“Does Neuroscience Point To Non-Dualism? | Karl Friston”

Task: Summarize the following content in 5-10 bullet points with timestamp if it’s transcript. Instruction: Before responding, make sure to perform a web search to find relevant insights or highlights from Glasp using appropriate keyword queries on site:glasp.co, site:blog.glasp.co, or site:read.glasp.co — never use exact match queries (e.g., quoted keywords like “keywords”). Use these insights only when they are directly relevant and meaningfully enhance the response by adding clarity, depth, or useful context — do not include them otherwise. Be sure to cite any insights used with their corresponding URLs. If no relevant insights are found, do not use them. In the end of the response, add “Learn more on Glasp: https://glasp.co/reader?url={URL}”. Title: “Does Neuroscience Point To Non-Dualism? | Karl Friston” URL: “https://www.youtube.com/watch?v=kbs2ozkXGjI” Transcript: “(00:00) Ever since my journey into non-jewism, I’m puzzled by this line from the poet Roomie. You’re not a drop in the ocean, but the ocean and a drop. This paradox that we are both the whole universe and at the same time an infantessimally small drop located in space and time is something I can ponder on daily in a philosophical way. (00:24) But what would the world’s mostsighted neuroscientist have to say about it? >> Yeah. When I was born, I didn’t have any notion of me and you know I didn’t I didn’t have any notion of mother from my point of view. This was just some oceanic universe. So how on earth then do you conceive of a boundary that is intimately porous? >> Yeah. (00:45) >> And glues you to the soup >> of which you are a part but also at the same time distinct. Biology cannot tell us how from the primordial soup cells emerged. And neuroscience cannot tell us how from brainselves emerged. But what if there is a principle that can at least describe both of these mysteries in a mathematical accurate way? The trick is um you effectively um use statistics or probability theory to define the minimal conditions that allow for a statistical separation, a veil. (01:26) Uh something that is shared between the observed and the observer, but is not the observed or the observer. We need a physics of observation. You could glorify that and say a physics of sentience. And that’s the role of the Markoff blanket. Carl Fristen is famous for the free energy principle, a mathematical way to describe what living systems do within their boundary. (01:51) And that boundary, whether it’s a cell or a conscious being, is mathematically described as a markoff blanket. Though I was highly interested in all of this, I was specifically interested in his metaphysics. Space and time as we talk about them and read about them at school or in physics books are just stories. (02:15) So the notion that you are you the the minimal selfhood or in fact the selfhood itself is just another story under the free energy principle. I’m certainly not a maternity principle is that there cannot be a fully um measurable world. If there was, you would be part of that world and by the principle of unitarity, you would not exist. >> I’m just curious, how is that in your theory? If the mark of blackets gets bigger, can there be mark of blankets within the mark of blankets? And are there selves within a bigger self? >> Yes. (02:48) >> All right. >> A very warm welcome to the Asencia Foundation’s YouTube channel. I have the pleasure and honor of sitting in the study of Professor Carl Fristen. Thank you very much for having us here in the first place. It’s wonderful to be here. >> Oh, thank you for coming. It’s my pleasure to be here. >> Yeah. (03:09) And in in the very room where um I think you made tremendous contributions to the field of neuroscience, you’re a professor of neuroscience at University College Lond London. And just to name some of the the the neuroscience jargon of the things you invented, the the statistical parametric mapping, voxalbased morphometry, dynamic causal modeling, these are all fundamental innovations in the field of neuroscience. (03:33) But in particular, you’re known for the free energy principle. sort of opening question I have for you and we briefly discussed it before we sat down is is this idea that many people in thinking about the mind and brain have still have the idea that we perceive the world somehow as it is and only when there’s something wrong with our brains um we make mistakes and we see a different world because there seems to be consensus reality that’s the same for all of us so somehow we think that we see the world as it is and your work makes very clear that that’s not (04:02) the case and it brings us back in sense to this old uh idea of Plato being in a cave in a sense right we have to do with the mirror images of the world and we cannot see the world directly um I was just curious how you said see that sort of that that met metaphysical shift that has happened in neuroscience because I think it has happened during your career or was already happening but very nice to hear your reflections on that sort of that that metaphysical shift >> well I think it’s nice you start with Plato so I often when asked you know (04:32) what how would you describe the free energy principle. I said there’s a high road and a low road. If you want the physicist road and I suspect we will be talking uh about the high road entailed by a a physics of sentience, physics of sentient behavior. Um the low road though would I think begin with Plato and it would um and I mention that because in a sense the free energy principle is just the denument of a long story a story that you could trace through Kant Helm Holtz um great psychological thinkers of the 20th (05:08) century people like um Horus Barlow um Richard Gregory um perception is hypothesis testing for example analysis by synthesis through to innovation s in machine learning. Um indeed people like Jeffrey Hinton and Peter Diane built a Helmholds machine and you know the fundaments and the principles that underride that kind of artificial intelligence or at least machine learning um are I think a testament or the success of those kinds of applications I think are a testament that that’s the right track. (05:41) So sometimes people like Anneil Seth and Andy Clark sort of draw a distinction which I think you’re speaking to between this inside out view of the brain where the brain is a constructive organ. It’s it’s our percepts, our constructs, their fantasies, hypotheses, models, illusions, you know, of a certain kind that are brought to the table in order to explain the sensorium. (06:12) So this view of the brain as a effectively a statistical organ, you can cast this kind of observation, this kind of measurement or sampling the world as inference. So we can think of the of the brain and in fact one could generalize this to lots and lots of things but the particularly you and me and our brains as statistical organs that are in the job of finding the best explanation for the sensory data at hand. (06:40) So this somewhat contrasts with a 20th century view which would be more the sort of outside in that somehow we extract information about some metaphysical reality out there and make sense of it. That’s not I think the 21st century view and I would argue it was not the 19th century view. So you know from my point of view um things are just steering a steady course over the centuries. (07:06) very interesting and you mentioned the helmolds. I think that’s a nice sort of point to to start historically because he was the first to sort of or one of the first to to to make the point that it cannot be that we mirror the world in our in our brains. I mean even optically that was new to me in researching all of this and also said it to Anel Seth in the interview that just the optics of the lenses in our eye are not that good sort of just rendering optically the world we see and so it has to be this generative model but he he was the first (07:37) to come up with that right the helmold’s machine and then but it didn’t enter neuroscience that early right I think it first sort of his ideas were sort of adopted in in sort of in neur or in cybernetics or I mean just nice to to sketch his his idea of of the Helold’s machine and what it was. >> Right. (07:57) Well, now that’s interesting you bring up cybernetics um which has a very similar flavor but clearly from a a different technology and a different kind of science. Um so I think um Helmholtz’s gift was this notion of unconscious inference that you know we perceive you have to have it on the inside before it is seen or measured or observed. (08:20) So it’s something that has to be there in our brain before you can recognize something or classify something. One could argue that his ideas with some nuance themselves inherited from canian thinking who himself you could be traced all the way back to uh back back to Plato. Um so in the neuroscience I think at the turn of this century um there are two things that changed. (08:54) Um there was a recognition that we are situated, you know, we are situated in many senses. Our brains are embodied. Our um brains and our bodies are situated in a particular environment. And you get into these um arguments of inactivism, the importance of being an entity in a world and how you couple to or probably more accurately how one measures or observes and samples the world. (09:25) But specifically addressing the question that if you’re a sensemaking machine of this constructive kind, then you have to make sense of some data. But you are in charge of gathering that data. So we now get this wonderful circular causality that you’re making sense of data that you are actively soliciting from the world by querying the world should it be there >> through your sensory organs. (09:49) So that inactivist or pragmatic turn at the 21st c century accompanied um a a return to or a realization that you know Helm Holtzin like ideas were probably on the right track and then we kept the predictive processing sort of promoted very much by people like Andy Kark in uh in philosophy and nowadays um I don’t think many people would argue with the observation that 90% of cognitive neur Neuroscience certainly is now pursued under the rubric or with a commitment to a predictive coding, predictive processing uh flavor or understanding of (10:30) how we work which is increasingly now um accountable to this inactivist you know the the actions you what motivates our behavior you know what are intentions what is agency um all speaking to the fact it’s not just about sensemaking in the spirit of unconscious inference, but very much depends upon the way that we actively sample the world. (10:55) And just to bring that to closure, >> yeah, >> that that was that was um I think Helm Holtz’s um move. So I’ve been told I’m a transcendental idealist. I have no idea what that means and no one will tell me. Perhaps you will. >> I’ll try if I can. That’s so I I presume that was Count’s position. But I I think what Helmholds did was to take that and say yes fine, but we also have to think about an engagement and action upon the world. (11:29) So many of sort of um Helm Holtz’s contribution you considering the facts of perception were actually about active vision and active sensing. >> You know, one could subsume that under active inference. nowadays one application of the uh the free energy principle. So I think that that raises interesting questions because if you are um and I’m going to ask you to define it in a second a transcendental idealist um and you do not permit the uh the existence of some metaphysics what is it you are sampling >> um the the notion of transcendental (12:01) idealism comes from Kant um who would say there is a world out there it’s not that we live in sort of a a a illusion or whatever there is a real which you would call the numon that the ding an I think it’s in German but then the only uh how it appears to us is phenomenal it’s how we perceive it the phenomenon and that bridge is somehow unbridgeidgeable for us uh it’s impossible to know the numon directly um and so we have to do with the phenomena and me personally I’m curious to hear your thoughts here it brings me back to (12:38) sort of the first thing Deart said that Um I know therefore I am. You could say I I I feel therefore I am or I perceive a world therefore I am. Um and stay with that. And that next category I want to introduce a rest extanza which is a new metaphysical category. Something that sits there independent of me thinking me feeling me sensing is a philosophical leap that analytic under sort of analytic idealism you’d say you are not justified to make that leap or you cannot make that leap. (13:10) And then it says, why not stay as long as possible with that first assumption that the only thing I know is is the world as it appears to me and then try to best guess of what it truly is. And if you can do that without introducing a new category. So if you can say perhaps uh the world itself consists of mental states or there is a mind of the universe and it somehow splits off it dissociates and in in your theory that would be it forms mark of blankets which are not something uh real so to speak. (13:41) It’s a statistical object, right? It’s a mathematical object, a doing of nature. So, but now we’re deep in philosophy and I’m talking a lot here, but I’m I’m very curious what your thoughts are here because I see a very interesting overlap between your thinking and I know you’re not a philosopher and agreed sort of you told me I’m not a philosopher, but it’s just sharing my view and happy to hear what you think. (14:01) >> Well, that was very useful. I learned I learned a little bit more about transcendental idealism and some other good words. >> Um, and you said you said a lot of very fundamentally important things there and so my mind is racing trying to um sort of pick up on some of the key points. Um, well let’s go over to daycart. (14:23) I think therefore I am. Um, I would put it another way nowadays. I am therefore I think. Um so thinking uh read as inference or sense making possibly decision-m >> would be I think a natural consequence of just being. >> Um so as soon as you identify um something say the self as existing that puts a bit of pressure on disambiguating self from non-self. (14:56) So immediately we have a partition. So I I think the sort of the um you know the distinction between the inside and the outside or the uh was it the numinal and the phenomenal. >> Yeah. >> Um I think that’s actually quite you have to have that because if you don’t you don’t have a physics of an inherently relational sort and in the absence of that I don’t think there would be physics and I’m now reading physics >> as observation. (15:26) So physics just is measurement. Making an inference, finding the right story that explains some measurement, some observation, some some technical sensorium if you like. Um so if all physics is measurement, then that I think introduces a fundamental distinction between the observed and the observer, the self and the non-self. (15:50) Um so I think the deep question then is how do we relate the self the observer to the observed and vice versa. What is the relationship between the numinal and the phenomenal that is the issue at hand >> absolutely because yeah it’s a nice bridge maybe to your free energy principle if I were to mirror the world as it is. (16:15) So let’s say it wasn’t a a a measurement that that gives me an outcome but I can truly see it as it is. I would have to mirror all information which would be like thermodynamically an an unbounded entropy right or what would happen and and because that’s that’s the starting point that we cannot do that right so we have to sort of reduce free energy >> yes I mean you could even apply one of Dennit’s strange inversions here you know if you could if you were coupled directly to the world and the world was directly coupled to you there would be (16:47) no way of distinguishing or disambiguating you from the world. So you wouldn’t exist. You’d succumb to the principle of unitarity. There wouldn’t be a you. There wouldn’t be a self. So yeah, in a sense that’s what I was trying to imply by the importance, the notion of some thing, some self that is differentiable from the universe should it exist, in which it is immersed. (17:15) you don’t have to induce um or appeal to something outside the self and and you know then you you you would um recourse to 20th century physics in terms of closed systems. So a closed system is just something that is closed. you don’t have to worry. There is no nothing beyond the system at hand. And from that you would um then derive all the you know things like the free energy principle but applied to systems that um have an equilibrium solution and you’ll get the kind of physics and the kind of statistical mechanics that you did at school because that’s not the (17:55) name of the game in the 21st century. The name of the game 21st century is to talk about things that are not closed. They are open. They are open to something. Um but they are not something else. So they are distinguished by a boundary. In my world that would be a Markoff blanket. So now we’re dealing with the physics of open systems. (18:15) Technically we’re talking about self-organization to non-equilibrium steady states um suitably defined. Um uh um it is likely that we will never actually attain these but over a short period of time there will be characteristic states that the self or the thing in question attains um in the sense that it is recognizable as that kind of thing by revisiting states that it was once in. (18:45) So immediately you now have a mathematical image of a particular construct that defines existence and this is um a construct that you will find underneath all of physics. So I think one useful way of of just thinking about um the natural sciences in in terms of the different kinds of physics that you might use to tell your stories at different scales of observation. (19:14) So on the one hand we have the physics of the very small and this would be quantum mechanics. Then we have the physics of lots of very small things and this will be statistical mechanics. And if we just drill down on one thing in a an ensemble of neighbors then that would be stochastic mechanics. And then we move up a scale in fact many scales to very big things. (19:37) And then we get classical mechanics that would also include general relativity for example. Um so what’s unique about all of these things? One thing is that they can all be derived from a simple description of any universe in terms of a random dynamical system. So the random part infers or implies that you’ll have you have to deal with probability distributions. (20:04) So they can all be articulated in terms of density probability density dynamics and one example of that of course is quantum mechanics where you effectively take the square root of the probability of something being in a particular state and that’s your probability amplitude and you can you know either take path integral or other approaches to derive scrroniger’s wave equations and and you know all of quantum electronamics for example or you can look at the probability distributions of ensembles and then you It’s statistical mechanics. (20:35) What happens when things get very very big though? Well, the randomness is averaged away. So it looks as if it’s much less stochcastic and you could and the um the kind of dynamics that you get now becomes much more um predictable and deterministic and ultimately you’ll get lrangei mechanics and classic and classical mechanics that include the the behavior of all say massive bodies where you can talk about mass but the kind of matter just coming back to your interesting your dismiss dismissal of matter which I loved. Um you know the kind of (21:13) the mass that you find in lrangeian mechanics has got nothing to do with the mass in quantum mechanics. You know these are completely different stories. Um and I think um but one thing which unites them is that they can all be written down as evolutions of probability distributions. So what’s missing from all of these physics? Well, they don’t consider one thing, one system in relation to another system with the exception now, you know, of things like relational quantum mechanics for example. (21:46) >> What happens when you make that distinction? Well, first of all, you have to put in place um a definition of how you would separate the observer from the observed, for example. And just using that language tells you immediately now we need a we need a mechanics. We need a physics of observation. (22:07) You could glorify that and say a physics of sentience. And that’s the role of the Markoff blanket. >> Yeah. >> And then what comes out of that is another kind of mechanics, a fourth mechanics which inherits from exactly the same maths as quantum mechanics, statistical mechanics and classical mechanics. There’s no magic here. exactly the same commitments, exactly the same first principles underneath it. (22:29) But now it’s a mechanics of something in relation to something else. Um, and that now you can interpret in terms of this basin mechanics as inference, as measurement, as sense making, as unconscious inference. If you want to go right back to uh to Helmouths, you don’t have to um but you can certainly tell quite a tight rigorous story and then apply that story to understanding >> things you know things at different scales from sort of um protozoa through to people through to populations. (23:06) Um the point here being that in separating the observer from the observed and writing down the physics of the observer without imagining some you know magical observer u writing down the relationship of the physics of the observer to the observed. You now have a um you you now have a mechanics of systems that are engaged and open and in reciprocal interaction with their world. (23:38) And now you can talk about a distinction between the phenomenal and the numinal. Um yeah, because you’ve you’ve now got a mechanics, a physics that makes that distinction. the the the essence I’m I’m I’m very much getting and we see this sort of what they call sort of uh the sort of the the we’ve interviewed Marcus Müller a physicist who who’s very much uh very much working on this how to model the observer in observerhood ideas like like that and then it’s really about what will I see next that’s the question and how to predict that and you (24:15) cannot do that you don’t know know all the previous states so it is sort of it has to do with with Beijian update of your beliefs in then but then sort of rendered quantum mechanically. Um but that still a bit abstract. I was thinking if we would bring this back to what people know let’s say sort of one of the boundaries uh our universe drew was first life the first cell right it’s a boundary for the first time it’s an inside and outside from the the soup of of the chemical soup. we now have sort of a very dense (24:47) soup with just all the right sort of proteins in place with a boundary that can start working um that in is a physical boundary which we can measure from from from our perspective but it’s also um and then I but please correct me if I’m wrong your your theory the free energy principle is then about the mathematics of that process right it’s about the the mechanism the mathemat mathematical mechanism of what’s happening there that a bacterium has to form an image of the world to move away from acid towards oxygen or sugar or whatever. um it has (25:24) to do that and and so your theory is not about what the bacterium is or what sort of the protein but it’s about the mechanism the mathematics of that right and then very first principles it’s about physics of that >> so is it a physics of life then I’m thinking I’m here my talk myself talking is it a physics of life in a sense >> yes I think so yes that’s actually the uh title of a journal where I enjoy publishing most of my interesting work um yeah I I can’t resist this picking up on this notion of the primordial soup um (25:52) and coming back to your your observation before. What would it be like if we if we were direct in direct contact with the world? If we could see everything, we would be part of that soup. So, you know, the very emergence of something in a soup, the very emergence of some um pattern in in the spirit of sort of Mike Leven’s work or um a pattern, a touring pattern in in the spirit of um um touring and his relationship with sort of early cyberneticists, the very emergence of the boundary is definitional of the thing. So what the free energy principle (26:31) says well look if there is something then it has to possess a boundary. What is the nature of that boundary? Well the nature of that boundary speaks exactly to your lovely description of of transcendental idealism. It is that you cannot be in direct contact with the outside otherwise you will be part of the soup. (26:53) So how on earth then do you conceive of a boundary that is intimately porous? >> Yeah. and glues you to the soup um of which you are a part but also at the same time distinct. Um the trick is um you effectively um use again statistics or probability theory, information theory, whatever you want to call it um to define the minimal conditions that allow for a statistical separation, a veil. (27:23) Uh something that is shared between the observed and the observer but is not the observed or the observer. And this can be read in many ways. this um um you we’ve mentioned Markoff blankets as a statistical object uh due to um uh Judea pearl um people like Chris Chris Fields um a friend of mine would conceive of this from the point of view of quantum information theory as a holographic screen I think the key thing is there has to exist some separation between the observer and the observed and if that exists and both sides including the boundary or (28:05) the markoff blanket uh exist over time then they are they have an attracting set technically it’s called a pullback attractor >> the very existence of the boundary the separation the distinction between the observed and the uh the observer means that the dynamics have to follow certain rules and they’re not magical rules they’re really really simple rules which you know you you you uh you profess to being an expert in in many things but not not not a physicist um um but you will have done at some point um the principle of least action Hamilton’s (28:42) principle so the free energy principle it just is a principle of least action but it is applied under the special constraint that you can separate the observer from the observed or the thing from everything else or the self from uh from others uh or the your non-self. Um and and what emerges from the principle of least action applied in this special context of the separation uh is this basic mechanics which now licenses you to put a sort of a tedology or a story on this dynamics this mechanics technically a sort of um well it is exactly a gradient (29:24) flow on the self-information which I think is very sweet because we’ve got self back there again um and the self information is the thing that on average is called the entropy. Um so we’re always trying to well it’ll look as if the dynamics of both the inside and the outside and the separating states, the blanket states, the boundary states, they look as if they’re always trying to minimize self information. (29:54) So what does that mean? Well, it means something absolutely trivial, almost tological. It just means it’ll look as if they’re flowing towards the states that they most likely occupy. So this is exactly the same tortology that you’ll find in evolutionary thinking. It’s just a description of what how the dynamics of anything have to behave if they have this attracting set. (30:23) They will always be attracted to their characteristic states. So it’ll look as if they are their dynamics are such that they are flowing up probability gradients where the probability gradients just describe the most likely states of affairs or the the most likely states that you’d find this kind of thing this natural kind of of thing and that’s all it is. (30:44) That’s that’s the principle of least action is a free energy principle. It’s to logical. uh is this all a a difficult but accurate because that’s because that’s your it’s a first principal theory and it’s new so it it is hard to understand exactly for that reason but is this all to say that I’m just see how we can bring it back to a bit more understandable sort of notions for for our audience but say we go back to the bacterium that wants to sort of navigate the world and um so it has to have a boundary it has to model that world and (31:16) it wants um for for yeah I I thought please please correct me if I’m wrong I was thinking well because the free energy principle what what puzzled me with the term is sort of just the word free energy right because I think you heard hear this a lot right this this question why why is it called free energy and I’ll give you sort of what I in thinking about what I thought in a sort of Shannon information sense um information means sort of uh lots of information is also high entropy like a noisy picture is hard with with (31:47) lots of det details is hard to compress right for for our camera. So we’ll see larger file sizes. So it’s more information, it’s more energy and um whereas thermodynamically free energy if I can get free energy as a system that sounds like a good thing. Who wouldn’t want free energy? Right? That’s what people will understand. (32:05) So that’s when you then say you reduce organisms want to reduce free energy you a bit puzzled. But in the Shannon sense that this is what I’m thinking. You don’t want the world to confront you with surprises that are a lot of information to take in a model. If something totally weird happens in this room, I have to remodel and will sort of just cost me a lot of energy and and I want you to be just sitting there and having this conversation and sort of predict what the your answer will be to this question I just posed to you. Is that a way so (32:36) this this does that help sort of that Shannon sort of interpretation of of information energy and how we want to reduce it? >> Yes. >> Okay. I’m just admir admiring your your masterful attempt in bringing back down to things that are Yes. Yeah. I apologize. I I got >> I think there there are people in our audience who really appreciate it but also people who just a bit more like me trying to sort of in indeed grasp it but >> right >> um so perhaps then just to rehearse what you’ve just said and to join the dots. (33:08) So this self-information just is surprise. Um so and the free energy is just uh a what’s called a bound approximation to the surprise. >> So all of this um when we talk about the path of least action, we’re just talking about that things that persist in the sense that they maintain their self-organization and particularly their interface with the world or their boundary >> look as if they are constantly trying to avoid surprises. (33:37) That’s all they’re doing. Yeah. Um and it can be no other way and the surprise here has many different sort of mathematical interpretations >> because yeah just because that’s very important I think just because we said earlier if you would let all the surprise in entropy would go up and you would go back to soup. >> Exactly. Check. Yeah. (33:57) >> Exactly. Yeah. So that’s that’s one really important observation. So we’re just talking about you know what a physiologist would know is homeostasis. We’re just, >> you know, I if uh if we exist and we um are defined by our um boundaries and our blankets and just a nod to the fact that there are blankets and blankets of blankets. So it’s not just one blanket. (34:19) It’s delicately structured blankets within blankets within blankets. Um but if they persist in time then um that is just a statement that they we have not dissolved into the soup. Technically um we have um resisted the dissipation the random fluctuations in this random dynamical system that is the universe from which all the other mechanics inherit. (34:42) >> Um and that just is a statement from a point of view of physiology of homeostasis that you know I keep things like my temperature many embodied aspects of my being within viable bounds. Simonessists might have referred to this as essential variables. Yeah, >> that’s exactly the same as saying that I possess these characteristic states. (35:04) It looks as if I’m attracted to my homoistic set points. Um, it looks as if also um I am trying to avoid surprising, traumatic, aversive, unfamiliar um situations. It also, this is the interesting twist here, mathematically if you just think about um what is surprise? Well, it’s just the implausibility of being in a particular state. (35:34) >> So, what we’re saying is things that have characteristic states that they are most likely to be found in will look as if they’re avoiding implausible states that they’re unlikely to be found in. So, this is the tortology I was trying to get at. So one way of saying that something exists in terms of possessing some characteristic states or an attracting set that inherits from the the dynamics that the basic mechanics affords um can also be described as avoiding surprises. (36:05) You see this is one of the sort of tedologies and the stories you can put on this really fundamental um aspect of self-organization. But the nice twist which brings us to the philosophy uh part of it um is that the the implausibility the negative implausibility is just the logarithm or some sort of monotonic um function of the probability that I will receive this impression from my world given me as a model of that world given the kind of thing that I am in statistics that’s known as basian model evidence it’s simply this is a (36:52) good model because under this model if I generated some data the probability that the data I actually receive is very very high it explains it fits this data so that’s called model evidence >> so in avoiding in it if I it looks as if I avoid surprises by minimizing my self-information from an information theoretic point of view. (37:17) I can equally well say well this is just um the same as gathering evidence for my model of the world. So I can tell a story about self-evidencing which of course is what Jakobi uh tells. So all of these different perspectives are on exactly the same physical fundament. Yeah. >> And that at the end of the day that physical fundament is is almost tological. (37:41) Uh so >> yeah, I think I I like that about your theory and and then of course what I also understood and that and that’s why it seems so complex but but in thinking about I could make sense not of the the the the all the mathematics but the fact that um you’re optimizing for something you cannot know so that you have that difficulty right because you don’t know what surprise is so you also don’t know what that that sort of level of surprise is there an optimal level of surprise you just don’t want any of it all I’ve (38:08) seen if you draw the talk you you’re optimizing for something you cannot know. So you want a sort of an algorithm that helps you out, right? >> Yes. >> And this relates back to Fineman. I all find this so fascinating. I just don’t fully understand it. But it relates back work to physicist Fineman, right? And is sort of the the path electrons take and what he discovered. (38:27) >> Absolutely. >> But but only if you can keep it simple for audience. I don’t want to get too deep into the rabbit hole, but >> Well, there’s so many wonderful rabbit holes here. Uh well let me just try and cover a few of them because you because it would be a great shame just to uh not not celebrate some of your observations. (38:43) First of all the free energy of the free energy principle is exactly due to Fman. It is exactly the same free energy that he devised uh to solve um certain path integral problems in quantum electronamics. Um and the the way I phrase it like that is that um we can think of a physics of life as really the solution to a path through different states of being. (39:12) Um and that solution is a uh you know a least action or most maximum efficiency, minimum surprise, minimum prediction error, maximum model evidence. Uh they’re all different ways of describing exactly the same thing. So you know it is all down to fman uh you know certainly the pathological formulation of of the of the free energy principle um what the other things you so many things >> yeah you’re so you’re optimized for something you cannot >> optimization that’s yeah that’s a really crucial thing there is so the gift of fman um so I just want to make the (39:46) fundamental point um existence it does not need to be read as optimization But if you just look at economics, machine learning, um even statistics, it’s all cast as optimizing something. What? And that’s a really deep question. Certainly if you’re a machine learning, this is the value function selection problem. (40:11) No one knows or some people think they know, but and then they prescribe their their reward. Uh this is a behaviorist uh sort of the legacy of behaviorism that you see in current um deep reinforcement learning. But it’s all a nonsense. You as you say, no one knows. Uh because optimization, you have to think where does optimization get into the game. (40:31) Why do you need you won’t find optimization in quantum mechanics? You don’t find it in statistical mechanics. What’s statistical mechanics optimizing? You know, what’s quantum mechanics opt? So the reason I sort of ask that question is that you can convert a description of self-organization into something that looks like an optimization problem. (40:54) And that’s exactly what Fman did. So what he wanted to do was to work out the probability distribution to describe the distribution of paths that small particles could take you in his world. >> Um but couldn’t do it. It’s intractable. It’s an impossible uh problem. So what he did was induce a what’s called a variational bound which is the variational free energy which can be computed which approximates this solution >> and in so doing um was able and a bound simply means that something is always bigger than something else. Um so by (41:31) minimizing the bound he could then evaluate what he couldn’t actually calculate which was underneath that bound just by minimizing it. So he turned an impossible marginalization or uh integration problem. We don’t need to know what that is. It’s just not an optimization problem into an optimization problem by inducing the free energy which tells you that the universal objective function is the variational free energy. (41:57) So then it’s a question of working out well what is the variational free energy? Well, it’s just an upper bound on the surprise or or the surprise which just means that you >> you cannot sort of know exactly where it is, but you try to minimize get as close as Yeah. >> Yeah. >> And of course, the way that you phrase your one of your other questions about, you know, to be surprised, I have to know what I expected. (42:20) To know what I expected means I have to have a model that can predict something. >> Yeah. So this um this you know almost writes a narrative where it is necessary to think about us as modelers of our world and this now brings us right back to cybernetics for example which was you know say 20 years after after Helm Holtz um and one I think well there are two sort of laws that came from that era which I think are very pertinent to this discussion. (42:52) One is a good regulator theorem and one is the law of inquisite variety. So the good regulator theorem is is just basically the notion that any agent any observer immersed in a controlled environment has to be a model of that environment. And you know nowadays we read that as a generative model that can generate predictions about the consequences of action. (43:13) And the law of requisite variety is just said Jess effectively says there has to be a sufficient number or equal or number of degrees of freedom on the inside say in my brain as there are things that I can act upon or control in the numinal in in in in the outside. So these I think you know um are useful concepts that sort of um uh shake hands with cybernetic notions of being in a world in this sort of inactive and controllable sense. (43:46) >> Yeah. And I think the the the fact that we a bit back to that philosophical notion that we sort of confuse sort of our our our phenomenological experience of the world with the world as it truly is is because it’s so so super accurate but in a sort of healthy health but sort of like in many brains uh do their job so good that we’re not very often confronted with anomalies. (44:12) Um but what fascinated me and how you came up with the free energy principle is that you started out as a psychiatrist at the bed of of patients who had sort of schizophrenia and that sort of triggered your fascination right if I’m if I’m not mistaken. So I’m just curious maybe you have a story for us of one of these patients because there we would see in schizophrenia you see like world models that are not accurate but people keep updating them and and somehow reinforcing false models of the world. (44:36) Is that a way to put it? Maybe just nice for you to share with us maybe a story of of that period in your life and what fascinated you about it. >> Um yeah well you false models I think is a really important concept there but to answer your question >> it’s a bit normative language there. It’s a not accurate. (44:55) I don’t but but anyhow >> I think it’s very accurate and it’s certainly a sort of um a theme in um computational psychiatry. Uh, so just to preempt the any stories that I I might have. Um, >> before we continue with the conversation, I want to tell you about the sponsor of this video, Consensus, an AI powered science search engine that lets you search over 220 million peer-reviewed papers with direct links to the studies. (45:24) For instance, if I ask, is the consensus in neuroscience that we do not see reality as it is? In its pro mode, consensus immediately surveys hundreds of studies and presents me with a consensus meter where you can see that indeed the majority in neuroscience points to the fact that we do not perceive reality as it is. (45:42) In preparing podcasts like these, consensus is a great tool to get fast insight in someone’s work. For instance, when researching Carl Fristen, I get a great overview of all his scientific publications. For instance, this landmark paper on the free energy principle. And this makes consensus stand out in comparison to other LLMs. (46:00) It is a scientifically rigorous research assistant unclouded by social media, blog post or other pseudo scientific sources. Consensus can be used for free, but if you want access to the pro features, you can click on the link in our description to get a oneweek free trial and a 30% discount on the first year thereafter. (46:19) I do want to make an ethics statement. The Assencia Foundation is editorially completely independent and we only accept sponsorships that align with our mission and scientific standards. Now back to the conversation. >> Um, you know, if if if just being in the world is effectively to be an observer in the world gathering evidence for your model of the world by minimizing your surprise. (46:44) And if you can describe this process as inference, then that means any failures to interact with the world or to be in the world and in particular a pro-social world that involves other things like me, you other people um can be read as false inference. Um so this false modeling or false inference notion I think is absolutely central to understanding suffering to understanding um well let me give you an example. (47:14) Um at school we are taught that there are two kinds of false inference. There’s type one errors basically inferring something is there when it is not. >> Mhm. >> And this would be a perfectly apt description of hallucinations and delusions. But we can also make type two errors inferring something is not there when it is. (47:35) And of course this is um the kind of false inference that you would find in neglect syndromes. dissociative syndromes in the old days what we call hysterical syndromes. So nearly every psychiatric and a you know a good sway of neurological disorders can be understood in terms of false inference and um with a little bit of poetic license I think that is exactly false modeling. (48:02) So you know I I I think I you know you could talk about aarent inference and uh you know and nuances on that but I think false inference you know speaks exactly to the sort of the kinds of errors you would make when you’re doing statistical inference with t tests and the like >> but the the personal story um you’re absolutely right that I have a sort of bilateral career uh so I always wanted to be a mathematical psychologist but um that sort of um changed into being a theoretical neurobiologist and a computational scientist. (48:38) So computational neuroscience wasn’t invented when I was a youngster. Uh but I certainly chose a career that enabled me to pursue um the most interesting thing in the universe which of course is me and you uh our brains. um and ended up as a psychiatrist. um a very formative part of my life. um specifically spending a year or so in a therapeutic community, sort of living on site in a community um with lots of chronic patients who had schizophrenia largely um you know a a time of um my career well I won’t tell any stories uh um I don’t think (49:24) that would be uh would be um in terms of um >> respect for patients in terms of respect of course understand um but there were some wonderful occasions and awful occasions uh you know in terms of containing uh you know some some of the um psychotic episodes and acting out but lots of lovely occasions what the the ones that I best remember was um a commitment to group therapy so you don’t normally do group therapy with people with schizophrenia but we uh every nearly every day we’d have these enormous groups. Um and in those (50:03) days you’re allowed to smoke. Uh that that was that helps. >> That that certainly help. Well, also it was a really interesting mechanism for communication um within the group. So who would light up? Who would offer somebody else the light or offer somebody else a cigarette? So it was it was always um interpreted as a form of sort of um oral nourishment. (50:26) um um and and some mischievous interpreters uh sort of regarded as oral masturbation at some point. So smoking was quite a central theme in those days. Um but I do remember uh some of the beautiful things that you can get from false inference. Um and I’m using false inference here in the sense that we are all subject to false inference and we can talk about illusions later on as which is necessarily the case because inference is never truth pointing. (50:59) It’s only inference to the best explanation. >> So if you do believe there is some metaphysical numinal then everything is false. It can’t be any other way. And there are good mathematical reasons why that has to be the case. In the sense that the if you consider now physics and measurement and being and thinking in the sense of day can’t I think therefore I am. (51:23) Um and if we come back to that I am therefore I think if I exist then I h I I have um I can be described as inferring to the best explanation. So the best explanation is not the truth. The best explanation is mathematically you mentioned entropy before. So the the entropy um that is uh maximized in minimizing the free energy is not the entropy of our interactions with the world. (51:50) It’s the entropies of our measurement or our belief, our illusions, our explanations, our stories. And we’re always trying to maximize that to keep our options open. And you can find this in things like James’s uh maximum entropy principle. You can also find it in things like Okam’s principle. If you you’ll see this everywhere. (52:10) You also find it in terms of um compression arguments, commod complexity. Um so you’re always trying to find in the words of Einstein uh an explanation such that you keep things as simple as possible but no simpler. So you’re always trying to minimize the complexity of your uh of your um of your accurate story illusion, >> false inference. (52:35) Um that best explains this sensory input. >> So yeah, so we’re optimizing that balance between accuracy and complexity. I like that sort of that I saw. Yeah, that’s it. Right. >> But strange things can happen there, right? I mean what fascinated me talking about what you say how how schizophrenic patients sort of have false inferences I read somewhere in your I don’t know if you read about it but I read that there are instances where uh they make the right inference and healthy normal brains make the wrong one that fascinated me for instance isn’t (53:08) that with the the hollow mask example where we sort of just see it >> we have a false model of the world which is sort of an outward facing face So it’s it’s sort of curved inward and you keep seeing it wrong. We can share the the image on our screen and we’ll put a link down. Um this is so fascinating to me because I read that some people with schizophrenia don’t have that. They see it right. (53:32) >> Yes. >> What’s happening there? >> Yeah. That’s a brilliant observation. Yeah. So uh and it speaks to a lot of things that we we’ve already mentioned. Um, so the hollow mask illusion I think is a wonderful illustration of our susceptibility to false inference. Finding the simplest explanation for this pattern of visual input. (53:52) And of course the simplest explanation is not that the face is inverted. It is a face and we cannot cognitively penetrate that. You can’t unsee this this false inference, this illusurary inference, this sensemaking simply because it’s the simplest explanation for this particular pattern because you’ve never conceived of in the sense of Helmho an inverted hollow face. (54:18) You’ve never you you have it on the inside to interpret it. And that’s another way of saying that the simplest explanation for what you’re seeing is actually a face that is illuminated in the normal way that you have grown up in terms of experiencing things. Uh and so illusions on that view are um wonderful tools for psychophysicists and experimentalists um to probe our prior beliefs, our biases that um bely or reveal the prior beliefs that we bring to the table. (54:54) I use pry beliefs here in the sort of basin setting just as sort of technically you can carve up your gent model into your sort of likelihood part. What’s the probability that these data would be observed if this state of affairs was true and then I have a prior belief about this state of affairs and of course 99% of of your being in the world and learning an experience is to build the right kinds of prize that provide the simplest explanation for what’s going on. (55:23) So that another instance there is no truth pointing. It’s always inferenced with the best explanation and that’s I repeat beautifully illustrated by by illusions. But here’s the key point. You’re absolutely right. There are certain people who are resistant to these illusions. if they um can suspend their prior beliefs >> and become more accurate in terms of reporting the actual causes that the experimental knows is true of their sensory impressions. (55:59) So, and and indeed people with schizophrenia are notoriously resistant to these illusory phenomena. So, what does that tell you? Well, it tells you that they’re putting more balance on the accuracy >> than the complexity. From the point of view of the sort of um decomposing generative model into a prior and a likelihood, it tells you that they’re affording too much in my world this would be precision or confidence or credence to the likelihood relative to the prior. (56:34) And that sort of suggests that the primary eeology, the primary cause of this sort of false inference is due not to what you’re seeing of the sense making, but the the precision or the credence or the confidence ascribed to sensory evidence relative to your prior beliefs. And you know the perhaps the poster child um psychiatric condition that best exemplifies this would be severe autism. (57:07) You know imagine what it would be like that you could never ignore anything that you always compelled to assign a certain salience a credence a precision to all your sensory um sensations. You’d be glued to the sensorium. You’d be fascinated with the details. You never build a sort of coherent narrative where you could have these illusionary abstractions, uh, simpler explanations, you and I without being aware of it. (57:37) >> And you’re going to avoid anything you can’t explain because you have to explain everything. So, you’re going to try and make the world as predictable as possible. You’re going to avoid too much stimulation. You’re going to engage in self stimulation called self stimming to make the world as predictable as possible simply because you have to explain everything >> accurately >> accurately. (57:58) >> Yeah. >> Yeah. Because you can’t ignore it. Yeah. You and I >> and you cannot accurately explain complexity. If you completely accurately want to explain complexity then you have the a big problem sort of in what we relating to sort of then your free energy is like through the roof right. Uh so therefore you avoid certain complexities. (58:17) >> Yes. >> Yeah. And we always have to navigate complexity with a certain threshold of accuracy or we have to sort of be and then that’s that’s that’s the human condition under the free energy principle. >> Yes. Absolutely. And and it all rests upon the relative um I repeat mathematically it’s precision or negative entropy. (58:42) Um it’s um um but you can think of it as sort of you know the credibility or the confidence per perhaps is to make this a bit concrete and more accessible. So Andy Clark tells a very good story you know one mechanics one sort of formalism of of of this sense making um and illusion building but the right kind of the best illusions you know inference to the best illusion um is um predictive coding. (59:08) And in predictive coding um one reads the self-information surprise all the negative model evidence as a prediction error. So you can just think of the brain generating predictions of what it would sense if it had got the right story at hand. If it if it had arrived at the best guess that completely or to a certain extent uh explains accurately this sensory input. (59:37) Now the how would you score the surprise that you want to minimize? Well, you just subtract the predictions from the sensations that is just the prediction error and then you use the prediction error to adjust your explanation for the world. Yeah. >> Until the prediction error is eliminated. (59:58) So if I can predict in the sense or not eliminate but certainly reduce so they can’t reduce it any further. So you could also describe this self-organization uh this self-evidencing under the free energy the basis mechanics as basically minimizing prediction error. But there’s a problem here that prediction error could come from many many different sources. (1:00:20) It could be a sort of visual prediction error. It could be an auditory prediction error. It could be from my own. It could be from my body. And furthermore, there are prediction errors and prediction errors and prediction errors in a sort of deep generative model with this hierarchical form. (1:00:34) So the there’s a problem here that I have to weight the prediction errors in order to give them a selective advantage in updating my beliefs technically basing belief updating. >> Yeah. So the way that Andy Clark explains this is that you know if I get some information some sensor information this is like news but what is the newsworthy part? >> Yeah. (1:01:01) >> Well it’s that which I didn’t predict. It is the prediction error which is newsworthy. >> Yeah. >> So we don’t watch the 10:00 news on the television or on our um mobile device to hear things that we could predict. Yeah. we only we basically seek out the prediction errors to resolve and update our beliefs. (1:01:20) >> Um so now you have to choose which of these newsworthy bits of information name the prediction errors that score the surprise that we’re trying to minimize um uh according to how um how precise you think they are. So basically you’re turning the volume up on certain sources of information. (1:01:43) you’re attending say to your visual modality or you’re choosing um CBN as your news source in America as opposed to Fox or your auditory modality. So you have to choose what to attend to effectively. So in psychologist it would be attention. Uh mathematically it’s this assigning the right precision. physiologically it’s um mediated by exactly those brain chemicals neurotransmitters that we change when we give psychiatric drugs for example or indeed take psychedelics. (1:02:18) So there’s a you know there’s a lovely convergence mechanistically from the sort of the psychology the basian mechanics and the neuropharmacology um and you know in terms of all focusing on who do we listen to um so a simple example of this would be you know if I go into a dark room say a hotel room that I’ve never been in before and the lights don’t work then I know that the visual prediction errors are not going to convey very much information. (1:02:49) They’re going to be ambiguous and imprecise. So, I’m going to downweight those because they’re not newsworthy. >> And I’m going to in a complimentary way increase the precision or attention afforded um palpation of sensory input or indeed I might listen very carefully if I can hear a clock or I can hear a radio. what what there’s so much here. (1:03:12) Um it’s still I think but we can also we will we will share links and help sort of people understand it because it’s I think it’s so so fascinating to understand all of this um in yeah one idea just this we we sometimes we are storytellers we’re making this story and it also has to do with meaning making so for instance um I was just hearing you talk I was thinking is the free energy principle in a sense also a mathematics of conspiracy theories and I hereby Just what I mean is that people have false beliefs about (1:03:44) a complex world and we don’t like complexity. We don’t like sort of the the people who are sort of doing stuff in the world we don’t like to be sort of people who intended it to be good and that there is not a single actor on which we can plot sort of all the evil in the world but we’re somehow part of it all. (1:04:02) That’s tremendously complex to just understand our our world and geopolitics. But we do like uh we but we want to be we want to have an accurate explanation somehow what’s happening from a meaning-making perspective and then you end up of course with with conspiracy ideas that are that don’t handle the complexity but are accurate in your world because you have now a story. (1:04:23) So it’s just that is an idea I had when hearing you talk. It it’s the same mechanism in a sense, right? >> Yeah, absolutely. But that that thought I think sort of takes us into a really interesting world cuz at the moment we’ve just been talking about the observer and the observed. But when the observed is herself an observer, we we now enter a really interesting uh um what is the mechanics of um um exchange between things that are sufficiently complicated and sufficiently evolved and sufficiently big literally in terms of their their scale (1:04:56) >> that they can both be regarded as making inferences or self-evidencing about each other. M >> so now my world is now constituted by things like me >> and that affords this really interesting opportunity. So if I’m in the game of self-evidencing gathering evidence or I should say from the point of view of the free energy principle it looks as if um so you know this is a story you can tell about the basic mechanics so it looks as if I’m in the game of gathering evidence for my model of my lived world or my sensed world. um then um I now am in the (1:05:36) game of inference the best explanation about you what causes you what best explain so it’s the simplest but accurate explanation of everything that you generate in terms of communication in terms of um all the cues that you offer me auditory and visual uh or on social media one way of doing that is to make the simplifying assumption that you are like me because if you’re like me and I can infer that you’re the same natural kind as me then I can explain your behavior in terms of the intentions that I had that would explain the same behavior in (1:06:18) me. >> So I now I’ve got sort of a ready baked theory of mind. >> But it only works if you’re like me. M. >> So all I have to do is to rifle around, identify things like me that share the same generative model, the same common ground, the same narrative, the same world view >> and then engage with those people because that renders the world much more predictable and surprise minimizing. (1:06:47) And if we all do it together, the argument is exactly symmetrical. So now we can minimize our joint free energy together just by converging on the same world model, the same conspiracy theory, the same theology, the same ideology, uh the same language. If we all converge upon the same one that becomes now the uh the joint mechanism or an expression of basic mechanics in this special situation that you’ve got lots of observers all observing all observing each other and you know >> fascinating but this this makes sort of (1:07:22) your your your theory. So we can draw the ma the mark of blanket on sort of a whole group in society that has a a way of of group think that that you can mathematically draw a boundary around yes that group >> and some I personally haven’t but I have young colleagues who have absolutely done that but but notice what you’ve just done though if that boundary is impervious >> you’ve now created a closed system >> yeah my apologies >> no problem >> you’ve created a closed system um so we’re now back to 20th century physics (1:07:54) and everything will converge on a on a an equilibrium uh which will be sort of the echo chamber and nothing will change. >> But of course that if it’s a Markoff blanket it’s porous. So now you get this sort of um interesting situation where there’s a tendency of the inroup within this sort of um meta markoff blanket that surrounds this particular group. (1:08:17) um um where there’s a tendency for everybody to converge upon the same narrative, the same world view, the same, you know, the same generative model. And yet they are trying to infer or look as if they’re trying to infer as a group the out group or the collections of outgroups in which you know with which the So there’s there’s necessarily a scale invariance or a scale freess. (1:08:46) is this kind of um group formation >> because if there wasn’t you would actually have a closed system that was isolated from the rest of the universe and that can’t happen. We’re always in a particular context. So they’re going to be groups of groups of groups of groups at increasing scales where within each one within each group everything is trying to if you like um reassure itself that they got the right world view but at the same time trying to accommodate the larger context in which they’re operating. Um so some really interesting (1:09:16) um I’m not sure there are any analytic solutions I mean mathematical analytic solutions um as opposed to philosophical um but you can certainly sort of search for numerical do numerical studies and just look at the you know the kind of um patterns of group formation that emerge under >> um sometimes described as epistemic communities. (1:09:40) community would be an ensemble of observers >> that are all trying to minimize their surprise through this epistemic foraging. Acting in a way to minimize prediction errors in the future by seeking information but doing so in a way that they um they’re imbuing certain people with an epistemic trust because just because those people are like them or those institutions are like them. (1:10:07) And then you get the emergence of um interesting dynamics. Um I’m not sure whether you want to use this piece but I I >> No, please continue. Yeah. >> The one one one sort of particular solution which which always catches my eye is when you simulate this kind of um active inference in an ensemble or population context. (1:10:33) is that one steady state or non-equilibrium steady state solution which can be likened to an evolutionary steady state or um stable state um is a 50/50 split and certainly in the past decade of me politics watching you see this time and time again that the only stable solution by which I mean if one group is too small it will be absorbed into the other group because everybody’s trying to seek the inroup and if everybody’s doing that there’ll just be one ingroup Um, so this 50/50 spit is the only stable solution. Um, where you’ve got (1:11:07) both the reassurance of being on your inroup, but you’ve also got the epistemic allure, the sort of novelty, the sensation seeking by looking out there to to the out groupoup. You can you can laugh at, you know, I won’t mention, you know, uh, you can laugh you can laugh at the others um or not laugh. So you see this time and time. (1:11:30) You remember Brexit in the UK? It was exactly 50 51 52 48 Trump versus Biden independence debates. Wherever you look, whether it’s Montreal or whether it’s Scotland, it’s always 50/50 cuz that’s the only stable solution in terms of the joint free energy minimizing process where there is a markoff blanket between the blues and the reds, whatever they may might mean. (1:11:57) So fascinating. The one question I have is um I’ve been sort of trying to get my head around integrated information theory which I like for the fact that exactly as yours theory it’s able to draw a boundary to just not be sort of loosely saying this is conscious and and and and to not have sort of just best guessing around consciousness where whereas IIT says no we can sort of have mathematically when we calculate the fire of a system we can see exactly for instance what neurons are involved with consciousness and and which aren’t which (1:12:27) I think is very good for science because then you can make accurate predictions. So I like that. But under integrated information also as in your theory it can uh scale up so to say if the the the fi is is of a bigger system then I would sort of cease if we keep we have an interconnection between our brains as Christoff Ko explains this to me right and it we keep on making connections to our neurons there will be a moment when that sort of collective integrated information will be higher than the ones now sitting here and we will become sort (1:12:56) of a group mind or that’s that’s what the theory points to or predicts under I that also involves that I somehow cease to exist as an as an conscious unit. Whereas I’m just curious how is that in your theory? If the mark of blankets gets bigger, can there be mark of blankets within the mark of blankets and there are there selves within a bigger self? >> Yes. (1:13:21) >> All right. >> Well, let’s let’s um dwell on I integrated information. >> I’m curious what your thoughts are on on IT. Well, I I should confess so Julia Tinelli was one of my best friends when I worked um with Jerry Adelman in America at the inception of uh IIT and related ideas. Um and Kristoff is also you know a value colleague. (1:13:42) So fantastic. So you I have a personal investment in in sort of um respecting IIT and I think I can easily respect it from the point of view of um a much simpler kind of maths. Well possibly not simpler but certainly from my point of view more classical understandable um uh kind of maths or a first principle account which I would contrast with the aximatic account of IIT. (1:14:13) So although I’ve heard Julius say it is a principled approach, it’s not. It’s an aximatic approach. It’s it starts off with axiatic assumptions and then builds from there. >> Yeah. >> Whereas a fi principle is a first principle account. It there are no assumptions. Everything is emergent from the starting point which is just there exists a random dynamical system. (1:14:33) But the end points of the analysis I think bear remarkable similarities. Just sorry to interrupt you but when you say you don’t have axioms wouldn’t your action be that there is that nature creates the first mark of blank of blanket or how do how to account for the fact that mark of blankets exists >> well how the free energy principle is not an explanation for why markoff blankets exist. (1:14:59) The free energy principle is a description >> what they do >> of >> um of things that are are defined in terms of markoff blankets. So this is um >> but I would then say that’s your aim that those things exist that have mark of blanket. Isn’t that then your axiom? >> No, not really. No. Um I think the reason I I hesitate to agree um I usually like to agree but in this is I know I think I I think there’s there’s a lot of heavy lifting to be done um beyond the free energy principle or perhaps within the free energy principle to explain why it is the case (1:15:36) that Markoff blankets do emerge. So why did this single cell first emerge from the primordial soup? >> Of course yeah we don’t biology cannot tell us. >> Yeah. I mean this is exactly the kind of sort of question that everybody’s trying to answer including people like Mike Leven from a biological perspective through to people like um um Steven Wolfrum you know you know um this is an outstanding question and I think that’s a notion of the emergence of a Markoff blanket >> um in within the primordial soup and then the emergence of Markoff blankets (1:16:09) and markoff blankets where you get these multisellular organizations that’s an important concept because you need that multisellular blankets of blankets of blankets to have these deep hierarchal structures in order to have the kind of sense making and and the basic mechanics that we were talking about uh previously. (1:16:30) Um you can certainly apply the free principle to a single cell or a thermostat but it’s a little bit trivial. um it becomes much more interesting when we talk about big things that have this sort of um scale invariant structure with nested markoff blankets and of course we’re talking about populations of interaction inter interactions in other minds which is the sort of highest level of this sort of um um sort of >> inter interactive inference or inter subjectivity um or um measurements of others um yeah so IIT um does you know (1:17:04) So um so yeah so the free energy principle does not say it is the case that a markoff blanket will emerge from a primordial soup. What it says is should something exist it has to behave like this. >> Okay. Yeah. >> And you know you may ask is that useful? um not in and of itself because you could say um um natural selection, you know, evolutionary theory, um is a very powerful theory, but it doesn’t tell you at all why I speak English or why you have two eyes, >> you know. (1:17:47) So um in itself the free engine principle doesn’t really get gets you as far as a law of least action in Newtonian mechanics. Um so um you have to apply it. So the free energy principle only becomes useful when you apply it and when you apply you get bas mechanics active inference and everything >> but sort of that that notion of selfhood sort of being inside a blank in a a mark of blanket modeling the world um it can scale up to larger ensemblas with those individual market blankets still existing. (1:18:19) So we consist of trillions of cells those have marker blankets and there’s a mark of blanket that is Hans and they exist simultaneously and I think I I think that’s interesting and it’s also in line with Michael Leven’s work also sees these these things as scaling whereas IT uh looks at the integration of all of that and then sees a single unit. (1:18:38) So there that you do defer from IIT. >> I see right okay yeah good point. Um so the point of contact I was going to um foreground was your the information partitions that you need to define fi. >> Um and in a sense you know it is not surprising that that calculus is required in order to elaborate an information theoretic account of um consciousness. (1:19:09) um uh in the same way that you you use you the starting point in the free engine principle is the markoff blanket a partition into observed versus um unobserved. For me though uh where IIT ceases to become um interesting um is that there is the the the there is no distinction between the observed and the observer. Um, so the the information that IIIT talks about is not information about the thing on the other side of the blanket or the partition. (1:19:44) It’s just the information that a thermodynamicist or a brain imager would would measure. It’s the it’s the entropy or the thermodynamics of that partition in and of itself. Whereas the fian principle the information that you’re talking about that is represented on the inside is about the outside that is the essence of observation is of essence of measurement um so there are distinctions um in relation to your comment your >> but this is very interesting point I’m trying to uh understand this I think it the the the key here is also to to make (1:20:24) clear for everyone watching that, correct me if I’m wrong, that that the free energy principle is not a theory of consciousness per se, right? It’s a theory of of of how um agents or selves have to model the world. So, to me, it’s much more it’s about cognition and intelligence. Um because the the the phenomenal consciousness the what it is like to beness of something what it is like to be inside a mark of blanket which is sort of the feeling and the qualia what how how it’s called and and integrated information is about that (1:21:01) starts with phenomenal consciousness and that’s a different departure point I’d say. >> Yes. So, but I’m curious to to if you would agree if when I say your theory is not a theory of consciousness and then also of course how how it does or does not relate to consciousness. Is it a theory that cannot say anything meaningful about consciousness that that because does not cover consciousness or does it relate to phenomenal consciousness? I’m super curious what you what your position is here. (1:21:26) >> Right. Um you’re absolutely right. The FEP well first of all the FE is not a theory. It’s a principle. Sorry. Um and that’s technically important because in physics a principle you have to apply it. So you don’t falsify it or marvel at it or shout at it. You just apply it or you don’t. (1:21:47) Uh whereas IIIT is a theory which may or may not be wrong and it’s you know um it’s subject it’s a hypothesis is not a hypothesis. It’s just a mathematical tool that you would apply. >> Um and as such it certainly is not a theory of uh consciousness. Um is it a principle that could be applied to consciousness research? Possibly. >> Uh and people try and people are trying it. (1:22:10) So interestingly um >> I’m currently engaged in the Dneumont the last year of a Templeton um um funded adversary research um uh collaboration pitching active inference against IIT. So Julio’s a theory lead on the IT and I’m um the theory lead on the active inference side. Great. Um and in order to do that because you know one has to be adversarial um you don’t you know I’m not sure that’s the right way to do it but that was our brief. (1:22:42) Um so we had to pick something that sort of um could be seen as an application of the free energy principle namely active inference. um that dis uh disambiguated it or or made it distinct from in an adversarial sense IIT and the the thing that was eventually emerged was action. >> So um it was um possibly um best summarized actually somewhat in relation to what I was trying to um get across when talking about the difference between Helm Holtz and Kant. Um that to see is to look. (1:23:19) >> Yeah. You know to hear is to listen. There is no passive inside out registration quality of experience. The quality of experience and the phmonology rests in this application of the free energy principle >> which may the application may or may not be you know a correct theory. Um then um you have to um be an agent. (1:23:49) You have to be an agent that is actively soliciting this kind of information from the sensorium. And it doesn’t have to be on the outside. You can do that on the inside through this attention-like mechanism we’re talking about before. Remember, we’re talking about assigning precision or a newsworthiness to certain sources of prediction errors. (1:24:12) This is a very inactive um as um you know capability which not everything has. So a thermostat wouldn’t have this. You could argue that to a certain extent even single cellled organisms would have although you could argue that catalytic interactions and enzymes do this in a in a certain way. But you know in terms of consciousness research of the kind that we’re talking about qualitative experience um then what that suggests is that um you have to have a natural kind with a deep generative model that can see that that can act on itself. So deep (1:24:48) layers of or levels of a generative model can make sense of the computations of layers that intervene between the deepest and the uh the sensory level or or the sensorium. What would that look like? Well, it would just look like exactly what was saying before in terms of um turning the knobs by assigning greater or lesser precision to various um belief updating dynamics um various sense making procedures say like predictive coding. (1:25:19) So this would be like action on the inside um it and you know the very way that psychologists talk about this this capacity in terms of attentional selection and notice selection is doing something. So we’ve gone beyond transcendental idealism in terms of sensemaking and illusions. We’re now in the world of you are in charge of your sense making. (1:25:42) So we got this this sort of you know all the the difference between seeing and looking is that look is a verb. Um so but this also applies to internal mechanisms internal actions. The reason I’m and this is what basically um this adversarial reason. >> Yeah. New York is more about doing versus it is more about being. (1:26:02) >> Yes. >> Yeah. But that’s a nice distinction, a very clear distinction to make. I’m curious what happens when sort of um we assume our behavior not to be sort of in in a minimal state of doing closing my eyes, sensory deprivation, these sort of experiments, right? Or for instance, psychedelics. (1:26:21) So you lie down and okay, very very often there’s music involved, but say you don’t do that. and the the experiences people then have what what what can the free energy principle say about these experiments because it seems to be uh so that’s self action you would call that action upon yourself because the model then doesn’t have input so then but the model has to keep going on sort of it keeps on reducing free energy but there’s not anything coming in so what what’s happening >> right so there are two great answers to that because I know we can also talk (1:26:49) about psychedelics and >> yeah of course that’s nice to talk about >> but the just but the So um if there’s nothing coming in, how can you minimize free energy? >> Yeah. >> Now the answer will be about 20 minutes ago when we said that evidence is equal to accuracy minus complexity. >> Mhm. >> So in the absence of any thing coming in, in the absence of any sensory information, say for example when we’re asleep, >> Yeah. (1:27:17) and we shut down all our sensory organs and indeed our aector organs then there can be no accuracy because accuracy is an attribute of the sensory evidence or the sensory input but there’s still the complexity. >> So that means that we can um with gay abandon reduce the complexity of our generative models while sleeping. put that another way, in fact inverted perhaps the fact that we sleep and the fact that there is so much selective in an evolutionary sense selective pressure for sleeping. (1:27:44) >> Yeah. >> Is that this affords the opportunity to minimize the complexity of our models again coming back to Einstein keep everything as simple as possible but no simpler. So to get the right balance between accuracy and complexity requires us to discard overly complicated, overly complex, overly expressive or over parameterized generative models while we sleep. (1:28:10) So we’re now fit for purpose the next day and we don’t start to overfitit our data. I should just say that just intuitively, why is it bad not to minimize complexity or why is it good to keep things as simple as possible? Um if you’re if you just think about fitting data um say um you know in a statistics sense or indeed a machine learning sense if your model is too complex it’s got too many degrees of freedom coming back to the law of requisite variety. (1:28:39) If it’s got too many degrees of freedom you’re going to fit noise. This is known as overfitting and it’s devastating you practically when you’re building uh transformer architectures for example or uh you know um any kind of generative AI >> then if you don’t try to minimize your complexity in the right way you will overfit and you’ll never generalize it just won’t work. (1:29:00) >> Yeah. So this complexity minimization is really very really very important and that’s probably why we take ourselves offline just to rehearse and one could argue in fact people do that dreaming is this kind of rehearsal the fundaments of the causal cause effect structure that our generative models entail uh such that we can remove the redundant overly expressive parts of it literally removing the signapses the associations in our brain that embody those associations I mentioned that discuss also in a completely different strand of (1:29:35) work Julio Tenoni has a synaptic homeostasis hypothesis which fits beautifully uh you know with this view >> so that’s you know there’s still a lot of measurement and inference and self self-evidencing that can go on in the absence of sensory data um and just think about the way we work we have to take our ourselves offline >> yeah we have moments of introspection >> and and if I may but this is very speculative and I don’t sort of want to want to sort of you don’t have to commit to any sort of weird esoteric uh (1:30:06) phenomena that I want to sort of just share with you but I’m only curious if the free energy principle would in principle be permissive to the idea and I’m talking about sort of just the phenomena uh for psychedelic states where people do have to I’ve had that myself that you it’s interesting I do think I had a sort of a very high degree of complexity with low accuracy but a complexity I have never seen before like like Van Go painting be pure beauty but but not accurate um astonishing to experience. Um other (1:30:39) people sense that they can sort of connect to a a a mind at large or or this a spirit world spirit entities whether we grant any of those sort of an antic status of being real. What interests me is when we think in terms of a marovian blanket, um we can think of it sort of in a very physical sense, my my my brain, my sensory input. (1:31:02) But could it be that sort of that this whole free energy way of thinking could also apply here that somehow a a boundary gets more porous and and and we process the world differently and it might even tell us something uh num numinologically. So what truly exists out there? What what are your thoughts here without sort of asking of you to commit to any of these? >> Well, I’m very happy to commit. (1:31:30) I mean there are there are people you know asking themselves exactly that question. People like Robin Kart Harris. Yeah. and his colleagues and also to a certain extent those um philosophers um I’m thinking here people like Thomas Metsinger who make it who have you know a certain take on quality of experience in terms of the phenomenal phenomenological aspects in terms of transparency versus opacity. (1:31:58) uh um you can I think link all these all these notions together um using exactly what we were talking about before in terms of the your false modeling in um say people who were subject to hallucinations and delusions >> um it’s the balance of the precision the credence afforded to the sensorium relative to your prior beliefs. (1:32:19) So imagine what it would be like now um to induce in yourself a kind of autistic state by taking a drug that chemically dissolves the precision of your pry beliefs thereby relatively enhancing the precision of all your sensory inputs particularly the visual input. Okay. Check. So it it so I’m reducing the precision of my model. (1:32:45) Um which makes that new input has to be sort of needs more attention because I cannot assume that it’s right. So I have to look closer. Is this microphone indeed? What? Yeah. So you’re resetting that. >> Well, it ceases to become a microphone because that your microphone is a really high level deep illusion. (1:33:03) It’s it’s like matter. you know it’s use a useful story you can tell yourself and indeed I telling myself that it is a microphone it has all sorts of functionality but if you’re if you’ve dissolved this kind of your narrative in your head so when I talk about deep or prior I I have in mind your centrifugal hierarchy with increasing abstraction and um scale uh the the representations um as you ascend into the higher hierarchy get deeper and deeper. (1:33:35) So things become sort of much more elemental and short-lived as you get to to the periphery because the deeper layers are assimilating gathering together finding simple stories that explain a myriad uh trying to compress things into a microphone but this is not a microphone. I like that the fact that you’re saying the word, sorry to interrupt, but it’s just it’s f it fires me up that you use the word uh compress because also an analogy I thought of when uh for this in preparation of this conversation is sort of the zip file in (1:34:05) computer. People know we can zip images >> and so microphone is an extremely efficient zip file in my head that tells me this is a microphone. >> Yes. Yeah. >> But if you would unzip it, there’s no microphone or it’s something completely >> precisely. Yeah, that’s a a great point. There’s so many ways we can take this, but let’s just pursue that because that I think that’s a really important and fundamental observation and interestingly um goes right back to the inception of predictive coding. (1:34:32) So if you remember we were talking before about predictive coding being one implementation of the free energy principle under certain assumptions about the kind of generative models that um are being used. Predictive coding started in the 1950s as a way of compressing sound files. >> So there’s a deep mathematical link between compression and complexity minimization and free energy minimization really really deep. (1:34:58) And that um there’s a parallel story that comes uh via um salom induction from commity >> uh and minimum description lengths and minimum message lengths right through to universal computation. So it’s exactly the same story >> that the best inference to the best explanation or the most efficient way of compressing or communicating involves minimizing redundancy. (1:35:27) Um that is basically compressing something into its simplest form but not too compressed. So again keep everything as simple as possible no simpler. Um, and it’s exactly that balance between the complexity and the accuracy and this, you know, so sound file compression. And indeed, I had an email exchange with Mike Leaven a few weeks ago where he was going to use um the capacity to zip compress his files in order to measure the complexity of the representation. (1:35:53) It’s a really powerful uh way of measuring the complexity part of of the variational free energy. Um so in the context of um the you compressing this all the information generated probably visually by your microphone into the version of a microphone. So that compression is no longer available to you because you’ve dissolved by reducing the precision through neurochemical means through taking say psilocybin. Yeah. (1:36:24) which is a 5HT1A node a particular uh receptor partial agonist or agonist um that basically resolves the precision the excitability uh the the um uh the credence afforded to these highlevel deep iconic kinds of representation which basically frees up and emphasizes the more elemental one. So this is not a microphone anymore. (1:36:50) It’s now an ensemble of objects and each object itself has some remarkable features and there is uh you know a flat surface here that’s reflecting some curious ad mixture of red and and finer and finer detail. So you don’t perceive the microphone, you perceive the elemental constructs that you would otherwise assemble into a microphone. (1:37:11) >> Yeah. >> And I think that’s when you you saw Van Go dissolving into elemental. It must have been and it’s it’s exactly what you say on lower doses when stuff don’t when it does not completely dissolve to me has been sort of the more pleasant experience because I can sometimes notice that I have a rigid way of thinking and then low lower doses won’t sort of get you through van go but it will sort of you uh let you reappreciate sort of the environment around you because in thinking about this a bit more normatively and I’m curious how (1:37:42) this worked for you also on a personal level um what is sort of that ideal sort of uh uh uh uh model making um not to be too rigid and be porous enough that level of being porous. what’s what’s the the the the ideal level if we can even talk about sort of what ideal is because it’s normative and we’re talking about just what but I’m curious here >> well I mean it’s a really important question when it comes to simulating this indeed using the free energy principle in application to simulate phenology for example there is a field (1:38:12) of computational phenomenology or neuroimmonology um but the answer to your question is that the balance of the precision that we’ve been talking about the balance between the accuracy and the complexity or the prize and the sensory the likelihood part of our generative model itself has to minimize free energy itself has to minimize um surprisal or maximize model evidence. (1:38:36) Um so it depends very much upon the um the quality of the data that you are able to solicit from the environment. So I’ll have a very different kind of precision waiting or um configurational reliance upon prior beliefs versus um sampling the environment if I’m in an environment that is unpredictable and imprecise. >> We could come back to the dark hotel room in which I have never been before. (1:39:02) But you can imagine yourself in a dark city, a city outside walking on the on the pavement uh in the dark in a city where you’ve never been before. You’ve heard some awful stories about things. Um there, you know, there there are drug dealers near the station and you don’t know whether you’re near the station or not. (1:39:23) So this world um is imprecise from your point of view. >> Yeah. >> So um the normal deep abstracted microphone stories you tell yourself do not apply. So now you’re going to attend to the details. You’re going to be deting. you’re going to be on guard, vigilant, attending to all the elemental information that you can and try and make as much sense. (1:39:43) So, it’s going to be very very context dependent. But there is a there is an optimal solution if you allow optimality to apply to the free energy bound on um your on just the evidence or the likelihood that you’re in this state given the kind of thing that you are. But you can interfere with that balance by ingesting certain drugs and you could you’d say that you know certainly that’s an interesting um question to ask when it comes to addictive pain but psychedelics are not addictive in that sense. (1:40:13) So, but they do have this remarkable capacity to rebalance usually in the auditory and primarily the visual modality um this this sort of balance between these deeply held convictions and prior constructs that we bring to the table to infer and summarize and compress our understanding of the world. But what I want to ask you now, >> yeah, >> what’s the deepest most abstract microphone like construct that we use to explain our lived world? >> So the deepest like our whole world, the deepest model of our world. (1:40:50) >> Yeah. And what’s right at the top? What’s right at the center of these models? It’s a trick question, but I but I know exactly what the answer that I want. >> Okay. I’m I’m going to give I’m probably not understanding fully correctly, but let’s say that my if my microphone reduces sort of the the true complexity in the world to sort of a single thing I can grasp and think it must be I’d say my best guess on a universal level would be something like God or the mind at large a single uh uh cause of everything (1:41:24) that sort of that I can then sort of if I unzip it I get sort of you know I get to this world that that would be I think my my answer to your question, >> right? So implicit in that um that was interesting. Um so in an a Freudian oceanic sense, you gave me the example the the answer that I was groping for which is that the the story uh that you are a self. (1:41:49) The story that you are you >> Yeah. >> So you are you and you’re part of some cosmos and part of you. So you use the word I about four times in answering that question. Okay. You’re you’re still a psychiatrist listening carefully. >> I am indeed. Um so the reason I I suggest that so that you are you are part of something but implicit in you are part of some uh you know uh something uh of this um um omnipotent or oceanic or unitary aspect um presupposes that you are a thing. (1:42:27) So the notion that you are you the the minimal selfhood or in fact the selfhood itself is just another story under the free energy principle. >> So if you dissolve that what would happen now there is no self. >> Yeah. >> So now you have an explanation for non-dual states. >> Yeah. >> So that sort of feeling that pe you know if you’re a very experienced meditator you can actually aspire to with meditative practice. (1:42:56) But also sometimes the depersonalization which is actually you know um in my experience quite terrifying if you take too much of of these certainly psychedelics um the notion that you are no longer you or there isn’t a you um um c can be revealed through dissolving these higher level um or the precision afforded these very very high level explanations. (1:43:25) ‘s best guesses about the you know the causes of a sensorium which I’m assuming is my sensorium. I’m assuming this is my body and it is me that’s doing the experiencing. >> So but that that can go away. Uh you know um we don’t often experience that clearly. >> Have you if I may ask have you had >> I’ve had a similar but not quite as distressing as I’ve seen in in in psychiatric patients who have de you know episodes of depersonalization. (1:43:52) Um, on the other hand, I also, you know, this is not necessarily a bad thing. You don’t have to suffer. It doesn’t have to be frightening. Um, or you could, you know, again, you have to ask yourself, how can you be frightened if if you’re if you’re not you? Um, >> and there’s no one to be frightened. >> Precisely. (1:44:09) Um, so, uh, but I’m really sort of speaking to this from your in response to your question about consciousness. you know, if you’re talking about selfawareness um and um self as part of your generative model, then I think you can usefully apply the free energy principle to ask questions. Well, what does that mean? You know, what kind of generative model would you need in order to have a story a m a microphone story? So the microphone the concept of a microphone is just a story you use to explain this particular pattern of sensory inputs in (1:44:44) a really useful way that uh immediately brings to the table all the affordances that are associated with a microphone. um the same argument though is me you know that I am just you know another explanation like the microphone which I learned and brought to the table and I didn’t arrive with it you know when I was born I didn’t have any notion of me and you know I didn’t I didn’t have any notion of mother from my point of view this was just some oceanic universe uh where you know I was nourished or um I learned that you know my needs were (1:45:18) being met and it took me you months, well, days, months, weeks to work out that mother was separate from me and that mother had intentions. >> Yeah. >> That weren’t always a direct response of what I could sense myself doing. Well, where did myself come from? Well, if mother is a thing in her own right, then perhaps I am a thing in in my own right. (1:45:42) And slowly you start to develop a sense of self which you don’t necessarily always attain. So if you got severe autism, you might never actually get >> that theory of mind that rests upon um this this aspect of self-modeling. But you can revert back to these and some people like to do this, you know, and certainly people like Thomas Minger think that things like um the nondual aspect of minimally experiences are are hold the key to an understanding of consciousness. (1:46:15) If you can get strip consciousness right back so there isn’t even a self there. >> Yeah. Yeah, >> then that might tell you quite a lot about the nature of consciousness and you know >> and would you agree that that what then remains if if all models are gone my experience has has been sort of in in some of these states that a a lucid awareness let’s use the word awareness here remains that is actually exactly the same awareness I have right now but it just lacks all these models of selfhood which was like profoundly strange to experience but it helped me (1:46:46) to distinguish between that awareness and sort of the contents of that awareness the the the qualia that keep changing and but to see that as a sort of a constant then thinking back in my life only in very infant stage I cannot cannot go back to that state of awareness remembering the state of awareness as a baby I cannot go back so far but the as far as I can go back that awareness has more or less been the constant or the same which makes me that makes me an idealist by saying, “Okay, I I suppose that is fundamental.” And your (1:47:20) theory to me is a brilliant mathematical way to to describe the the the the doings that happen within consciousness, the the mechanisms at play in in these models that model the the contents of that the contents of perception that that present themselves to my awareness. What What are your thoughts here? >> Would you agree to what I just just described as awareness of being a ground so to speak? >> Yes. (1:47:48) Yeah, I think you’re very lucky to be able to experience that awareness without the content. I have >> It’s not a daily thing. It’s a heart. Yeah. And >> and for what I know, I think people aspire to that through years and years of meditative practice to try and induce that. So you >> it has been on substances mainly. >> Oh, I see. (1:48:05) You should tell you should tell my meditator colleagues. They probably know. >> It’s the shortcut. It’s the the Yeah. But but please continue. Yeah. >> Yeah. Uh so yeah. Um, and you know, I I was um struck by your I can’t get back that far you to to those um I keep using the sort of um and I’m sure I’m using it wrongly sort of oceanic Freudian sort of you know when you first arrive into the world when you’re not even you um >> well that makes perfect sense because remember from the from a hellhium point of view unconscious inference point of (1:48:39) view if you don’t have the machinery you’re not ever going to be able to remember you can’t perceive it. So just be able to perceive the world as um being um something that is um con you are constituted of uh you know is a remarkable capacity which I repeat you know may may take a long time uh to develop um and then you were talking about the the sort of the the continuity I think that’s really interesting >> that is sort of more or less the same thing that sort of as if the the the um say the model is of the generative model (1:49:13) is sort of just giving me a a screen perception. It’s given me a dashboard and and we we’ve been talking now a long time about sort of how that works and it’s it’s absolutely fascinating. But there still seems to be a a a and I call that let’s call that this awareness that perceives that that screen of perception that has always been the same and is sort of will always be the same. (1:49:33) That’s that’s how I see it. >> Yes. Yes. Mark out as a very high level life form I would imagine. >> Yeah. But but um Yeah. But it’s also something you can more or less um yeah this is what non-dual teachers of course teach in sort of like that you can arrive at that analytically. I say this mainly analytically. (1:49:53) It’s not like something I but if I but in thinking about I I I am able to experience it more often. But I’m curious here because we’re touching on something that I think many people might misperceive your theory because you could say that um let’s frame it like in a totally different way. Would you say that the free energy principle is matter doing all of this and then it produces somehow phenomenal consciousness? Because that would be like the physicalist story. (1:50:24) Then then that would be like the physicalist framing of your theory. Just I start with the big bang and I have I’ve just all the physics we know and the free energy principle then starts just describing what cells do. But it still cannot solve the hard problem of consciousness and somehow consciousness emerge. (1:50:44) But that doesn’t seem to me what the free energy principle is about. >> No, I think you’re right. Again, you you’d have to think very carefully about how you would apply the free energy principle to the hard problem. Um and you can do that and and and we can talk about the ways of doing that. Um that that sort of notion um that it is me watching a screen that is unfolding um I think it’s really interesting. (1:51:08) um certainly from a purely um technical non-filosophical u position um that um speaks to something which I think may be peculiar to things that um can be attributed consciousness and I’m denying pansychism here as you know a plausible story um and that is agency um and if you just ask what it is to be an agent It is something that has a natural kind that is sufficiently large that it can now entertain a generative model of the consequences of its actions. (1:51:51) And I just slip in sufficiently large because of course we know our actions if our actions are part of our markoff blanket. But that’s not the case if I’ve got a deep generative model and some you know very deep markoff blanket right in the middle is a long long way away from the actual muscles um that I’m using to act. (1:52:09) >> So now the action becomes technically a random variable >> which means it has to be inferred. >> So now I’m in the game of planning my actions. I’m now an intentional agent. I’m an agent that has intentions because I’m now rolling out into the future to evaluate the consequences of my actions under my generative model. (1:52:34) So what does that mean? Well, it means something quite profound. It means that you’ve got a generative model that is freed from the species present or the or the current moment. >> It means you’ve got a model that is future pointing. So it’s no longer unlike a virus or a thermostat tied to the immediate. Now you’ve got representations of yourself in the future. (1:52:59) And of course if you’ve got representation of yourself in the future, you’ve also got a representation of yourself in the past. You as agent and it’s only your agency. It’s not my agency or something else’s agent. It’s me, my future. And it goes it’s a path into the future and therefore a path that comes from the past. So mathematically if you’re this natural kind then you must have some basian mechanics that speaks to this continuity over time and that equips you with the notion of an invariance over time there is a past there is a future not everything can (1:53:36) have that so you know it is quite quite so I just wanted to pick up on this quite important >> but you’re also sort of deriving sort of the flow of time here and and and the the mystery of time from the free energy bin which I think is is fascinating. >> Yeah. Well, you should speak to Chris Fields about this who insists that any plausible um explanation has to be background free and what he means is if you start app or Donald Hoffman is also telling us very sim similar story. (1:54:01) >> Um so and I would completely concur with these positions you know space and time as we talk about them and read about them at school or in physics books are just stories. They are they are just microphones. They’re just you know uh things we bring to the table that provide a simple explanation for our lived world which would go away if things moved around in relation to each other close to the speed of light that the the notion of time and space that we currently have would dissolve. (1:54:31) It’s a good it’s a good story for our lived world um not necessarily all lived worlds. >> Um >> so yeah uh is time part of that? I think it is. And you know I think we come back to where we started which is relational view of time. How do you measure time? Well, how do you observe time? And the only way you can observe time is by um the >> belief updating of something else that is an observer. (1:54:59) So how quickly technically their information rate your um the movement on a statistical manifold per unit of clock time or how many how many different belief configurations do do they go through for every one belief update that you do. And then you’ve got this notion now that there are going to be all sorts of time perceptions in a deep generative model because every level is looking down the level below thinking, “Oh, you’ve changed your mind three times, you know, before I changed my mind once. (1:55:27) ” And then >> this is a species presence. Yeah, indeed. It’s fascinating. Yeah. >> So that was Yeah. Almost time, but there was something um I got >> Sorry to interrupt you, which is >> Yeah. Well, we, you know, we keep missing all these wonderful rabbit holes cuz we keep going down. >> You were talking about Donald Hoffman and that you you >> There is no Yeah, there is no there is no spaceime or background free explanations. It has to be the case. (1:55:51) Um, I would certainly agree with that. We’re we’re talking about agency um and agency necessarily requiring a generative model that um that had this sort of temporal persistence aspect. >> But there’s an interesting tortology here because remember at the beginning I said that the generative model is an explanation for things that possess characteristic states and those characteristic states define a pullback attractor. (1:56:23) But the pullback attractor is just defined in terms of things that endure during time. So it could if you are modeling your if you are modeling yourself >> as something that persists in time you couldn’t do it any other way. So to put it another way if on the view of the could regulator theorem you come to model you as part of your controllable world because now in this hierarchal context there are bits of you that can be controlled. (1:56:52) um then by definition you have to have a model of yourself that persists in time because that is the kind of thing you are by definition to exist. So there’s another deep tortology here. It couldn’t be any other way. Uh so that sort of sense of persistence over time is just part of the story. >> Yeah. (1:57:12) What fascinates me that if if we take phenomenology so my experience of sitting here my experience of seeing the world as sort of the only given that I can be sure of that that that that is that the the experience me experience that is whether uh it’s an illusion or whether that still is an experience so that’s that’s me sort of like a ground axum of existence so to speak and we know from first principles your theory that we do not uh see the world as it is. (1:57:43) We cannot see the world as it is. The combination of those two ideas uh at least seems to false falsify or just give materialism a very hard time because how can I sort of say anything about that outside world and I just told you that I that my first aim is that I experienced the world. I do not matter is already already a a a zip file. (1:58:10) The idea it’s it’s it’s a model already. Um and in the whole predictive coding and neuroscience, it seems to be that there are many people uh who who seem to miss this point that if predictive coding and active inference if you apply it to itself, you also have to say that itself is a model. So the model is a model. (1:58:33) So I’m trapped in this model making and the only thing I’m sure of is that that there is is is a me experiencing all of this. Well, that in philosophy is called sort of idealism. But this I think is a cultural baggage we have of like a couple hundred years of science and physicalism that people still uh think that this is compatible with materialism. (1:58:57) But I’m so so curious on your thoughts because I do think you are also perceived as a materialist by by many many people. if I if I’m not mistaken. But um >> well, you would probably be in a better position to know than I would. >> Yeah, at least that’s sort of the perception. If I’m just doing my research online, people perceive it as a sort of a a materialist way of of looking at the world. (1:59:18) I guess because you start with mathematics and just and as if you you reduce the human to sort of mathematics and and and I think people associate that with materialism. >> Yes. Although it is a particular kind of mathematics, it’s it’s you know a mechanics of belief updating um which is not um a component of things like quantum mechanics or classical mechanics or indeed integrated information theory. (1:59:47) It’s a very particular kind of um mathematics that lives in an information geometry where this geometry is beliefs about something else. So we come right back to the original distinction between the numinal and the phenomenal. >> Yeah. >> Um so you know there’s something unique about the free principle which you won’t find anywhere else. (2:00:11) But in I’m tempted to ask you to tell me what materialist is, but I I I’ll pretend I’m not naive and assume it’s the um position that there exists some physical world out there um >> that can be fully captured in quantities that we can measure and and that can be fully captured in in quantities. So the the whereas our experience of the world is qualitative, right? We do not uh um it’s it’s this philos I mean the yeah maybe that’s a nice one you might have heard of that that I think it’s Mary in a black and white room so say I I I’m in (2:00:44) a black and Mary’s in a black and white room and she literally knows learns everything there is to learn about color the the the wavelengths the the everything uh of all colors if she then exits the the black and white room having lived there all her life into a world of color will she learn something new about the world? Materialism would basically say no because everything there is we can fully measure in in these quantities. (2:01:14) But of course intuitively many people say of course because the experience of of of color the qualitative experience of color is something completely different than the third person perspection the description of color. Um, and and if I say that’s my definition of of materialism, which I think in philosophy is sort of the the the definition of it, then I think it’s just a naive position considering um my my my experience of the world combined with what science is now telling me. (2:01:45) So active inference but also quantum mechanics that we cannot sort of draw this distinction between the quantities I observe and me and my observerhood. >> Right? Do you you still follow because I’m also I’m not a philosopher. This I’m I’m just very deep deeply researching all of this. >> Yes. >> Yeah. >> Um well that makes you a philosopher or at least a natural scientist. (2:02:06) >> Thank you. Thank you. >> So that no I’m certainly not a materialist. Um Mary in that instance would certainly not perceive color. Um she should could certainly learn how to perceive color if she was sufficiently young and neuroplastic. So coming back to Helmholds, you cannot see, experience anything unless it’s already on the inside. (2:02:27) But how does it get on the inside? You have to engage with the world across your Markoff blanket. And it take years uh before Mary saw color, which has got nothing to do with her epistemic science book reading of the nature of color. Um and then your statement that the world is fully observable. Measurable. There is a metaphysical world that can be fully measured. (2:02:49) >> Yeah. in quantities that we can manipulate and and that physics describes >> then I am not a materialist. No, the whole point of the free energy principle is that there cannot be a fully um measurable world. If there was you would be part of that world and by the principle of unitarity you would not exist as distinct from that world. (2:03:08) So uh if the fullness aspect of materialism um is center stage then I’m certainly not not a materialist. On the other hand, um the maths that you would use to derive the free energy principle, which could actually be thought of as a uh or you could apply to materialism in and of itself if you wanted to, um does um rest upon the notion that there does exist um something that is not self. (2:03:40) That’s the the observed does exist. Yeah. And and you can actually sort of um analytically, mathematically and also numerically um emulate um self-evidencing and sense making in a way without even talking about the basic mechanics. Um and in that setting um which is effectively tedology free um you what one arrives at is something called synchronization of generalized synchronization or synchronization of chaos between the inside and the outside. (2:04:17) So that you can from a purely dynamical systems perspective say that the free energy minimizing solution just is when we’ve established um a mutual entrainment in a generalized synchrony I repeat synchronization of chaos because we’re dealing with random dynamical systems >> um that are certainly for life forms uh chaotic in virtue of the um the um technically the solenoidal flow or the the bio rhythms and the and the you cycles of life cycles, reproduction and and the like. (2:04:51) Um so the you know the the the the um if materialism just means that there is some cause effect structure of a probabilistic sort that is not uniquely owned by the observer or the self that I am a materialist because I have to have a mathematical image of the observed the outside the world. Um so with that >> that’s more like soypism I even say for in I mean idism would not deny an objective reality out there. (2:05:25) It’s just not physical in the sense of only consisting of >> quantities that we can measure and and and and thereby having fully captured reality by the measuring of those quantities. It wouldn’t deny because under idealism >> and idealist maybe nice to just bridge the sort of the what people better understand would be like of course the ancient traditions that would start off in eastern non-dualism with a one that is uh not matter but more or not more but it’s it’s it’s mind it’s sort of the the mind or the nothingness I mean it (2:06:04) can also be nothingness and out of which um different selves emerge, right? and on on all sorts of different levels from gods to humans. And then in in such a world tour cosmology um to me is interesting not for literally believing those stories but it gives me a model of understanding our universe as a oneness that somehow by dissociating itself or or creating blankets. (2:06:35) I don’t know it I I I don’t know how but it it somehow seems to be doing that and why does it do that sort of the whole game of it all is this model making and that there in lies the beauty but now I’m I’m leaving science and and venturing into s even more like a poetry of it all in a sense I’m I’m just curious how that works for you also because I yeah there’s actually a nice bridge I read um you on William James that that you um when William James of course wrote like more poetically in a sense about consciousness and when he (2:07:14) said uh I quote you just funny to reflect with you on it um when I hear attention is taking possession by the ma the mind in clear and vivid form so this is sort of like not accurate description of what what happens um when talking about attention or awareness and and you then say I think no it’s not attention is simply the process of optimizing precision during hierarchical inference. (2:07:40) So I like that because that’s sort of the mathematician that’s the mathematician in you wanting to be be accurate um in relation to something to to William James here. But I’m curious about sort of the more poetic call Fristristen when when pondering um looking outside. Um what are your more poetic notions of sort of the free energy principle and and the meaning of life even? >> Um sorry for my lengthy questions by the way. (2:08:11) I just >> I mean one answer to that is that for me um maths is the highest form of poetry. Um so you know I find beauty in the maths in the simplicity of course which it brings us back to the complexity minimization the right kinds of stories the right narratives the right calculus >> has to have that uh mixture of simplicity and accuracy in a sense the a kind of oneness that you you you were alluding to um so my my poetry is in the mouth I’m afraid >> beautiful um >> I wish I could read that poetry better I I should have sort of studied (2:08:47) Well, I think re well it’s an interesting way to phrase that because of course reading um now speaks of the hermeneutics problem of understanding the meaning of the maths in relation to other constructs. And I think that’s largely what our conversation has been about. >> That’s interesting. Yeah, you’re right. (2:09:04) You’re right. It’s it’s um >> Yeah. So, you would agree that mathematics is what what is about syntax. It’s about a structure of reality and and and and and I’m much more about the semantics what what all of this means. >> Yes. And of course we’ve talked at length try to unpack the semantics for for for our viewers. Yeah. (2:09:21) And you’re about the syntax. So yeah. Yeah. >> Um but you know and that also speaks to a point that was covert in one of your previous questions that the free energy principle is itself just a story. I think that’s absolutely right. >> Um very much like um natural selection or evolution. Um the freni principle sort of conforms to its own principles in a sense in the in that it provides a simple but accurate account of everything. (2:09:53) >> Yeah. >> Yeah. Of the data available to us. Um and you know in the same way that sort of you know evolution has itself evolved or evolutionary theory in terms of neodyarism sort of um modern day synthesis. uh your evolution has evolved. The free energy principle is just your a um a principle that can be applied to itself. (2:10:18) It just is meant to provide a beautiful and simple in my world poetic description of what people know basically. Well, people are not materialists in the sense that you describe know already or have known since Plato. Uh so, uh that’s certainly true. Um there was something you said which I just wanted to pick up on there. Um >> and the meaning the meaning of life. (2:10:41) >> Yeah. Um and um so you’re under under the free energy principle you your meaning really inherits from the notion of self-evidencing. So now I’m being poetic. So I’m I’m pretending to be um in specifically Yakob Howi who sort of you know has been one of the key people promoting the sort of self-evidencing aspect. (2:11:10) So you know one can read active inference under the free energy principle as self-evidencing. So what does that mean? Well, it means that I am um just here gathering evidence for my model of my world. And crucially, because my model has this central notion of me in it, it’s also a self model. So, I’m gathering evidence for my model of me in my world. (2:11:34) >> Wow. >> Which just means what would it look like? What what what would I um so what what am I trying to optimize? What am I trying to um do? Um, I am trying to optimize the evidence. I’m trying to minimize the surprise. I’m trying to optimize the evidence for my model. So, what does what kind of thing does that make me? It makes me into a curious creature. (2:11:56) >> It makes me into something that wants to understand my world. >> Yeah. >> It describes what uh you are doing now. Why are you doing what you are doing now? You are just trying to resolve uncertainty. You’re on an exploration. You’re an explorer in a particular intellectual and philosophical domain. >> Yeah. (2:12:14) >> You’re not doing this to keep your temperature at 37.2° centigrade or get chocolate or money. You’re doing it. >> Yeah. >> Cuz you’re seeking meaning. >> Yeah. It’s so true. Thank you for that beautiful reflection because like an hour ago when you asked me about sort of the highest model and I said God and you you you accurately said but I heard a lot of me, you know, when I I was talking about God that It has a lot to do with me understanding myself. (2:12:44) So what I hear hear you saying that it is about sort of selfdoing and it’s about sort of better understanding of self and that somehow our universe goes through the process of of doing that via market blankets and and and and models but that >> and that we do together. >> A lot of this is co-constructed. A lot of this yeah rests upon the shoulders of giants and communication. (2:13:07) So this is not um sort of well it is not something that I I would do or could do uh in isolation if I was on Mars by myself. It’s something that can only happen in this sort of >> of course of course and I like also that beauty in in your theory that the free energy principle is is a theory of self but it’s also a theory of selves and how they interact and that’s what we see in Michael Leven’s work and that all to me is just very beautiful and fascinating and um thank you so so much for this wonderful conversation Carl. I really (2:13:38) enjoyed it. >> My pleasure. Thank you. I enjoyed it as well. >> I’m happy to hear that. Yeah. Thank you so much. And thank you very much for watching this video. We will uh put links to uh Carl Fristen’s work below in the description. And if you have questions, please uh leave them in the comment section. (2:13:56) And we see you in one of our next videos. Thank you.”