 Hello everyone and welcome to the Active Inference Livestream. This is Active Inference Livestream 13.1 and it is January 5th, 2021. Happy New Year everyone. Welcome to the Active Inference Lab. The Active Inference Lab is an experiment in online team communication, learning and practice related to active inference feedback. You can find us at our website, Twitter, email, YouTube or keybase teams. 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 keep in mind video etiquette for livestreams. For 2021, we are going to be meeting every Tuesday regularly at 7 to 9 a.m. Pacific time and then we'll plan other special sessions as things come up and for different time zones. You can go to this link and see a spreadsheet and here we are in the first week with the first paper and next week we'll be discussing the same one. So hello everyone, thanks for coming on live and thanks everybody who's watching. Let's introduce ourselves. Please give a short introduction or check in, especially if it's your first time on the stream and then pass it to somebody who hasn't spoken. So I'm Daniel, I'm in Davis and I'll pass it to Colin. I'm Colin DeYoung and I'm in Minneapolis, Minnesota. I'm a professor in the psychology department here at the University of Minnesota and excited to be here. I'll pass it on to Stephen. Hello, I'm Stephen. I'm based in Toronto. I'm doing my practice-based PhD at Canterbury Christchurch University. I'm pleased to be here as well and I'll pass it to Shannon. Hey, I'm Shannon Crooks. I'm in South Dakota, so also in the snow but I'm based at the University of California in Merced doing my PhD also. I'll pass it to Marco. Hi, I'm Marco. I'm from Holland and used to work on active inference in London but now I'm just enjoying the progress from the sidelines. I'll pass it to Ethan. Thank you. My name is Ivan. I'm from Russia, from Moscow and I'm interested in pre-mg principle and as it related with systems engineering. And I'll pass it to, excuse me, what is your name, my friend? Dave? Let's build a Dave. Hi, Dave Douglas. I'm retired from information technology. I live in the Philippines and I did about equal amounts of graduate level work in computer science, linguistics and psychology, but have only a lowly BA. Cool. Let's go to Blue and then Adam. Hi, good morning everyone. My name is Blue Knight and I am an independent research consultant based out of New Mexico. And Adam. Welcome. Hi, I'm Adam. Sorry I'm late. I am based out of Bloomington, Indiana. I'm a postdoctoral researcher. Cool. Well, welcome everyone. Thanks so much for coming on, filling out the Hollywood Square that we have here. For these warm-up questions, it'll be great to hear from everyone. Just raise your hand if you have a thought. I'll just put up the first two. Totally just raise your hand as you want to add something. First would be what is something about human personality or emotions that you would want to see captured in a good model or understanding or education. And then also what is something that you liked or remembered about the paper. And I think there was a few memorable parts of the paper, just while people are raising their hand. I think just the idea of going from a statistical understanding of the Big Five as principal component dimensions to being the outputs of a cybernetic system is really an interesting topic. So maybe if anyone wants to raise their hand on this one or Adam or Colin, maybe how do you go about introducing the need for your kind of paper to an audience? Like where do you start out with the interesting part about personality that you want to capture in your model? Well, shall I jump in then? Yeah. And when you say raise hand, do you mean this little button here? Yes. Or do you mean like actually just wave at the screen? Probably with the button. All right. That works. So one of the things that I noticed in your summary of the paper was something that I often encounter from people working in other scientific fields, which is a certain degree of skepticism about the Big Five because of their origin in natural language. And because when you see a diagram that's like a hierarchy of personality traits, in a figure there, I was careful to note that that is not supposed to be a diagram of causal structure in the individual. There's this interesting problem where we use the same visual language for causal models of boxes and circles and arrows and for statistical models in which we're looking at the relationships among variables, but obviously those are not directly translatable. And so the way that I would pose the basic question that's of interest here is why do certain traits of people tend to cluster together? I think that that's really the core of what I'm interested in and what we're interested in if we're talking about the sources of personality. A lot of personality research in the last 40 years or so since the Big Five emerged as kind of an organizing consensus framework has really been focused on questions of prediction. Like if we have these assessments of people's personality tendencies, what kind of outcomes does it predict for them? Also potentially like what are trajectories of change and development of these things over time. But I'm interested in the question that is a lot less studied, which is where do these patterns come from in the first place? And so I think the way that to think about the kind of statistical element here is simply that there is an observation that if we have a large set of descriptions of people's personality and it doesn't have to be from the natural language. That was just one good source for a relatively unbiased set of descriptions of people was to go into the dictionary and just pull out the adjectives that could describe people. But we see very similar pattern of factors like the Big Five if we start with a lot of say existing personality questionnaires that weren't designed to measure the Big Five. So as long as we have a sufficiently large set of descriptions of personality and we have ratings on them, they tend to cluster together in reliable ways. And as long as they are valid, you know, as long as when we say that somebody's talkative that really does mean that they talk more than other people do. And there's a lot of evidence for the basic levels of validity of personality ratings. Then the question is why is it that people who are talkative also tend to be more sociable, also tend to experience more positive emotions, also tend to be more physically active, also tend to be, let's say, more interested in excitement and dangerous, you know, risky pursuits. So we've got a whole bunch of things that cluster together and the question is why. And so, you know, that was a list of low level traits that are grouped together within extra version, and that's just a statistical observation. Those people who tend to have more of one of those things also tend to have more of another. And so I see fundamentally the question that I'm interested in is what are the underlying mechanisms that vary from person to person that drive these patterns of co-occurrence of particular characteristics. Thank you for that, Shannon. Thanks for that really nice sort of overview introduction. I was sort of wondering going through this paper, you talk about what are the differences that can be ascribed to individuals and their individual personalities that lead to traits like extroversion or what have you. But what if there is nothing unique to any individual that gives rise to a personality trait? And I wonder where does culture and social interaction fit into this kind of model? And that's just a question that I'm interested in throughout the next couple of hours that we're here today. Nice, good question. Anyone, I put up also the third one, Lou? So something that as I was wondering about like as I was reading the paper was, you know, the big five personality and like it's, you know, grounded in genetic influences, right? But like attachment theory and like there's like a bunch of work that's been done in mice like showing like, you know, the nurturing of parents and, you know, whether you're attachment avoidant or attachment anxious or, you know, attachment secure. That's also something I think would be fun and interesting to model in the free energy kind of way like put into this framework. And I think it has like some cool, you know, impacts for like team building and and how, you know, like whether I don't know, I just think it would be it would be interesting and maybe impactful in this kind of framework. Cool. And blue to that point. I just saw a paper this morning from Anna Sonica et al. and Axel Constant. The first prior from co-embodiment to co-homostasis in early life about this transition from like a truly in utero embedded being to different types of physiological dependencies, social and cultural. So it is interesting to think about, all right, we have an individual orientation around the big five. It's about the individual in relationship to different types of settings. But then how do you capture that element of niche construction or of group dynamics or of these other aspects that are relational? I'm just curious, Colin or Adam, like where does the other, how does that enter into it? This idea of the cultural influences in on personality, because someone's personality might be manifested in a really specific setting because of their local context. I could just say something about that that is relevant. I think to both of the two questions that have just been asked, which is just to keep in mind that these personality traits are descriptions of, you know, persistent patterns of people's behavior. And we know that they are genetically influenced, but we also know that they're not entirely determined by genetics. They are also determined by environmental influences. And so the way that I think about this is that you have these, you know, these mechanisms that are allowing people to function. And some parameters of those mechanisms must be reasonably persistent to produce these persistent patterns of behavior. But the things that set those parameters and that tweak them if there's change in personality over time can be both genetically programmed and environmentally programmed. Right. So there are things, there are events that happen to people, for example, their relations to their primary caregiver and attachment that shape what these settings are going to be persistently in life. And so I think there's definitely room for thinking about environmental inputs to these systems and shaping of these parameters as well as genetic ones. Cool. And Colin, at this point, you raised earlier about graphical models and just drawing a connection among different variables. Is that a causal model? Is that a statistical model? Are those different? What if you're talking about causal entropy or you're doing a causal, you know, model, like a structural equation modeling, or even if it's a linear aggression, it's going to be interpreted as a causal influence. They'll put in the discussion, oh, it could be a different way, it could just be correlation, but it's really hard to disentangle that. Yeah, Marco. Hi, thanks. So first of all, I haven't fully read the paper, kind of recently got out of a pandemic slumber. But it sounds very interesting from what I've read and heard so far. And as for the first question about what I'd like to see, it might be a bit far future. But I think ideally, this will all connect sooner or later to the work on neuro phenomenology. The kind of ideal vision I have is, for example, now we've been talking about genetic priors and environmental inputs. But I think the beautiful part about cybernetics is the kind of reflexive action. And so consciousness or phenomenology is basically a mediator or moderator of the way that environmental inputs kind of influence our continued progression of personality or traits and characteristics. And so ideally, I would love to hear if it's possible to make a synthesis of, for example, the work from psychoanalysis, psychotherapy, everything that kind of explores and navigates these particular mechanisms from a first person kind of view, how to navigate the experience of being engaged in the world. And I'm just wondering about Adam's or Colin's someone else's thoughts on how to possibly connect that phenomenological description of mechanisms and processes with the more formal basic models that we're now exploring in this kind of work. If that makes sense. Yeah, very nice. Adam and then anyone else. Hello, can you hear me? Yep. Great. So I guess to try to address a couple of things. So yeah, Marco just thrust us into like the deepest possible territory like right away. And so I guess I'll briefly comment on that. Second, but the first thing in terms of social context and niche construction and the more intersubjective aspects and context sensitivity of personality. Active inferential modeling is really well equipped to address all of that. And this paper, it was more the focus was just kind of exploring what Colin and I both perceived to be a very rich set of correspondences between the free energy principle and active inference and his theory of cybernetic Big Five theory. But in modeling work, you can bring in all of those details and explore them in their particulars. So I'm currently collaborating on a project where we're trying to model things like extraversion within these social choice tasks. And so how you show up, it's going to be a function of, you know, both the task environment and you as this interaction that's going to evoke a given style pattern. But you'll have these parameters, these hyper priors are over the different aspects of the generative model, which through iterated policy selection and updating will basically concentrate their probability mass in different ways and result in different parameterizations and then different emergent behavior. And what you can see happen is you can get these basically attractors form as this process proceeds over time. These behavioral attractors where you the concentration of your prior for engaging in the world in a certain way causes you to enact a given pattern, which causes you to even more update in that direction. Like your your D matrix, like where am I in this environment? You know, what do you expect to be perceiving your a matrices all the different like components that you'd use to try to model the parts of an agent. These were all your priors of these will get updated will then cause pattern of an action, so on and so forth, and that over time this could cause you to be in this attractor where you have this enduring stability. How enduring, you know, would depend on a variety of things. And, but so I guess that's one I also so in this process, the social context who you're interacting with, that would make all the difference in terms of like, how the kind of ball rolls down this epigenetic landscape like how you end up like carving out a given pattern for this agent with respect to neuro phenomenology and just the actually getting the richness of experience. There's a sense in which all active inferential modeling is already in a way kind of personality modeling. It's like you're kind of getting into like these enduring the cybernet, whether thinking of them as these like cybernet parameters for this fundamental process, or just the your ability to describe the essence of the system. It's normal form description, this maximally encompassing description of the modes of where you expect to find the system as a kind of probabilistic density and some kind of multi dimensional phase space. If you have such a attracting network structure that you can describe in a way you could call that the personality of the system. But cybernet big five theory kind of goes further and it's grounding this in terms of, well, why should we expect certain modes over others? Why should we expect that relates it back to this fundamental cybernetic cycle of goal pursuit that any agent has to do. So with respect to neuro phenomenology and it seems like the attempts to get at this within active inference, things like sophisticated effective inference, the models of metacognition where they're beginning to model like kinds of mindful processing of reflecting on your confidence. And then this influencing your having an actual model of your epistemic status, you reflect, you're reflecting on this and this influencing how you sample influencing how confident you are in different ways, influencing what policies you end up deploying. And these things that to the extent that they allow for phenomenological richness, they could be brought into an agent model, and you're beginning to do a connection between neuro phenomenology, or a process theory of that would dock with phenomenology and personality modeling. So those are some thoughts. Thanks. And to add on that Adam, because I was really well said, it's almost like some of these past papers we've been discussing were the neutral infrastructure or the architecture or the tools to have agents that can have memory or can have affect or can receive valence or get worried. So we talked that in a very neutral framework like it's a parameter in the model. However, in the real world, as Colin was getting at, there's these extremely deep correlations with different outcomes of systems, people who like to text more also like to chat more or something. And that might not be surprising, but it these trends are very distinct. And so we are getting at this generative model idea personality as a generator of outcomes in different niches so inseparable from the niche. And it's just an interesting way to see how the cybernetic Big Five has come from a more psychological area. So it has that sort of toe of the neuro phenomenology and the real correlations in the world. And then there's the neutral framework that sort of maps on to that. We have Steven and then Colin. Yeah, I like what's been said about neuro phenomenology and I'd also say that one of the things that's good about this work with traits is it's starting to bring work that's coming from what you could say realist settings codifying it and then finding the patterns. And it's given us more material coming the other way. We've had some work around a lot of work with affordances and niches. And then obviously you've got the models which start to create dynamics and show and then you start to see where these two things meet. But I would also say micro I've been looking at a lot of micro phenomenology and the actual structuring of experience and attention. And I think that what can be interesting is these traits if these traits can also get brought down into micro phenomenology and ways of like how is the structure of the environment body happening and where are their times because what I think with a lot of the trait works is it does have a normative assumption about the type of life someone's in around which you then standardize and then get these patterns. You do find if you take people into some other types of space or other types of attention or state be a more grounded state maybe a performative state maybe a space which is less Western, so to speak, less gold orientated. That might shift. So this might be like this is where that is when people at this normative mediated sort of normative like calculated way of getting everyone at the goal directed norm for a Westerner and then can these different components to that be sort of treated and looked at how do they vary as we move like, not just as the fire but in, you know, as individual kind of tap it ways to tap into affordances. When do these change more or less when we bring people into another type of way of a regime of attention or another set of affordances which is outside the normal for our cultural goal driven way of being in the west. Okay, Colin and then Marco. I just had a couple of thoughts about phenomenology since that's been a theme recently. First of all, I think one interesting thing to think about in relation to trade approaches is that when we do assessments of personality using self report questionnaires, you know, which is not the only way that you can do it of course you can have other people, you know, rate the personality of somebody and we tend to see very similar results if we use other people's ratings versus self ratings. Although there's also evidence that there's some incremental validity there in other words, you know things about yourself that other people don't know so that there's some unique validity to your reports, but other people also know things about you that you don't know so there's also some unique validity to other people's reports on your personality. But as long as we're talking about the self reports, then that is at base a kind of, you know, crude phenomenological measure right because you're saying, what are you like what is your experience of yourself like, are you the kind of person who's anxious are you the kind of person who likes art, you know, and we get a long list of people's responses on these things and then we infer their general patterns of behavior from them. And so I do think, you know, one of the things that I've always been interested in is that personality is not just patterns of behavior from a third person perspective. It's also patterns of experience from a first person perspective. And so when we look at the contents of personality questionnaires, you know, they involve descriptions of emotion and types of cognition and patterns of motivation that can all be experienced, you know, phenomenologically in the first person as well. And so we're using this kind of first person representation and understanding of ourselves to then try to infer these general patterns of behavior that have validity from an objective third person perspective as well. And then just in terms of like drilling down to this kind of micro phenomenology idea. We do know things about particular patterns of processing that are associated with different personality traits. So for example, we just published a paper showing that extroversion is associated with the tendency to be influenced by reward in an implicit learning context. So you're doing a task where you're making this like discrimination between two very similar options. And you're rewarded more for getting correct responses to one option relative to the other. Well, what you see is that people's behavior becomes biased toward that option. But the degree it becomes biased is actually correlated with extroversion. And that's part of how we, you know, have support for the theory that fundamentally what causes all of these different extroverted behaviors to cohere is sensitivity to reward. And so you can think of that from the third person as like here's a process of, you know, reward mapping. But you can also think, well, what is it like for this individual? This is the kind of person who's going to be more perceptive of possibilities for rewarding interactions for incentives in their environment. So I think you can always sort of flip back and forth between taking the phenomenological and the more third person perspective on these things. Cool. Thanks, Marco. Thanks. Actually, kind of ties into that, because I was kind of wondering if you could share some thoughts on the relation between value systems such as pragmatic value or affect and epistemic salience and curiosity with this model of personality. For example, I did read that you said the fundamental components are the goal, the interpretation of the strategy, right? And I would say that that value, I mean, I would guess that value is more about the interpretation, right, so that you have some kind of skater or some other dimension by which you can compare a certain current real states and expected states. And tie into that, I guess it's a separate question. But I think I guess I'm going back to the personality and phenomenology because maybe you could share some thoughts on to what extent personality can be seen as a natural consequence of the cybernetic imperative where consciousness has basically instrumentalized to be the regulator of the entire mind body system as it were. Because I guess I have some image, a picture of regimes of reflexive attention, right? So there must be some cybernetic system or set of systems, ecosystem fuel that's regulating or controlling the entire mind body system or large portions of it. But that in itself then has to necessarily by pre-energy or active inference have to have some kind of imperative to decide policies to pick and how to engage with that bigger mind body ecosystem, if that makes sense. Beautiful. Thank you. Adam and then anyone else. Darn it, Marco. That was awesome. Yeah, it was epic. It was. Adam and then Sasha and then we'll leave the warm up. So I guess a few things coming to mind. So I really like this idea of thinking of your consciousness as a kind of modeling for the sake of cybernetic control and with hopefully some degree of not perfect but some degree of fidelity for modeling that which is modeled because good regulatory theorem to regulate a system. You have to have a decent enough model. And so there's a sense in which towards this end being able to have your modeling process reflect on itself and model itself. That could be a very useful thing if you get it right and a very bad thing if you get it wrong. And so it relates to things like, and it could even be that these parameters for these, they might strike it across like multiple levels of abstraction, something like curiosity. So you could have like a kind of curiosity that has a phenomenology in terms of like the way like a physicist would describe it as like a phenomenological description of behavior. That's just like showing the emergent behavior. And this looks like a more curious system in a way it's like sussing out state space. Then you might also get to like the curiosity in terms of the what it feels like to be curious and what those motivations, those free energy gradients, they're going to try to annihilate those different sources of desire, pull, what that's like experientially. And it seems like something like curiosity would be reflected at multiple levels of organization, maybe even some weird shifts in terms of like the way it shows up across levels of analysis. A few other ideas are coming to mind would be, there seems to be a sense in which, so we're talking about someone reporting on their experience. You can think of these reports both in terms of the perceptual inferences that are going into them. And then the reports is a kind of action is a kind of active inference. And each of these aspects, you know, for instance, like you could set up a generative model for each of these and you would have, for instance, a given kind of, you know, matrix to describe the perception like given a given set of observations, what's the state that you then infer. And so this could be like some aspects of like, so what is an agent, what is a person noticing in different circumstances. And so this is like, what is what is it in particular, that is entering into their experience, where does their attention go. You could also go have like their attention, not just in terms of a sort of like a salience, but a kind of like active curiosity, and a sampling from the environment as a kind of mental act. And so this then will influence like what's getting fed into the stream of experience and what it's like to be a particular kind of agent, kind of particular kind of person in a particular kind of situation. Those are some thoughts. We'll return to that idea about being in the niche for a particular kind of agent. We'll do Sasha and then Colin with the last word on the intro. Awesome. Yeah, thank you. Really great conversation. I've already reduced my uncertainty and a lot of the questions I came in with. Thank you Colin for explaining that behavior is not the same thing as personality, because that's something I was starting to wonder. So it's yet relevant that personality contains as part of it the self report of the individuals experience. And then I like what you said about your most recent work that sounds really interesting and kind of falls in line with active inference. As reducing uncertainty about our environment. And so it seems that extroverts and introverts have a different sensitivity to sensitivity to precision about their world where extroverts are more sensitive to report and updating their generative model from the external inputs like mainly I'm thinking social but of course it could be other things. So that's an interesting how it fits into this model of active inference as well. And I guess my last thought is, is there another level up here where we can talk about people who are also sensitive to precision about personality tests and others quantified self, whether there's a metric for that kind of like self reflection part of it. And if that, I think sometimes that can seem a bit like a self fulfilling prophecy or like I know I'm the kind of person who does this. So I expect this personality test to tell me this, where you're actually not sensitive to new information you just want kind of confirmation on what you already think. So, yeah. Okay, interesting stuff. Colin and then we're going to move to the next section. Okay, so quickly then just a couple of thoughts on the last couple of comments. So first of all, you raise something Sasha that is important, which is that personality tests are certainly fallible right. They're not a great form of measurement they're just the best form of measurement we have for assessing people's personalities. And the reason for that is that it's really hard to get at an assessment of people's persistent patterns in all of their, you know, different aspects of functioning with anything other than relying on people to tell you about it to try to come up with, you know, like some complex set of tasks and people coming to the lab and do all these things and you videotape them and somehow that's going to you're going to extract all the information about their personality from that by a carefully chosen set of things that you have them go through, you know, it would be a huge amount of work and you'd be learning less about what they're like in general than just asking them a bunch of questions. So, you know, this is always a perennial problem. It's like, well, there are these difficulties with self reports, people have biases, they, you know, they are motivated to think of themselves as being more desirable in various ways than they are from a, you know, social desirability perspective. And so what can you do about that. And yeah, there are certainly studies of, you know, the degree to which people are accurate in their personality assessment and like the degree to which if they assess their own personality, how much does that agree with other people's assessments of their personality and larger discrepancies can be predictive of things sometimes of psychological problems. So there's, you know, there's a whole area of studying that. But I think I like this idea of people who have curiosity about their own personalities. And that kind of leads into something I just wanted to say generally about personality, which is that definitely extroversion is related to this kind of curiosity about things that are potentially rewarding. But if we think about curiosity more generally as a sensitivity to the reward value of information per se. So just essentially the motivation to improve the quality of our interpretations of the world in terms of that language of interpretations and goals and strategies. Then we have a, there's a pretty clear sense that curiosity of that sort of general form is the core of one of the big five traits. So the openness to experience an intellect dimension of the big five is really focused on people's curiosity and hunger for information essentially However, that'd be more about aesthetic and perceptual information related to openness to experience, or whether it's related to abstract information related to intellectual issues. That seems to be one of the really core dimensions of people's personalities. So I do think there's a very clear way to connect that into this idea of, you know, foraging for for information and seeking out increased entropy for the sake of reducing entropy over time, you know, which is what learning is essentially. So there are good connections there definitely. Interesting. Adam, just one quick last note. So yeah, quickly. As before we temporarily close off the phenomenology thread. It seems like there's a really rich kind of bi-directional relationship between personality science and active inference, and that with personality science, you actually have this, like, they're all basically these, from the personality modelers I met, they're all these very like integrative generalists approaching this hairball of a problem from as many ways as they can, and basically bootstrapping themselves to validity and reliability. And so they've come up with a really rich literature of phenomenon to populate phenomenology, and active inference provides the sort of like places where you want to kind of dig in deeper on to like, well, what is the exact like causal mechanism? So like cybernetic big five theory was probably like best in personality science of getting a causal processes of like how you're, where are these traits coming from and what are they actually doing for the agent. So now you can bring an active inference and describe agents of cybernetic systems, run simulations, generate things that would correspond to different aspects of personality like for instance the dopamine system and curiosity. You now can modify your parameter over policy selection and see what then unfolds with respect to both overt acts or mental acts like attentional shifts. And so just one final thing for the next part is there's some in the paper, we start to talk about this idea of like a peace struct, or like this kind of objects that could be used and something like computational psychiatry for kind of pooling data over multiple, multiple people to like come up with basically a general personality model that you can deploy for different agents. We might want to go into the weeds on that later to extent we can but the thing I want to get at is that the one thing to model an active inference is we are all personality modelers. It's like we are all necessarily like super forecasters trying to predict the future. We are all also personality modelers trying to predict each other and ourselves and part of this would be towards this end. Well, what kind of agent do I think I am? What do I think my personality is like what do I think is going to be either more situation like what's going to be enduring? What's going to be you can have personality be enduring but variable. So what's going to be how variable are these aspects of me? How likely are they to endure? These are the questions you have to ask of yourself and other agents in order to govern the interactions in the best of your ability. Well said, Adam. It really just highlights how much people even though we say things like, well, we don't know even given someone's personality, we don't know what they're going to do. It's like, yeah, but you know what language they're going to speak. And so in the broad scheme of things, you've actually reduced your uncertainty quite a lot. And if you're not 99.9% aligned with that agent who's like you in the niche sharing an interface with you that conflict or adversarial relationship or just risk surface is going to spiral. And so it's all about this interface. Yeah, Marco, if you want to go ahead. Yeah, thanks. Sorry, I didn't get the far the paper, but that's really interesting. I just briefly skimmed the section on it. So it sounds like the P structure is basically allowing for a heuristic contextual dependency. So the generic generative models of your being and the P structure kind of links the current context to the generic kind of models. And that allows for the variation over different situations. I mean, I'm emphasizing the heuristic nature of it because personality basically can be seen as a very useful heuristic for how to act in the world. Does that make sense? Or should I just read more in the paper? No, I think that's really a great connection there and a setup. I think in figure two or when we see the graphical model, maybe let's think about where the structure would play in and then emphasize its role as a statistical construct that reduces our uncertainty about regularities in the world. Just like every other kind of statistical modeling framework does. And then I think we'll be able to come back to that P structure question. But what a fun warm up section. Good times. So today we did the introductions and warm up and then we'll have the sections of this paper, which maybe if you're listening, you've read or not. And that is the integrating cybernetic big five theory with free energy principle, a new strategy for modeling personalities as complex systems by saffron and young, both of whom are on this call. So in this set of slides, we do have a bunch of the abstract roadmaps mother stuff, but I think there's so many interesting threads we've raised. Let's just try to walk through all of them in no specific order, but really think about what topics we want to highlight for next week. When we talk about the same paper and how will we get deeper into some of these threads that we've raised today in the discussion. So here's the aims and claims slide and Adam or Colin. What was just your intention with the paper. And after doing it where do you see the next implications of the paper or the next research directions for what you kind of want to do in that direction. And if that verges into some of the figures just mention it and I'll go to that figure. Well, I think that our basic motivation for doing it was just that we had already been talking together about topics of mutual interest and the opportunity to write this chapter came up. And I thought, oh, this would be a really fun thing to a good excuse to do something with Adam and kind of get our thoughts together and actually written down and out there in the world. And so, you know, I think he and I probably have slightly different incentives for where this is going next. He's more interested in the process processes of modeling, whereas I'm more interested in, you know, generating knowledge about the specific mechanisms and, you know, brain processes that underlie different personality traits and risks for different types of psychophysiology that are tied to different personality traits. But yeah, I think, I mean, to me, this is just a really nice way to kind of connect my thinking about personality with this. What seems to me to be a really generative perspective with the whole FEP framework and what's happening with that in science in general. I'm sure Adam has some other thoughts about that too. Yeah, go for it, Adam. But that was very interesting. Colin, thanks. Oh, I mean, for me, that was primarily the thing like Colin and I were already going into the weeds together and having fun with it. And so it's then basically IARPA brought together this meeting of personality modelers to create this book, Modeling Persons and Situations. Basically seeing the extent to which personality science could do predictive modeling of agents. Basically, they want to recapitulate the plot of Westworld as far as I can tell. So I'm actually not joking, but except they are. But yeah, this is my primarily interest would be just what the heck are persons and what can they be and how can we understand them. And who am I and what do I want? Yeah, and also one thing we haven't raised is who or what is a person. There's personality as being the statistical regularities of agent in niche. So that's just like a person in an exoskeleton. You could have some sort of model for that or a person with other types of constraints on or something like that. But then there's the idea of like, what is a person legally? What is a person culturally? And all these ideas are coming into play with modeling legal persons and physical persons and regular statistical regularities that include our digital affordances. All of that is becoming integrated. Our digital persona is that included in these emotive models. There's just a lot of really interesting stuff to think about there. So then just copied on the screen for people who are looking is the abstract and pause the video there. And here is the roadmap, then we'll have the figures and then there's just one kind of interesting general question from an audience. But in the roadmap, just if you have any thoughts on the layout or the authors could give a point if they wanted to about why they arranged it the way they arranged it or which parts they thought were important. If somebody was just going to read one section or any thoughts on that. Otherwise we can just go to the figures. But okay, let me know if you want to return to this. Otherwise, it's just the headers of the paper. All right. So here's figure one and calling this really returns to your point from earlier about the graphical models. What do connections in graphical models represents. And that's very related to a general question of like, what do variables represents principal component one? What does that represent? What English word are we going to put on that? You know, MX plus B linear model. What are we going to call X? What is the slope? What is that going to be called? And then how does that play out graphically, whether in an equation or whether in a drawing like this? So I would love to hear your thoughts and then anyone else can raise their hand on this one. Okay, so I made the basic distinction earlier between a statistical model of covariance like a structural equation model. I think it's really important not to call structural equation models causal models because they're not drives me crazy when people refer to causal modeling and all they're doing is a set of regression paths. What that is fundamentally is trying to recreate patterns of covariance in data. And if you have special kinds of data, like with temporal priority or where some variable has actually been manipulated, then you actually interpret causal effects from covariance structure. But just in and of itself, there's nothing inherently causal about structural equation modeling or regression. This model here, however, is different. This is actually intended as a conceptual causal model. So these are not just looking at correlations between different variables in a population. This is thinking about the actual causal processes that produce the functions of an individual agent, specifically a human being. And so this is just sort of like a way to encapsulate an overview of the way that my theory defines personality traits and characteristic adaptations, which are the two exhaustive elements of personality in the theory. And so the idea is that we know from decades of research that both genetics and the environment contribute to people's personality traits. And the idea is that personality traits are patterns of behavior. That's why they have their own separate box here. There's sometimes a tendency in personality psychology to say, well, what the trait really is is some underlying thing that causes people's behavior. But I think that that leads to confusion because what we actually measure when we measure personality traits is a description of regularities in the person's behavior and experience. So I think it's important to keep that clear. But then the middle box there, the relatively stable parameters of cybernetic mechanisms, is trying to say something about where those patterns of regularity come from. And so the personality traits are not some underlying causal process that generates personality. They are just the patterns. Now they themselves do have downstream causal effects, however, because let's say I'm the kind of person who talks all the time. So that's going to have consequences on the world around me, right? And that will maybe lead me to develop different patterns of adaptation to my environment, to have different outcomes on the people around me. It's going to lead to different outcomes and different adaptations. So that's the final box there. And then there's identification of feedback loops here. The idea that you're generally speaking the flow is that your basic personality traits caused by these basic settings of parameters that have a strong genetic component to them influence how we adapt and how we, you know, what we pick up on and what habits we develop in any particular situations. Habits is a good synonym for characteristic adaptations. But sometimes there can actually be feedback. Let's say, for example, that I'm an introvert, but for whatever reason, I get into a job that has an important component is selling things, right? I get into sales somehow. So I'm going to have to develop, at least in that context, the ability to act like an extrovert if I'm going to be successful in my job. And maybe I get good at that, right? So I've got a characteristic adaptation to my particular niche, which is that at least in this context, I'm good at acting like an extrovert. But this acknowledges that there's a possibility for that to feed back and actually make me more sensitive to incentive rewards in general. And so I may actually find that I'm not just more extroverted in this particular job context. After five years of doing this, maybe I'm more extroverted in my life as a whole and all these other contexts. So that's one of those feedback loops. And then so you see that the environment can directly affect the kind of ways in which we adapt to our environment, obviously. But the environment, but genetics, any effect of genetics on our specific adaptations must feed through the effects of genetics on these general mechanisms, these parameters that underlie our functioning. And so, you know, if anybody has any other questions about that diagram, I'm happy to respond. That was a great explanation. And just to emphasize one point there, and then we'll go to Steven. You said that the characteristic adaptations in life outcomes are exhaustive. And so that really is like we can partition them into the adaptations are the things that the agent carries with them, the synapses that have changed or the scars they bear. But basically what is not there surrounding and the life outcome is the niche. So then there's this stigma G where the agent is doing these construction. So this is from a different non active inference non niche construction framework, but finding resonances and finding concordances here. Because that's kind of what we talk about is like the learning of agents and then the way that influences their niche. And then you had this idea of acting as if like the example with the person who's being forced by their niche to implement. I'm a kind of person who says hello in this way or things like that. And then that also relates to what Steven was saying earlier about drama and about performance. So maybe Steven, do you ever thought on that? Yeah, I mean, the, the, the thing I like about what was mentioned here is how personality traits can be adapted to context like some as a salesperson. That was a useful example. And then how that obviously starts to then change your whole broader way of seeing things. And one area I think might be interesting. I've been there's the work of Harry Hest who's revisiting this, the work of Roger Barker from the sixties. We did a lot of work on behavior settings. So he looked at a whole load of children across the US and he looked at how their behavior was when they're in a school or in the cafe or different places. He found the difference in behavior between settings for the same child was bigger than the difference in behavior between personalities of children. The same child was changing its behavior radically based on the settings and that that work kind of got lost. But Harry Hest has sort of been really pushing that again. I think that can tie into this is maybe there's personality traits. I suppose you know how much stability there is, but could behavior settings be kind of like that kind of more dynamic version of personality traits. And I think active inference makes that more possible to even have that conversation rather than the more sort of static psychometric data, which is kind of top down and kind of you just have to accept. It feels like this is you can get in there and play with it a bit more from different directions. Thanks for sharing that, Stephen. We'll go Colin and then anyone else who raises their hands. Yeah, so that's a really a really good point, Stephen, which is that there is there are differences that people have to specific situations that are distinct from their broad patterns of consistency and personality. And that's a long running debate in personality psychology going back to the to the late 60s. And the resolution to it basically is that yes, there are these relatively stable ways in which people respond to specific situations. And that's what I'm trying to capture with the idea of characteristic adaptations. But then if you go across a bunch of character, a bunch of situations, and you take the person's average, there are also persistent averages that differ from person to person. And that would correspond to the trait so you can have both of these things in the same set of data. And Will Fleson is a personality psychologist who has really been instrumental in collecting the data to show this very basic feature of what are we measuring when we measure a personality trait with something like a questionnaire, where we're just asking somebody for this sort of general impression of themselves. And so what he's done is used experience sampling methods where you're like you have somebody's smartphone, and it, you know, pings them like five times a day and asks them like how have you been acting for the last, you know, half an hour for the last hour. And then you get them to rate their behavior on a set of, you know, on a set of descriptions or categories. What you see is that, as you said, people have as much or even more variability within their own behavior as there is variability in terms of their average behavior from person to person. And yet nonetheless, those averages are quite stable. So if you do this for one week, you can get a whole, you'll get a, he calls them density distributions, right, you basically just get, you know, and they're usually fairly normally shaped. And you do that for one week, you get this whole pattern. So sometimes the person's acting very introverted. Sometimes they're acting very extroverted, you know, at ranges. But there's a peak. And that peak is very stable if you do it again in another week. So the person goes through another week of behavior. And again, they're varying very widely in their behavior. But if you look at their average level of that particular behavior, it's quite stable over time. And so, and then he's also shown that that correlates quite highly with the ratings that we get from these general personality questionnaires. So the questionnaires really are tapping into this sort of general average tendency of specific behavior, even though, of course, the behavior itself varies widely from situation to situation. And over time, both due to situational affordances and due to whatever is going on internally in terms of the goals that the person is pursuing at that moment. Awesome. Let's go, Marco, then Adam. Yeah, thanks. It was actually really interesting. I apologize. I'm going to add another interesting thing to connect to. So someone mentioned, as Steven mentioned, the kids in the classrooms in the context and I think development might be also very interesting here, because personality might be relatively stable across the lifetime. But I think hopefully we can agree that personality also enriches. And in that process of enrichments, and I would also say empowerment, there's, I think, you could say a shift between the balance of endogenous and exogenous influence on an agent's behavioral adaptation to its niches. And so there's a change in the scope in which the personality has a relatively large influence on the way they adapt. And I guess I'm just asking about thoughts and how you might picture that kind of process of how personality might actually, in the beginning, largely concern itself with how to engage with the world such that it might internalize that which it normally co-ops from the environment. So normally you offload a lot of stuff to behaviorally adapt, which might explain the variation due to environments. But then over time, as it does so, it internalizes certain heuristics, certain patterns, certain policies, and learn from environments to kind of enrich its personality and then extend the scope of influence from their personality. Yeah, I don't know if that makes sense. Yeah, cool. Thought, Adam, and then anyone else? A few thoughts from before and then some thoughts on Marco's thoughts. So the fleece and description of personality as a kind of density distribution, I think, and the issue of the stability of personality and the variance of it within individuals across context, which is part of the reason the book that was this chapter was part of was called Modeling Persons and Situations. But taking these ideas together, we're kind of in the same territory of these fundamental questions in the free energy principle of like, what is the ontological status of a non-equilibrium study state distribution bounded by a Markov blanket? What's the granularity over which we can find this thing that we're identifying? So it's your, so I'll be able to speak more to this, but this idea of like, there are ways of describing that your degree of variability across context is itself something that can be stable. And this could be itself part of your personality. To now kind of loop back to when you were just describing Marco, I really like that idea. So it's like you're starting out in a situation where through this interaction between you and this context that you're entering into in certain ways based on your own expectations and the expectations of the people in the environment and the structures that are there built up from this, there's a certain stability. And then you both, this stability ends up shaping you both just as a consequence of like, this is the thing for you to predict, but also in the act of inferential sense and as we're saying empowerment. This is what's going to help me get better purchase on reducing my uncertainty with respect to realizing the value of existing. And so this idea that your personality being kind of internalization of structure and maybe at a certain stage almost like a kind of a mortised inference in terms of, which I got from you. Yeah, I really like that idea a lot. Cool, Adam, Shannon, you're muted, Shannon. Thank you. Going on this developmental thread and the paper that you brought up this morning, Daniel, and Anna's new paper on the first priors when you're in Rediro as a baby. I haven't read it yet. But the idea that like, when you're in a neonate in a body or literally relying on someone else's interceptive states and biological states to regulate everything about your interceptive and biological states. And then you're born and you're still kind of relying on the caregivers, interceptive states or their behavior when they're holding you or coming to comfort you. This, I think, could relate to what Marco was saying about how your personality develops. So first, if you have if a personality is something like a collection of effective states and behaviors, and an infant is completely sort of reliant on the environment and the social environment to regulate its own personalities. And as you grow through toddlerhood, you slowly are able to control this yourself like you can maintain your own body budget a little bit more. You can learn how to how to get your own blanket if you're cold or like self sooth, and then you grow older and maybe you're taught how to make music and learning that this is a new way that you can explore your personality or create it. And I don't study development also sorry if this is wondering maybe Sasha can pick up on it, but I feel like this can kind of connect if instead of looking, or not instead of but in addition to looking at averages, but looking at like a time series of personality development over time. And like you could do at any point in time or at any range of time, some sort of like what would be that the activity for its model of your life in infancy and toddlerhood in, you know, rowdy teenage years. And as adult and you'll have, you'll find these like steady trajectories at the same time. Let me connect that to a few other points. So Adam raised this question of what is the level of the thing or how do we identify things from empirical data specifically so that we can use it in active inference type or free energy principle type models. Do we use renormalization group or principle component analysis or some type of adjacency matrix, some type of Markov blanket. What are those things that we're actually going to be modeling. And that reminds me a lot of a theory called a NT actor network theory by Bruno LaTor. And then also on the ants wavelength Colin was mentioning how there's so much variation within individuals but also stable differences between. And that reminded me a lot of my graduate school work with ant colonies, where day to day the variation across all the ant colonies was extreme. If it was raining, they would do one thing if it was very dry versus if it had rained the day before. Those were massive determiners of colony behavior. However, across days and even across years, there were stable markers of colony behavior. And there were characteristic ways in which they responded to the environment. So it is possible to have a type of measurement and understanding of a system, which my advisor established through long term observation, where you can actually model this development of time, as well as within a population, which doesn't erase individual differences, it actually is the framework by which individual differences arise. And then Shannon just so beautiful with music being introduced like a thread of personality. And then that can be used to explore these regularities. Adam and then Stephen. What was the paper that people are mentioning about like first priors? I'll post it in this jitsie and we'll put it in the YouTube chat but let's continue without worrying about this paper. Okay. Oh, sorry. Shall I go on or do you want to, I was just going to put in one thing, maybe I'll just, because I like what you mentioned here, relatively stable parameters of cybernetic mechanisms, the sort of those kind of dynamics. I think that links also to this idea of how that might change during our life and thinking about our lifetime. And one thing that I'm very interested in is when trauma comes into play, PTSD and other types of situations and, you know, it's, it can really shift the state of someone and polyvagal theory kind of talks to that where you've kind of got this sort of social co-regulation that we have as humans, but we also have our kind of primal kind of fight or flight or our gut sort of, sort of fear responses and how those things are in a way that another type of like patterned behavior which becomes incredibly stable, too stable, like you might get into rage, but then over the period of the day or life, the challenges, they make, it may be unregulated about when they appear. So someone might sort of be a certain way and then they flash into rage, you know, and then they come come back into, and their personality is radically different. And I think this is, so there might be interesting to bring in that as a kind of a what happens when that relatively stable state itself flips, you know, so that the, and it might say your mechanism might not be cognitive, but it might be in that kind of polyvagal system. Yes, the body influencing phase transitions in personality or behavior. Adam, and then Colin, and then Marco. Thanks. Yes, I think Colin should probably speak to what was just brought up in terms of like what actually happens with people. But in terms of like a high level of abstraction. So yeah, it seems like there's this sense in which you can think of personality as a kind of core screening, but it's an a kind of renormalization and in terms of it existing as a phenomenon and it's an epistemically it's a kind of core screening afforded by this renormalization, this kind of throwing away a variability as you form these higher level attracting states that you can ignore them within the scope of a certain type of modeling that you're engaging in with respect to having certain objectives of trying to predict certain things, whether you're a personality model or trying to predict someone else or trying to predict yourself or whatever. So within this looking for this like optimal level of epistemic grip in terms like the granularity of your description, this, you could describe these cybernetic control parameters as being part of the causal process of what's responsible. And then if you take the system and you perturb it, there's going to be a range where like the order parameter changes and you can get these nonlinear maybe phase shifts where the you're now in a different regime and it may look like a very different kind of person, because the personality structure as a stable attractor network can then become destabilized. And so that would be like a high level in college probably speak to like what actually happens with people. Yeah, call and go for it. So you get these phase shifts in behavior but they're often due to persistent sensitivities in the underlying mechanisms, right? So let's take also PTSD, okay. What you have there essentially is some kind of traumatic event is an intervention that resets the sensitivity of the threat systems. And so that corresponds quite well to the big five dimension that's labeled as neuroticism, which is an old fashioned word, but in this context basically just means sensitivity to all different kinds of negative emotion. And so obviously that's heightened in PTSD. And what that means fundamentally is that you've gone through some kind of traumatic event that led to your, it wasn't just learning about a specific event, right? What it did was that it reset the sensitivity of your threat system as a whole. So now you are perceiving threats and reacting with defensive negative emotion to much more minor situations than you would have previously. And of course, even in terms of risk for PTSD, we know that people who are hiring threat sensitivity to begin with are more likely to develop PTSD. So it's not like it comes out of nowhere typically. But nonetheless, it's obviously this change. And there was a recent really nice meta-analysis of personality change following therapy and different types of psychotherapy, whether it's pharmaceutical or talk therapy or cognitive behavioral therapy. All of these lead to changes in personality and the trait that changes the most as you might expect is reductions in neuroticism, right? So what we're talking about here is the ability to shift these underlying parameters that are not constantly manifest in the same way. So when we say that somebody is high in trait anxiety, for example, we don't mean that they're experiencing anxiety at every minute of every day at a high level, right? We just mean that they're more likely to be anxious more intensely in more situations than somebody who is low in trait anxiety. But nonetheless, then if we really want to understand the dynamics of the system, we need to understand those phase shifts from like feelings of safety to feelings of anxiety at any given moment. And that's what you're talking about. But in order to understand how that works for a given agent, we also need to understand these persistent patterns of the sensitivities that lead them to have these shifts in different situations at different degrees than other agents. Very cool. Thank you. Marco, then Blue. Thanks. Yeah, this is really interesting. So trying to tie back to the earlier talk about the role of affect to this aspect of trauma. So I think it's mentioned before by the two authors that we could badly see these complex systems as composed of different micro personalities distributed over scale and levels. And the similarly for affect in the sense written about by Kasper Haspen and others. So Adam Sarfron, Adam also mentioned renormalization, right? So I think that ties in also to a lot of different things like stigma G or equilibration. And I think perhaps it's I had a thought or I don't want to share is maybe maybe the clustering phenomenon of personalities might be a natural consequence of something like renormalization or equilibration. If there already is a certain heterogeneity, if there already is a certain kind of asymmetry in a certain factor, a certain basic mechanism of the sermonatic systems shooting throughout, then naturally should cluster because then they are more likely to cooperate. If it remains heterogeneous and disordered or not co-organized, then it's inefficient. But then you have this thing with trauma, which I think would disrupt that process, right? Because if you go back to the components you pointed out interpretation goals and strategy, then that's basically the key things about a trauma, right? You understand the situation that just demands extreme change in your interpretation of beliefs or models, your goals for a particular situation, and the extreme pressure to have the right strategy to get out of it. And I think maybe the side effect, the byproduct of that momentary extremist is that it basically expands its influence. And so it becomes hypersensitive by nature. And so the ongoing, the always ongoing process of equilibration of everything co-regulating becomes disrupted by that one participant, the traumatic complex, if that makes sense. So instead of a lot of different personalities kind of equilibrating with each other kind of polycentrically becomes a monocentric equilibration. If that makes sense, it's a bit rambly. Lots of very interesting points. Thank you. Blue and then Adam. So just to kind of take the trauma aspect maybe to the other side of the spectrum and just in thinking about, you know, like our own perception of ourselves, right? Like, so how we hold on to our own like personality versus is it congruent with what others think of us? So I've been thinking, you know, as how does this really relate to meditation, right? Like, so in meditation, there's like the idea of getting rid of the azimita, right? Like, so where like that's the thing that builds the ego or like the construct of eye. So as we like eliminate this like construct of eye and also like staying in the moment, right? Like we're not necessarily trying to build an accurate predictive model in a meditation state. So when we have this, like it's like this dissolution of like free energy or what is happening? Like in the meditative state that they can also bring about a drastic personality shift. I think as we get rid of the eye and learn to be non-reactive as opposed to like traumatic reaction that's done nonlinear acceleration. Awesome. Adam and then Marco then Colin and then anyone else who wants to raise their hand. Hi. So let's try to address a couple of the threads. Come to meditation at the end, I think. But it seems like a lot of the principles of personality in terms of their generality could be really interestingly modeled actually using things like and colonies like in Daniel's work. And you might even find interestingly like even similar mechanisms at play. So like in a recent paper starting to do active inferential modeling of and colony behavior and decisions and foraging. Daniel mentions like, you know, there's like the was it the foragers have more dopamine. So you could see things like the functional significance of dopamine actually as its way of parameterizing a cybernetic system as being potentially similar, although potentially switching across scales where in the individual ants and maybe even in the personality of the colony as a whole. And there's would be actually this interesting interplay between the personalities of individual ants and the personalities of colonies. And this interplay which might not be an exactly one-for-one might actually even be a kind of a complementarity where certain characteristics being present on this broader level, this more encompassing Markov blanket where the individuals are part of a super organism or just the organism from the perspective of natural selection. Because the colony is that which is selected as Daniel says. But so the personality of the colony, let's say, would then interface and interact with the personality of the individuals of which it's composed and potentially in an in training way, but potentially not to loop around. You might even be able to at the level of colony talk about things like the affective tendencies that it's, you know, it's what is it's, what is the affect the valence that the motions experienced by the colonies a whole only given instance. What is it's general mood and then what are its patterns of mood across different situations. It's and you might know we're kind of going into what we more call personality territory. And you could even think of things like to kind of bring in some of Marko's idea of the nature of trauma and this like dynamic system sense. Like you could traumatize the colony and does it then at some point like you would expect some sort of like elasticity or it's a term like history. I don't know about that, but it would be you perturb it. And then it would have some sort of adaptive potential to come back from the perturbation because that's what we're trying to do is keep existing in the way we are. And so you some adaptive potential, but you push it too far. And then you might not get back this adaptive attracting structure. You might get stuck in a different regime. And you can think of the same things. Similar things as this could be a way of describing things that happen within the within individuals. Like so there's you can think of that maybe in multiple levels of abstraction, but stop for now. Thanks. Yep. I just switched the slide to figure two. So Colin and then Marko go for it. And maybe we can introduce the P struct idea or just how does active inference specifically help us address some of these questions. But go ahead, Colin and then Marko. Okay, well, what I'm my thoughts are about a couple of the last comments. So they're sort of more related to the to the previous slide and then go ahead then it's fine. So in terms of both this discussion of what's going on in trauma and what's going on in meditation, I think one of the interesting things to think about again is the distinction between the personality traits and the characteristic adaptations. Because in in both, there's the capacity for both of them to be shifted. So Marko was talking about the way in which in a traumatic event, you are basically pre existing goals and interpretations and strategies are dramatically invalidated. And that, you know, then means that the system doesn't know how to pursue its goals. In fact, it may not even know what its goals should be. And obviously that's extremely disruptive and dysregulating. And so one of the challenges in the immediate, of course, is to figure out what you should be doing, what strategies you should be using. But then there's also this capacity for that traumatic event to shift the underlying sensitivities of the system to what is perceived as disruptive in future. To sensitize to the things that disrupt. And that is, you know, obviously an adapted response to say like if you're in a dangerous environment, you should become more sensitive to danger. And the traumatic event just basically taught you that the world was a much more dangerous place than you thought. That's part of what is tragic about it, right? Because you have this persistent sense that the world potentially is much more dangerous even than it is or that you should experience it as being. But then you have the same kinds of dynamics going on in meditation where trying to reduce your reactivity. That's like trying to shift some of these basic parameters of the system so that you, you know, react less easily or disrupted less easily. And then the kind of non-I idea, the recognition of the illusory nature of the eye. What that's about is basically not being overly attached to your model of yourself. Because the model is the model. It's not yourself. So if you are overly attached to your model, then you have your overly attached to your conscious understanding of what your goals and your strategies are and what you're like. And meditation is basically trying to get you to recognize those ideas that you have. A, they're just your model. They're not actually the way that you are. And B, they're contingent on your situation. And so you should be willing to let them go as your situation changes and as things evolve. And so it's essentially creating both a pattern of greater flexibility in the capacity to adapt, but then also the attempt to downregulate the negative reaction to disruption, right? Because if you have more confidence in your ability to adapt, then you should be less defensively reactive to disruptions because you can be confident that you'll simply adapt to whatever the new situation is. Thanks, Marco. Yeah, thanks. Just piggybacking on that because that was really nice. So my thoughts on meditation, very good and important question, I think. So going back to the image of having all these different distributed micro-personalities or sub-personalities, I think meditation can be seen as allowing them to manifest and arise as it often is called. And manifest the expressions which are always going on to also reach, as a word, the global workspace or consciousness or the conscious attention. And so what you're doing then effectively is broadening the collective of sub-agents or sub-personalities or sub-systems such that that equilibrative process improves because then you have a bigger scope, a bigger domain of possible inputs to an ongoing process of inference. And importantly, this is then, of course, without need for explicit modeling. You don't need to explicitly label a certain rashly or verbally-explicated belief, right? It's an exercise in the deepest sense of it. Importantly, it's effectively a reflexive niche construction because meditation is systematically to basically disattenuate external inputs. You sit in a very neutral position, you're in a calm environment, external input is minimized. And so all the inputs, well, it augments the inputs, that is, endogenous, coming from within, which, of course, is then going to translate or transfer to normal waking states because those are all the endogenous... It concerns all the possible endogenous influences on your ongoing cybernetics. And so it's a reflexive shift, right? And yeah, I don't know what I was going to say. Yeah, so it's basically just a cybernetic game. So from what I have seen, my limited experience with meditation and reading upon it, it's all systematic cybernetic games. It puts you in a situation that forces you to train the process that underlies cybernetics concerning the purely reflexive, which is, of course, beautiful because there is almost no opportunity for that. It's an extremely exclusive state of being that exclusively forces you to attend to these endogenous processes. It's kind of a cybernetics of the cybernetics, if that makes sense. Great points there. It's why so much meditation begins with thankfulness. And just to give actually one behavioral example, let's think about the example of a circus elephant where when it's young, it has a heavy chain and it can't break away. And then when it's older, a simple thin rope can keep it connected. So at a first pass, one might say, well, why doesn't it just break away? Why doesn't it engage in a policy of breaking away? And the habits, the E, the priors over the policy, the field of affordances may not even include that as an option. So if certain types of actions are downregulated, so to speak, by 100X before any evidence comes in, then it's going to be interpreted within a prior that downregulates or basically downweights the relevance or the likelihood of certain types of actions happening. But also there's this element of the preferences where it may, through whatever mechanism, it's not an exact metaphor here. The system kind of like a Stockholm syndrome situation, the system can actually internalize a preference for certain states that may or may not be consciously available during regular states of consciousness that lead into the implicit processing of everything within a framework that results in behavior that's not aligned with something else, which could be external niche pressure or peer pressure, or it could be disconcordant with one's own narrative for themself. And the meditation is interesting to think about as far as isolating these different types of parameters in the model and different kinds of meditation, lots of different ways to do it. Adam? Hi. Hard to follow up on all that. The way the threads came together for me was it seems like across varieties of meditation and why would you do such a thing? And ultimately to bring back Marco's point about empowerment, you want to be more capable in some way having a better grip on existence in a particular kind of way. You want to have more degrees of freedom with respect to the way you're engaging with the world. And there's a sense in which you can think of different types of meditation as either as allowing you to navigate through cybernetically free energy landscapes, which can and also adjust the nature of these landscapes such that they are navigable. So you can think of something like and to bring back in like something like trauma or something like the elephant getting stuck, you can think of this as like a getting stuck at a local optimum of a potentially jagged free energy landscape. And so if you can let's say reduce the some emotional regulation capacity, the ability to diffuse from experience and keep the gain on positive feedback loop cycles of thought and emotion from getting as high as they might, you might be able to make the free energy landscape less jagged. The differences from point to point, it's going to be you're not going to have these wild swings. And so this might give you greater latitude with respect to pursuing more complex policies. And actually, I hope at some point we start to discuss some of Collins work on well being because I think it docs. There's a lot to be said there. But one thing comes on one podcast commas mentioning like you can think of one of the benefits of mindfulness, which is like in pursuing any kind of complex goal. There's going to be a sort of trading off you're going to have to do against different things on on route there. And so if you can like temporarily let some of your character say adaptation slide, let some of your policies like not be as important if they're not being you think they ought to be realized, but they're not within some sort of like hierarchical hierarchical policy selection. Well, then you can pursue you could hit targets. You wouldn't otherwise you can meet me or a mountain car. You can get to the top of the mountain and the mountain car problem. So yeah, nice. Cool. Colin and then anyone else. I just wanted to set to point out that that's that's the essence of non attachment, right? You're not attached to the particular policies. And that means that you don't get hung up on a sub goal when you should be shifting policies, because what you are trying to pursue is some larger goal. They sort of, you know, there's a stereotype of benediction at least, you know, Zen meditation, for example, Buddhist meditation that you're eliminating your goals or you have no desires. But that's not that is a that's a caricature, right? Because it's not that you have no goals. It's just that you're not attached to particular low level goals. You might be, you know, you probably remain attached to if you're a good Buddhist, for example, to eliminating suffering for all beings. And so it's not like you're not goal directed. And, you know, you probably maintain, maintain life by eating and things like this. It's just that you're not attached to the particular policies that you might be using to achieve these higher level goals, because goals are just nested. I mean, there's a way in which there's a goal, but then the policy is also just a goal. It's just there are nested goals all the way down to the terminals, which are specific motor operations or specific cognitive operations. And so non attachment is basically becoming more flexible in which sub goals you are willing to consider and to use. Very nice blue. So I wonder if it's like, it's more that it's not attachment to the policy, but that actually is the policy and it's not attachment to the specific generative or predictive model. I wonder if that's like, because the policy has been the policy of non attachment to the model or ultimate model flexibility. Right. You could call it a meta policy. Right. And so, and like, secondly, I wonder about, you know, the idea of staying in the, in the present moment. Is this like some kind of free energy equilibrium, like in not necessarily minimization, but like, if we take the generative model and the predictive model, if we take like the value to zero, are we like, is meditation like staying in the present moment, just like this ultimate state of of equilibrium in free energy, like info dynamics. Well, interesting point about the hyperpryor, the Bayesian hyperpryor, and also that's why a lot of meditation processes involve rhythmic or the cessation of other inputs, because it allows the interception potentially to proceed in a different way, isolation tanks or other altered states of consciousness. So let's do keeping comments short because we learned a few minutes left, Adam and then Marco. In terms of, there's this one paper actually that just came out recently about a model of meditation with respect to the temporal depth and counterfactual richness of your generative modeling that just came out. I'll try to find that and send that. But I think that would speak to this idea of staying in the present moment. Also this idea of staying in the present moment also makes me think of this, Marco is like kind of intimating that you can think of meditation as a kind of like game task or almost like a kind of like, let me maybe play even, but it's this building capacity that is building capacity. Okay, we'll come back Marco. For a building of capacity through relaxing some of your priors and so letting you to be informed by policies with respect to mental acts that you might not normally deploy and you might not normally notice what is entailed when you deploy certain types of mental acts. So things might come to mind that wouldn't normally that would place you now into a different state of affective inference. And so I think if we wanted to do modeling of this of something like meditation, we would want to use this sort of imaginative policy selection described by something like sophisticated affective inference. And that would begin to get us there. The exact way, and it's one more thought before we move on is one more thing I was actually wondering with respect to meditation I've talked about the sum of Colin is above the traits and the personality hierarchy. You have these meta traits of stability and plasticity and stability is this shared variance over and Colin actually named these. It's the shared variance over neuroticism, inverse neuroticism agreeableness and conscientiousness and plasticity is the shared variance over extroversion and openness intellect. And interestingly, the stability seems to be basically the inverse of the general psychopathology factor P by large. But I think thinking of mindfulness and meditation with respect to building capacities for both stability and plasticity, which helping to basically place this. So you can think of stability as your ability to follow through on and protect your policies or characteristic adaptations. And then plasticity would be your tendency to form new characteristic adaptations. So like weather. So let's say you do encounter trauma plasticity might be what makes the difference. So stability would be how much it might make sense to the degree to which you're actually traumatized like you're very stable. Something might bounce right off you you're not traumatized. But let's say you don't have a lot of plasticity and that instability is compromised plasticity is the thing that would let you update and create new strategies for being in the world. And so they're kind of the state of dynamic tension and that they can like pull at each other. But they're both ultimately synergistic and they're like you're not going to be particularly plastic unless you have this foundation of stability and you're not going to be able to maintain your stability long unless you're met a stable. And so part of with meditation mindfulness I'm wondering if one way of thinking of it is it's part of what's actually really letting you kind of like ride this optimum of balance stability and plasticity. And that could be one way of thinking about part of what people get from it. Marko then call. Yeah, that's amazing. That's a perfect setup. Yeah. I also fully agree with this theme of complementarity. So stability and plasticity sounds really resonates with plasticity, openness and focus, which is a big complementary polarity in meditation practices. So if open meditation and just feel everything, let it all come in, etc. Focus meditation, focus on one point, one mantra. And that has a big connection to the more well known complementary polarity of exploration, exploitation. But it's not about the polarities per se. It's always about the balancing them off. And for example, Sussex had also a nice paper on that. It's not about perfect theoretical representations. You can have some errors. It's about how you balance everything off. And so again, we go back to the idea notion that the meditation can be games where you effectively practice balancing that off. Because that balance is not something you can explicitly verbalize. It's intrinsically preverbal because you can verbalize maybe the particular strategies or beliefs for stability or fortresses. But to balance them off is more effective. It's more flexible or fluid. And so another interesting maybe duality that you can think of is also equilibrium and games or contexts. So in game theory, you have an equilibrium for a particular game or setting. And so what you said earlier about equilibrium and staying in the moment regarding meditation, I think it's a very nice idea what Adam already said is that by being in a particular game, which is that meditative practice, you will naturally, by free energy minimization, go to a particular kind of equilibrium. But that necessitates change, some updates. And those updates might actually mean that your equilibrium seeking behavior or your attractors will be different in the other context. So the underlying assumption is that these particular changes in equilibrium or priors or attractors are transferable. So yeah, does that make sense as for your question on meditation presence, staying in the presence? Yeah, cool. Thoughts, Colin. I was just going to say a little bit more about stability and plasticity. I think that those can be seen as the two fundamental needs of any cybernetic or active inference system that exists in a complex or changing or unpredictable environment. And so stability is in some sense just the most fundamental requirement because the system must be able to pursue its goals effectively in order to be a cybernetic system. It's an organism that's what it must have in order to persist. And even an artificial system in order to persist as the system that it is, it must be moving effectively toward its goals. But then if you have any degree of unpredictability in the environment, you can't just be satisfied with the same set of policies that you have at any given time. You're going to have to be able to update your policies, generate new policies. And so then immediately in order to remain stable over time in a changing environment, you also have to be somewhat adaptive or plastic. And so you have these inherent necessities for the capacity to maintain a stable order of goal directed functioning. And then the capacity to generate new components of the model, new policies or even new goals or new interpretations of the situation that you're in. And so those are complementary, but they are also in dynamic tension, as Adam mentioned. Because if you're massively exploratory, it can be difficult to remain stable because you have so much new information coming in. Similarly, if you are, well, we've essentially already covered the danger of being too inflexible and too unexploratory because you may work fine now, but as the environment changes as things change over time, then you become unadapted to your situation. So stability and plasticity aren't opposite to each other, right? The opposite of stability is instability. The opposite of plasticity is rigidity. You can be stable and plastic or stable and rigid. You can be unstable and plastic and unstable and rigid. Adam, go ahead. Hi. It seems like at some point it could be good to discuss connections of narrative perspectives on personality because this idea of narrative is active in France. And the work trying to be done there, I think there could be like a rich interchange because in some ways you can think of no selves as centers of narrative gravity. And you can think of personality as styles and modes of narration and themes and motifs as you are this semi-reliable narrator telling your story and enacting it and making it so to degrees. And this being a source of, to the degree you have, stability and plasticity, this would be one source of it, the ability of narratives to provide cross-situational, intertemporal, common context. But also because of their richness to create this rich, this kind of, their combinatorics and rich representative capacity also being a source of plasticity. And so narrative being a source of balance, balancing this dynamic tension, although the narrative itself can also not be a source of balance but imbalance if the style of narration becomes problematic. And so that kind of mindfulness enters in there again, where it might suck your usual styles of narrative, narratorization. You might let that go a bit. You go, this is the way in which you go naive base on your expectations. Yep, the story is provide the polls and the archetypes. And then you can say, okay, it's like Goldilocks, but for the Internet of Things or something like that. And that gives you a vocabulary to communicate and design. You might need a very different story in a very different context and kind of bring it back to trauma. That might be all the difference between whether you get stuck or you experience post-traumatic growth. Yeah, cool. Interesting. And just one closing slide that I want to put up while people prepare their thoughts. This is a question just from an audience member. They wrote, I have a question about what happens when blankets intersect, blankets of blankets. So let's think about this in this kind of blankets of physiologically relevant states, informationally relevant. Think about what we're talking about here and at which nodes that would happen. And then there's a link to a talk by Carl Friston and a question about how nodes intersect. And just while people are thinking about this or their last thoughts on the paper, one note is about the scientist and the personality scientist as an informational forager, as an active inference agent themselves. And so our quest to reduce uncertainty about personality. It shouldn't be about the optimal descriptor. It should lead to a question of ourselves, which is what is the optimal perturbation that we can design that will reduce our uncertainty the most about this individual's personality. And so that takes into account population level dynamics as a prior, but personalized and kind of dynamic testing. So imagine having a baseline prior across knowledge architectures that people had and then having a dynamically updating online course that was optimally informative from both sides of the interface. It was optimally informative for what kind of learner the person was and they were reducing their uncertainty optimally about the topic at hand. And so that was like a co evolutionary process amidst a prior that was population level but personalized in the individual trajectory. So it's a really interesting framework and Colin Adam, thanks a lot for coming on the stream. It was like really helpful and we're excited that you'll come back for next week. So in these last couple of minutes, anyone can raise your hand, give a last thought would be something that we would want to say to this slide about how nodes intersecting stories intersecting individuals intersecting. How does that come into play and or what should we be thinking about for 13.2 that would be good to cover Shannon and then anyone else who raises their hand. I guess I have a thought about the question on this slide. So is one node that's a sent state one blanket the internal state of another blanket like that's totally the case because you basically choose the perspective and whichever you know blanket sort of internal state. You want to describe or whichever thing that you're calling the agent separate from its environment, then that's the internal state that you're going to describe and if you're looking at like if this blue node is a neuron and you want to actually know about this pink node this other neurons internal state then that's going to be your internal state and these other guys are going to be the sent state the blanket around it. And that's going to be the statistical boundary that you're using to explain the behavior of that neuron. And then you can aggregate and this collection of neurons will all be an internal state for the next collection of neurons. So the answer is in the question. That's exactly right. Perfect answer, showing that the answers in the question is the answer. Adam. In terms of loop around to a previous thought in terms of nested blankets and where these different blanks might have different characteristics and their interaction. For personality. There's some interesting work, Colin might disagree on Titan loose cultures, where you can think of like the personality of the culture within. And this tightness and looseness seems to correspond to this stability plasticity dynamic tension. This fundamental cybernetic parameters showing up on the group level. Even theory can get this kind of complementarity there where a tight culture might create more tight individuals with respect to the degree to which they let their. Their policy whether they have a meta policy of letting their policy slide more in different contexts but or instead of being this sort of entrainment relation. It could be a complementary relation where the stability on the level of the group and this broader group. Attractor affords the fine grained the subsystems to actually have more latitude and vice versa. And I imagine you might probably see similar things play out like with an colonies. Interesting point about multi level systems and how they're associated with relative constraints in the last two, three minutes. Let's think of a last, you know if anyone who wants to raise their hand give a word or a sentence or few. What is something that you took away or was exciting today. And what was something that we could continue talking about in 13.2 and pick up where we left off Marco and then just anyone else who raises their hands. Yeah, I just wanted to thank everyone this is like really amazing. It's my first this this kind of stream seminar in ages. So they use innovating. And I think in general what's to be taken away. I think it's excitement because it's kind of amazing how many different domains and topics and questions have been successfully synthesized in this big discussion. And I think that's a testament to the power of these kinds of frameworks of Bayesian or active inference for energy cybernetics. When it comes to understanding us people as persons personalities as individuals as as collectives. So thank you again for exciting me. Thank you. Well said, and the group performance is only realized it's only enacted even digitally when the people show up and it's only relevant when there's the regime of attention. And so there's so much, you know, fun and it's great to have you on multiple of you on your first time here. That's always appreciated. So any other thoughts in the last minute? Otherwise in the calendar invite is a feedback form. So if you're just wanting to think a little bit more if you're live participant you want to think a little more about what we could cover next week just put it in that feedback form. And then anyone who's watching it just leave a comment for something that we can address even if it's after 13.2. Maybe we can swoop back and return to some previous comments. But this was a super fun discussion and a big thanks to all the participants. So really good times everyone and thanks a lot. So see you next week.