It's kind of what I find exciting about exciting about this because you can kind of see how different forms of precision control and awareness would lead to different kinds of model optimization and you can calculate the free energy you know based on on how different precision profiles are defined and evolving and kind of. Over time that's something that you would hope to be able to learn and then the question is like how do you actually learn that well by paying attention. By paying attention to how it's changing and that's not always easy and sometimes things in life you know make us pay attention to it you know start to notice like I'm the kind of person who gets. Yeah thank you exactly this figure here that gets impacted in these ways but I think there's a really important thing to say around the role of a teacher in this as well somebody who can point these things out. And in doing so direct your attention in a way that you wouldn't have thought to previously into noticing like hey there's this pattern in my mind that I hadn't thought of as a pattern I just thought of this. Well I hadn't even thought of it is just happening and then as soon as I start to see it as a pattern is something I can start to model and then you've got this kind of emergence of a of a higher level. Structuring or understanding or grasping or modeling expectation over these patterns that were previously unconscious. Thanks I'm going to just read a comment with no response needed from the live chat and then go to Scott. So Joshua Benjamin wrote all psychic ability seems to be subconsciously driven oneness of the active and subconscious mind would be magical to say the least. People that already have their active and subconscious mind working as one also seem to have identical palmistry between their left and right palms. Interesting observation Scott. Yeah thank you. I'm sorry I was delayed and coming in but I'm always thrilled as soon as I drop into the conversation it's like being with family so this is so wonderful what a great conversation. I wonder if you could comment a little bit on the attention versus intention. It feels like part of what we're talking about now let me let me comment on that a little bit and it's falling up on Dean's exploration there. I was thinking I always talk to people about the aha moment if you know the if you know the punchline to a joke and someone tells you a joke it's just not funny because you know the punchline. And so the aha moment is the same as that moment of Eureka when you get out of the bathtub and run around yelling Eureka but you do you understand something. It's a paradigm shift like Thomas Cune in the structure of scientific revolutions talks about that you shift from not knowing something to all of a sudden knowing. So those I'm an addict for those moments I love ignorance as a result so I wonder if you could talk a little bit about going from attention to intention. I am coming to the belief from working with all you folks and others in similar explorations that the mind actually doesn't exist at all in the brain the mind is always in society and in language always. The brain gets tuned into a local mind by exposure by perception. And then that become the attention turns into intention you adopt like a bower bird you adopt nascent models from the externality there's no on board feral consciousness in my assertion here. So you develop it and cultivate it based on your exposure to the environment as you build it up the attentions your teachers out there in the world whether they're formal teachers or informal. Can help you convert attention to intention because the models get developed based on their their explanatory power of your environment right if they get cast aside if they're not as helpful in closing free energy gaps for instance. So I just wanted to throw those things out there and following up with Dean to me the paradox of the frizzle between consciousness and unconsciousness is not something that we can resolve it's something that we can observe and maybe manage in some ways with these models. But the paradox is one thing I always told my kids is the only time you see realities when you're seeing paradox and anytime you're seeing the absence of paradox you are seeing a model. And when people say they want to resolve paradox it means they want their model to win. Here it feels like as Dan was alluding to that there are multiple models that are in place simultaneously and the multiple models maybe can give us a sense of the shift from attention to intention. Anyway your comments are welcome. Thanks Scott. Thank you for that lots of interesting points there. I'll just pick out one of the things you said there around insights and these aha moments and from and this isn't present here again and it's working kind of falling into a Dean's trap here a little bit of running forward with it. But if you imagine here that what what's being afforded by you know paying attention in new ways to dynamics and and precisions you know within our generative model and how and how the all these things are evolving. You know we're gaining in model kind of structure and complexity ultimately and one of the bits of work I find really interesting from Carl and other colleagues is around you know insights and they've been talking about it in the context of this. And so I think that's one of the things that I think is really interesting from the perspective of psychedelics specifically as these moments of model reduction and vision model reduction whereby you lose a bit of model structure and there's this liberating aspect of that and there's this sort of it's not that you've sort of learned something new you've kind of unlearned something and in that you you gain in model fitness as well you know there's this kind of. Improvement of the state of affairs by by by doing that and so you could imagine a scenario whereby. In playing this game of paying attention to how things go and how things are in the mind specifically your model structures kind of gaining complexity but you're also identifying places that can be reduced and you get this sort of. Cyclical growing of complexity and reduction which matches on really really nicely and I have to be really careful sometimes to not try and like recapitulate things that I like from sort of practice theory into science and really be you know grounded in in the in the act of inference but nevertheless there's you know my heart kind of wants to say that kind of nicely matches to the cycles of practice to the cycles of of insights and growth that does come in this sort of non non linear and cyclical way. Thanks Lars. So this doesn't really like a question but it's more of a comment so Scott was kind of talking about inverting attention or intention to attention but I think that that's also something that's cyclical so you can convert attention to intention and intention to attention you intense pay attention attention gives you an intention. Sorry that's like way tongue twister but where does imagination come into play there. There's attention intention but what where's the exploration the creativity and does active inference allow for maybe this component on attention intention and imagination. Cool question. I think of imagination a lot in terms of generative practices so like state generation practices like loving kindness gratitude forgiveness these kinds of things and for me the way that I think about that is you are using your imagination to generate the goal observation the goal state right which then allows you then to set a prior over that to aim for so if you evoke in your mind the a you know may you be happy so I'm imagining a situation where you are happy and now that is a possible future state that I can put a prior over and once I've done that then the machinery of active inference starts you know coming into play to achieve that kind of goal and that's a very very hand wavy and very loose and again like I can I can sense that people out there might be cringing to this kind of way of talking about about these models but you know an answer to the question is kind of how I how I think about at least that form form of imagination. That's some point to stuff thinking about how our cultural environment scaffolds our preferences and our tilios our end directness, which is something that came up in navigation now it's coming up in cognitive navigation and setting of these higher order parameters. Scott. I just wonder why is it cringy from the math perspective the because I'm a humanities guy and I'm the math always makes me out so I want to understand though what is it about the your math person steep in the math. What is it about what you just said that makes you uncomfortable with the math and how might the math be why is it stretching the math or how might the math be extended to to accommodate that cringiness problem. Yeah, no thanks Scott. Good question. The reason I say that is because these models can pretty much model anything right you can you can adapt them to make almost any point that you want. And so, you know, if I start saying, oh, this is how it is and this is what imagination is and does well then I can sort of ad hoc play with the model until it until it until it makes that point. And what I've really tried to do here in this paper is be very careful with that and to allow the dynamics to emerge intrinsically from moves in the mathematics that are like allowed and and and don't make many assumptions. Because really all we're doing here is is having a higher level that is parameterizing precision is on the lower level, and then we're just running it and seeing what kind of dynamics emerge. And interestingly, the dynamics that emerge are aligned with with expectations, but if I go too far the other way, which is really easy to do with this right because you have, especially in a practice meditation context there's really detailed phenomenological models of all kinds of mental dynamics, and it be very easy to say, you know, just gerrymander active inference to describe all of that. And so that's why I'm trying to just be a little bit careful on how I talk about it. Want just to follow up the one of the things. So this may be rationalization, but so I was I'm a lawyer so I was a rhetoric tition. And so when we have incommensurables in the law we use rhetoric to glue them together. And it's not crazy glue crazy glue doesn't fill gaps. It's rhetoric is epoxy. Right. It's glue glue stuff together and fills the gaps right crazy glue just stick stuff together. And so the reason I'm talking about that is, is it again I asserted before that the mind actually exists in language that's very weird and loose language I'm using there. The mind exists in language that exists in material culture. We're born feral. And we get a brain gets tuned to a local mind. Let's go with that for a second as a posit. So you're you have nothing on board when you're born with is it I guess another way asking the question, can the math construct consciousness from a feral mind with no social exposure. That's my question and I'm beginning to believe that nothing. Yes there's imagination of course is a lot of lateral thinking, but all the things are borrowings from the externality is my assertion now that's not the imagination you can mix and match them different ways. But if you're isolated and have no question is I guess will the math generate a consciousness without any external input. And I think the answer is no. And so therefore it feels to me like the brain is a mind instant instance. And the mind actually doesn't it's a paradigmatically weird and different. I don't maybe there are people who say this, I don't know I just don't know enough about the mind studies. But it feels like we can allow ourselves to not feel cringy about the extension of the math now again I'm not the math guy I'm the cringy cringy side of this, but by if we recognize that the source of the mind stuff is external. Maybe that's okay. And it's processed in the brain. Anyway, I don't know if that's something he needs responding to but it's to me this is very interesting because this is that metaphysics jump. And is it metaphysical or not. I don't know. I mean, I don't I don't have much resistance to what what you're saying there I don't think these things happen in a vacuum and I think, you know, just intuitively, you know, that kind of a cognitive structure would not emerge without some kind of, you know, societal or back and forward multiple nodes in the network, you know, working working together on this and, you know, anecdotally a line that always sticks with me. And from the Buddha of all people is, you know, there's two things that are required for awakening there's wise attention and the voice of another. And it kind of speaks to that need for interaction. Now, the reason the reason I the reason I'm what I really meant with I'm saying cringy is that I think there's actually a huge advantage in just going slowly and method like and systematically through this because what it allows us to do is be very, very precise in what we mean with our words, because I can talk about attention. And that might mean several things that are from people but it almost doesn't matter what the definition is of it I can just point here in and say well whatever you this is what I mean by this word. And I can keep doing that, building up the complexity of it and I can start to have very detailed descriptions and articulations of mental dynamics that don't rely on interpretation of the words, because they are referring to something quite specific. Thanks. And it is really important to have that, perhaps, Carl first and would say a deflationary view. So the deflationary view of this figure 12 the generalized deep generative model of mental action it's like, I mean, big if true, the deflationary view would be that just like a linear model like a linear aggression, Bayesian modeling can be used on any kind of data. So just because there's a word that you may or may not have familiarity in front of it like frequency statistics Bayesian statistics, it doesn't get around the fundamental problems of data modeling. And it actually has a lot of similarities with hierarchical Gaussian models of control, which are common, at least related to this idea of nesting slower levels of policy inference with some, again tricks that are really interesting, like free energy calculation all those other things that we talk about other times. But that kind of model never took on a philosophical or metaphysical baggage, even when it did appear to correlate, or even have predictive power, for example predicting what somebody would buy. And it just it's a choice that we make. What is our cultural priors? What are our personal preferences over connecting that metaphysical jump that Scott mentioned maybe some people, they think it's a six inch jump, and they can jump 20 feet and other people they have a belief that it's a bigger jump. So that's why it's such an interesting area to talk about because there is some hyperinflation going on. And then there's some extreme deflation and there's everything in between as well. Yeah. And if I could just. Yeah, yeah. The I thank you. Yeah, for articulating that so well, because I know I'm also guilty of this in my own mind when I'm thinking about this privately of just sort of running running this forward and losing track of, of what these are ontological commitments in in this whole discussion and I think Maxwell, one of the coauthors he sometimes refers to, you know, hard for Stonians and soft for Stonians and I kind of speaks as deflationary and inflationary and inflationary point. But for me in this in this work, what I'm most excited about isn't so much a an ontological proof or conclusion of hey this is what it is and this is what the mind is doing this or or anything that strong. What I'm more excited about what I think is more relevant to our times is that even if this is a facsimile or a very kind of loose description or a, you know, what am I trying to say, approximated framework of something is going on. If it allows us to articulate intelligently like how these mental dynamics are going on. It really helps, you know, I teach meditation as well. And one of the big obstacles that you come up against is like yeah but why should I do this like how does it work and what what what's the point. And if you have something like this that allows you to articulate why and how and the dynamics and what's going on, then that breaks down that barrier and allows for in a conversation that's at least grounded in something closer to, you know, biologically inspired computational neuroscience rather than a phenomenological report, not the space for that but for a lot of people this is what would get them through the door. Thanks, Ars. Scott, and then anyone else with a raised hand. So I just wanted to go back. That's this is fascinating and I wanted to go back to that rule. The idea of rules of general application like is this reducible to is our elements of this that are interesting reducible to formulas is what I'm where I'm going with this query now. It made me think about you have the society have different societies with different standards to different music scales different recipes of food and things that people are familiar with right they're raised in that context so Chinese food is and music or familiar to a Chinese person. If you're born in a different culture you don't those are unfamiliar. So I wanted to talk a little bit about rules of general application because in law so I come from law and law, you know, we got technical specs now and here we got frist in analysis and trying to find out general rules of general application. That's fine that's good stuff and they and the one of the things that's interesting but rules general application is how generally are. So in in law you have this thing called equity which corrects for rules of general application I just want to read you this Aristotle quote that guides me a lot here. Aristotle said the source of the difficulty is that equity though just is not legal justice but a rectification of legal justice. The reason for this is that law is always a general statement, yet there are cases which it is not possible to cover in a general statement. Hence, while the equitable is just and is superior to one sort of justice, it is not superior to absolute justice but only to the error due to its absolute statement. This is the essential nature of the equitable. It is a rectification of law where law is defective because of its generality. The reason I'm raising that here is, you know, we can find general rules. They don't have to work in every edge case. Right. And if we find edge cases that undermine the general rule, it doesn't mean the general rule is no good. Right. It just means we need some tweaking at the edge cases or something right and one of the things that's funny in equity. The idea is you have to bring in someone, a judge or someone who can come in and say, what did they mean when they were coming up with the original law? Not what does the original words say, but what do they mean to do the substance of it, right? And similarly here, what I guess I'm arguing for is the social aspects of the mind probably don't need to be divorced from the model entirely. The hand waving aspects of the mind, the metaphysical aspects of the mind because they are, they I think are always the foundation of the possibility of a mind. Right. We come into social, you've come into life feral and then you are, and then you learn that there's a society and other people and you have self and other, you learn that as a child. Now maybe you would learn it out in the wild, maybe not. I guess what I'm saying though is the model it feels like doesn't ultimately have to be mathematically 100% rigorous because the borrowings in the model that are made from society intrinsically have caused leakage into that model being affected by the externality. So I guess both the perception and the other model contents feel like they're unavoidable. Anyway, just a few ruminations there and do with what you should with the CUD that I just served up. I mean I think this is what you're speaking to here is a pretty big topic in the active inference community in general and this idea that you know the math is not the territory. And depending on who's speaking we kind of get lost in that a lot of the times like are we agents with generative models or are agents generative models and to what extent is this a description and to what extent is it. You know that's I'd rather not wait in there to be honest. No I get it and I like your statement of math is not the territory it's nice because really that's nice way to think of the math as a tool to understand the journey or whatever but that's a nice I like that statement you just made about the math is not the territory because it often feels like and that may be something that is helps with the invitation you know the space can be threatening to non math people because it does need a rigor I get it. But there are things that are being said by the math that need to be conveyed to non math people. You know they're important things when I first dropped into the space I said oh my God all the contracts I ever made are synthetic Markov blankets. They are there I'm creating a there there that wasn't there it's not an observation of a biological system that's natural it's an artificial of creation of a thing a place where these rules apply for internal and external operations. Right it's an it's an artifice so we do that a lot in the world but so that was my first kind of impression but I don't I'm never going to understand the math enough to have the rigor to be able to. Delve into that and understand the implications of active inference math for the law it just won't be available to me. So if into the extent that without going to metaphysics were able to create narratives that are honest to the math. Realistic and also available to the non math people that's going to be something that will lead to a lot more both understanding and misunderstanding I get it but but a lot more awareness and adoption of the models I think which will lead to more. I think better systems better policy because there's a lot to be learned from this and it's the same thing as an engineer knowing how to make a nuclear power plant work and then other people down the line knowing how to sell power but they don't know how to make a new nuclear power plant work. Right you need all steps of the supply chain of this to be delivered as a social good I guess is what I'm saying so that's a big I'm putting it all on your shoulders you guys. But I mean that's the hope here I mean that's I think what's at stake a lot of the time is that you know I think about this stuff in terms of a practice practice content sharing and communicating on how how best to you know improve our mental well being given given what the insights of this might afford and given the structure in terms of communication that it might allow and it's already impacted the way that I teach is already impacted the way that I practice personally. But then there's a whole field of computational psychiatry as well which is which is very closely related to this I mean the kinds of well being I'm talking about are the kinds of dysfunctions that the computational psychiatry people are talking about so there's there's a lot of hope to that and it doesn't take somebody to understand. You know how to do a gradient descent on free energy to be able to benefit from that. Definitely typo but this is a slide we had from the dot zero just like there's so many directions that this area verges towards and just how we even make sense of this field where is active inference going to connect the pieces just I don't know any of these elements on the list that was interesting to you. Give a thought on or anyone else could just raise their hand. The high the high for scanning pieces really fascinating because there's a there's a obviously a deep tradition of you know these kinds of things being learned in community and in interaction with other people. And that is not obvious how that how that happens and I'm looking forward to you know that whole space developing a little bit and being able to. Model you know what it'd be like to learn. Know what's the difference in learning through an app and learning from a teacher. Very interesting so I see Scott's hand and then anyone else. So the other place is just in in the crass world of markets you know you have information differentials are differentials. And the markets are driven by information differentials I've asserted for a number of years that Cardo's equation and thermodynamics the hot and cold differentials necessary for a heat engine to function. The that that is equivalent to the differentials in information differentials in markets you don't have market action if everyone has the same information and that's that idea that markets generate information price information availability things like that. So the crass part of this and that tends to make things get swept away in the world of craziness when you get things into markets because anything goes and you know it's just a big mess. Having said that the impact of this kind of analysis will be very appealing in markets because it does have a scale independent aspect to it. And so it allows for insights at different levels of different markets and meta markets and related supply chains that if everyone is doing this analysis then it leads to a certain kind of interoperability because you have a risk and risk evaluation interoperability right. And so the value in markets if everyone started doing active inference style of risk analysis you'd be able to make observations about at different scales because you'd have data coming out of every scale that was generated by models that were similar. Right so that kind of aspiration it's what I'm looking at with the I'm working with banks now etc and I'm not talking about active inference or Carl Friston. But I'm starting to set that up as best I can in the early stages of it because it feels just like a natural thing we are biological systems are markets and our institutions are ultimately doing reproductive and nutritional opportunities at large scales. You know and so there's a lot of biology that's still bound up and what we do as organisms in our scaled structures and so it's exciting to watch because this thing is going to come out of left field. And it's like quantum computing everyone thought that was a thing in the lab for a lot of years and now people are like oh geez here we go that's going to be a big deal in other things like business and governments and dot dot dot. And I mean you just imagine legislative activity informed by a active inference style of model where you really do probe and we get feedback right what if we had feedback from legislation every year saying this is good or not good more this less of that right. I mean that's the kind of thing there's been aspirations that for years putting lobbying aside that would lead to a great efficiency in things like legislative activity and enforcement activity things like that. So I'm very excited that these things are going to get taken in the directions where they weren't intended and I think it behooves the people who really understand it early on to set up some nice chunky solid notions and paradigms. So when they're adopted they don't drift too far from where they should be right because the things get picked up and taken in all sorts of wacky directions out in the world. So it's kind of that's what I'm talking about simplifying in a way that still does justice to the what's going on there. And as a person who's teaching meditation that's exactly the same kind of mindset you need going into these crazy markets right is something of common intention because the markets have different agendas and intention. If we want the active inference intention to be carried forward we need to make sure that we have robust presentation of the model. So when it's picked up that model can withstand the abuses that happen out there in the rhetorical markets. Anyway again more just load on your shoulders. Any thoughts there blue or Dean or Dave. I didn't mean to bring things to a screaming halt sorry I can go in a different direction. No all good. What's a figure or a word that you might like to go to Lars or when you're communicating this to different audiences like more from the maybe non active inference side what is your entry point. Is it one of the keywords here or one of the figures here or how do you enter do you show the action perception loop. Do you just go straight to the you know Bayesian graphical model. Good question. I think this this this crowd is a particular one but I think when I talk about this paper I kind of do it in in two steps starting with figure. Well when we make when we look at precision so early on in in just talking about attention so yeah exactly here. And just starting here because it's quite straightforward as a model of perception and attention ultimately and you can kind of see here intuitively what this precision is about and we can. Yeah let's do this let's kind of build it up and we'll go straight to go up to figure 12 which is where I'm really excited about going in the future which is this kind of generalization of this this across and what it might mean. And the game that we have to play here a little bit is a little bit dicey in that in that we're trying to make intuitive phenomenological interpretations of maths right. And that's not not easy and requires a lot of discussion back and forth but that's why we started here with attention because it's the one place where it's less controversial and more widely sort of defined and so. We're starting with attention here is just the precision on you know the likelihood matrix and you can kind of see just intuitively I mean maybe this spending on who you are listening to this is like that's already kind of obvious yeah. You know higher precision on the likelihood matrix means you know is related to attention but just to kind of motivate that a little bit more subjectively. You can tell that at the center of your vision you know as opposed to the periphery you're a bit more confident about what you're seeing like if you see a glass on the table at the periphery of your vision. You might not be sure if it's a glass or a mug right but you can still see it and then when you look over to it. Then you can be more confident in the causes of your sensory observations right where is this cases. Just lost Lars for a second but just to sort of while he's reconnecting. That's one sort of experiential you're in a generative model type thought experiment or really commands for covert action for attention which is to realize that the color detection outside of the center of vision as well as the resolution is very low and there's a significant blind spot. So there's three features of your visual field for healthy vision that make it so that the visual field is not just seen homogenously like a sort of silicone light detector like detector device in a camera. So just a starting point for saying okay something else is happening such that attention isn't being paid it's being normalized in some way to not just increase the resolution outside of the high resolution. Detection zone and colorize it in many cases but also to normalize the absence and paste over the blind spot. So then that can be extended to other sensory observations. So what do you mean by normalize the absence and paid over the blind spot that part like it feels normal to what is being perceived under just again this is the whole interesting question about what is perceived as normal at what scale. Oh it's weird there's a door here. How weird what about the door. So I don't want to be I don't want to be normative with the way that I'm talking about the mental states but there's different ways that that anomaly can be detected. And there's certain sensory illusions that are really illusions because the alternate perception which turns out to be factually different than the actual sensory. Oh that is different but that difference structurally or inaccuracy or resolution is just seen as part of not salient but not salient part of the organisms experience phenomenologically not to say unimportant at a nested level of analysis. Gotcha. Hey Lars welcome back we're just talking about generative model vision so continue on to perfect thanks. Hopefully it doesn't drop out again. So that that's kind of motivating me the the phenomenological translation let's say and then and then the next step is to then say well OK we are somehow in control of that precision. And here I leaned on existing work by Jaco Blumnowski and Friston and others whereby that notion of control and mental action. Then you can you can talk about that as a deployment of precision it's some kind of a precision modulation is the word I'm looking for. And so then then the natural the kind of next step there is to you know make that possible within the active inference framework in the same way the other action is also possible. And that's where the second level comes from in figure four I think it is because then this is what then allows that to be the case. And really all that we're all that we've done here is to say OK attention is this kind of thing this is this kind of precision. And we can motivate that in all kinds of different ways both computationally and intuitively and we know we can control it so what would it mean to be able to control it. Well the same thing that it means to control anything within this framework and then here you go this is what kind of makes that makes that possible. And again luckily I'm really standing on the shoulders of giants here and in that you know a lot of this stuff was already worked out. You know Casper it's all an active effective inference already had this in place it just didn't have a policy policy set over this higher level yet. And so this is cool but like I said in the beginning the the real aim of the game or what really allows for attentional freedom in the kind of practice context is not so much the attention itself right. But it's your awareness of where the attention is so then then the question for me was well how do we pay attention to attention. But once we have this this this structure well then that becomes kind of straightforward and straightforward. I mean you have to make a little bit of a leap but then you say well we pay attention to it in the same way that we pay attention to the sensory states. And that's where you get the third level and where these states now or the degree to which you are aware of your attention. And you know if I if I just ask you you know how aware are you of where your attention is right now. When I asked you a question probably not so aware just afterwards probably a little bit more aware. And so there was a transition there that happened in that high level state which is then modulating the precision on the lower levels. Now this kind of this kind of a move computationally mathematically is internally coherent and allowed but it's also not unique. There's no reason why you can't do the same treatment to other precisions in the model. The other ones that we're most familiar with from the literature are the precision on G model precision and the precision on B which are usually which is usually related to sort of volatility beliefs sort of unexpected uncertainty. I think I'm saying that right. And just as a side note we made the decision to kind of change the notation here just because we're talking about so many different precisions in the literature until here precision on a was zeta. I think a precision on B was omega whereas here now we've just called all precisions gamma and with a subscript based on which which parameter they're referring to gamma previously was was really just referring to the precision on G. So in case that's causing any confusion. So the the idea then what I think is exciting is to then think about all right well if we do the same treatment right what does it mean to have this higher level state now as one as to a superscript to which parameterizes other precisions within the model because that now goes from an attentional state to something quite different. So we can talk about these individually and from a practice perspective this is what this is what I'm most enthusiastic about because there's a lot more to meditation practice for instance and also just sort of cognitive mental or catalog of mental action then paying attention. You know there's lots of other things that we're doing so for example the notion of acceptance or equanimity would be the technical term where does that live what was it what does it mean to be in a state of equanimity. Right and here we could define that as a as an S as a factor of S to state which is modulating the precision over our preferences. Gamma C right to kind of give an idea here of how we can start talking about this so a state of acceptance would be you know or state of sensitivity would be the opposite is to what degree are we allowing our preferences to impact our actions effectively. And so you know in the in the example that you said in the beginning there Dean of like sitting still you know how important is that I think yeah that we can you then a big part of that is acceptance right is being able to be OK with the discomfort that's arising from you know whatever whatever is going on. And that's also something that we're explicitly teaching that is a that is a instruct that is an instruction that is important and a big part of what makes it all work and when I say it all work it's leading to happiness right like it's leading to well being. And so you can start to you can start to do that game with these different different preferences you know if you are these different precisions and see what it would mean to affect a policy over them and then you have to do this like I said kind of delicate work of translating that into you know a phenomenological kind of interpretation that I don't think that's impossible and I think a lot of it lends itself quite well intuitively to be able to do that because of this sort of Bayesian space that we're in of beliefs and beliefs of beliefs and different kind of forms and the way that we talk about and the point there for me is that you can start to figure out what is causing or what a kind of instruction you would need to alleviate different kind of dysfunction or unpleasantness so for example. Ryan Smith put out and colleagues put out a great paper recently on gut inference and the point there is they're trying to develop a model for. You know the ways in which inferences from the gut could go wrong and try to see if that could be related to overeating and one of the things that has come out there is that the one of the ways in which people's. You know well being is effective comes from this tendency for priors to get frozen or like not be updated very well and in that context it means that you know people are. Effectively I mean the reason for that is that you're you're selecting attention away from uncomfortable sensations of hunger right. If you're continuously selecting away because you have a preference against hunger sensations then over time the priors around hunger and your behavior and hunger gets frozen because you're not updating updating those with any new observations. So what's the solution there right like how can we actually help somebody in that situation and well this this gives us a bit of a hint is that maybe what needs to happen there is a training in mindful acceptance of sensations of hunger because that's where in this sort of precision. Dynamic and cocktail of interactions things are going wrong and if we can kind of translate. Practice instructions or internal instructions into into this this framework then it gives it gives avenues for alleviating different forms of different ways in which things can go wrong here. Thanks just one comment. Thanks one comment and Scott and then can you mute Larson speakers at a different speaker set up for you. You said about instructions that might alleviate some non-prefable state and that really recalls our discussions on instructionism and interactionism and how active inference lets us think about one extreme with the instructions like the cues and then on the other extreme who knows how many ways it goes with what it looks like. To interact and co develop based upon a total model or even parts of a total model rather than to up high level X in blood that means you need to take the X blocker or if it's too low you take the supplement like that could be approached here. Oh parameters too high lower it. It's that sort of biomarker driven approach that maybe we can move beyond when we have a rich generative model and then we can underlie go into underlying factors informed by information rather than just trying to follow up on mathematical parameters and constraint people that way. So Scott if you want to say anything or Scott. Okay Scott. So that last part of the equanimity brought me back to equity the equity thing again I was reading from this book the Aristotle thing before and it's it's interesting because in a way it's a correction to too much. Sensitivity I guess and you know the correction of the general rules and Aristotle's words and it got me thinking you know and and this is something we may have talked about I'm just trying to remember the phrasing. It feels like what we're talking about there is that the models themselves are engaging in a peer to peer discussion with each other to refine each other right to the models. Yeah it's never one model it's right I always tell people if I'm trying to explain active inference I say if you're in the woods when you don't have a flashlight and it's dark you're not going to run full speed ahead you're going to stick your hand out that's the active part and feel if there's a tree. And if there's a tree you're not going to run forward straight into the tree you so you now you change your model there's a tree in front of me that's how I maybe that's a bad way to explain it but that's how I explain. But this the thing is is not just a tree is also a model of the rain and there's a wolf chasing you and there's and you're hungry is 17 things going on. So what it feels like is in intention we bring our models to bear in a peer to peer discussion internally and do we do we already use active inference to model the internal model discussion among models. Is that already been something we've talked about in earlier session I don't remember but we know we have a zillion different things going on that are agenda in our heads that motivate behavior right. And so when we're trying to correct behavior or address behavior or become aware of behavior. We bring the other models to bear to let us do that now again they're all borrowed ultimately from the externality. But once they're on board can active inference or does active instance already has it been brought to bear to that internal discussion among the models that then informs the decision making thanks. I believe the closest that we've seen one example is this nesting which we'll get back to when we return. But it was Mark Miller at all with their discussion of happiness with the drives for like low thirst you know low overwhelming this for one stimuli versus another there's like a sort of domain specific optimization. And then there's this cross domain optimization that has to do with this like euda monia well being component whereas the sort of hedonic drive is the domain specific. That's the Department of Thirst Department of Hunger and then those are being optimized in potentially recursed levels as well. So that's one way it's gone was that mapping onto classical philosophical concepts of well being success and then another direction has been also linking it up a little bit more graphically I think in this type of work. But nice insights Scott. Just one on that just the response that when the imp of the perverse which is that Edgar Allen Poe notion that people act against their own self interest to show they have power. It's called the imp of the perverse and the idea is that people don't take vaccines because they want to show that they have efficacy or whatever. So the it's interesting what you just said explains that nicely because the state that they're seeking is a state of perceived efficacy. And it may and it's overwhelming perhaps some other things like the health thing or whatever so that's kind of an interesting notion we just raised. All I say is just that we always got to keep that realism instrumentalism so you said there's 17 things going on well not if you didn't model them. That's a claim about how it quote is out there and that's a second order question. So if it's an if it's an illusion to someone causal process inside of some people's head. That's different than saying I fit a model that has this hidden variable and that is this intersection of mathematics and qualitative experience that is so tantalizing and so important to how we think about it and frame it. So it's nice insights though blue. So what Scott just said made me think of you know Scott Scott was talking about how our own models interact with one another right like the desire for thirst or hunger or you know to regulate temperature whatever. But but really like where can we start to model each other's models because I think that that in this like collective kind of dynamic and that was something that I brought up in the dot zero like this collective phenomenon like my model of Daniel. Updates every time I see Daniel he is as I expect him to be but then you know suddenly like if he were to be you know purple or something that I would have like the mismatch right. So our models of each other are constantly interacting and where where can we start to maybe look at this or think of this in what aspect. I love that point because what's what break what comes to mind for me is is in this context of as you are. I mean to respond to both of your points really like the what we're talking about here is a model of our model right as you move up in the hierarchy you are modeling your own model you're getting some beliefs over how the precision in the lower levels are are evolving. As you've got a model of your your model as you go up but what you're also learning there is the how your mind works and how. You know different states lead to different like affect for instance and as you continue to do that is you continue to learn how your own mind works and you get more and more awareness of how things go in your mind. Well then naturally that you you are also learning about how other people's minds work as well and there's always this kind of tension between. So compassion in practice and awareness in practice or tension there's this kind of duality between the two and they kind of seem to come come together out of a model like this whereby. You're training in awareness of your own mind and what you end up discovering is all how difficult it is actually sometimes for things to go well and. Fundamentally that the one thing is driving it all is your own desire to be happy. As you learn that as that really becomes part of your own model then that naturally projects on to everybody else and you kind of get this rebound of oh but then everybody else is also trying to be happy and it's hard for them to. And you get this kind of natural compassionate response then that comes it comes as a result. That I think also is related to this idea that Scott was talking about before which is that none of these things happen in the vacuum like my own well-being is for so so intertwined with your well-being. And so all these things kind of grow then together. Nice it's almost like in in Scott's forest one element is there's other people and maybe non people entities in the forest and then another aspect would be like you can modify your niche you can you can walk different ways you can use technology. And it's sort of that tension between how sort of slam dunk it is for a very constrained case. Wanting and sort of hinting that it could generalize that more because of the ease at which you can integrate not just within a domain like the mountain car or just thirst but across domains. Especially in a way where all the benefits of the first level relatively speaking like the computability and the interpretability of parameters if one believes that then it exists also at the second level and higher level so like figure five you have the three layers. Or six but then 12 the generalized model like you only had two would did it I guess implicit hint that you just do it again. Yeah. Yeah. It's just so pretty cool. Dean. I mean just to just to speak to that point I mean you can and I have and you can start to think about you know the the third level the sort of third level equivalent of attention was sort of awareness of mental states. Right. So now it's like you're not you're not just aware of where your attention is but you're aware of how much you know to what extent do I want here to what extent am I certain about my environment what extent am I happy to what extent am I doing things automatically to what extent am I being influenced by my memories and priors. Right. That to go through the whole gambit there. But then you could do the same thing with B. The third level gamma B. So what is what is that state telling you. So well to what extent is my mind volatile. So how confident am I about the transitions of my mental states. Right. And let's say you take a psychedelic or something and all of a sudden your mental states are all over the place. Well then maybe that precision at the third level would drop because now you don't have the ability to predict how your mental dynamics are moving or see you know to what extent am I am I wanting or not wanting particular mental states as opposed to sensory observations. You know like how much how much attachment do I have to joy and fear right now is what you'd be inferring at that level and you can kind of keep keep going keep going through that. Thanks Dean. Yeah so I know we're getting close to the end here and I just kind of wanted to bring this up so my history just a little bit of context was that I was a context generating programmer for high school kids and I wanted to take them from a place of subliminal to liminal to understanding what the edges were what the limits were. So my my basic job was unnesting learners having them transition but having them take a model outline along with them so they weren't kind of going naked into the forest. So what one of the things that we used to talk about in this transitioning across basically the deep generational transformation was using a street metaphor starting with perceiving. So which is which kind of is unplanned and moving to thinking about thinking which is plan and attend and attend and plan. And then this was the critical piece realizing the inside the lines versus the outside the lines and with the example that we use is that there's a kind of a crosswalk logic calculation problem. So you use logic to push the button that it gets the flash or going before you enter the crosswalk. You stay inside the lines you even have perfect gating and balance so that you can look at your phone while you're crossing the street through those lines of immortality. And now you're dead because there was a car driver driver in a car who was also looking at their phone and then an accident occurred. So the question you had then was did you follow the logic. Did you use the rules. And the answer was yes. And were you aware of the free energy gap. And the answer was well not anymore. Because logic tends to point at the J walker and ask what are they thinking. I mean they're taking the risk here. Those J walkers are probably thinking about the opportunity that there's no cars coming at you right now or on the street. They're not present. So they're thinking about the context and the directionality of the transition and they're also thinking about the rate of travel. How fast can I get across the road. And will I get a ticket. Will there be somebody who's going to catch me in and pull me over. All together who's making a decision based on more information from which to update. And I think this is the big question around precision. After all our parallel lines in the street and in the mathematical operations context supposed to represent logic and precision. I think they are. And in this case and based on this slide right here. This is why I like to talking about math is not the territory. Math does allow for the sketching out of territory. And sometimes we can get inside the lines and think that we are using the logic. When in fact we're actually going to be more precise and actually use what I would think to be logic based on more information. Once we get outside those lines. So when I think when we get into the point two of this. I think that's going to be a really interesting point. Because Stephen talked about it was Stephen you were wondering Lars who was the person who's talking about being really still. That was Stephen before he had to take off. I wanted I'm kind of wanting to think about this as being really active and being really contemplative and being really precise and then be able to sort of tie that in with the the the metacognition piece because I think it's the transition here that all of us are trying to parallel or trying to sort of piggyback on your paper. And I think it's I think it's I think that's the real value in this conversation. I also think that we can get outside the lines and learn more than we often do while we're inside the lines because you know I'm immortal now I've got these two white lines protecting me and that's not always necessarily the case. Yes. Yes. Exactly. I mean this what you're talking about there for me speaks to this power of suspended belief or uncertainty in our own maintaining a sense of uncertainty or in our own model so that no single belief can become so rigid and shrine that it's resistant to updating in the future. And if you spoke to Inesh and Mark Miller recently you know there is kind of down the same line that a lot of a lot of dysfunction a lot of low states of well being comes from that rigidness and that freezing of beliefs that then you know leads all kinds of afferent behavior afterwards because you're trying to fulfill the expectations of belief that isn't being updated. The the what's on offer here is a mechanism to describe how you might remedy that solution by deliberately controlling or inquiring about the the metacognitive belief. So gamma B in this case over my priors and if I can control that state transition well then it gives me the opportunity to relearn that B matrix at the bottom because I've injected a little bit of uncertainty. And now I'm more able to sample information that is counter to what I know and allows for you know just better optimization ultimately. Yeah, take the model along. Thanks Dean and awesome insight there so just in the last couple minutes to Dave any textual questions live chat any final questions and then Scott and then blue if you'd like to make a last comment. So go ahead Scott. So that Dean we were just saying there goes to that what I was trying to fuss with on that equity point before where it's the rules of general application. And what I realized while you were saying the reason the equity is so fascinating to me is what they did in law and equity is they said okay the rules of general application don't always lead to the right result. And what they've now done is collected all the remedies for the bad results and can characterize those into groups. So you may have a rule of general application and then there's something out there in equity called unclean hands which means if I did something wrong I if I'm a burglar I can't sue you if I trip over your coffee table. Right that's unclean hands right or the doctrine of unjust enrichment means if I find some money and or if I get some extra money you pay me too much give me an extra $20 bill. I say I keep it because you gave it to me but it wasn't meant to be it's unjust enrichment so equitably you should give it back or I find someone's wallet all this stuff right. So the reason I when you were just talking maybe you realize you know it may be that we apply the rules of general application and then we find some exceptions. And then when we start or some ways they don't apply the right way in active inference and maybe we start to cluster those bad results let's call them or unintended results or something to start to cluster them and characterize those like they did in equity. So that's what I was groping for before is the rules of general application can have great value and they need those corrections and those corrections themselves may have clusterings and groupings that themselves can be managed. Thanks great discussion by the way thank you so much Lars is fascinating stuff. Yep thanks Scott blue. Yeah I'm looking forward to more next week hopefully if you're available and maybe kind of probing the different types of meditation and I'm not sure how familiar you are with different kinds of meditation but I've done a bunch and highly variable and maybe whether or not they all fit this model or not is something I've been kind of thinking about and I'm going to think a little bit more about that over the next week. So thanks for the great discussion. Thank you. Thanks blue I think it will be good to see if anybody who's listening in the intervening week wants to join or if they just want to watch along next week so they can ask questions. And I think exploring that aspect of group active inference agents by design or out there in nature that we've been discussing a lot you know when is it is it can we model it that way. And then asking well can't we have nested metacognitive models of groups. What does that mean is it a different philosophical bridge to two exits on the same freeway. What is going to be the relationship there. So great times. Thanks Lars and if you have any final comments otherwise we'll end it. No thank you this has been really great and a really fun conversation thank you for your kind of openness and great questions and I just really enjoy the energy here it's not being that difficult at all if you see what I'm saying so thanks. Awesome. Peace out everybody see you next week.