Audio Player

Starting at:

Karl Friston: The Most INTENSE Theory of Reality Explained

August 16, 2024 • 1:11:15 • undefined

Channel: Theories of Everything with Curt Jaimungal

⚠️ Timestamps are hidden: Some podcast MP3s have dynamically injected ads which can shift timestamps. Show timestamps for troubleshooting.

Transcript

Enhanced with Timestamps

180 sentences 9,417 words

Method: api-polled Transcription time: 71m 49s

[0:00] The Economist covers math, physics, philosophy, and AI in a manner that shows how different countries perceive developments and how they impact markets. They recently published a piece on China's new neutrino detector. They cover extending life via mitochondrial transplants, creating an entirely new field of medicine. But it's also not just science they analyze.

[0:20] Culture, they analyze finance, economics, business, international affairs across every region. I'm particularly liking their new insider feature. It was just launched this month. It gives you, it gives me, a front row access to The Economist's internal editorial debates.

[0:36] Where senior editors argue through the news with world leaders and policy makers in twice weekly long format shows. Basically an extremely high quality podcast. Whether it's scientific innovation or shifting global politics, The Economist provides comprehensive coverage beyond headlines. As a toe listener, you get a special discount. Head over to economist.com slash TOE to subscribe. That's economist.com slash TOE for your discount.

[1:06] This episode is brought to you by State Farm. Listening to this podcast? Smart move. Being financially savvy? Smart move. Another smart move? Having State Farm help you create a competitive price when you choose to bundle home and auto. Bundling. Just another way to save with a personal price plan. Like a good neighbor, State Farm is there. Prices are based on rating plans that vary by state. Coverage options are selected by the customer. Availability, amount of discounts and savings, and eligibility vary by state.

[1:40] Professor Carl Friston. It's wonderful to speak with you again. I think this is the fifth time that we've spoken on this channel. We've also spoken in person off air in London. That was beautiful to meet you. Where was it? The Royal Society. It was the Royal Society. And it was wonderful to see you. Yes, it was wonderful. You said, meet me at where did you say it was some statue? It was the I thought it was a pub. It was the Duke of York steps.

[2:07] Douglas Goldstein, CFP®, Financial Planner & Investment Advisor

[2:17] Yeah, anyhow, it was great. Great to see you. Why don't you start your presentation, sir? Sure. But just as context for the audience, this is for a lecture series on theories of everything called Rethinking the Foundations of Biology. What lies beyond Darwin is the question. In this episode, Carl Fursten is just taking more of the Rethinking the Foundations of Biology itself. So please. Thank you. Indeed, I am. And for a reason.

[2:46] That reason is, I'm going to appeal here to Chris Fields, that there is no bright line between physics and biology, psychology and what could even argue philosophy. So the foundations of biology should by definition therefore be the foundations of physics and everything else. And I'm going to leverage the notion of everything else by just thinking about the nature of things, a foundation of thing-ness,

[3:16] The story I'm going to tell is a characterization of what it is to be some thing and the behaviors that that thing must possess. So this is a slightly deflationary foundational account of existence in the sense it just describes those things that persist in time over a suitable time scale. The other

[3:43] aspect of this foundational account. I refer to it here as the physics of sentience. It's really just the physics of things that self-organize technically to a set of characteristic states or attracting states and attracting set. The other aspect of this is that there's no new physics here. There's no new biology here and there's probably no new psychology.

[4:12] with the foundations that are shared by all of physics ranging from quantum physics, quantum mechanics through to stochastic or physical mechanics right through to classical mechanics. We're just going to start right at the beginning and ask the question what is it to be a thing and having answered that question.

[4:42] What it is to be a thing and what characteristics must this kind of thing be and I reiterate the physics interpretation of this is not new and it's actually almost tautological. The final point I want to make is I did notice that there was beyond Darwin in many senses the account that I'm going to briefly rehearse

[5:12] over the next few minutes is very closely allied with Darwinian thinking at two levels. First of all it shares the same almost tautological aspect in the sense that we are just describing things that exist and that existence is defined in terms of things that are there.

[5:40] What arises or emerges from this approach is a mechanics that you could say was the very mechanics of Darwinian selection. So I'm not going to go beyond Darwin, I'm going to meet Darwin and effectively tell a very very similar story but using the language

[6:09] Language of Theoretical Biology. What I'll do in the next few minutes is divide this discussion into three parts. First part is going to be just a consideration of the statistics of life with a special focus on Markov blankets and the ensuing Bayesian mechanics. I'm then going to tell the same story but using the rhetoric and concepts that a

[6:39] cognitive scientist might bring to the table specifically unpacking the mechanics that has been established in the first bit in terms of predictive coding neural networks and then if we've got time sort of move to a consideration of different kinds of things, different kinds of particles, different kinds of people, different kinds of institutions and draw a distinction between

[7:09] certain things that do not have an authentic kind of agency and other things that might have a natural intelligence or agency and try and articulate that distinction in terms of in fact Markov blankets. Okay so the first bit then

[7:31] I'm going to start with a question posed by Skrodinger. How can the events in space and time which take place within the spatial boundary of a living organism be accounted for by physics and chemistry? Now I'm not going to answer that question but I am going to draw attention to the notion of a boundary. The boundary I think is essential in terms of individuating or distinguishing something from everything else.

[7:58] Or indeed, no nothing or no thing. And I got to read that boundary as a statistical object, specifically a Markov boundary or Markov blanket. So what's a Markov blanket? Well, imagine we had a little universe where these cyan circles represent states and the arrows or the edges represent a causal influence. So this state influences this state.

[8:28] If I identified some set of states, say my internal states, then the Markov blanket comprises the parents, the children and the parents of the children. And the role of this set of states, this blanket plays is that it

[8:46] It separates me statistically from all the other states in the universe. If I wanted to know the dynamics of my internal states, given everything else in the putative universe, then I'd only need to know my blanket states. In other words, it provides a statistical veil or insulation that surrounds and shrouds my internal states and separates and individuates my internal states from my external states. Technically, it just means that the

[9:15] dynamics on the inside are conditionally independent of the dynamics on the outside given the blanket states. I'm going to make a further move here. I'm going to introduce a by partition of the blanket states into sensory states and active states where the sensory states influence but are not influenced by the internal states

[9:40] while the active states influence but are not influenced by the external states. So we've got this interesting very simple partition of the states of any universe from my perspective in terms of blanket states comprising sensory and active state that together with my internal states

[10:01] Could be regarded as the states of a particle, so it could be a small particle, it could be a person or as you've disintimated an institution. It doesn't matter the rules or the causal structure and the Markov blanket should apply in a scaling variant and possibly even scale free sense. And just to provide you with a couple of examples of different scales, here's my favorite system, the brain.

[10:28] Where we can record the internal states of the brain as my neuronal dynamics my connectivity that influence the active states that then change external states that could include my body that in turn reciprocate by influencing the sensory space my sensory epithelia.

[10:48] The difference between scale invariance and scale-free means what? Technically your renormalization group flow or your RG flow has some quantitative conservation as you move from scale to scale as opposed to the functional form of the dynamics as

[11:18] described by the Lagrangian or indeed the functional form of the equations of motion just being conserved. So, yeah, scale-free, if you like, is a sort of special case of scale invariance where there are extra constraints on what is conserved as you move from one scale to the next scale. And also just briefly, conditional independence for the Markov blanket is different than being completely independent. Absolutely. And this is quite crucial.

[11:47] So clearly if two sets of states were independent in the sense that they never influenced each other, you'd be describing two separate universes and you would only be interested in focusing on one universe or the other universe. So you're only interested in terms of systems or states that can be distinguished, individuated via conditional

[12:16] It's an open boundary. So if you take yourself back to your schoolboy physics, what we're talking about now is

[12:26] What constitutes the heat bath in a sort of classical sort of physics view of say a canonical micro ensemble. We're interested in what's beyond the heat bath and indeed how do we communicate through a heat bath which we can now treat as a Markov blanket to what is beyond that. So we're talking specifically about the way of foregrounding the importance of

[12:53] Mark of blanket is removing away from 20th century physics in the sense of equilibrium into the world of non equilibria that definitively require the system to be open. So where's the openness here? Well, the openness here is the openness of the system to the outside through a vicarious exchange with the outside.

[13:19] where the the if you like the the conditional independence is maintained by this two-way traffic so the inside influences the outside

[13:29] The inside influence is the outside through the active states. So the system is open and therefore when we're talking about self-organization of these open systems, we are effectively talking not about equilibria, we are talking about non-equilibrium steady states and I'll illustrate one of those.

[13:51] In the next slide just to give a bit more intuition to this so this is all about sort of self organization of non equilibrium systems that are open open in a way that allows you still to preserve the difference between the thing the internal states of the thing and the states external outside that kind of thing.

[14:14] This is just another example here, say a single cell organism with its intracellular states will be the internal states, the actin filaments would be the active states that push the cell surface on sensory states into the external milieu that reciprocates by changing cell surface receptors that changes the intracellular states. So just another example

[14:37] of the preservation of these conditional independences and thereby the functional form of the dynamics and thereby making this effectively apt for application of the apparatus of the renormalization group which brings us back to the scaling variant aspect of this kind of partition.

[14:59] So you said the filaments are the active states, but why do you call them active states and not an active part of the organism? What I mean to say is that a state I imagine is the entire organism. But then you would say that this part is active. You don't call that an active state, at least just in the terminology that I'm familiar with. Sure. So I'm using states here in the context of

[15:23] A state space so your an organism would be a collection of a very large number of states so that the state of an organism would be a point in a high dimensional state space so states here are collections of

[15:47] multiple different states and this Markov bank effectively is providing a partition of these multiple states in a particular way that allows you to distinguish. So you certainly could say that the active states for example could comprise

[16:08] A list of the particular position of all my actuators on my muscles the reason i mentioned active filaments is the active filaments of cells other things that push the cell surface around the actually cause movement so normally speaking and.

[16:26] Your active states are simply those that afford the capacity to move of the kind that you would expect to see in biotic motion for example. Is that distinction quite clear or is there some ambiguity between what's an active and an inactive state?

[16:46] Right, well the definition of active states is only in relation to this Markov blanket partition. So you can label the notion of an inactive state introduces a different kind of semantics to it which doesn't exist at this stage. So you can certainly have an active state or an internal state

[17:15] around an unstable point attractor so it's not moving so you could say you're from a dynamical perspective it's inactive because it's not changing. At this stage I'm just using the phrase active states just to denote a subset of waves of states in some state space that has this particular dependency relationship or influence relationship and specifically here

[17:43] the kind of states that influence the rate of change of external states. So this is explicitly as a dynamical formulation.

[17:53] Hola, Miami! When's the last time you've been in Burlington? We've updated, organized, and added fresh fashion. See for yourself Friday, November 14th to Sunday, November 16th at our Big Deal event. You can enter for a chance to win free wawa gas for a year, plus more surprises in your Burlington. Miami, that means so many ways and days to save. Burlington. Deals. Brands. Wow! No purchase necessary. Visit bigdealevent.com for more details.

[18:21] It might actually be more intuitive given this little diagram here. This is a starting point for everything that follows. In fact, let me ask you now to forget about the Markov blanket for a moment. What we'll do now is just do a 101 crash course in all of physics.

[18:46] And then what we're going to do is put the Markov blanket back into play and see what emerges.

[18:52] So this in effect is a reflection of what I was saying in the introduction that the kind of physics that I'm going to describe here is no different from any other kind of physics other than it is committed to a careful distinction between the states of something and the states of everything else that is not part of that thing, not a particular state where the particular states include the Markov-Bankhead states and the

[19:21] but if we just start where one could argue all of physics starts including quantum mechanics and with a random differential equation or random dynamical a description of a random dynamical system in terms of its motion which is some lawful function of where it is in state space plus a random fluctuation then we can sort of cartoon this system so i've just got two states here on the x and the y axes

[19:51] And as the system develops, it is traces out a trajectory or a path in this two dimensional state space. And again, appealing to this sort of scale environment aspect, this can be any scale you want. So for example, it could be electrochemical oscillations in a single cell in my body.

[20:14] It could be my cardiac cycle during a heartbeat, different aspects of the state of my heart. It could be me getting up in the morning, doing my emails, having a cup of coffee and so on. It could be Christmas, Easter. The key thing is at any scale or for any description of the universe at this scale in terms of some state space,

[20:42] The object in question here means that the system I will keep revisiting at any scale states that I have once been in. So that's the kind of thing that I am trying to describe. Technically, a random dynamical system

[21:02] that possesses an attracting set or a limit set that itself is a random variable, namely a pullback attractor. So I'm interested in systems in which things can exist in virtue of there being this tendency to revisit states that you were once or the system was once in. Is that clear?

[21:26] Yes, so anything that you do on repeat, whether it's your heart that has a beat or you're drinking a cup of coffee, either in the day or in multiple sips within the same five minute period. Now, what I don't get is what is a pullback attractor compared to regular attractor? Well, a regular attractor, if one should exist.

[21:44] normally is read as an attracting or limit set for a deterministic system but as soon as you introduce random fluctuations into the equations of motion or the dynamics because this thing itself is a

[22:01] a random variable the limit set has is again itself a random variable intuitively you can think of winding back in time and then winding forwards again given a particular realization of these random fluctuations to produce these these pullback attractors.

[22:24] I'm sure that doesn't help. What I often read is for technical reasons we refer to a pullback attractor but I think in spirit you can regard this just as an attractor in the vernacular sense. It's just a set of states

[22:44] that effectively the system looks as if it is attracted to in virtue of the dynamics keeping bringing you back to this manifold. But this manifold can be very very space filling and incorporate the itinerant dynamics that you were just referring to. So you know that repetition I think is probably more crucial than one might realize when one first

[23:13] We're talking about certain things that will have biotic aspects to them. So we're talking, for example, about things that have oscillations in them.

[23:30] Using the example of a single cell, electrochemical cell in my brain, I could regard this as a description of fast camera oscillations, for example, in my hippocampus. But we can also read this tendency to revisit this particular kind of itinerancy that is associated with these pullback attractors as biorhythms. And you can even go as far as life cycles.

[23:59] I'm just sort of bring you back to the title of your series Beyond Darwin replication is just another facet of the behaviors that these kinds of systems have to possess.

[24:16] Hmm. So this is another, if you like perspective on we don't need to go beyond Darwin. Darwin had everything that was necessary. Just reproduction replication is just one way of this kind of itinerancy on a pullback attractor is manifest. Interesting. Okay. Let me see if I got that correct.

[24:38] In life you reproduce or you replicate and this replication you can see as an oscillation in the same way that you would drink the same cup of coffee multiple times. Absolutely. I mean what you've just described is a lovely illustration of the scaling variance of the mechanics at hand and all you're saying is that there has to be an attracting set and this attracting set is a set of states

[25:07] In a system that is open and the openness comes along with the Markov blanket, which we'll get to in a second. Sure. And I also assume that in the same way your heart doesn't beat the exact same way every single time, even though it looks identical to us with our eyes, it has a slight variation. Maybe there is no such thing as two heartbeats in the same way. There's no such thing as two snowflakes that are exactly akin, but sorry, that are exactly equivalent. They are akin.

[25:35] Absolutely. So I assume and we don't have to get to the details, but I assume that in your model, it doesn't have to be exact repetitions. There can be some variation. Absolutely. Otherwise, you don't have Darwinianism. Absolutely. Yeah. And in fact, it can never be exact. You can never revisit, I should have said the neighborhood. So you revisit the neighborhood. And that is, if you like the essence of this pullback attractor that inherits from these random fluctuations. So okay, you just understood. Yeah, you just you just

[26:05] You're revisiting regimes of state space that can be described probabilistically. So unlike a point attractor, classical attractor, indeed even milliner attractors, we're not talking about

[26:22] being attracted to the same spot we're talking about to the vicinity in the neighborhood of. So this is where the probabilistic perspective comes in because I can now interpret this pullback attractor as where the density of the trajectories that are never coincident, they never converge, they never cross, or perhaps they do cross,

[26:52] But we can interpret the density of these trajectories, these paths, as a probability that you will find me in this state at any one particular time if you sampled me at a random time. So exactly your observation that a man never steps in the same river twice, two snowflakes are never the same, is accommodated by this probabilistic perspective on these pullback attractors.

[27:21] That's effectively the root or the basis of the move that gives rise to this Bayesian mechanics. Because once you've got in mind a description of the system in terms of the probability density of the system being in any state, you can now just appeal to all of physics to describe

[27:49] evolution not of the state in state space but of the probability distribution and this boils down to something really very simple really interesting because we've just said that there is effectively a pullback attractor we could also describe this as a non-equilibrium steady state it's not equilibrium because talking about open systems

[28:12] But there is a steady state in the sense that this repeating this replication, this itinerant revisiting of neighborhoods means that the density, the probability density reading of this object is itself unchanging.

[28:30] So I can just go along and take off the shelf your favorite density dynamics, this could be Richard Feynman's pathological formulation, it could be a master equation and kinetic models, it could be the Fokker-Planck equation, whatever you like, they're all at this level expressions of the same thing, they just basically express the rate of change of this probability density in terms of the amplitude gamma of random fluctuations

[28:59] and the flow, the flow through state space at any point in that state space. So I've just written that down in terms of the Fokker-Planck equation. But note, because we want to describe things that have possessed this attracting set or have a non-equilibrium steady state solution, the rate of change of this probability distribution is zero, which means I can now solve the density for the density dynamics.

[29:27] Perhaps I should just say that this is also one expression of a time independent Schrodinger equation, so I can just get the solution to that. And that allows me to do something quite interesting. It allows me to express the dynamics, the flow of the states as a function of where I am in state space as a mixture of gradient flows

[29:50] on the gradients subtended by the log of this probability, basically a potential erosion. In my world, we would call this self-information, which is nice because it's all about self-organization, also called by people like Tribus, surprise or more simply surprise or the negative of it is surprise or surprise and self-information. This partition of these two kinds of

[30:20] or a generalization of the Helmholtz decomposition and all it's saying is that the flow at any one point on this attracting set, this pullback attractor, this manifold can be divided into two parts. One is a gradient flow up log probability gradients or down potential gradients

[30:48] Intentional or probability contours, so often called

[30:58] solenoidal flow in virtue of that or divergence free flow or conservative flow whereas this part is the dissipative dynamics of flow and it has curl free aspects to it. So that circular flow I think I sort of highlight that

[31:19] simply because of our discussion previously about the importance of replication, repetition, revisiting the neighborhood of states that we were once in, those states that are characteristic of the kind of thing that I am. This rests upon the non-linearities in these systems, these random dynamical systems that manifest as this seronoidal circular flow

[31:45] I think that's quite important though it wasn't something i was going to foregrounded this you do realize that we're not gonna do this in thirty minutes or an hour.

[32:01] The designated time will keep going as long as you can keep going and then the audience will have questions. Anyhow, we can always come back for a part two. I think we're going to have to come back for a part two at this rate. It's going to probably be part three. I think we're going to have to come back for a part three because you're drilling down some of the really key issues which speak to the nature of the of

[32:29] The kind of systems that we're trying to account for, so not only do we have this sort of solenoidal aspect and this itinerancy and the randomness that

[32:45] is underwritten by pullback attractors but also what implicitly we've spoken about is the fact that these systems exist at multiple scales and each scale provides a context in the scale below and they all have this sort of itinerant non-equilibrium steady state aspect but the steady state is never actually attained but is slowly going towards these solutions at different timescales so an important aspect of

[33:15] this construction is that there is a scale above so we're talking before about my this being my heartbeat that scale is going to be much much longer and larger than the scale associated with the depolarization of the myocardium or any particular cell in my myocardium that could be responding to

[33:42] Inputs and outputs are much faster timescale say oscillating very very fast frequencies much more quickly than the slow second to second evolution of the system at the level of the heart itself and what that means is that from the point of view of one cell.

[34:06] The context in which it is operating is largely unchanging by appeal to an adiabatic approximation, which means that there is a solution at that temporal scale of the single cell to its fast dynamics, so it can attain its particular attracting set at this phase in the cardiac cycle.

[34:29] however of course the phase of the cardiac cycle is itself slowly changing so the pullback attractor from the point of view of single cell is slowly moving all the time and of course exactly the same maths underwrites darwinian selection so you've got your phenotype which is the single cell you know from the point of view of evolution or transgenerational dynamics

[34:58] The phylogenetic time course or time scales. The organism is changing very, very quickly as it grows and behaves and acts and decides and develops and dies. But from the point of view of the phenotype, the change in the genotype

[35:21] is so slow that's irrelevant so the phenotype the creature the organism at a very small scale now can be described as if it is conforming to its own little pullback attractor the states that i described before about getting up and doing my emails and having my cup of coffee that's only true while i am alive

[35:44] or why I am in a position to read emails and drink my cup of coffee. So that attractor manifold has a if you like an expiry date on it but because of the adiabatic approximation we can assume that at any given scale the scale above is changing so slowly that we can assume the existence of the solution to these dynamics and therefore

[36:14] and apply the solution of the Pocaplank equation so i'm i'm wittering on about that because i think there's a beautiful connection with darwinian thinking here once you put this density dynamics of the kind that you would you know pursue in terms of um

[36:31] quantum mechanics if you read this as a Schrodinger equation. You can pursue this and what you end up with is a variational or a density dynamic approach to natural selection in and of itself. Lots of people have written about this and it's certainly a current theme as far as I know in theoretical biology.

[36:56] Reading things like say the replicator equation as effectively exactly the same mechanics that people

[37:08] Doing inference and basic filtering would use so there's an opportunity to have a completely distinguish just talking about the mathematical and formal correspondences between the story i was going to tell about basic mechanics and the story that somebody

[37:30] Taking a Bayesian perspective on Darwinian dynamics would tell it's the same story basically but anyways in order to tell that story I have to make a link now between this sort of universal dynamics expressed in terms of a Helmholtz decomposition which we can read as

[37:55] an admixture of gradient flows up non-probability gradients up adaptive fitness or marginal likelihood if you like and this circular replicator like dynamics that is associated with the conservative or the devotions free flow or dynamics. How does that read as inference and Bayesian filtering or

[38:25] That's where the Markov blanket comes in. So if we now go back and just remember that previously we'd just been talking about any arbitrary and exceedingly large state space X. Now we come to partition X into the external, internal and intervening blanket states and furthermore just focus on

[38:52] two of those subsets namely the internal states of any given system say me and my active states and then by construction if you remember the unique thing about the internal states and the external states is that they are not influenced by the external states however that Helmholtz decomposition still has to be present which means that i can write down

[39:22] The dynamics of my internal states say or my neuronal dynamics and my motion of my actuators, my active states, my muscles and autonomic reflexes as in terms of this Helmholtz decomposition where the gradients are supplied

[39:43] The reading of this and it is just an interpretation

[40:00] is in terms of perception and action respectively and the perception part now yields to an interpretation in terms of sense making and Bayesian inference which is why I was talking about the emergence of Bayesian mechanics from

[40:18] The master equation on this time independent situation equation or the fucker plank equation if you're given the Markov blanket petition. In other words, if something exists in the sense that it possesses this distinguishing aspect in terms of possessing a Markov blanket, then it must comply with this

[40:44] these its internal and active states sometimes referred to as autonomous states of this particle or person must conform to this dynamics here so now the game is how would one interpret this how will you describe this to your students or your professors or your children and there are lots of ways that you can do that so I've just listed a few here I'm sure you will have come across all of these

[41:14] But just to again reinforce this point that this is foundational in the sense that it is assumed or described by many theories of self-organization and the particular accounts of behavior, things that actually move. So what is this quantity? Long probability of my sensory states given a Markov blanket. Well, we've just said

[41:44] that these states belong to an attracting set the pullback attractor so these are the sensory states that are attractive to me in the sense that they score those kinds of states i expect myself to to be in if you know if i was darwinian or darwin i would say that these are the adaptive states they are the the states that

[42:10] I repeat our characteristic of me as a surviving phenotype given I've got this pullback attractor and it is permitted by the context in which or the universe in which this attractor is manifest. So it just scores the attractiveness or the value to me of being in this particular sensory state here or the value in this sensory state space.

[42:34] So all this equation is saying is that both perception action is going to try and maximize value and that can now be read as reinforcement learning. If I was an engineer it would be called optimal control theory and if I was an economist it would just be expected utility theory where the reward the control objective or the utility just is this log probability of being in a state that is characteristic of the kind of state that I find myself in.

[43:04] The negative of this thing is what i was referring to before which is the self-information also known as surprise and surprise.

[43:12] So this equation says that i'm trying to it looks as if i am trying to minimize my surprise or self-information and from that we can read off the principle of maximum mutual information or the informax principle it could be the minimal redundancy principle of horospalo and the free energy principle. So where does the free energy principle get into the game? Well the variational free energy

[43:37] Is a proxy for the fact it's actually an upper bound on the self information so when people talk about the free energy principle as systems trying to minimize the free energy what they mean is basically systems that.

[43:55] Conform to this foundational Helmholtz decomposition of their dynamics where you've just substituted the free energy for the Lagrangian, the potential self-information or the surprise.

[44:09] Just a moment, sorry. Professor, you keep saying, well, before we explain this to our children, some non-trivial task, to put it lightly, you keep saying given a Markov blanket, now what's giving us the Markov blanket? Is it the causal structure of the universe? Is it our internal versus active states, or does the Markov blanket give rise to those instead?

[44:30] I understand there's some interplay, I understand that, but what gives us initially the Markov blanket? Right, well, I think you're now touching upon the tautology that I

[44:43] Introduced right at the beginning. So when I say given I just mean technically conditioned upon. So that bar in the notation in my world just means given or conditioned upon something. So more explicitly everything I am saying depends upon

[45:09] the existence of a Markov blanket. So your question I think is slightly deeper than that. So the question I heard was does the Markov blanket emerge from this kind of dynamics or does this kind of dynamics or these kinds of dynamics create a Markov blanket?

[45:39] All we are doing is describing the dynamics of any system that possesses a Markov blanket. So this is a little bit like it's deflation in the following sense. So what the three energy principle is saying is not that in order to survive, you must maximize value or adaptive fitness or marginal likelihood.

[46:08] That's not what the free energy principle is saying. What it's saying is if you survive in the sense of possessing a pullback attractor, then you will always be described as if you are maximizing value, adaptive fitness or marginal likelihood.

[46:30] That is the tautology that I was appealing to in the Darwinian sense. So we're assuming all we're doing is trying to convince a physics basically what this ends up being is a variational principle of least action that can be applied to something where the thing in question is very carefully defined technically in terms of a Markov blanket. So by saying that something exists

[46:58] Then the principle applies simply because the thingness is defined stipulatedly by being able to differentiate. The time average of this self information is entropy. Which means that these equations could also be read as a holy grail of self organization in the sense that they look as if.

[47:20] On average they're trying to minimize entropy and then from that you can elaborate synergetics of the kind that Herman Haken has established and if I was a physiologist it's just an expression of homeostasis.

[47:35] It's just keeping your essential variables, your physiological variables, within viable bounds that are valuable for me, that are characteristic of my particular states. So resisting the dispersion of the random fluctuations by these gradient flows that look as if I tried to minimize self-information on average, trying to minimize the entropy of my sensory states.

[48:05] There's a final interpretation which licenses the rhetoric of basic mechanics. So if I was a statistician, what I would describe this quantity as is just the probability of some sensory data

[48:26] Given conditioned upon my mark off blanket then i can now read as a model so my my blanket and my internal states now can be read as a model of the external states so i can now equivalently describe this Helmholtz decomposition of the dynamics

[48:49] In terms of the Bayesian brain hypothesis or evidence accumulation and a popular one in the neurosciences and the cognitive sciences is predicted coding. So this has been taken even further in philosophy that you can read these equations as essentially doing a gradient descent on model evidence.

[49:14] Model evidence is also known as a marginal likelihood. Why marginal? Well, because you've marginalized out all the causes of your sensory states or your sensory data, your sensory impressions, the impressions on the sensory sector of your Markov blanket, by which I mean

[49:32] Those causes that are the external states are not part of this probability density. So you've effectively integrated out or averaged away or marginalised them. So this is also known as the marginal likelihood or the model evidence. The likelihood of these data given me as a model of the world, the external states that caused these data. That has been

[49:59] Described or summarized in philosophy and self-evidencing so i'm minimizing my self-information just is Maxima acting and um perceiving in a way where perceiving is associated with my internal dynamics in a way that maximizes my the evidence for my model of

[50:22] the world or the external states. Put that another way, this self-evidencing just means that it looks as if I'm going around gathering evidence from my model of the world that just is me. So I'm just gathering evidence for the fact that I exist. So that's the most deflationary but poetic reading of this Bayesian mechanics. So

[50:49] That would be, that gives you just a flavor of the different ways in which you can understand ways of describing the dynamics of things where the thing in virtue of its existence cannot have any other dynamics and that can be sort of neatly summarized in terms of the existence of a

[51:17] The system's states, the systemic states into internal blanket comprising sensory and active and external states that are hidden behind the Markov blanket from the point of view of the internal states. And because the active states change, but are not changed by the external states, they look as if they're going to reduce the entropy of the blanket states.

[51:43] This means that action will appear to maintain the structural and functional integrity of the Markov blanket and one can read this in terms of self-assembly in computational chemistry or in biology as self-creational or autopoiesis in a very elemental form. Finally, internal states appear to infer the hidden causes of sensory states by increasing

[52:10] Bayesian evidence or model evidence or marginal likelihood and actively influence those causes and in my world we refer to this as active influence taking this self-evidencing perspective

[52:27] Afforded by the interpretation of the underlying self-information as effective the log of the or the negative log of model Bayesian model evidence. So that was the first half of the talk. I now notice that we've been

[52:48] Well over our allotted time. I'm going to suggest that we stop here and then you can wax a lyrical and ask questions for a few minutes and then we should rebook part two.

[53:01] To tell the same story, but now just using the concepts and the rhetoric and the the constructs from neurobiology and psychology. Okay, so then this is actually a natural stopping point because the latter half of the talk is using the first half of the talk to explain neurobiology and psychology. Absolutely. Yeah, we're just going to basically, you know,

[53:27] Yeah, interpret the same gradient flow held as decomposition, but through the lens of somebody looking at electrophysiology and sense making in animals and indeed psychology or possibly even neurophilosophy. Yeah, absolutely.

[53:42] OK, so one of my questions was this Fokker Planck equation. I'm unsure what you're saying is the relationship between that and other density equations like in quantum mechanics, Schrodinger's equation or the Feynman path integral. Are you saying that you can derive the Feynman path integral from the Fokker Planck equation or that the Fokker Planck equation is a density equation of the same sort that follows the calculus of variations? Like, is that what unites them? Yeah, just that there's minimizing something or maximizing something.

[54:12] Yeah, well I think the latter expression, there's just ways of articulating the same thing. So you can either write down the dynamics in terms of a random dynamical system, in terms of a Langevin equation, specify the equations of motion in terms of these flows and the statistics on random fluctuations, and then it is a fairly trivial matter to

[54:42] move to either a Fokker-Planck description of the implicit probability densities of this random dynamical system that can be articulated either in terms of a pathological formulation or in terms of a Fokker-Planck equation.

[55:00] Okay, so then it's the claim that the Langevin equation is the ultimate equation that gives rise to the other equations. I think in terms of what is the seed that gives birth to the rest. I think, well, for me, I mean, you know, if you spoke to other people, you may get a different answer. But certainly for me, the seed is the Langevin equation. It is

[55:24] Just a description of a random dynamical system. It's random because you've got these random fluctuations that in my world distinguish themselves from movements in state space

[55:42] In virtue of just being very, very, very fast, so fast that they cannot be observed. So you can't actually observe them. All you can do is write down a probability distribution, say the mean and variance. So for me, it would start here. Given that I can then write down the Fokker-Planck equation. So specifically, the amplitude, the statistics, the sufficient statistics of the random fluctuations just are the gamma here.

[56:12] And I know F. So if I know the statistics, the random fluctuations and the flow, then I can now just articulate this in terms of the Fokker-Planck equation that is just a function of the statistics, the random fluctuations and the flow. And from that, you can or you could express it in terms of a pathological formulation.

[56:40] So that is the seed and then we move to the Fokker-Pank equation and then make a very very simple and important move. The move is that this is equal to zero because the things that we want to understand are

[56:58] Possess a solution to the focal bank equation that they have a solution to the density dynamics in the same way that the time independent scrolling equation, which would be the equivalent if you were working. Well, perhaps I can just show you what I meant by that, which will be the equivalent in quantum mechanics. You're just looking at the solutions with a little preview of what we might have the opportunity to go through in sessions three. But what I wanted to show you

[57:28] Well this is just another intuition as to the importance of this Helmholtz decomposition in terms of self-organizing systems countering the random dispersion due to random fluctuations by imagining somebody dropping a blob of ink into a cup and the ink molecules gathering themselves up by flowing up the concentration

[57:58] The special cases of the solution to the Fokker-Planck equation that obtain

[58:27] When the amplitude of the random fluctuations goes away, you're just left with the solenoidal flow and that's just classical mechanics. So that's just things where there are lawful relationships between position and the velocity and mass and the like that emerge just because of this circular solenoidal conservative dynamics. Or the random fluctuations can get so big when you get very, very small because you're

[58:56] under the renormalization group or your random fluctuations haven't yet been averaged away because you're too small and then you get sort of thermodynamics where it's all about the gradient flow and from that you can derive fluctuation dissipation theorems and statistical or specifically stochastic mechanics. For

[59:20] If you wanted to go quantum, you just factorise this probability density in terms of complex roots and you get from this the time independent Schrodinger equation.

[59:36] Who was the first to point this out? See these three different colors that you have here, the three different derivations. Who was the first to point out that they're all reflections under different assumptions, four actually, that they're all under the Langevin equation, which is then decomposed in various ways.

[59:52] This is the Bayesian mechanics that is exactly the same as these things, but you've just now split your x's into internal and active states. I've shown the fourth one because the Bayesian mechanics is just the same as that. I don't know, the last time I was sitting in a lecture theatre

[60:18] So you've at least independently stated this? Oh yeah, but I'm assuming it's common knowledge. Certainly. I mean, you'll know better than I will. Certainly the equivalence between the pathological formulation and the Fokker-Planck equation and the time independent Schrodinger equation.

[60:46] Sorry, sir, to maximize on the time as this is a pun, actually, at the time that we have left, I want to talk about maximization or minimization. You said that. Let's take Darwin as an example.

[61:12] Survival of the fittest sounds like if you're the fittest, you will survive. But then you said, well, if you were to survive, it would appear as if you were the fittest. And those two aren't the same statements. And rather, what's happening is the latter. So help me understand how those two sentences are different. I think they're only different in the sense that the notion that I survive because I have a high adaptive fitness

[61:42] implies a teleology. So you can have a completely teleologically or telonomy or teleology free account of self-organization under the free energy principle and also quantum mechanics and stochastic and classical mechanics. Okay. So you weren't saying that, let's say we take a, let's say we take this cup and

[62:06] If I was to let go, it would just fall and it would fall because it's minimizing action. Now for people who are listening, who don't understand what action is, let's say it's minimizing energy. We don't need to concern ourselves with the difference between two different types of energy being action. It's minimizing something, whether it's energy or action. So that's one way of saying it, that this cup minimizes its action. But then another way from what I understood of what you were saying is that I will only perceive objects that minimize their action.

[62:33] In other words, this cup could have done plenty of other actions or there could be many other cups, but my perceptual system for whatever reason is only tuned to what minimizes and therefore it's not that what survives is the fittest is only that I'll see what's maximizing the fitness as surviving or I'll only see this cup move in this direction because it minimized something. I think that's, um,

[62:57] The things out there conform to a path of least action because they exist and therefore we will perceive them as pursuing paths of least action.

[63:27] So that's all the free energy principle is saying. It's just saying that we pursue paths of least action. And what could one apply that to? Well, adaptive fitness, for example. So Darwinian thinking.

[63:43] Adaptive fitness is just the thing that supplies the potential in the action, which is time times potential. So you're maximizing the adaptive fitness or you're minimizing the negative adaptive fitness. The nice point of content that comes into the way you perceive the cup though, I think reflects the fact that

[64:04] When you put the Markov blanket in place, don't forget about doing that. You can't talk sensibly about perceiving apples unless you've got an observer and something that's being observed. So you've immediately invoked a Markov blanket. So if you're talking to Stephen Wolfram, he'd call this the observer problem. To solve the observer problem, to even talk about it, you've got to have a Markov blanket. So there has to be a distinction between the observer and the observee. The measurement problem, the same thing.

[64:33] quantum physics. You have to have the Markov blanket explicitly in play. So once you've got the Markov blanket in play, then the adaptive fitness is just the measure of synchrony or isomorphism between the inside and the outside. So literally, if you take this basic mechanics perspective, the adaptive fitness is the ability of you to fit into your environment, by which I mean

[65:01] The sensory impressions provided by your environment, the feedback that the environment is giving you was the most likely for the kind of thing that you are, the phenotype that you are. So adaptive fitness is just the marginal likelihood of the environment providing that sensory, the way that the environment influences you by the sensory states.

[65:24] is the measure of adaptive fitness and it's just the fit of your model the ability of your model to fit this world so it all i think um it does actually come back to you say i can only perceive falling apples because apples only fall yes if i lived in a universe where apples didn't which did not comply with a principle of least action which is not impossible you know i couldn't live in a very um

[65:50] Small universe where random fluctuations cause things to depart enormously from paths of least action and I would still be able to model that but you know in this instance I would be able to perceive random fluctuations of the apple or itinerant dynamics which violated say conservative dynamics. So I have one more question and I know you have to get going. If the free energy principle is true by definition of tautology

[66:21] Then why is there so many papers published on it? There has to be something more to the free energy principle than just being a tautology. So in other words, Kant has analytic truths, but there must be something here that's synthetic. Otherwise, I don't understand how it can be rich if it's not purely analytic. Well, I never said it was rich. I know you're not saying it's rich. The fact that we've even spent over an hour on it.

[66:47] and that there's more than one slide and that there are papers and that you're the highest cited neuroscientist in the world implies that it's rich. So help me understand that. I can. There is an answer, but given I've already now got three minutes left, I think that's an excellent, an excellent place we should start. I'll give you a clue. Um, you can use the free energy principle in the same sense. You can use Hamilton's principle of least action to design automobiles, trajectories,

[67:17] physical artifacts.

[67:31] Free energy flows or the base basic mechanics to reproduce self-organization and in principle, intelligent kind of self-organization. So you can create and simulate self-organization because you know the principles that underwrite the dynamics. So that's a clue, but I can unpack that in more detail the next time. Let's uncompress it next time. And I see Hamilton's equation as synthetic.

[67:58] Not just analytic, but we can talk about the distinction next time, sir. Okay. This was so much fun. It went by like that. Thank you. I love talking to you. So I'll see you hopefully in a few days or weeks depending upon your itinerary. Yes. Okay. Brilliant. Thank you, sir.

[68:18] Firstly, thank you for watching. Thank you for listening. There's now a website, curtjymungle.org, and that has a mailing list. The reason being that large platforms like YouTube, like Patreon, they can disable you for whatever reason, whenever they like.

[68:33] That's just part of the terms of service. Now, a direct mailing list ensures that I have an untrammeled communication with you. Plus, soon I'll be releasing a one-page PDF of my top 10 toes. It's not as Quentin Tarantino as it sounds like. Secondly, if you haven't subscribed or clicked that like button, now is the time to do so. Why? Because each subscribe, each like helps YouTube push this content to more people like yourself

[69:00] Plus, it helps out Kurt directly, aka me. I also found out last year that external links count plenty toward the algorithm, which means that whenever you share on Twitter, say on Facebook or even on Reddit, etc., it shows YouTube, hey, people are talking about this content outside of YouTube, which in turn

[69:18] Greatly aids the distribution on YouTube. Thirdly, there's a remarkably active Discord and subreddit for theories of everything where people explicate toes, they disagree respectfully about theories and build as a community our own toe. Links to both are in the description. Fourthly, you should know this podcast is on iTunes. It's on Spotify. It's on all of the audio platforms. All you have to do is type in theories of everything and you'll find it. Personally, I gained from rewatching lectures and podcasts.

[69:46] I also read in the comments

[70:06] and donating with whatever you like. There's also PayPal. There's also crypto. There's also just joining on YouTube. Again, keep in mind it's support from the sponsors and you that allow me to work on toe full time. You also get early access to ad free episodes, whether it's audio or video, it's audio in the case of Patreon video in the case of YouTube. For instance, this episode that you're listening to right now was released a few days earlier.

[70:30] Every dollar helps far more than you think. Either way, your viewership is generosity enough. Thank you so much. Think Verizon, the best 5G network is expensive? Think again. Bring in your AT&T or T-Mobile bill to a Verizon store today and

[70:58] Jokes aside, Verizon has the most ways to save on phones and plans where you can get a single line with everything you need. So bring in your bill to your local Verizon store today and we'll give you a better deal.

[71:19] Amazon, Pizza Hut, Audible. How'd they get so big without soul destroying complexity? On Founders Mentality, the CEO Sessions, we're going to find out. Who's number one? It's a customer. Whose Walmart is it? My Walmart. If you looked at Audible, it was kind of like growth, growth, and then growth. It separates Amazon and AWS from anyone else. Join me, Jimmy Allen, partner of Bain & Company, to hear surprising stories from the world's greatest leaders. Subscribe to Founders Mentality, the CEO Sessions, now.

▶ View Full JSON Data (Word-Level Timestamps)

{
  "source": "transcribe.metaboat.io",
  "workspace_id": "AXs1igz",
  "job_seq": 4842,
  "audio_duration_seconds": 4309.07,
  "completed_at": "2025-11-30T23:25:29Z",
  "segments": [
    {
      "end_time": 20.896,
      "index": 0,
      "start_time": 0.009,
      "text": " The Economist covers math, physics, philosophy, and AI in a manner that shows how different countries perceive developments and how they impact markets. They recently published a piece on China's new neutrino detector. They cover extending life via mitochondrial transplants, creating an entirely new field of medicine. But it's also not just science they analyze."
    },
    {
      "end_time": 36.067,
      "index": 1,
      "start_time": 20.896,
      "text": " Culture, they analyze finance, economics, business, international affairs across every region. I'm particularly liking their new insider feature. It was just launched this month. It gives you, it gives me, a front row access to The Economist's internal editorial debates."
    },
    {
      "end_time": 64.514,
      "index": 2,
      "start_time": 36.34,
      "text": " Where senior editors argue through the news with world leaders and policy makers in twice weekly long format shows. Basically an extremely high quality podcast. Whether it's scientific innovation or shifting global politics, The Economist provides comprehensive coverage beyond headlines. As a toe listener, you get a special discount. Head over to economist.com slash TOE to subscribe. That's economist.com slash TOE for your discount."
    },
    {
      "end_time": 95.572,
      "index": 3,
      "start_time": 66.834,
      "text": " This episode is brought to you by State Farm. Listening to this podcast? Smart move. Being financially savvy? Smart move. Another smart move? Having State Farm help you create a competitive price when you choose to bundle home and auto. Bundling. Just another way to save with a personal price plan. Like a good neighbor, State Farm is there. Prices are based on rating plans that vary by state. Coverage options are selected by the customer. Availability, amount of discounts and savings, and eligibility vary by state."
    },
    {
      "end_time": 127.142,
      "index": 4,
      "start_time": 100.606,
      "text": " Professor Carl Friston. It's wonderful to speak with you again. I think this is the fifth time that we've spoken on this channel. We've also spoken in person off air in London. That was beautiful to meet you. Where was it? The Royal Society. It was the Royal Society. And it was wonderful to see you. Yes, it was wonderful. You said, meet me at where did you say it was some statue? It was the I thought it was a pub. It was the Duke of York steps."
    },
    {
      "end_time": 136.817,
      "index": 5,
      "start_time": 127.568,
      "text": " Douglas Goldstein, CFP®, Financial Planner & Investment Advisor"
    },
    {
      "end_time": 165.947,
      "index": 6,
      "start_time": 137.09,
      "text": " Yeah, anyhow, it was great. Great to see you. Why don't you start your presentation, sir? Sure. But just as context for the audience, this is for a lecture series on theories of everything called Rethinking the Foundations of Biology. What lies beyond Darwin is the question. In this episode, Carl Fursten is just taking more of the Rethinking the Foundations of Biology itself. So please. Thank you. Indeed, I am. And for a reason."
    },
    {
      "end_time": 195.759,
      "index": 7,
      "start_time": 166.305,
      "text": " That reason is, I'm going to appeal here to Chris Fields, that there is no bright line between physics and biology, psychology and what could even argue philosophy. So the foundations of biology should by definition therefore be the foundations of physics and everything else. And I'm going to leverage the notion of everything else by just thinking about the nature of things, a foundation of thing-ness,"
    },
    {
      "end_time": 222.483,
      "index": 8,
      "start_time": 196.101,
      "text": " The story I'm going to tell is a characterization of what it is to be some thing and the behaviors that that thing must possess. So this is a slightly deflationary foundational account of existence in the sense it just describes those things that persist in time over a suitable time scale. The other"
    },
    {
      "end_time": 252.637,
      "index": 9,
      "start_time": 223.2,
      "text": " aspect of this foundational account. I refer to it here as the physics of sentience. It's really just the physics of things that self-organize technically to a set of characteristic states or attracting states and attracting set. The other aspect of this is that there's no new physics here. There's no new biology here and there's probably no new psychology."
    },
    {
      "end_time": 280.299,
      "index": 10,
      "start_time": 252.927,
      "text": " with the foundations that are shared by all of physics ranging from quantum physics, quantum mechanics through to stochastic or physical mechanics right through to classical mechanics. We're just going to start right at the beginning and ask the question what is it to be a thing and having answered that question."
    },
    {
      "end_time": 311.749,
      "index": 11,
      "start_time": 282.022,
      "text": " What it is to be a thing and what characteristics must this kind of thing be and I reiterate the physics interpretation of this is not new and it's actually almost tautological. The final point I want to make is I did notice that there was beyond Darwin in many senses the account that I'm going to briefly rehearse"
    },
    {
      "end_time": 340.026,
      "index": 12,
      "start_time": 312.142,
      "text": " over the next few minutes is very closely allied with Darwinian thinking at two levels. First of all it shares the same almost tautological aspect in the sense that we are just describing things that exist and that existence is defined in terms of things that are there."
    },
    {
      "end_time": 369.65,
      "index": 13,
      "start_time": 340.299,
      "text": " What arises or emerges from this approach is a mechanics that you could say was the very mechanics of Darwinian selection. So I'm not going to go beyond Darwin, I'm going to meet Darwin and effectively tell a very very similar story but using the language"
    },
    {
      "end_time": 399.172,
      "index": 14,
      "start_time": 369.906,
      "text": " Language of Theoretical Biology. What I'll do in the next few minutes is divide this discussion into three parts. First part is going to be just a consideration of the statistics of life with a special focus on Markov blankets and the ensuing Bayesian mechanics. I'm then going to tell the same story but using the rhetoric and concepts that a"
    },
    {
      "end_time": 429.377,
      "index": 15,
      "start_time": 399.565,
      "text": " cognitive scientist might bring to the table specifically unpacking the mechanics that has been established in the first bit in terms of predictive coding neural networks and then if we've got time sort of move to a consideration of different kinds of things, different kinds of particles, different kinds of people, different kinds of institutions and draw a distinction between"
    },
    {
      "end_time": 450.64,
      "index": 16,
      "start_time": 429.753,
      "text": " certain things that do not have an authentic kind of agency and other things that might have a natural intelligence or agency and try and articulate that distinction in terms of in fact Markov blankets. Okay so the first bit then"
    },
    {
      "end_time": 478.148,
      "index": 17,
      "start_time": 451.101,
      "text": " I'm going to start with a question posed by Skrodinger. How can the events in space and time which take place within the spatial boundary of a living organism be accounted for by physics and chemistry? Now I'm not going to answer that question but I am going to draw attention to the notion of a boundary. The boundary I think is essential in terms of individuating or distinguishing something from everything else."
    },
    {
      "end_time": 507.602,
      "index": 18,
      "start_time": 478.524,
      "text": " Or indeed, no nothing or no thing. And I got to read that boundary as a statistical object, specifically a Markov boundary or Markov blanket. So what's a Markov blanket? Well, imagine we had a little universe where these cyan circles represent states and the arrows or the edges represent a causal influence. So this state influences this state."
    },
    {
      "end_time": 526.237,
      "index": 19,
      "start_time": 508.063,
      "text": " If I identified some set of states, say my internal states, then the Markov blanket comprises the parents, the children and the parents of the children. And the role of this set of states, this blanket plays is that it"
    },
    {
      "end_time": 555.367,
      "index": 20,
      "start_time": 526.527,
      "text": " It separates me statistically from all the other states in the universe. If I wanted to know the dynamics of my internal states, given everything else in the putative universe, then I'd only need to know my blanket states. In other words, it provides a statistical veil or insulation that surrounds and shrouds my internal states and separates and individuates my internal states from my external states. Technically, it just means that the"
    },
    {
      "end_time": 580.367,
      "index": 21,
      "start_time": 555.913,
      "text": " dynamics on the inside are conditionally independent of the dynamics on the outside given the blanket states. I'm going to make a further move here. I'm going to introduce a by partition of the blanket states into sensory states and active states where the sensory states influence but are not influenced by the internal states"
    },
    {
      "end_time": 601.049,
      "index": 22,
      "start_time": 580.896,
      "text": " while the active states influence but are not influenced by the external states. So we've got this interesting very simple partition of the states of any universe from my perspective in terms of blanket states comprising sensory and active state that together with my internal states"
    },
    {
      "end_time": 627.875,
      "index": 23,
      "start_time": 601.459,
      "text": " Could be regarded as the states of a particle, so it could be a small particle, it could be a person or as you've disintimated an institution. It doesn't matter the rules or the causal structure and the Markov blanket should apply in a scaling variant and possibly even scale free sense. And just to provide you with a couple of examples of different scales, here's my favorite system, the brain."
    },
    {
      "end_time": 648.114,
      "index": 24,
      "start_time": 628.524,
      "text": " Where we can record the internal states of the brain as my neuronal dynamics my connectivity that influence the active states that then change external states that could include my body that in turn reciprocate by influencing the sensory space my sensory epithelia."
    },
    {
      "end_time": 677.568,
      "index": 25,
      "start_time": 648.626,
      "text": " The difference between scale invariance and scale-free means what? Technically your renormalization group flow or your RG flow has some quantitative conservation as you move from scale to scale as opposed to the functional form of the dynamics as"
    },
    {
      "end_time": 707.654,
      "index": 26,
      "start_time": 678.097,
      "text": " described by the Lagrangian or indeed the functional form of the equations of motion just being conserved. So, yeah, scale-free, if you like, is a sort of special case of scale invariance where there are extra constraints on what is conserved as you move from one scale to the next scale. And also just briefly, conditional independence for the Markov blanket is different than being completely independent. Absolutely. And this is quite crucial."
    },
    {
      "end_time": 736.254,
      "index": 27,
      "start_time": 707.944,
      "text": " So clearly if two sets of states were independent in the sense that they never influenced each other, you'd be describing two separate universes and you would only be interested in focusing on one universe or the other universe. So you're only interested in terms of systems or states that can be distinguished, individuated via conditional"
    },
    {
      "end_time": 746.664,
      "index": 28,
      "start_time": 736.578,
      "text": " It's an open boundary. So if you take yourself back to your schoolboy physics, what we're talking about now is"
    },
    {
      "end_time": 773.08,
      "index": 29,
      "start_time": 746.903,
      "text": " What constitutes the heat bath in a sort of classical sort of physics view of say a canonical micro ensemble. We're interested in what's beyond the heat bath and indeed how do we communicate through a heat bath which we can now treat as a Markov blanket to what is beyond that. So we're talking specifically about the way of foregrounding the importance of"
    },
    {
      "end_time": 798.831,
      "index": 30,
      "start_time": 773.524,
      "text": " Mark of blanket is removing away from 20th century physics in the sense of equilibrium into the world of non equilibria that definitively require the system to be open. So where's the openness here? Well, the openness here is the openness of the system to the outside through a vicarious exchange with the outside."
    },
    {
      "end_time": 809.258,
      "index": 31,
      "start_time": 799.121,
      "text": " where the the if you like the the conditional independence is maintained by this two-way traffic so the inside influences the outside"
    },
    {
      "end_time": 831.63,
      "index": 32,
      "start_time": 809.65,
      "text": " The inside influence is the outside through the active states. So the system is open and therefore when we're talking about self-organization of these open systems, we are effectively talking not about equilibria, we are talking about non-equilibrium steady states and I'll illustrate one of those."
    },
    {
      "end_time": 854.275,
      "index": 33,
      "start_time": 831.63,
      "text": " In the next slide just to give a bit more intuition to this so this is all about sort of self organization of non equilibrium systems that are open open in a way that allows you still to preserve the difference between the thing the internal states of the thing and the states external outside that kind of thing."
    },
    {
      "end_time": 876.357,
      "index": 34,
      "start_time": 854.599,
      "text": " This is just another example here, say a single cell organism with its intracellular states will be the internal states, the actin filaments would be the active states that push the cell surface on sensory states into the external milieu that reciprocates by changing cell surface receptors that changes the intracellular states. So just another example"
    },
    {
      "end_time": 898.763,
      "index": 35,
      "start_time": 877.381,
      "text": " of the preservation of these conditional independences and thereby the functional form of the dynamics and thereby making this effectively apt for application of the apparatus of the renormalization group which brings us back to the scaling variant aspect of this kind of partition."
    },
    {
      "end_time": 922.517,
      "index": 36,
      "start_time": 899.138,
      "text": " So you said the filaments are the active states, but why do you call them active states and not an active part of the organism? What I mean to say is that a state I imagine is the entire organism. But then you would say that this part is active. You don't call that an active state, at least just in the terminology that I'm familiar with. Sure. So I'm using states here in the context of"
    },
    {
      "end_time": 947.09,
      "index": 37,
      "start_time": 923.848,
      "text": " A state space so your an organism would be a collection of a very large number of states so that the state of an organism would be a point in a high dimensional state space so states here are collections of"
    },
    {
      "end_time": 968.114,
      "index": 38,
      "start_time": 947.756,
      "text": " multiple different states and this Markov bank effectively is providing a partition of these multiple states in a particular way that allows you to distinguish. So you certainly could say that the active states for example could comprise"
    },
    {
      "end_time": 986.254,
      "index": 39,
      "start_time": 968.319,
      "text": " A list of the particular position of all my actuators on my muscles the reason i mentioned active filaments is the active filaments of cells other things that push the cell surface around the actually cause movement so normally speaking and."
    },
    {
      "end_time": 1005.657,
      "index": 40,
      "start_time": 986.578,
      "text": " Your active states are simply those that afford the capacity to move of the kind that you would expect to see in biotic motion for example. Is that distinction quite clear or is there some ambiguity between what's an active and an inactive state?"
    },
    {
      "end_time": 1035.657,
      "index": 41,
      "start_time": 1006.698,
      "text": " Right, well the definition of active states is only in relation to this Markov blanket partition. So you can label the notion of an inactive state introduces a different kind of semantics to it which doesn't exist at this stage. So you can certainly have an active state or an internal state"
    },
    {
      "end_time": 1062.892,
      "index": 42,
      "start_time": 1035.998,
      "text": " around an unstable point attractor so it's not moving so you could say you're from a dynamical perspective it's inactive because it's not changing. At this stage I'm just using the phrase active states just to denote a subset of waves of states in some state space that has this particular dependency relationship or influence relationship and specifically here"
    },
    {
      "end_time": 1072.619,
      "index": 43,
      "start_time": 1063.251,
      "text": " the kind of states that influence the rate of change of external states. So this is explicitly as a dynamical formulation."
    },
    {
      "end_time": 1100.111,
      "index": 44,
      "start_time": 1073.541,
      "text": " Hola, Miami! When's the last time you've been in Burlington? We've updated, organized, and added fresh fashion. See for yourself Friday, November 14th to Sunday, November 16th at our Big Deal event. You can enter for a chance to win free wawa gas for a year, plus more surprises in your Burlington. Miami, that means so many ways and days to save. Burlington. Deals. Brands. Wow! No purchase necessary. Visit bigdealevent.com for more details."
    },
    {
      "end_time": 1126.732,
      "index": 45,
      "start_time": 1101.34,
      "text": " It might actually be more intuitive given this little diagram here. This is a starting point for everything that follows. In fact, let me ask you now to forget about the Markov blanket for a moment. What we'll do now is just do a 101 crash course in all of physics."
    },
    {
      "end_time": 1132.108,
      "index": 46,
      "start_time": 1126.971,
      "text": " And then what we're going to do is put the Markov blanket back into play and see what emerges."
    },
    {
      "end_time": 1161.766,
      "index": 47,
      "start_time": 1132.841,
      "text": " So this in effect is a reflection of what I was saying in the introduction that the kind of physics that I'm going to describe here is no different from any other kind of physics other than it is committed to a careful distinction between the states of something and the states of everything else that is not part of that thing, not a particular state where the particular states include the Markov-Bankhead states and the"
    },
    {
      "end_time": 1191.34,
      "index": 48,
      "start_time": 1161.766,
      "text": " but if we just start where one could argue all of physics starts including quantum mechanics and with a random differential equation or random dynamical a description of a random dynamical system in terms of its motion which is some lawful function of where it is in state space plus a random fluctuation then we can sort of cartoon this system so i've just got two states here on the x and the y axes"
    },
    {
      "end_time": 1213.456,
      "index": 49,
      "start_time": 1191.749,
      "text": " And as the system develops, it is traces out a trajectory or a path in this two dimensional state space. And again, appealing to this sort of scale environment aspect, this can be any scale you want. So for example, it could be electrochemical oscillations in a single cell in my body."
    },
    {
      "end_time": 1241.681,
      "index": 50,
      "start_time": 1214.019,
      "text": " It could be my cardiac cycle during a heartbeat, different aspects of the state of my heart. It could be me getting up in the morning, doing my emails, having a cup of coffee and so on. It could be Christmas, Easter. The key thing is at any scale or for any description of the universe at this scale in terms of some state space,"
    },
    {
      "end_time": 1262.039,
      "index": 51,
      "start_time": 1242.398,
      "text": " The object in question here means that the system I will keep revisiting at any scale states that I have once been in. So that's the kind of thing that I am trying to describe. Technically, a random dynamical system"
    },
    {
      "end_time": 1285.93,
      "index": 52,
      "start_time": 1262.329,
      "text": " that possesses an attracting set or a limit set that itself is a random variable, namely a pullback attractor. So I'm interested in systems in which things can exist in virtue of there being this tendency to revisit states that you were once or the system was once in. Is that clear?"
    },
    {
      "end_time": 1304.155,
      "index": 53,
      "start_time": 1286.323,
      "text": " Yes, so anything that you do on repeat, whether it's your heart that has a beat or you're drinking a cup of coffee, either in the day or in multiple sips within the same five minute period. Now, what I don't get is what is a pullback attractor compared to regular attractor? Well, a regular attractor, if one should exist."
    },
    {
      "end_time": 1320.862,
      "index": 54,
      "start_time": 1304.906,
      "text": " normally is read as an attracting or limit set for a deterministic system but as soon as you introduce random fluctuations into the equations of motion or the dynamics because this thing itself is a"
    },
    {
      "end_time": 1342.79,
      "index": 55,
      "start_time": 1321.237,
      "text": " a random variable the limit set has is again itself a random variable intuitively you can think of winding back in time and then winding forwards again given a particular realization of these random fluctuations to produce these these pullback attractors."
    },
    {
      "end_time": 1364.224,
      "index": 56,
      "start_time": 1344.224,
      "text": " I'm sure that doesn't help. What I often read is for technical reasons we refer to a pullback attractor but I think in spirit you can regard this just as an attractor in the vernacular sense. It's just a set of states"
    },
    {
      "end_time": 1393.37,
      "index": 57,
      "start_time": 1364.582,
      "text": " that effectively the system looks as if it is attracted to in virtue of the dynamics keeping bringing you back to this manifold. But this manifold can be very very space filling and incorporate the itinerant dynamics that you were just referring to. So you know that repetition I think is probably more crucial than one might realize when one first"
    },
    {
      "end_time": 1410.111,
      "index": 58,
      "start_time": 1393.814,
      "text": " We're talking about certain things that will have biotic aspects to them. So we're talking, for example, about things that have oscillations in them."
    },
    {
      "end_time": 1438.848,
      "index": 59,
      "start_time": 1410.555,
      "text": " Using the example of a single cell, electrochemical cell in my brain, I could regard this as a description of fast camera oscillations, for example, in my hippocampus. But we can also read this tendency to revisit this particular kind of itinerancy that is associated with these pullback attractors as biorhythms. And you can even go as far as life cycles."
    },
    {
      "end_time": 1456.954,
      "index": 60,
      "start_time": 1439.445,
      "text": " I'm just sort of bring you back to the title of your series Beyond Darwin replication is just another facet of the behaviors that these kinds of systems have to possess."
    },
    {
      "end_time": 1477.892,
      "index": 61,
      "start_time": 1456.954,
      "text": " Hmm. So this is another, if you like perspective on we don't need to go beyond Darwin. Darwin had everything that was necessary. Just reproduction replication is just one way of this kind of itinerancy on a pullback attractor is manifest. Interesting. Okay. Let me see if I got that correct."
    },
    {
      "end_time": 1506.988,
      "index": 62,
      "start_time": 1478.234,
      "text": " In life you reproduce or you replicate and this replication you can see as an oscillation in the same way that you would drink the same cup of coffee multiple times. Absolutely. I mean what you've just described is a lovely illustration of the scaling variance of the mechanics at hand and all you're saying is that there has to be an attracting set and this attracting set is a set of states"
    },
    {
      "end_time": 1534.94,
      "index": 63,
      "start_time": 1507.381,
      "text": " In a system that is open and the openness comes along with the Markov blanket, which we'll get to in a second. Sure. And I also assume that in the same way your heart doesn't beat the exact same way every single time, even though it looks identical to us with our eyes, it has a slight variation. Maybe there is no such thing as two heartbeats in the same way. There's no such thing as two snowflakes that are exactly akin, but sorry, that are exactly equivalent. They are akin."
    },
    {
      "end_time": 1564.787,
      "index": 64,
      "start_time": 1535.162,
      "text": " Absolutely. So I assume and we don't have to get to the details, but I assume that in your model, it doesn't have to be exact repetitions. There can be some variation. Absolutely. Otherwise, you don't have Darwinianism. Absolutely. Yeah. And in fact, it can never be exact. You can never revisit, I should have said the neighborhood. So you revisit the neighborhood. And that is, if you like the essence of this pullback attractor that inherits from these random fluctuations. So okay, you just understood. Yeah, you just you just"
    },
    {
      "end_time": 1582.363,
      "index": 65,
      "start_time": 1565.265,
      "text": " You're revisiting regimes of state space that can be described probabilistically. So unlike a point attractor, classical attractor, indeed even milliner attractors, we're not talking about"
    },
    {
      "end_time": 1611.8,
      "index": 66,
      "start_time": 1582.637,
      "text": " being attracted to the same spot we're talking about to the vicinity in the neighborhood of. So this is where the probabilistic perspective comes in because I can now interpret this pullback attractor as where the density of the trajectories that are never coincident, they never converge, they never cross, or perhaps they do cross,"
    },
    {
      "end_time": 1641.51,
      "index": 67,
      "start_time": 1612.722,
      "text": " But we can interpret the density of these trajectories, these paths, as a probability that you will find me in this state at any one particular time if you sampled me at a random time. So exactly your observation that a man never steps in the same river twice, two snowflakes are never the same, is accommodated by this probabilistic perspective on these pullback attractors."
    },
    {
      "end_time": 1669.172,
      "index": 68,
      "start_time": 1641.834,
      "text": " That's effectively the root or the basis of the move that gives rise to this Bayesian mechanics. Because once you've got in mind a description of the system in terms of the probability density of the system being in any state, you can now just appeal to all of physics to describe"
    },
    {
      "end_time": 1691.886,
      "index": 69,
      "start_time": 1669.445,
      "text": " evolution not of the state in state space but of the probability distribution and this boils down to something really very simple really interesting because we've just said that there is effectively a pullback attractor we could also describe this as a non-equilibrium steady state it's not equilibrium because talking about open systems"
    },
    {
      "end_time": 1709.582,
      "index": 70,
      "start_time": 1692.193,
      "text": " But there is a steady state in the sense that this repeating this replication, this itinerant revisiting of neighborhoods means that the density, the probability density reading of this object is itself unchanging."
    },
    {
      "end_time": 1739.565,
      "index": 71,
      "start_time": 1710.179,
      "text": " So I can just go along and take off the shelf your favorite density dynamics, this could be Richard Feynman's pathological formulation, it could be a master equation and kinetic models, it could be the Fokker-Planck equation, whatever you like, they're all at this level expressions of the same thing, they just basically express the rate of change of this probability density in terms of the amplitude gamma of random fluctuations"
    },
    {
      "end_time": 1766.271,
      "index": 72,
      "start_time": 1739.957,
      "text": " and the flow, the flow through state space at any point in that state space. So I've just written that down in terms of the Fokker-Planck equation. But note, because we want to describe things that have possessed this attracting set or have a non-equilibrium steady state solution, the rate of change of this probability distribution is zero, which means I can now solve the density for the density dynamics."
    },
    {
      "end_time": 1790.06,
      "index": 73,
      "start_time": 1767.568,
      "text": " Perhaps I should just say that this is also one expression of a time independent Schrodinger equation, so I can just get the solution to that. And that allows me to do something quite interesting. It allows me to express the dynamics, the flow of the states as a function of where I am in state space as a mixture of gradient flows"
    },
    {
      "end_time": 1819.94,
      "index": 74,
      "start_time": 1790.282,
      "text": " on the gradients subtended by the log of this probability, basically a potential erosion. In my world, we would call this self-information, which is nice because it's all about self-organization, also called by people like Tribus, surprise or more simply surprise or the negative of it is surprise or surprise and self-information. This partition of these two kinds of"
    },
    {
      "end_time": 1848.404,
      "index": 75,
      "start_time": 1820.179,
      "text": " or a generalization of the Helmholtz decomposition and all it's saying is that the flow at any one point on this attracting set, this pullback attractor, this manifold can be divided into two parts. One is a gradient flow up log probability gradients or down potential gradients"
    },
    {
      "end_time": 1857.142,
      "index": 76,
      "start_time": 1848.831,
      "text": " Intentional or probability contours, so often called"
    },
    {
      "end_time": 1879.753,
      "index": 77,
      "start_time": 1858.575,
      "text": " solenoidal flow in virtue of that or divergence free flow or conservative flow whereas this part is the dissipative dynamics of flow and it has curl free aspects to it. So that circular flow I think I sort of highlight that"
    },
    {
      "end_time": 1905.23,
      "index": 78,
      "start_time": 1879.991,
      "text": " simply because of our discussion previously about the importance of replication, repetition, revisiting the neighborhood of states that we were once in, those states that are characteristic of the kind of thing that I am. This rests upon the non-linearities in these systems, these random dynamical systems that manifest as this seronoidal circular flow"
    },
    {
      "end_time": 1921.374,
      "index": 79,
      "start_time": 1905.23,
      "text": " I think that's quite important though it wasn't something i was going to foregrounded this you do realize that we're not gonna do this in thirty minutes or an hour."
    },
    {
      "end_time": 1948.951,
      "index": 80,
      "start_time": 1921.578,
      "text": " The designated time will keep going as long as you can keep going and then the audience will have questions. Anyhow, we can always come back for a part two. I think we're going to have to come back for a part two at this rate. It's going to probably be part three. I think we're going to have to come back for a part three because you're drilling down some of the really key issues which speak to the nature of the of"
    },
    {
      "end_time": 1965.367,
      "index": 81,
      "start_time": 1949.343,
      "text": " The kind of systems that we're trying to account for, so not only do we have this sort of solenoidal aspect and this itinerancy and the randomness that"
    },
    {
      "end_time": 1995.265,
      "index": 82,
      "start_time": 1965.657,
      "text": " is underwritten by pullback attractors but also what implicitly we've spoken about is the fact that these systems exist at multiple scales and each scale provides a context in the scale below and they all have this sort of itinerant non-equilibrium steady state aspect but the steady state is never actually attained but is slowly going towards these solutions at different timescales so an important aspect of"
    },
    {
      "end_time": 2022.227,
      "index": 83,
      "start_time": 1995.674,
      "text": " this construction is that there is a scale above so we're talking before about my this being my heartbeat that scale is going to be much much longer and larger than the scale associated with the depolarization of the myocardium or any particular cell in my myocardium that could be responding to"
    },
    {
      "end_time": 2046.323,
      "index": 84,
      "start_time": 2022.227,
      "text": " Inputs and outputs are much faster timescale say oscillating very very fast frequencies much more quickly than the slow second to second evolution of the system at the level of the heart itself and what that means is that from the point of view of one cell."
    },
    {
      "end_time": 2068.968,
      "index": 85,
      "start_time": 2046.544,
      "text": " The context in which it is operating is largely unchanging by appeal to an adiabatic approximation, which means that there is a solution at that temporal scale of the single cell to its fast dynamics, so it can attain its particular attracting set at this phase in the cardiac cycle."
    },
    {
      "end_time": 2097.892,
      "index": 86,
      "start_time": 2069.531,
      "text": " however of course the phase of the cardiac cycle is itself slowly changing so the pullback attractor from the point of view of single cell is slowly moving all the time and of course exactly the same maths underwrites darwinian selection so you've got your phenotype which is the single cell you know from the point of view of evolution or transgenerational dynamics"
    },
    {
      "end_time": 2120.981,
      "index": 87,
      "start_time": 2098.234,
      "text": " The phylogenetic time course or time scales. The organism is changing very, very quickly as it grows and behaves and acts and decides and develops and dies. But from the point of view of the phenotype, the change in the genotype"
    },
    {
      "end_time": 2144.377,
      "index": 88,
      "start_time": 2121.476,
      "text": " is so slow that's irrelevant so the phenotype the creature the organism at a very small scale now can be described as if it is conforming to its own little pullback attractor the states that i described before about getting up and doing my emails and having my cup of coffee that's only true while i am alive"
    },
    {
      "end_time": 2174.002,
      "index": 89,
      "start_time": 2144.582,
      "text": " or why I am in a position to read emails and drink my cup of coffee. So that attractor manifold has a if you like an expiry date on it but because of the adiabatic approximation we can assume that at any given scale the scale above is changing so slowly that we can assume the existence of the solution to these dynamics and therefore"
    },
    {
      "end_time": 2190.845,
      "index": 90,
      "start_time": 2174.411,
      "text": " and apply the solution of the Pocaplank equation so i'm i'm wittering on about that because i think there's a beautiful connection with darwinian thinking here once you put this density dynamics of the kind that you would you know pursue in terms of um"
    },
    {
      "end_time": 2216.237,
      "index": 91,
      "start_time": 2191.118,
      "text": " quantum mechanics if you read this as a Schrodinger equation. You can pursue this and what you end up with is a variational or a density dynamic approach to natural selection in and of itself. Lots of people have written about this and it's certainly a current theme as far as I know in theoretical biology."
    },
    {
      "end_time": 2227.449,
      "index": 92,
      "start_time": 2216.237,
      "text": " Reading things like say the replicator equation as effectively exactly the same mechanics that people"
    },
    {
      "end_time": 2250.333,
      "index": 93,
      "start_time": 2228.234,
      "text": " Doing inference and basic filtering would use so there's an opportunity to have a completely distinguish just talking about the mathematical and formal correspondences between the story i was going to tell about basic mechanics and the story that somebody"
    },
    {
      "end_time": 2275.572,
      "index": 94,
      "start_time": 2250.606,
      "text": " Taking a Bayesian perspective on Darwinian dynamics would tell it's the same story basically but anyways in order to tell that story I have to make a link now between this sort of universal dynamics expressed in terms of a Helmholtz decomposition which we can read as"
    },
    {
      "end_time": 2304.957,
      "index": 95,
      "start_time": 2275.845,
      "text": " an admixture of gradient flows up non-probability gradients up adaptive fitness or marginal likelihood if you like and this circular replicator like dynamics that is associated with the conservative or the devotions free flow or dynamics. How does that read as inference and Bayesian filtering or"
    },
    {
      "end_time": 2332.039,
      "index": 96,
      "start_time": 2305.247,
      "text": " That's where the Markov blanket comes in. So if we now go back and just remember that previously we'd just been talking about any arbitrary and exceedingly large state space X. Now we come to partition X into the external, internal and intervening blanket states and furthermore just focus on"
    },
    {
      "end_time": 2362.432,
      "index": 97,
      "start_time": 2332.978,
      "text": " two of those subsets namely the internal states of any given system say me and my active states and then by construction if you remember the unique thing about the internal states and the external states is that they are not influenced by the external states however that Helmholtz decomposition still has to be present which means that i can write down"
    },
    {
      "end_time": 2383.114,
      "index": 98,
      "start_time": 2362.892,
      "text": " The dynamics of my internal states say or my neuronal dynamics and my motion of my actuators, my active states, my muscles and autonomic reflexes as in terms of this Helmholtz decomposition where the gradients are supplied"
    },
    {
      "end_time": 2399.616,
      "index": 99,
      "start_time": 2383.609,
      "text": " The reading of this and it is just an interpretation"
    },
    {
      "end_time": 2418.49,
      "index": 100,
      "start_time": 2400.23,
      "text": " is in terms of perception and action respectively and the perception part now yields to an interpretation in terms of sense making and Bayesian inference which is why I was talking about the emergence of Bayesian mechanics from"
    },
    {
      "end_time": 2444.684,
      "index": 101,
      "start_time": 2418.916,
      "text": " The master equation on this time independent situation equation or the fucker plank equation if you're given the Markov blanket petition. In other words, if something exists in the sense that it possesses this distinguishing aspect in terms of possessing a Markov blanket, then it must comply with this"
    },
    {
      "end_time": 2474.633,
      "index": 102,
      "start_time": 2444.957,
      "text": " these its internal and active states sometimes referred to as autonomous states of this particle or person must conform to this dynamics here so now the game is how would one interpret this how will you describe this to your students or your professors or your children and there are lots of ways that you can do that so I've just listed a few here I'm sure you will have come across all of these"
    },
    {
      "end_time": 2503.677,
      "index": 103,
      "start_time": 2474.906,
      "text": " But just to again reinforce this point that this is foundational in the sense that it is assumed or described by many theories of self-organization and the particular accounts of behavior, things that actually move. So what is this quantity? Long probability of my sensory states given a Markov blanket. Well, we've just said"
    },
    {
      "end_time": 2530.35,
      "index": 104,
      "start_time": 2504.036,
      "text": " that these states belong to an attracting set the pullback attractor so these are the sensory states that are attractive to me in the sense that they score those kinds of states i expect myself to to be in if you know if i was darwinian or darwin i would say that these are the adaptive states they are the the states that"
    },
    {
      "end_time": 2553.575,
      "index": 105,
      "start_time": 2530.674,
      "text": " I repeat our characteristic of me as a surviving phenotype given I've got this pullback attractor and it is permitted by the context in which or the universe in which this attractor is manifest. So it just scores the attractiveness or the value to me of being in this particular sensory state here or the value in this sensory state space."
    },
    {
      "end_time": 2582.517,
      "index": 106,
      "start_time": 2554.053,
      "text": " So all this equation is saying is that both perception action is going to try and maximize value and that can now be read as reinforcement learning. If I was an engineer it would be called optimal control theory and if I was an economist it would just be expected utility theory where the reward the control objective or the utility just is this log probability of being in a state that is characteristic of the kind of state that I find myself in."
    },
    {
      "end_time": 2591.766,
      "index": 107,
      "start_time": 2584.019,
      "text": " The negative of this thing is what i was referring to before which is the self-information also known as surprise and surprise."
    },
    {
      "end_time": 2617.927,
      "index": 108,
      "start_time": 2592.415,
      "text": " So this equation says that i'm trying to it looks as if i am trying to minimize my surprise or self-information and from that we can read off the principle of maximum mutual information or the informax principle it could be the minimal redundancy principle of horospalo and the free energy principle. So where does the free energy principle get into the game? Well the variational free energy"
    },
    {
      "end_time": 2634.292,
      "index": 109,
      "start_time": 2617.927,
      "text": " Is a proxy for the fact it's actually an upper bound on the self information so when people talk about the free energy principle as systems trying to minimize the free energy what they mean is basically systems that."
    },
    {
      "end_time": 2648.933,
      "index": 110,
      "start_time": 2635.162,
      "text": " Conform to this foundational Helmholtz decomposition of their dynamics where you've just substituted the free energy for the Lagrangian, the potential self-information or the surprise."
    },
    {
      "end_time": 2670.947,
      "index": 111,
      "start_time": 2649.497,
      "text": " Just a moment, sorry. Professor, you keep saying, well, before we explain this to our children, some non-trivial task, to put it lightly, you keep saying given a Markov blanket, now what's giving us the Markov blanket? Is it the causal structure of the universe? Is it our internal versus active states, or does the Markov blanket give rise to those instead?"
    },
    {
      "end_time": 2682.517,
      "index": 112,
      "start_time": 2670.947,
      "text": " I understand there's some interplay, I understand that, but what gives us initially the Markov blanket? Right, well, I think you're now touching upon the tautology that I"
    },
    {
      "end_time": 2709.309,
      "index": 113,
      "start_time": 2683.507,
      "text": " Introduced right at the beginning. So when I say given I just mean technically conditioned upon. So that bar in the notation in my world just means given or conditioned upon something. So more explicitly everything I am saying depends upon"
    },
    {
      "end_time": 2738.473,
      "index": 114,
      "start_time": 2709.701,
      "text": " the existence of a Markov blanket. So your question I think is slightly deeper than that. So the question I heard was does the Markov blanket emerge from this kind of dynamics or does this kind of dynamics or these kinds of dynamics create a Markov blanket?"
    },
    {
      "end_time": 2767.09,
      "index": 115,
      "start_time": 2739.138,
      "text": " All we are doing is describing the dynamics of any system that possesses a Markov blanket. So this is a little bit like it's deflation in the following sense. So what the three energy principle is saying is not that in order to survive, you must maximize value or adaptive fitness or marginal likelihood."
    },
    {
      "end_time": 2790.299,
      "index": 116,
      "start_time": 2768.49,
      "text": " That's not what the free energy principle is saying. What it's saying is if you survive in the sense of possessing a pullback attractor, then you will always be described as if you are maximizing value, adaptive fitness or marginal likelihood."
    },
    {
      "end_time": 2817.978,
      "index": 117,
      "start_time": 2790.742,
      "text": " That is the tautology that I was appealing to in the Darwinian sense. So we're assuming all we're doing is trying to convince a physics basically what this ends up being is a variational principle of least action that can be applied to something where the thing in question is very carefully defined technically in terms of a Markov blanket. So by saying that something exists"
    },
    {
      "end_time": 2840.452,
      "index": 118,
      "start_time": 2818.609,
      "text": " Then the principle applies simply because the thingness is defined stipulatedly by being able to differentiate. The time average of this self information is entropy. Which means that these equations could also be read as a holy grail of self organization in the sense that they look as if."
    },
    {
      "end_time": 2855.367,
      "index": 119,
      "start_time": 2840.811,
      "text": " On average they're trying to minimize entropy and then from that you can elaborate synergetics of the kind that Herman Haken has established and if I was a physiologist it's just an expression of homeostasis."
    },
    {
      "end_time": 2884.275,
      "index": 120,
      "start_time": 2855.691,
      "text": " It's just keeping your essential variables, your physiological variables, within viable bounds that are valuable for me, that are characteristic of my particular states. So resisting the dispersion of the random fluctuations by these gradient flows that look as if I tried to minimize self-information on average, trying to minimize the entropy of my sensory states."
    },
    {
      "end_time": 2905.691,
      "index": 121,
      "start_time": 2885.06,
      "text": " There's a final interpretation which licenses the rhetoric of basic mechanics. So if I was a statistician, what I would describe this quantity as is just the probability of some sensory data"
    },
    {
      "end_time": 2929.087,
      "index": 122,
      "start_time": 2906.305,
      "text": " Given conditioned upon my mark off blanket then i can now read as a model so my my blanket and my internal states now can be read as a model of the external states so i can now equivalently describe this Helmholtz decomposition of the dynamics"
    },
    {
      "end_time": 2954.138,
      "index": 123,
      "start_time": 2929.462,
      "text": " In terms of the Bayesian brain hypothesis or evidence accumulation and a popular one in the neurosciences and the cognitive sciences is predicted coding. So this has been taken even further in philosophy that you can read these equations as essentially doing a gradient descent on model evidence."
    },
    {
      "end_time": 2972.278,
      "index": 124,
      "start_time": 2954.616,
      "text": " Model evidence is also known as a marginal likelihood. Why marginal? Well, because you've marginalized out all the causes of your sensory states or your sensory data, your sensory impressions, the impressions on the sensory sector of your Markov blanket, by which I mean"
    },
    {
      "end_time": 2999.377,
      "index": 125,
      "start_time": 2972.688,
      "text": " Those causes that are the external states are not part of this probability density. So you've effectively integrated out or averaged away or marginalised them. So this is also known as the marginal likelihood or the model evidence. The likelihood of these data given me as a model of the world, the external states that caused these data. That has been"
    },
    {
      "end_time": 3022.415,
      "index": 126,
      "start_time": 2999.599,
      "text": " Described or summarized in philosophy and self-evidencing so i'm minimizing my self-information just is Maxima acting and um perceiving in a way where perceiving is associated with my internal dynamics in a way that maximizes my the evidence for my model of"
    },
    {
      "end_time": 3048.848,
      "index": 127,
      "start_time": 3022.637,
      "text": " the world or the external states. Put that another way, this self-evidencing just means that it looks as if I'm going around gathering evidence from my model of the world that just is me. So I'm just gathering evidence for the fact that I exist. So that's the most deflationary but poetic reading of this Bayesian mechanics. So"
    },
    {
      "end_time": 3077.363,
      "index": 128,
      "start_time": 3049.189,
      "text": " That would be, that gives you just a flavor of the different ways in which you can understand ways of describing the dynamics of things where the thing in virtue of its existence cannot have any other dynamics and that can be sort of neatly summarized in terms of the existence of a"
    },
    {
      "end_time": 3102.961,
      "index": 129,
      "start_time": 3077.688,
      "text": " The system's states, the systemic states into internal blanket comprising sensory and active and external states that are hidden behind the Markov blanket from the point of view of the internal states. And because the active states change, but are not changed by the external states, they look as if they're going to reduce the entropy of the blanket states."
    },
    {
      "end_time": 3129.787,
      "index": 130,
      "start_time": 3103.763,
      "text": " This means that action will appear to maintain the structural and functional integrity of the Markov blanket and one can read this in terms of self-assembly in computational chemistry or in biology as self-creational or autopoiesis in a very elemental form. Finally, internal states appear to infer the hidden causes of sensory states by increasing"
    },
    {
      "end_time": 3146.92,
      "index": 131,
      "start_time": 3130.64,
      "text": " Bayesian evidence or model evidence or marginal likelihood and actively influence those causes and in my world we refer to this as active influence taking this self-evidencing perspective"
    },
    {
      "end_time": 3167.875,
      "index": 132,
      "start_time": 3147.21,
      "text": " Afforded by the interpretation of the underlying self-information as effective the log of the or the negative log of model Bayesian model evidence. So that was the first half of the talk. I now notice that we've been"
    },
    {
      "end_time": 3181.527,
      "index": 133,
      "start_time": 3168.319,
      "text": " Well over our allotted time. I'm going to suggest that we stop here and then you can wax a lyrical and ask questions for a few minutes and then we should rebook part two."
    },
    {
      "end_time": 3206.544,
      "index": 134,
      "start_time": 3181.869,
      "text": " To tell the same story, but now just using the concepts and the rhetoric and the the constructs from neurobiology and psychology. Okay, so then this is actually a natural stopping point because the latter half of the talk is using the first half of the talk to explain neurobiology and psychology. Absolutely. Yeah, we're just going to basically, you know,"
    },
    {
      "end_time": 3222.517,
      "index": 135,
      "start_time": 3207.159,
      "text": " Yeah, interpret the same gradient flow held as decomposition, but through the lens of somebody looking at electrophysiology and sense making in animals and indeed psychology or possibly even neurophilosophy. Yeah, absolutely."
    },
    {
      "end_time": 3251.527,
      "index": 136,
      "start_time": 3222.756,
      "text": " OK, so one of my questions was this Fokker Planck equation. I'm unsure what you're saying is the relationship between that and other density equations like in quantum mechanics, Schrodinger's equation or the Feynman path integral. Are you saying that you can derive the Feynman path integral from the Fokker Planck equation or that the Fokker Planck equation is a density equation of the same sort that follows the calculus of variations? Like, is that what unites them? Yeah, just that there's minimizing something or maximizing something."
    },
    {
      "end_time": 3281.203,
      "index": 137,
      "start_time": 3252.073,
      "text": " Yeah, well I think the latter expression, there's just ways of articulating the same thing. So you can either write down the dynamics in terms of a random dynamical system, in terms of a Langevin equation, specify the equations of motion in terms of these flows and the statistics on random fluctuations, and then it is a fairly trivial matter to"
    },
    {
      "end_time": 3299.855,
      "index": 138,
      "start_time": 3282.261,
      "text": " move to either a Fokker-Planck description of the implicit probability densities of this random dynamical system that can be articulated either in terms of a pathological formulation or in terms of a Fokker-Planck equation."
    },
    {
      "end_time": 3323.695,
      "index": 139,
      "start_time": 3300.077,
      "text": " Okay, so then it's the claim that the Langevin equation is the ultimate equation that gives rise to the other equations. I think in terms of what is the seed that gives birth to the rest. I think, well, for me, I mean, you know, if you spoke to other people, you may get a different answer. But certainly for me, the seed is the Langevin equation. It is"
    },
    {
      "end_time": 3342.278,
      "index": 140,
      "start_time": 3324.326,
      "text": " Just a description of a random dynamical system. It's random because you've got these random fluctuations that in my world distinguish themselves from movements in state space"
    },
    {
      "end_time": 3372.5,
      "index": 141,
      "start_time": 3342.841,
      "text": " In virtue of just being very, very, very fast, so fast that they cannot be observed. So you can't actually observe them. All you can do is write down a probability distribution, say the mean and variance. So for me, it would start here. Given that I can then write down the Fokker-Planck equation. So specifically, the amplitude, the statistics, the sufficient statistics of the random fluctuations just are the gamma here."
    },
    {
      "end_time": 3399.599,
      "index": 142,
      "start_time": 3372.961,
      "text": " And I know F. So if I know the statistics, the random fluctuations and the flow, then I can now just articulate this in terms of the Fokker-Planck equation that is just a function of the statistics, the random fluctuations and the flow. And from that, you can or you could express it in terms of a pathological formulation."
    },
    {
      "end_time": 3418.592,
      "index": 143,
      "start_time": 3400.52,
      "text": " So that is the seed and then we move to the Fokker-Pank equation and then make a very very simple and important move. The move is that this is equal to zero because the things that we want to understand are"
    },
    {
      "end_time": 3448.063,
      "index": 144,
      "start_time": 3418.968,
      "text": " Possess a solution to the focal bank equation that they have a solution to the density dynamics in the same way that the time independent scrolling equation, which would be the equivalent if you were working. Well, perhaps I can just show you what I meant by that, which will be the equivalent in quantum mechanics. You're just looking at the solutions with a little preview of what we might have the opportunity to go through in sessions three. But what I wanted to show you"
    },
    {
      "end_time": 3477.688,
      "index": 145,
      "start_time": 3448.353,
      "text": " Well this is just another intuition as to the importance of this Helmholtz decomposition in terms of self-organizing systems countering the random dispersion due to random fluctuations by imagining somebody dropping a blob of ink into a cup and the ink molecules gathering themselves up by flowing up the concentration"
    },
    {
      "end_time": 3507.244,
      "index": 146,
      "start_time": 3478.08,
      "text": " The special cases of the solution to the Fokker-Planck equation that obtain"
    },
    {
      "end_time": 3536.084,
      "index": 147,
      "start_time": 3507.483,
      "text": " When the amplitude of the random fluctuations goes away, you're just left with the solenoidal flow and that's just classical mechanics. So that's just things where there are lawful relationships between position and the velocity and mass and the like that emerge just because of this circular solenoidal conservative dynamics. Or the random fluctuations can get so big when you get very, very small because you're"
    },
    {
      "end_time": 3559.821,
      "index": 148,
      "start_time": 3536.442,
      "text": " under the renormalization group or your random fluctuations haven't yet been averaged away because you're too small and then you get sort of thermodynamics where it's all about the gradient flow and from that you can derive fluctuation dissipation theorems and statistical or specifically stochastic mechanics. For"
    },
    {
      "end_time": 3575.503,
      "index": 149,
      "start_time": 3560.179,
      "text": " If you wanted to go quantum, you just factorise this probability density in terms of complex roots and you get from this the time independent Schrodinger equation."
    },
    {
      "end_time": 3592.312,
      "index": 150,
      "start_time": 3576.015,
      "text": " Who was the first to point this out? See these three different colors that you have here, the three different derivations. Who was the first to point out that they're all reflections under different assumptions, four actually, that they're all under the Langevin equation, which is then decomposed in various ways."
    },
    {
      "end_time": 3618.729,
      "index": 151,
      "start_time": 3592.398,
      "text": " This is the Bayesian mechanics that is exactly the same as these things, but you've just now split your x's into internal and active states. I've shown the fourth one because the Bayesian mechanics is just the same as that. I don't know, the last time I was sitting in a lecture theatre"
    },
    {
      "end_time": 3646.459,
      "index": 152,
      "start_time": 3618.729,
      "text": " So you've at least independently stated this? Oh yeah, but I'm assuming it's common knowledge. Certainly. I mean, you'll know better than I will. Certainly the equivalence between the pathological formulation and the Fokker-Planck equation and the time independent Schrodinger equation."
    },
    {
      "end_time": 3672.108,
      "index": 153,
      "start_time": 3646.8,
      "text": " Sorry, sir, to maximize on the time as this is a pun, actually, at the time that we have left, I want to talk about maximization or minimization. You said that. Let's take Darwin as an example."
    },
    {
      "end_time": 3702.073,
      "index": 154,
      "start_time": 3672.568,
      "text": " Survival of the fittest sounds like if you're the fittest, you will survive. But then you said, well, if you were to survive, it would appear as if you were the fittest. And those two aren't the same statements. And rather, what's happening is the latter. So help me understand how those two sentences are different. I think they're only different in the sense that the notion that I survive because I have a high adaptive fitness"
    },
    {
      "end_time": 3725.725,
      "index": 155,
      "start_time": 3702.466,
      "text": " implies a teleology. So you can have a completely teleologically or telonomy or teleology free account of self-organization under the free energy principle and also quantum mechanics and stochastic and classical mechanics. Okay. So you weren't saying that, let's say we take a, let's say we take this cup and"
    },
    {
      "end_time": 3752.807,
      "index": 156,
      "start_time": 3726.169,
      "text": " If I was to let go, it would just fall and it would fall because it's minimizing action. Now for people who are listening, who don't understand what action is, let's say it's minimizing energy. We don't need to concern ourselves with the difference between two different types of energy being action. It's minimizing something, whether it's energy or action. So that's one way of saying it, that this cup minimizes its action. But then another way from what I understood of what you were saying is that I will only perceive objects that minimize their action."
    },
    {
      "end_time": 3777.244,
      "index": 157,
      "start_time": 3753.046,
      "text": " In other words, this cup could have done plenty of other actions or there could be many other cups, but my perceptual system for whatever reason is only tuned to what minimizes and therefore it's not that what survives is the fittest is only that I'll see what's maximizing the fitness as surviving or I'll only see this cup move in this direction because it minimized something. I think that's, um,"
    },
    {
      "end_time": 3806.749,
      "index": 158,
      "start_time": 3777.705,
      "text": " The things out there conform to a path of least action because they exist and therefore we will perceive them as pursuing paths of least action."
    },
    {
      "end_time": 3822.449,
      "index": 159,
      "start_time": 3807.125,
      "text": " So that's all the free energy principle is saying. It's just saying that we pursue paths of least action. And what could one apply that to? Well, adaptive fitness, for example. So Darwinian thinking."
    },
    {
      "end_time": 3844.206,
      "index": 160,
      "start_time": 3823.541,
      "text": " Adaptive fitness is just the thing that supplies the potential in the action, which is time times potential. So you're maximizing the adaptive fitness or you're minimizing the negative adaptive fitness. The nice point of content that comes into the way you perceive the cup though, I think reflects the fact that"
    },
    {
      "end_time": 3873.012,
      "index": 161,
      "start_time": 3844.667,
      "text": " When you put the Markov blanket in place, don't forget about doing that. You can't talk sensibly about perceiving apples unless you've got an observer and something that's being observed. So you've immediately invoked a Markov blanket. So if you're talking to Stephen Wolfram, he'd call this the observer problem. To solve the observer problem, to even talk about it, you've got to have a Markov blanket. So there has to be a distinction between the observer and the observee. The measurement problem, the same thing."
    },
    {
      "end_time": 3900.828,
      "index": 162,
      "start_time": 3873.319,
      "text": " quantum physics. You have to have the Markov blanket explicitly in play. So once you've got the Markov blanket in play, then the adaptive fitness is just the measure of synchrony or isomorphism between the inside and the outside. So literally, if you take this basic mechanics perspective, the adaptive fitness is the ability of you to fit into your environment, by which I mean"
    },
    {
      "end_time": 3924.462,
      "index": 163,
      "start_time": 3901.323,
      "text": " The sensory impressions provided by your environment, the feedback that the environment is giving you was the most likely for the kind of thing that you are, the phenotype that you are. So adaptive fitness is just the marginal likelihood of the environment providing that sensory, the way that the environment influences you by the sensory states."
    },
    {
      "end_time": 3949.957,
      "index": 164,
      "start_time": 3924.804,
      "text": " is the measure of adaptive fitness and it's just the fit of your model the ability of your model to fit this world so it all i think um it does actually come back to you say i can only perceive falling apples because apples only fall yes if i lived in a universe where apples didn't which did not comply with a principle of least action which is not impossible you know i couldn't live in a very um"
    },
    {
      "end_time": 3980.503,
      "index": 165,
      "start_time": 3950.998,
      "text": " Small universe where random fluctuations cause things to depart enormously from paths of least action and I would still be able to model that but you know in this instance I would be able to perceive random fluctuations of the apple or itinerant dynamics which violated say conservative dynamics. So I have one more question and I know you have to get going. If the free energy principle is true by definition of tautology"
    },
    {
      "end_time": 4007.637,
      "index": 166,
      "start_time": 3981.015,
      "text": " Then why is there so many papers published on it? There has to be something more to the free energy principle than just being a tautology. So in other words, Kant has analytic truths, but there must be something here that's synthetic. Otherwise, I don't understand how it can be rich if it's not purely analytic. Well, I never said it was rich. I know you're not saying it's rich. The fact that we've even spent over an hour on it."
    },
    {
      "end_time": 4036.681,
      "index": 167,
      "start_time": 4007.807,
      "text": " and that there's more than one slide and that there are papers and that you're the highest cited neuroscientist in the world implies that it's rich. So help me understand that. I can. There is an answer, but given I've already now got three minutes left, I think that's an excellent, an excellent place we should start. I'll give you a clue. Um, you can use the free energy principle in the same sense. You can use Hamilton's principle of least action to design automobiles, trajectories,"
    },
    {
      "end_time": 4051.254,
      "index": 168,
      "start_time": 4037.21,
      "text": " physical artifacts."
    },
    {
      "end_time": 4078.319,
      "index": 169,
      "start_time": 4051.596,
      "text": " Free energy flows or the base basic mechanics to reproduce self-organization and in principle, intelligent kind of self-organization. So you can create and simulate self-organization because you know the principles that underwrite the dynamics. So that's a clue, but I can unpack that in more detail the next time. Let's uncompress it next time. And I see Hamilton's equation as synthetic."
    },
    {
      "end_time": 4097.688,
      "index": 170,
      "start_time": 4078.746,
      "text": " Not just analytic, but we can talk about the distinction next time, sir. Okay. This was so much fun. It went by like that. Thank you. I love talking to you. So I'll see you hopefully in a few days or weeks depending upon your itinerary. Yes. Okay. Brilliant. Thank you, sir."
    },
    {
      "end_time": 4113.592,
      "index": 171,
      "start_time": 4098.643,
      "text": " Firstly, thank you for watching. Thank you for listening. There's now a website, curtjymungle.org, and that has a mailing list. The reason being that large platforms like YouTube, like Patreon, they can disable you for whatever reason, whenever they like."
    },
    {
      "end_time": 4140.06,
      "index": 172,
      "start_time": 4113.865,
      "text": " That's just part of the terms of service. Now, a direct mailing list ensures that I have an untrammeled communication with you. Plus, soon I'll be releasing a one-page PDF of my top 10 toes. It's not as Quentin Tarantino as it sounds like. Secondly, if you haven't subscribed or clicked that like button, now is the time to do so. Why? Because each subscribe, each like helps YouTube push this content to more people like yourself"
    },
    {
      "end_time": 4158.558,
      "index": 173,
      "start_time": 4140.06,
      "text": " Plus, it helps out Kurt directly, aka me. I also found out last year that external links count plenty toward the algorithm, which means that whenever you share on Twitter, say on Facebook or even on Reddit, etc., it shows YouTube, hey, people are talking about this content outside of YouTube, which in turn"
    },
    {
      "end_time": 4186.8,
      "index": 174,
      "start_time": 4158.78,
      "text": " Greatly aids the distribution on YouTube. Thirdly, there's a remarkably active Discord and subreddit for theories of everything where people explicate toes, they disagree respectfully about theories and build as a community our own toe. Links to both are in the description. Fourthly, you should know this podcast is on iTunes. It's on Spotify. It's on all of the audio platforms. All you have to do is type in theories of everything and you'll find it. Personally, I gained from rewatching lectures and podcasts."
    },
    {
      "end_time": 4206.749,
      "index": 175,
      "start_time": 4186.8,
      "text": " I also read in the comments"
    },
    {
      "end_time": 4230.094,
      "index": 176,
      "start_time": 4206.749,
      "text": " and donating with whatever you like. There's also PayPal. There's also crypto. There's also just joining on YouTube. Again, keep in mind it's support from the sponsors and you that allow me to work on toe full time. You also get early access to ad free episodes, whether it's audio or video, it's audio in the case of Patreon video in the case of YouTube. For instance, this episode that you're listening to right now was released a few days earlier."
    },
    {
      "end_time": 4255.998,
      "index": 177,
      "start_time": 4230.367,
      "text": " Every dollar helps far more than you think. Either way, your viewership is generosity enough. Thank you so much. Think Verizon, the best 5G network is expensive? Think again. Bring in your AT&T or T-Mobile bill to a Verizon store today and"
    },
    {
      "end_time": 4279.633,
      "index": 178,
      "start_time": 4258.012,
      "text": " Jokes aside, Verizon has the most ways to save on phones and plans where you can get a single line with everything you need. So bring in your bill to your local Verizon store today and we'll give you a better deal."
    },
    {
      "end_time": 4309.07,
      "index": 179,
      "start_time": 4279.633,
      "text": " Amazon, Pizza Hut, Audible. How'd they get so big without soul destroying complexity? On Founders Mentality, the CEO Sessions, we're going to find out. Who's number one? It's a customer. Whose Walmart is it? My Walmart. If you looked at Audible, it was kind of like growth, growth, and then growth. It separates Amazon and AWS from anyone else. Join me, Jimmy Allen, partner of Bain & Company, to hear surprising stories from the world's greatest leaders. Subscribe to Founders Mentality, the CEO Sessions, now."
    }
  ]
}

No transcript available.