Audio Player
✓ Using synced audio (timestamps accurate)
Starting at:
Karl Friston on the Perils of Investigations into Consciousness, Derealization, and the Free Energy Principle
February 21, 2022
•
2:14:05
•
undefined
Audio:
Download MP3
✓ Synced audio available: Click any timestamp to play from that point. Timestamps are accurate because we're using the original ad-free audio.
Transcript
Enhanced with Timestamps
309 sentences
18,716 words
Method: api-polled
Transcription time: 131m 10s
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.
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.
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.
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 we'll give you a better deal. Now what to do with your unwanted bills? Ever seen an origami version of the Miami Bull?
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 Miami Verizon store today and we'll give you a better deal.
Recall that you can click on the timestamp in the description to skip this intro, as there will be a lengthy introduction. This episode may be the most important Theories of Everything video. Namely, we talk about the perils of attempting to understand consciousness, theories of everything, even the UFO phenomenon. These are aspects which almost no one talks about. Because of this, I've written some notes, so I'm going to proceed to read. You'll find that those who think of themselves as gurus tend to believe that almost all their lessons apply to almost everyone at almost all times,
though it doesn't take into account where you may be psychologically. Certain spiritual teachers or people who consider themselves to be enlightened suggest that you should understand their point of view and if you resist their truths, it's because of deleterious fear and they do so without the consideration of the mental state of you, also without a conception of salutary fear slash uneasiness. This episode serves as a cautionary tale for those who are seriously trying to investigate
fundamental reality, how to identify when you're on a nourishing track versus being off track, etc. Again, this may be the most important of all the theories of everything content. The push to not feel fear, slash not feel the ego, quote unquote, slash, etc. can be dangerous, not because the lesson is false, though it may be, but because of the social pressure around those words. In fact, there's a universally frowned upon quality called being egotistical ego is embedded in it, and also to be seen as retreating is the same as
being seen as being defensive which is the same as being fearful which is seen as being lower on the spiritual hierarchy perhaps it's true that you should feel less disquietude about certain facts though perhaps your disquietude is justified for where you are now and that's okay it's 100% okay it took me about one year to realize that in fact perhaps more than one year it took me over one year to realize that these spiritual lessons
tend to not have the physics notion of path dependence. Perhaps you shouldn't uncritically adopt an Eastern practice in a Western world and vice versa, or perhaps certain truths are dependent on the path that is where you are now rather than being independent of anything. Frankly, it's embarrassing to reveal myself as so selfish and malicious and cowardly and to do so so unguardedly to Carl Friston and even to you as I'm a
Relatively thin-skinned person, at least currently I am, but hopefully in the end it's a net benefit. The dangers of going on to journeys of investigations into consciousness, the self, computational reality, and even the phenomenon, as it can lead to a spiral down some viciously dark paths,
that are at least ostensibly irrevocable while having certainty that your new founded conception of reality is indeed truthful. While I don't like to state personal opinions because I generally don't have many strong views and the views that I do have are fickle, changing week by week sometimes, my current conclusion is that if it's not life-affirming, if it's not loving, if it's not something you can tell others without others being horrified and
Frightened for you then use that as a sign that what you're experiencing or what you're feeling or what you think is not truthful Please seek help don't suffer alone You're not alone and you can verify that simply by talking to as many people as you can if you're listening to this podcast Then it's likely because you at least have some trust in it. So trust that statement or at least test that statement out
Now onto the guest. Carl Friston is a professor of neuroscience at University College London and is one of the most cited individuals in his field. He's also the inventor of the free energy principle and is someone who's made advances in the field of active inference that's matched almost by no one. He may be one of the most insightful and brilliant people on the planet. I've been lucky enough to speak to Carl before in a substantive four-hour podcast, linked below, which is best watched at least twice to properly understand.
This episode is an outlier in the Toe Channel because this will serve as an introduction to that episode. As many after watching that were still unclear as to what the free energy principle is. If you feel like you're lost, that's okay. Let that be a theme. Understanding is less like a light switch and more like a stovetop. That is, it's not like all of a sudden you have an insight and then you understand some phenomenon. You went from zero to one.
It's more like you get a gradual increase in understanding where a modicum here makes more sense and then a smidgen here. As Wheeler says, and I like to quote, the point isn't to drink from the fire hose, but instead to get wet, trusting that your conscious befuddlement is part of the process and unconsciously more is becoming clear. Click on the timestamp in the description if you'd like to skip this intro.
My name is Kirchheim Ungel. I'm a Torontonian filmmaker with a background in mathematical physics dedicated to the explication of the variegated terrain of theories of everything from a theoretical physics perspective, but as well as analyzing consciousness and seeing its potential connection to fundamental reality, whatever that is. Essentially, this channel is dedicated to exploring the underived nature of reality, the constitutional laws that govern it, provided those laws exist at all and are even knowable to us.
If you enjoy witnessing and engaging with others on the topics of psychology, consciousness, physics, etc., the channel's themes, then do consider going to the Discord and the subreddit, which are linked in the description. There's also a link to the Patreon, that is patreon.com slash KurtGymungle, if you'd like to support this podcast, as the patrons and the sponsors are the only reasons that I'm able to have podcasts of this quality and this depth
Given that I can do this now full-time, thanks to both the patrons and the sponsors' support. Speaking of sponsors, there are two. The first sponsor is Brilliant. During the winter break, I decided to brush up on some of the fundamentals of physics, particularly with regard to information theory, as I'd like to interview Chiara Marletto on constructor theory, which is heavily based in information theory.
Now, information theory is predicated on entropy, at least there's a fundamental formula for entropy. So, I ended up taking the brilliant course, I challenged myself to do one lesson per day, and I took the courses Random Variable Distributions and Knowledge Slash Uncertainty. What I loved is that despite knowing the formula for entropy, which is essentially hammered into you as an undergraduate,
It seems like it comes down from the sky arbitrarily. And with Brilliant, for the first time, I was able to see how the formula for entropy, which you're seeing right now, is actually extremely natural. And it'd be strange to define it in any other manner. There are plenty of courses, and you can even learn group theory, which is what's being referenced when you hear that the standard model is predicated on U1 cross SU2 cross SU3. Those are Lie groups, continuous Lie groups. Visit brilliant.org slash tau, T-O-E.
to get 20% off an annual subscription, and I recommend that you don't stop before four lessons. I think you'll be greatly surprised at the ease at which you can now comprehend subjects you previously had a difficult time grokking. The second sponsor is Algo. Now, Algo is an end-to-end supply chain optimization software company with software that helps business users optimize sales and operations, planning to avoid stockouts, reduce return and inventory write downs while reducing inventory investment.
It's a supply chain AI that drives smart ROI headed by Amjad Hussein, who's been a huge supporter of this podcast since near its inception. In fact, Amjad has his own podcast on AI and consciousness and business growth. And if you'd like to support the Toe podcast, then visit the link in the description to see Amjad's podcast because subscribing to him or at least visiting supports the Toe podcast indirectly. Thank you and enjoy.
Let's do an introduction once more. So what is the free energy principle specifically, specifically its relationship to a theory of everything as that's what this channel is about? Right. So my usual response here is do you want the high road or the low road? Let's take the high road, given that we're talking about theories of everything. So
The free energy principle is a principle of least action. So it's a method really. It's put most simply a method that allows you to identify the most likely dynamics, the most likely way a process will unfold. What is the process that we're talking about here? Well, we're talking about things like you and me, particles or people. So the first thing to note is that you can
apply standard variational principles of least action to any system, specifically a random dynamical system of the kind that can be described by say a Langevin equation that underwrites most of the physics we know. But the special thing about the free energy principle is it tries to understand the relationship of something in relation to everything else. So you're starting from the premise that something exists and then you have to think carefully
What is a thing and how do I distinguish it from something else? So immediately you start, you have to introduce a partition between the states of things that comprise states that are internal to the thing and states that are outside the thing. And thereby in think carefully about how those states are coupled. And what happens is that you have to introduce a further set of states known as blanket states.
So you've got this notion now of taking the states of any universe, any system that you want to try and understand,
and partitioning it or carving it into four subsets, internal, external, and then sensory and active states that constitute the blanket states where the active states mediate the effects of the internal on the external, the inside on the outside, and the sensory states conversely mediate the influences of the external on the internal. So with this particular partition, which you wouldn't need if you wanted to now just move on and develop
say quantum mechanics or statistical mechanics or Lagrangian classical mechanics we're now in the special case of dealing with a system that has this partition and in this special case those conventional variational principles of least action translate into or can be used
to associate the paths of least action, the most likely paths that the system will take, particularly the states of the particle or person that we're talking about. We can associate those paths of least action with paths that minimize something called variational free energy. So that's basically it. And then the story is, well, what's variational free energy? Well, variational free energy is
As with all actions in physics, it's just a path integral or a time integral, a sum, an accumulation of a functional, usually known as a Lagrangian, where in this instance the functional in question, this variation of free energy, can be read as exactly the same quantity that statisticians
This Marshawn beast mode lynch prize pick is making sports season even more fun on prize picks whether
Football fan, a basketball fan, it always feels good to be ranked. Right now, new users get $50 instantly in lineups when you play your first $5. The app is simple to use. Pick two or more players. Pick more or less on their stat projections. Anything from touchdowns to threes and if you're right, you can win big. Mix and match players from
any sport on PrizePix, America's number one daily fantasy sports app. PrizePix is available in 40 plus states including California, Texas,
Florida and Georgia. Most importantly, all the transactions on the app are fast, safe and secure. Download the PricePix app today and use code Spotify to get $50 in lineups after you play your first $5 lineup. That's code Spotify to get $50 in lineups after you play your first $5 lineup. PricePix is good to be right. Must be present in certain states. Visit pricepix.com for restrictions and details. The sensory, the path through the sensory states under a particular model that would try to explain
hypothesis that would try to explain those data. So you've got this interpretation or certainly the fact that the functional form of this variational free energy that is being minimized by the passive least action of the autonomous the active and the internal states of something a particle or a person now looks as if it is trying to minimize a quantity which provides an approximation or a negative
approximation to marginal likelihood or model evidencing. So in other words, it looks as if the most likely paths are trying to maximize model evidence. So one can read that in many different ways. One can read that. And as Jacob Howey has read it, a philosopher and neuroscientist who has a commitment to understanding to this kind of formulation as self-evidencing.
literally interacting with your environment, actively inferring, engaging with your experienced world, your eco niche in a way that looks as if you're trying to solicit the most evidence for your model of that world. This quantity is also found in machine learning. It's known as an evidence lower bound. I should qualify or just clarify that
In machine learning, this free energy is the reverse of the free energy used in physics. So in machine learning, they try to maximize the negative free energy, aka an evidence lower bound or ELBO. While in physics, you're always articulated in terms of minimizing a potential energy or minimizing, in this instance, a variation of free energy. So you'll find that exactly the same functional, the same quantity
of the goodness of some model of how you think your sensations, your sensorium was generated in terms of this variational free energy that leads you to an interpretation of people, particle, persons, inferring, accumulating evidence
making inferences about the states of the world that are generating their sensations. So that in a nutshell, or perhaps not a nutshell, is a sort of formal explanation of the free energy principle. I did slip in at the beginning, very much like things which you may be familiar with when you were at school, say Hamilton's principle of spatial reaction. It is
just a method. It allows you to identify, predict, simulate, engineer those trajectories, those paths, those dynamics, those processes that will minimize this action functional or the path integral of the variational free energy. So what that means is that if you now understand or have at hand the functional form
of your variation of free energy that depends upon posterior beliefs or beliefs about the outside world under a geratin model, quite simply a probability density over the external causes of the consequences that are the sensory states. Then you can use that methodology, use that principle to start to simulate, to build sentient artifacts
that behave in the way that we assume under the free energy principle that you and I are behaving. Let's see if I can make a nutshell of the nutshell for some people who are more familiar with physics. So let's say you have a ball, you want to know the dynamics of the ball, you can understand this with Newtonian force and that's useful for certain calculations or you can think of it as minimizing a Lagrangian and then that actually provides a more useful for most cases way of calculating the trajectory of the ball. So then
Right there, when I said ball, we've identified the ball. Now here, you're not just limiting yourself to balls, you're saying let's say cells instead of talking about people, because we can break it down simply, at least for me, for myself, I find it much more easy to think of a cell. Okay, so then we want to know what are the dynamics of the cell? How is the cell, a single cell going to act? Okay, firstly, the first step would be identifying that cell. And when you mentioned the word
Markov blanket that's a way of identifying what this object is because in order to identify what an object is it's often useful to identify what it isn't at the same time what it is and isn't so that Markov blanket delimits the cell now is that correct so far absolutely yeah so you're adding the cell is a perfect example here you've got you know a cell is just defined by the boundary which would be the surface
inside the intracellular states would be the internal states, and then the milieu in which that cell lives, which is usually in a bunch or a family of other cells, would constitute all the external states. So you've made the very first step just talking about a cell, a thing. You've defined it in terms of this partition into internal, external, and the sensory states that bound and separate the inside from the outside.
You may be asking, well, where have the active states gone? For a cell, I usually like to associate those with something called actin filaments that just lie underneath the surface, the sensory states, and actively push the cell surface, say, the filia, the mechanisms that cause motion of the cell, should it move around.
those parts, those states that are responsible for cellular motility. So if we're now thinking about a moving cell or a swimming cell, for example, then you need to supplement that partition with active states. So that's absolutely the first step. And then you're going to go on and now translate the ball. What's the ball in this metaphor?
You also mentioned the word bi-directional earlier. And when I was learning about this, the way that I like to understand this was that in computer science, generally computer scientists think and can only think in terms of you have an input and then there's a black box and you have an output. And that's essentially computer science in a nutshell. It's a bit more complicated than that, but it's in a nutshell. It's you have an input, you perform some calculation and you get an output. The reason why you said bi-directional, at least the way that I understand it,
So please correct this is it's because of actions in that model where there's input black box output. That's generally if we want to use the term embodied cognition more on the cognition side. So you're just this black box where you're impressed with sensory information. The input is your senses. You're sensing all of what's happening. And then this output now.
is the actions that generally a computer sits there but we're creatures in this world at least we think we're creatures in a physical world the black box has a model and then one way of maximizing or showing that your model is correct is by updating your model or it's by changing what you've sensed in order to match the model now to change what you've sensed is to act so it's useful to think of two black boxes so there's one that's inputting to the other
and then the output goes into the other, which is external, this is the internal, this is the cell, it senses, and then it acts upon the world, and then that changes the world, which then changes what you sense. So is that the reason that you said bi-directional? Yeah, no, I think that's a really important observation, yeah. So, you know, before we were talking about these states that separate the inside from the outside, and I made a big thing that they had comprised sensory and active states, and that there is a two-way traffic here,
And that two-way traffic instantiates exactly what you're talking about, which is a circular causality. So, you know, I use the word self-evidencing and described it as soliciting evidence for sensory data that provides evidence for your model of what's going on. But in soliciting, this is something that a computer would not normally do, a passive or a sessile
would not be in charge of nor would it terribly worry or indeed be equipped with the capacity to select the inputs, to select the data, say it had to analyze. So just having a statistical package, you give it some data and it produces an outcome, say a p-value or some kind of inference. So a lot of computing and a lot of machine learning, you give it data and it provides a prediction or a classification.
That's not the kind of computational process that is implied by the free energy principle. It's much more, as you put it, inactive. It has the aspirations to provide a mathematical image of the action perception cycle, where crucially, you have to go and decide what data to base your inferences on by acting upon the world.
That sort of inactive aspect is the thing that distinguishes this kind of formalism, this calculus, this Bayesian calculus, which is quintessentially inactive. It's not just a good computer. It can be made a good computer just by removing active states if you want to. But that would be a limiting and a rather biophysically unrealizable and unsustainable kind of artifact.
The whole point here is that there is a reciprocal exchange of the world. So your notion of having two black boxes talking to each other, I think, is a really nice notion. And indeed, when applying this principle of least action, the free energy principle, to interesting scenarios, one ends up very often putting two
sets of internal states each equipped with Markov blankets talking to each other. So the active states of one blanket or one particle or one person now constitute the sensory states of the other and vice versa. So you've got now two systems trying to model each other and if their game is to effectively provide or solicit
the most evidence for their model of the rest of the world, which is just another artifact very much like you, who's trying to do exactly the same thing. What will inevitably emerge from that joint minimization of free energy or the maximization of marginal likelihood is the kind of active engagement that renders, say you and me as the two black boxes, renders as mutually predictable.
So what we will do is we will come to share the same generative model, the same narrative, so that I can predict exactly what you're going to say next, given that's the kind of thing that I would say, because that's the sort of world that I'm trying to model. So immediately you have this picture of the minimization of the joint variational free energy
or the maximization of the joint marginal likelihood for shared models. And one gets from that some quite compelling and interesting formulations of multi-agent games, communication, synchronization between conspecifics, that all rest upon trying to resolve uncertainty
Can any system, let's say it's more than two. So in the example that I gave and that we were talking about, there's one cell and then the external world. But obviously the world is much more complex than that. And let's just imagine slightly more complicated. There's two cells and then an external world.
Is it always the case that from the perspective of one cell, you may as well just treat the external world, even though it comprises an external world plus another cell out there. Can you always treat that as its own singular black box such that we can always understand the dynamics as two black boxes? Or is there ever a need for a third, a fourth, a fifth? I think there probably would be a need if the external states are themselves
So if my universe is composed of external states that can themselves be carved into natural kinds, you know, like other artifacts and objects and air, planets, liquids, natural kinds, as soon as you talk about a natural kind, you're talking about a kind of thing, as soon as there is a thing that's in play, there's a mark of blanket. So what that tells you is,
In principle, it would be possible to take all the states that constitute, say, an eco-niche.
find, identify the thing of interest, say me and my internal states and then I would identify my blanket states and then I got the rest of the states. But now I can start again, I can find another set of internal states and identify its blanket states and then take those states off the table and then start again and recursively tile all the external
Because there's a statistical insulation or separation of internal
I don't need to know your internal states because everything that is knowable about your internal states is on the surface, on your blanket states. So now the picture that emerges is an ensemble of things, natural kinds, that are coupled to each other via their blanket states.
No individual particle or person or thing will ever have access to the internal states of anything else. But there will be coupling and influences that are mediated by the blanket states, the active states, and possibly the sensory states. Mathematically, you can actually have the sensory states also influencing the outside. But for simplicity, let's assume that sensory states do not influence external states.
You mentioned something extremely interesting, and I want to pick up on this. You said that the sensory states can influence the external states. At least that's not how I understand it. So how can a sensory state influence an external state? So I said that mathematically it is that is possible. So if you if you think about, you know, from a pure from the point of view of somebody who's trying to understand
the statistical behavior, the probabilistic behavior of a set of dynamics described as differential equations or coupled systems, then the name of the game really is to ask your
What statistical dependencies that define a Markov blanket, specifically a Markov blanket, renders the internal states conditionally independent of the external states, given the Markov blanket. So that's their role. That's what is implied by a separation of the inside from the outside, that all you need to know is the blanket states, because given the blanket states, the inside and the outside are conditionally independent.
So the question now is, what kind of sparse coupling among all the states at hand give rise to that conditional independence that definitively specifies the Markov blanket? It so happens that all you need is the following rule. Internal states can only influence themselves and active states.
External states can only influence themselves and sensory states. That's all you need. So with that condition in place, you now immediately have
a Markov blanket or a particular partition. From that follows the conditional independence and therefore the interpretation of the autonomous dynamics, which is a description of the trajectories or the paths of the processes taken by the active and the internal states of a particle. So that means that it is actually mathematically allowable
in the sense that you do not destroy that Markov blanket or the statistical properties of that Markov blanket. It is allowable for the sensory states to influence active states. And you may be thinking, well, that's a bit silly. There's no way in which my photoreceptors or my cochlear in my ear materially affect the outside. I think that's absolutely true. So for things like you and me,
then you would not normally expect the sensory states to influence the external states. But that's not the case for much simpler things. When you look at much simpler things where most of the dynamical influences, the coupling that we're talking about that you'd write down in terms of the differential equations, the Langevin equation, most of those influences are basically short range.
So we can preclude action at a distance, for example. If you're a single cell, unless you're next to me, I don't really know what you're up to. I'm having to exclude here all sorts of interesting things about long-range electrochemical communication and electrical gradients and the like, but let's just take a bag of cells that
They just diffuse stuff locally. They touch each other and that's the only way that they can influence each other is by being proximate in some Euclidean space. So everything's short range. Now in this situation, you have a very different kind of structure. So here, the sensory states are the cell surface.
And when these active states that lie underneath the cell surface change the environment, they do so by pushing the cell surface into the external milieu and cause changes in the sensory states, the surface states of other cells. So being a cell is probably a good example of where your sensory states, which just are your surface states, actually do all the heavy lifting in terms of changing the outside.
the spatial relationship with all the other cells that you're co-inhabiting a particular organ, for example. So it's a little bit counterintuitive, but I think a really interesting sort of thought experiment and actually a practical consideration when it comes to looking at the different ways in which
insides coupled to outsides and the nature of interactions between the internal states of something and the external states of something and we sort of take it for granted that we can do action at a distance. Literally I can talk to you in Canada while you're talking while I'm in London. That is a beautiful example of this action to distance that means that we have quite a special
a kind of coupling between our sensory states and our active states that is not a gift of very simple cellular like structures, an automata that can't interpret signals or be subject to or be sensitive to
Just so you know, or for people who are listening and if this wasn't obvious enough, the free energy principle is famous for, or perhaps infamous, for being somewhat inscrutable. For people who have a physics background or math background, what would be the minimum prerequisites? This is how I like to conceptualize the podcast before I interview someone. I think about, okay, what are all the pieces of knowledge I need to know prior to interviewing them?
So obvious example is let's say QFT. You need to understand classical mechanics, Lorentz of variance, quantum mechanics, perhaps bundle theory if you want to understand Yang-Mills and so on properly. So what would be the prerequisites for someone to properly understand the free energy principle, let's say if they have a math or physics background? It wouldn't be, it would be probably much less than you'd expect to encounter at university.
I think you'd certainly need to know sort of, you know, the basics of variational calculus and it would be really nice if you understood where a Lagrangian came from and to understand the simplicity of action, you know, basically, you know, exactly Hamilton's principle of least action and how that inherits from patheticals of Lagrangians and how that arises from
dynamical systems theory of a very simple sort where you can write down any random dynamical system as, you know, a stochastic differential equation or a Langevin equation. I think that's probably quite sufficient, you know, what you would need, I think, to
appreciate all the interpretational richness afforded by what happens to the paths of least action when you have a Markov blanket, you need to know a little bit of probability theory, you need to know what a conditional distribution is, you need to know what have an intuition as to what is meant by model evidence for example or marginal likelihood. So with I would imagine
a week's foraging on Wikipedia, you could probably acquire that if you were fluent already in sort of empathological-like formulations of the kind you can find on Wikipedia. So you certainly wouldn't need to know what sort of bundles were, you wouldn't need category theory, you wouldn't need to know anything about gauge theories, you
Your example of Wikipedia is
is on point because the way that I conceptualize this interview, this podcast and the way that we spoke about it prior to going on air via emails is it's going to be tendril-like or gossamer-like where you have these words imagining what you're saying is like a Wikipedia page and you say certain keywords which would be if you were a Wikipedia page underlined in blue and worthy of their own investigation in and of themselves. So some of these I'm going to pick out and interrupt you
sometimes and I apologize for that like I did earlier in order to in order to properly set the prerequisites because we had a four-hour behemoth before and that was meant to serve as an introduction to the free energy principle however there are some people who were still uncertain as to exactly what the free energy principle is so this will serve as an introduction to that introduction okay hear that sound
That's the sweet sound of success with Shopify. Shopify is the all-encompassing commerce platform that's with you from the first flicker of an idea to the moment you realize you're running a global enterprise. Whether it's handcrafted jewelry or high-tech gadgets, Shopify supports you at every point of sale, both online and in person. They streamline the process with the internet's best converting checkout, making it 36% more effective than other leading platforms.
There's also something called Shopify Magic, your AI-powered assistant that's like an all-star team member working tirelessly behind the scenes. What I find fascinating about Shopify is how it scales with your ambition. No matter how big you want to grow, Shopify gives you everything you need to take control and take your business to the next level. Join the ranks of businesses in 175 countries that have made Shopify the backbone
of their commerce. Shopify, by the way, powers 10% of all e-commerce in the United States, including huge names like Allbirds, Rothies, and Brooklynin. If you ever need help, their award-winning support is like having a mentor that's just a click away. Now, are you ready to start your own success story? Sign up for a $1 per month trial period at Shopify.com
Go to Shopify.com slash theories now to grow your business no matter what stage you're in Shopify.com slash theories.
Razor blades are like diving boards. The longer the board, the more the wobble, the more the wobble, the more nicks, cuts, scrapes. A bad shave isn't a blade problem, it's an extension problem. Henson is a family-owned aerospace parts manufacturer that's made parts for the International Space Station and the Mars Rover.
Now they're bringing that precision engineering to your shaving experience. By using aerospace-grade CNC machines, Henson makes razors that extend less than the thickness of a human hair. The razor also has built-in channels that evacuates hair and cream, which make clogging virtually impossible. Henson Shaving wants to produce the best razors, not the best razor business, so that means no plastics, no subscriptions, no proprietary blades, and no planned obsolescence.
It's also extremely affordable. The Henson razor works with the standard dual edge blades that give you that old school shave with the benefits of this new school tech. It's time to say no to subscriptions and yes to a razor that'll last you a lifetime. Visit hensonshaving.com slash everything. If you use that code, you'll get two years worth of blades for free. Just make sure to add them to the cart.
Plus 100 free blades when you head to h-e-n-s-o-n-s-h-a-v-i-n-g dot com slash everything and use the code everything. You mentioned marginal likelihood slash model evidencing and I believe you see those two as the same and I'm unclear as to what any of those two are. So can you give an explanation? What is model evidencing slash marginal likelihood?
Excellent question. So this is where you would need to be slightly fluent in the concepts and constructs that statisticians would bring to the table. So if I now want to use probability theory to articulate the job of a statistician trying to make sense of data, just note
You'd start off by understanding the problem in terms of estimating under uncertainty, sort of assimilating data whilst quantifying and accommodating
uncertainty due to random effects. So, you know, the random effects here can be read as the random fluctuations or the vena fluctuations or the innovations in some stochastic process from the point of view of a physicist, from the point of view of a statistician. That randomness is inherited as random effects. Observation noise, for example, would be one if you were dealing with
a state space model, which is really a model of a dynamical process that a statistician might use to do data assimilation given some time series data. There may be a distinction between observation noise or sensor noise and the noise or the random fluctuations on the states that can't be directly observed, and that would be known as state noise or system noise,
But from the point of view of the statistician, this becomes noise or uncertainty or random effects. So how do you deal with this? Well, basically, you're confronted with a problem of some data and you have a hypothesis or a model about how those data were generated by some latent states, sometimes known as hidden states and hidden
Here, I think, just as to try and connect it to the Markov blanket, they're hidden behind the Markov blanket. So the statistician can, if you can think of it like being in internal states, they're surrounded, they're enshrouded by a Markov blanket. The data is impressing themselves on this sensory veil. There's something on the outside that are latent in the sense that they're hidden behind this veil. So they only have access to the data. They don't know how
the data were caused so for example they might have a hypothesis that this drug treatment materially affected some pathophysiology that reduced the symptoms that were measurable or recordable by the doctor but it's only the observable data that the doctor can get from the patients that the statistician has at hand now the statistician now has to infer was there a drug effect
And what she will do, she will create a generative model that if there was a drug effect of a certain size, then I'd expect a separation of the data, for example. Now, that generative model at the end of the day can always be written down as a joint probability density over the causes and consequences where the causes are the latent or the hidden states
that you are trying to estimate, and the consequences of the observable data that you have at hand. Given that generative model, you can now estimate, given some data, the most likely hidden or latent states or causes of that data, basically the treatment effect that you hypothesized. So that's the posterior distribution. That's the probability of some hidden state, say the external states, the Markov blanket,
given conditioned upon the data. So that's the holy grail usually of a statistical analysis and you know what your most classical inference for example those based on t-tests or f-tests right through to more sophisticated and informed log-odd ratios such as Bayes factors. These are just scoring
the evidence for an effective interest expressed in this latent state space in terms of the probability of one model relative to another model and usually in science that would be the altered hypothesis versus a null hypothesis and then you can look at the ratio of the likelihood of this model or more specifically the likelihood of the data given this model divided by
likelihood of the data given another model, usually the null hypothesis. So to come to our example of the statistician trying to infer whether the drug has had a remedial effect or not, what she would do is compute the probability of her data with
an effect and then compute the probability of her data if there was no effect of the drug. So the data sets from the treatment and the non-treatment group were treated as from a single group. And then she would use that to make an inference that there was indeed an effect. So that probability, the probability of the data given a particular model, a specific model, be it the alternate hypothesis or the null hypothesis, that's the marginal likelihood
So why is it called a marginal likelihood? Well, what you've done is you've estimated, given your generative model, the probability of the data having marginalized or averaged away the thing that you don't know, which is the effect size, the treatment effect. So the probability of the data given a model is the average probability of the data
given that model and the effect size averaged under your beliefs about the effect size. So that's why it's called a marginal, a marginal likelihood. That's a great question because it actually leads one to the genesis of variational free energy on one reading in the context of Richard Feynman's pathological formulation of quantum electrodynamics.
He had this marginalization problem. So he wanted to, or one could read the history as him wanting to evaluate the marginal likelihood of paths of, say, small particles, which would necessarily involve computing or marginalizing
integrating of extremely high dimensional probability distributions which is intractable. So in order to solve the marginalization or the integration problem to ensure everything sums to one effectively to get your partition function, he introduced the variation free energy and that was a really clever move because he converted what was an intractable
a non-realizable and incomputable marginalization or integration problem into a tractable optimization problem simply by introducing this variational free energy which was a bound on the marginal likelihood. So that's why it's called an elbow or an evidence lower bound simply because it now becomes a computable object or mathematical functional
that you can evaluate given some data and a generative model that is always in machine learning lower than the log of the marginal likelihood. In my world, the physics world, it's the other way around, but let's stick with the machine learning world. So the negative variational free energy of the physicist is always smaller than the thing you want to actually evaluate. So if you just optimize by pushing up
the elbow or the variation free energy until you can't make it any larger. And under the assumption that the bound, the times that the bound is high, you've not got a good approximation to what you were chasing, which is the marginal likelihood. And once the statistician has that marginal likelihood, she can repeat the process for another model. So the null hypothesis with no treatment effect. And then she can take the ratio of the free energies, the variation free energies,
and start to compare the models using base factors and then go back to the doctor and say, look, I am pretty confident that that drug really did have an effect. And I can say that because the evidence in your data suggests that the marginal likelihood of a hypothesis or a model that included an effective drug was 52.6 times greater than the marginal likelihood if I assume there had been no change.
So that's where the notion of a marginal likelihood comes from. So the marginal likelihood just is the evidence for a model. It just is the probability of some data in the present context, you know, sensory or from the part of the free energy principle, the sensory states.
or the sensory trajectories, say time series, given a particular model marginalizing away all your beliefs about the particulars, effect size, and the parameters, and all the quantities that were responsible for generating those data under the structure afforded by the model. Is that clear? I mean, these are the fundaments of
machine learning, data simulations, system identification, and you could actually argue this is one way of nuancing the scientific process, not quite in using the rhetoric of Karl Popper, but certainly has this Popperian aspect to it, that every bit of scientific inquiry
including the way we live our lives, is just about hypothesis testing. And we just generate alternative explanations for the evidence at hand. And then we try to evaluate the evidence for this hypothesis relative to that hypothesis, evaluate the marginal likelihood of the data at hand for this hypothesis versus that hypothesis, this model versus that model. And then we usually commit to the model with the greatest evidence. This just is.
evidence-based scientific progress. You're having committed to or selected, technically known as Bayesian model selection, this hypothesis. You then think about, okay, what's the next space of hypotheses or what's the next portfolio of ideas or models that I want to now go and gather evidence for?
I think this analogy is quite nice. I'm glad you let me tell this explosion because a good scientist then, the good one, will generate new hypotheses and then the problem contending the scientist who's just chasing these marginal likelihoods or model evidences for her models of her fielding of inquiry is to design a good experiment.
So here we come back to the other side of the coin that takes us beyond machine learning as a black box, give it some inputs, give it an output. We now come back to the design of our experiments, how we actively secure and solicit the evidence that will best disambiguate
among our hypotheses and I repeat usually when you're writing scientific papers it's a null hypothesis versus an altered hypothesis but in real life and in real scientific thinking people you know normally have eight hypotheses or a hundred hypotheses and they're scoring all of them by evaluating the marginal likelihood of the data that they have carefully solicited by Bayes Optimal Experimental Design
to maximally disambiguate in terms of the marginal likelihood on the model evidence between all of their hypotheses. So that sort of brings another aspect of probability theory to the table which would be an important complement for somebody wanting to understand Bayesian statistics. The base aspect of course is that
As soon as you start to talk about conditional distributions of the kind that a marginal likelihood is, it's conditional in the sense that it's a probability of some data conditioned upon a model or given a model, having marginalised away all the parameters of that model. And building a posterior, the probability of my hidden states, the parameters of my model,
given the data falls out of Bayes rule. So we're talking about Bayesian inference here and just straightforward probability theory that can be articulated in terms in terms of Bayes rule. But there's another really important application of Bayes rule which speaks to this sort of inactive situated statistician where statistician now
is not simply in the service of the doctor supplying her with data and asking a particular question. Now, the statistician is in charge of designing the perfect experiment that gives her the data that allows her to make the most efficient and statistically powerful inference about whether there was a treatment of bed. And that's the problem of optimal experimental design.
There is an answer to that, it's called the principle of Bayes' optimal design, first articulated by people like Lindley in the 1950s, and then re-articulated by people like David McCain in active learning. And it is, if you like, the other side of the Bayesian coin when it comes to
Bayesian assimilation of statistical assimilation of data and inference. And one could also argue it's the other side of the coin when it comes to things like Bayesian decision theory as well. But that's a slight distraction. I think we just stick with the big idea here that, you know, there are good. Let's look at it like this. If it is the case that the marginal likelihood is just the probability of some data given a model,
And your job is to evaluate that quantity and usually optimize the marginal likelihood or maximize the evidence for your model. You can either change the model or you can change the data. And that means that you've got sort of two problems to solve, as it were, if you cast this as an optimality problem, as a maximization of an evidence lower bound,
I can either change my mind about you or I can try and change you by querying you, asking a question, performing an experiment upon you to get some better
better data from you that resolves my uncertainty about you. Is there a relationship between changing your evidence versus changing your model and the psychoanalytic repression? So repression would be saying, you know what, that data doesn't exist, let me ignore it, versus the perhaps a cognitive behavioral therapy technique would be to say, no, go out and perhaps update your model of it, view it differently, view the maybe it's our arachnophobia, for example. I think it's a beautiful analogy. And
Hear that sound? That's the sweet sound of success with Shopify. Shopify is the all-encompassing commerce platform that's with you from the first flicker of an idea to the moment you realize you're running a global enterprise. Whether it's handcrafted jewelry or high-tech gadgets, Shopify supports you at every point of sale, both online and in person. They streamline the process with the internet's best converting checkout, making it 36% more effective than other leading platforms.
There's also something called Shopify Magic, your AI-powered assistant that's like an all-star team member working tirelessly behind the scenes. What I find fascinating about Shopify is how it scales with your ambition. No matter how big you want to grow, Shopify gives you everything you need to take control and take your business to the next level. Join the ranks of businesses in 175 countries that have made Shopify the backbone
of their commerce. Shopify, by the way, powers 10% of all e-commerce in the United States, including huge names like Allbirds, Rothy's, and Brooklyn. If you ever need help, their award-winning support is like having a mentor that's just a click away. Now, are you ready to start your own success story? Sign up for a $1 per month trial period at Shopify.com
Probably beyond an analogy, there are people in computational psychiatry who think exactly like that.
You can find articles, trends in cognitive science, reviews of the kind, you know, some that deal with, you know, how do you resolve cognitive dissonance? How do you understand wishful thinking? So examples, I think that's a perfect example of this way of maximizing
You shouldn't try to change the evidence to match your model, you should update your model.
And in fact, there are some political implications. I'm sure people shout that on the streets across various political divides. So is this an argument to say that updating the model is actually the one that should be primary? Because the way that I understand from how you've explained it is that both are equally valid. However, when I look across the world and I look at how we think just in terms of common sense, psychological advice, at least to us in the 21st century, it sounds like you should update your model. But I think you're right to say that it would be
disingenuous from a mathematical point of view to privilege the model versus the data. So how does one choose between those mathematically? I don't think you can in the sense you can't split a coin. A coin has two sides to it. The whole point of this inactivist take on sense making
is that you have to optimally go and listen to the right kind of news, talk to the right kind of people in order to solicit the right kind of data that's going to maximize the evidence for your model, whilst at the same time you are updating your model, sometimes known as basing belief, updating
or certain basic model selection, for example. So the two things go hand in hand. I think one way of resolving or answering your question is just to say in a political or perhaps even sort of communication science sense or these sort of the social aspects of it. What you're saying is that
There is now, once you acknowledge that you also are in charge of getting the data, then you also have to contend with the problems of how much weight you afford to that data versus that data and where you get your data from. Which news channels do you forage? Who do you listen to? Who do you trust? Who do you ignore?
And we do this all the time. We do this in a very skillful way, so personally, in terms of deploying our attention. So if you're an engineer and you were thinking about assimilating visual information using a Kalman filter, for example, then one of the most artful challenges or artful resolutions
that the Kalman filter brings to the table is to get the Kalman gain right. How much weight do you afford? The prediction errors that supply the upper term of a Bayesian or a Kalman filter. And we do that on the fly all the time, right down to the level of turning on and off the attention or the precision or the gain afforded
visual information. So my favorite example is, um, saccadic eye movements. Um, so just to demonstrate that, you know, one can do this experiment at home. You can just look, um, from one side of your visual scene to the other side and back and forth and back and forth with saccadic eye movements. Uh, and you will not perceive the world jumping around. However,
If you reproduce the same visual impressions just by palpating your eye in the spirit of Helmholtz. What's that palpating eye? What does that mean? Well, you can literally do this now. If you just take your finger and place it gently on the outer aspect of your eyeball and just gently press in until you see the world shift around and move around. So you can see visual motion.
that is a consequence of optic flow introduced in this instance by you pushing your retina, translating your retina with the input fixed. And that is basically exactly the same from the point of view of the retina, the back of your eyeball, as moving the eyeball with the world stationary. But notice, you only saw motion of the world
when you were moving your eyeball. When you move your eyes in the normal way, you don't see that motion. And the reason is that your brain ignores visual input during eye movement. And that's called saccadic suppression. So I think it's a beautiful example of the fact that every moment we are selectively ignoring information that is not newsworthy. We are selecting it. We don't know we're doing it, but we do it all the time.
Of course, at a much more abstract and epistemic level, exactly the same principles apply when we choose which newspapers, which Twitter accounts to follow, which friends to talk to, which universities to go to. Is it going to be a Scholarpedia or a Wikipedia? All of these decisions that we make in terms of securing the right kind of informative, precise data
that will literally have a salience in the sense that it complies with the principles of optimum Bayesian design in the sense that it will resolve the most uncertainty about what I don't know. So this is absolutely crucial, this sort of being in charge of the data.
even before thinking about the consequences of your belief updating, just knowing that you have to solicit my debt and of course from a sociopolitical point of view what we're talking about now is how to disambiguate between fake news and true news, how to
solicit data in a way that precludes as being isolationist or precludes as succumbing to cult-like ideologies, you know, precludes fundamentalism of a dysfunctional sort, or why is it dysfunctional? Well, usually because that kind of fundamentalism does not speak to and does not
So what you will find is that that dialectic between changing your mind on the basis of some evidence and changing the evidence by changing where you look for that evidence, that balance is at the heart of action and perception. And indeed, you could argue that all of action
is simply in the service of gathering better data. What do you mean by better? Those data that minimise your surprise, minimise your expected surprise or your uncertainty, resolving uncertainty about the way you think
This sounds like a deeper issue than just a political problem. It sounds like an issue for the process of suggesting that the conclusions that one has come to was from a dispassionate assessment of the evidence.
So you hear people say this, generally people who like to think of themselves as extremely rational people who've come to their conclusions because they have objectively analyzed some situation. It sounds like that is a doomed project because we're all sampling different data without realizing it. Is that the case or no? Yes, no, I think that's absolutely right. I don't think there's anything bad about it. I mean, that's just the way things are. So how does one overcome that? What do we do? How is it that the scientific process seems to work?
Well, it works because the operational definition of working simply is maximizing the marginal likelihood. So you are a liberty to actually maximize the marginal likelihood by either changing the data or changing the model. So notice that
in this instance, you're not going to make any false inferences, provided that all the data you solicit is always very consistent with your model. You won't need to change your mind. If you've got the wrong model, the data will provide disconfirmatory evidence and you may well try and get data from another source. And if you can do that and continue engaging until your model is fit for purpose, what you have basically done is find a niche for yourself
where your ideology, your conceptions, your understanding of the way the world works is perfectly fine. So you've found a cultural eco-niche or possibly a physiological eco-niche or an information eco-niche that suits you and you will have a high adaptive fitness and survive and persist in that eco-niche. If you can't, you move to another one. And so now we have a model of speciation. We have a model of
shared narratives, shared convictions that can be communicated, so supply and teach each other, that is self-assembling and self-constructing, co-created simply because you are looking for, at the same time you're looking for information that resolves the uncertainty about your models of how the world works,
which makes it look as if from the outside, you're basically changing the world to make your hypotheses come true. And that, of course, is just another way of describing action. So if we come away from the sort of the exchange of information as a scientist or somebody
In a sense that is you changing the data to match your model. So if you
So this falls out of the free energy principle because, you know, if it's the case that the autonomous paths, the dynamics that constitute our active engagement with the world, if they're trying to minimize free energy, and we in this instance minimize free energy as a prediction error or surprise, remembering that it's the negative of the marginal likelihoods, it's something if
the probability of these data, given my understanding of my construction of the world at this point in time, is very very low. Then it will have a very very small marginal likelihood or have a very very high free energy and therefore it will be surprising given my model of the world. Some people like to articulate that surprise in terms of a prediction. What I predicted was going to happen
is nothing like what I actually sense the data I actually got. So imagine now you apply that sort of concept that the autonomous states that include the active states are just in the service of minimizing prediction error. How would you understand that in terms of elemental movement and behavior?
And one way of understanding that is that basically your models of say how you're going to move are actually realized by creating data that is consistent with your predictions. So I have in my head a model, a hypothesis, the notion that I'm going to raise my arm. And all that's happening in that notion is that that's my belief about state of the world.
And I'm telling my spinal cord and my musculature and my muscles, that's how I expect the world to be. And then my reflexes are driven by the prediction errors, driven by the free energy gradients, literally like the ball rolling down the hill, to realize, to change the world so it supplies the kind of data that I predicted. I just wanted to mention that notion of
movement as simply a realization of predicted endpoints, intentions to move, is physiologically fully licensed by things like equilibrium point hypothesis in motor control and motor physiology,
But it goes back even further to idea motor theory. So this was in the Victorian age, how people understood movement, that basically it was a physical manifestation of beliefs about the way my body should be posed or what my body should be doing. So to me, that is, if you like, an elemental and quite fundamental example of the importance of changing the data, not the model.
So if you changed your model in the context of, I'm going to stand up now, and you looked at the data and nothing is happening, you would immediately change your mind and say, oh, I'm not standing up. What would that look like? That would look like Parkinson's disease. That would be a failure to properly attenuate the evidence that I'm not moving when I strongly believe that I am about to move.
And of course you can understand now the pathophysiology of Parkinson's disease that's due to a failure of neuronal message passing and particularly the weight or the gain afforded proprioceptive information in that light. It's a failure of active imprints, it's a failure of the active states to properly respond to
The evidence that or change the evidence at hand. So this is a lovely example of ignoring selective ignoring just like the second suppression with the eyes.
That enables movement. If you didn't do that, we couldn't move. So what I'm trying to say is that, you know, right from the inception of these kinds of this kind of thinking in the inactive domain in terms of idea motor theory, the there is a there is I think a fundamental importance of, you know,
of the other side of the coin from changing your mind, which is basically changing the data which causes you to change your mind through sense making. And without that, we wouldn't have action. So you can almost say that action and perception are the two sides of the same coin. The perception is changing your mind by changing your model to maximize the marginal likelihood given a particular model.
Action just is changing the data by acting upon the world to maximize the marginal likelihood, basically maximizing changing the data. So the probability of the data given the model is so active inference. So that coupling of action perception is essentially trying to realize your beliefs about or your models of your lived world. And if you're successful in doing that, you've got the right eco niche. You will have a high marginal likelihood.
you know, from an evolutionary perspective, you will have a high adaptive fitness from a social perspective, you will be, people will listen to you and teach your job into these podcasts, suggesting you do indeed solicit that kind of information that causes people to, you know, to go to your channel, to attend to your stuff, simply because they think they're going to get good stuff and they're going to get uncertainty resolving stuff.
Okay, well, perhaps let me undermine my own credibility here. So a few weeks ago, and I think about a year ago or so we talked, and I had some experience about, and I was worried and you said, Kurt, don't be worried. I had some experience where I thought I heard a voice. It was in a hypnagogic state. So I was just about to fall asleep when
Many people
that shouldn't be a state that I attribute much credence to but I remember being afraid because I heard my wife either say okay or just some one syllable word and then I was so worried and I worked myself up into anxiety thinking do I have schizophrenia oh my gosh there was no other voices since then but I was extremely anxious and then you told me don't worry and even Kurt even if there is there's medication and many people go through psychosis and or psychotic breaks and so on but luckily I didn't luckily there was nothing wrong with me
And your words did calm me, so thank you for that. However, and coincidentally, perhaps not, but coincidentally, a few weeks ago, I had a strange experience, Carl, and I wasn't sure if I was going to bring this up on the podcast for one reason, because I don't want you to think that I'm someone who's beset with psychological issues. But on the other side, I am driven by a filmmaking adage, which says that the more personal the pain, the more widely applicable it is.
And the opposite is also false. The opposite is false. So if you try to speak generally, you end up speaking to no one. What happened a couple of weeks ago was I had this sense and it was so and even to talk about it now is a bit frightening.
I had this sense that I was like, oh shoot, solipsism. Is everything in my head just that thought? And I've, because I'm on this podcast and I've entertained many ideas, it's extremely taxing to do so because I try to take different people's points of views seriously. And perhaps I shouldn't, but I feel like in order to give them a fair shake, I need to hear that sound.
That's the sweet sound of success with Shopify. Shopify is the all-encompassing commerce platform that's with you from the first flicker of an idea to the moment you realize you're running a global enterprise. Whether it's handcrafted jewelry or high-tech gadgets, Shopify supports you at every point of sale, both online and in person. They streamline the process with the Internet's best converting checkout, making it 36% more effective than other leading platforms.
There's also something called Shopify Magic, your AI-powered assistant that's like an all-star team member working tirelessly behind the scenes. What I find fascinating about Shopify is how it scales with your ambition. No matter how big you want to grow, Shopify gives you everything you need to take control and take your business to the next level. Join the ranks of businesses in 175 countries that have made Shopify the backbone.
of their commerce. Shopify, by the way, powers 10% of all e-commerce in the United States, including huge names like Allbirds, Rothy's, and Brooklynin. If you ever need help, their award-winning support is like having a mentor that's just a click away. Now, are you ready to start your own success story? Sign up for a $1 per month trial period at Shopify.com slash theories, all lowercase.
Go to Shopify.com slash theories now to grow your business no matter what stage you're in Shopify.com slash theories. Embody them in some way, shape or form. So if they say, well, all is mind, for example, that's the idealist.
I'm trying to imagine what would that be like and if someone's a materialist well what does it mean that this was dead and it somehow emerged that we have consciousness either way the point of that is to say I've been feeling like for almost a year and a half now just in a void not sure what is true and what's not true that's fine I can deal with that except that that led me to this spiraling of such it was probably the most terrifying experience I've ever had in my life and it took me weeks and I'm still recovering from it where I felt like shoot
Am I all that exists? And then just that thought alone? I'm like, I don't want to believe that that could be the case. And I remember looking at my wife and thinking, is my wife even real? How much of an imbecile and how ungrateful and how foolish is it? And how that I would even think that I feel bad that I would even entertain that thought. Oh my gosh, I'm scared. I don't want to go to the
Hospital because I felt like if I was to truly accept that thought I could I was on the brink of just losing it into an insensate spiral. I became extremely scared of myself and I think that I'm the type of person I am an anxious person and I am the type of person that tends to obsess over thoughts.
So then I started to obsess over this and for the past couple of weeks, I was so scared, so scared of my own mind. So almost like a hypochondriac for mental disorders where I didn't want to even look up what's the definition of schizophrenia or psychosis was because I was afraid that I would see the signs everywhere. For example,
A few days ago, or about a week ago, remember, generally we all speak to ourselves in our own heads. And sometimes I use the phrase, so I don't like that, or I like that. I use the word I in my own internal monologue, but sometimes the word, yeah, you can do this, you can do this. So sometimes they use I, and sometimes they use you. And I imagine we all do that. It's colloquially okay to exchange those two. However, then I thought, oh, shoot,
What if I felt distanced from my thought? What if the one that's saying you is a thought speaking to me like a schizophrenic may hear? Oh my gosh. Oh my gosh. I don't want to hear this. And I was so afraid and I felt derealized and depersonalized. When you mentioned Parkinson's earlier, that the difference between the model and then the data, that didn't strike me as a symptom of Parkinson's. The way you were describing it struck me as a symptom of derealization. Anyway, I'm bringing this all up and actually feel
Good about speaking about it. It somehow feels Curative to even voice these concerns. I Well, actually, I don't know I don't I don't know why I'm I'm voicing this mainly to Mainly, okay. How about this? I just verbally cascaded incoherently perhaps
What are your thoughts on what I just said? The reason I'm saying this one more time is that people in the comment section, when they view different theories of what consciousness is, what reality is, that's the theories of everything channel, in a nutshell, I see many of them in a similar place where they feel lost. For example, I've heard that some people who view Yoshabok's ideas that this is all an illusion or this all simulated consciousness, some people become suicidal.
That's actually extremely terrifying. Luckily, I'm not suicidal, though I'm scared that I could because I've heard that other people could. I'm not sure if my problem is just obsessive thoughts or if it was indicative of something else and I was so afraid. And it's such a scary feeling, Carl, to be afraid of one's self, to not even be able to sleep at night because you're afraid of your own thoughts and what that may lead to.
Does any of that make sense? And I know that sounds like it's completely off field. Perhaps it's not. No, no, I think I think it follows on very gracefully. But you've taken it right now to some of the most important aspects of modeling, which we all contend with, which, of course, is I mentioned before, you know, getting models for our
our lived worlds. But of course, the most important part when we're a baby of the world is our body. So the first thing we have to do is to build a model of our body and work out what things that we are able to move, hence motor babbling and rattling. Probably more important than that, of course, is exchanges with the mother
developing a sense that mother is an object and crucially an object that is separate from me. And that's quite a skillful move and it takes a lot of my belief updating and changing your models and building a coherent explanation for all your sensations, both the political touch and the physiological intercepted consequences of being suckled, for example, and being cared for.
And I'm taking this route to the key thing, which is basically models of self and selfhood. So if we are coming back to this picture of some mathematical universe that I've now tiled with multiple Markov blankets encapsulating or defining lots of particles, and basically we
inhabit a universe that is constituted by things like us and initially you know mum and dad and brothers and sisters and then you know peers at school. That means that our models of our lived or experienced world have to have models of other in it which means you need to be able to disambiguate between sensations caused by others and sensations
That brings with it something quite special, which is a model that encompasses selfhood.
So you wouldn't need this if you lived in a world that did not involve an exchange with other conspecifics. But because we do live, our world is constituted by other things like us, we need to be able to contextualize and make the inference, I did that, or you did that, or it's my turn, or it's your turn, or you've got that intentional stance, and I've got this intentional stance.
So just having as part of the generative model of the hypothesis, you know, the hypothesis that I am me and you and you is something which characterizes one would imagine very high life forms. Now we go even higher to basically philosophers and people like you. So if you've got, if you have, if you spend your life worrying about
What is the self? Does the self even exist? Well, precisely. That seems to be the essence of your existential angst and all its attendant, interceptive and emotional consequences. Do I exist? I think really usefully you brought to the table the notions of depersonalization and derealization
If this is for physicists, then can you explain to your audience what those things are? Because I think that it would be really useful for people who do not know what... Well, to be quite frank, Carl, I didn't want to look up this exact definition because I know that I'm a valedictorian and I would just...
obsess over myself having these issues. So I can give an explanation of my experience and then perhaps you can delineate what the psychopathology is. Okay, I'll make an aside first. Some people say that what we need to do is realize that the self is an illusion. You can hear this in the more eastern end. And I know that many people who watch this channel also feel like, oh, they take these people seriously, like I took them seriously. And I don't know if that's exactly true.
Not whether or not the truth of the self being an illusion is true, but whether or not that is indeed useful for everyone. It may not be. It may be. At least for me, it was extremely harmful and perhaps still is. And it may be dependent upon where you are in psychological development, spiritual development. And sometimes these lessons that are told to us by philosophers or spiritual gurus
or people who seem like spiritual gurus, perhaps they should be taken with such a grain of salt. And I've come to the conclusion that if it's not life-affirming, then perhaps don't follow that. Perhaps use that as a clue that that's a pathological path. And don't do anything that can be drastic. Don't give up. Because I felt like I was in a state where I could
snap and feel like, well, this is all a dream. I remember feeling, looking at my arms and feeling like, am I even behind, like right now I'm feeling great and sorry that I don't seem like it, but I'm speaking with you. I feel safe. I love my wife. I've actually whatever I'm feeling happy right now. And I feel like I would a month ago, which is I'm behind my eyes and this is a person speaking a week, two weeks ago, three weeks ago, I was feeling like,
All of that could be a model. I could emulate anyone else's head. I could be anyone else. This could all be in my head. So debilitating, destabilizing, walking around feeling like, is this a dream? Constantly checking my fingers, because that's one of the signs you can check for lucid dreaming. How many fingers do you have? And then also feeling bad, like, why the heck am I thinking this is a dream? Shoot, what if it is a dream?
Spiraling, spiraling, tumbling, tumbling thoughts. Okay, that's what I had slash still have to some degree, though not right now. I wouldn't say that I'm over it because it was such a recent experience. I'm not going to say that I've solved it. I feel like it's tapering off. Luckily. Anyway, that's what I had. I would characterize that as derealization in the sense that I didn't feel like it was real. I don't know if it's depersonalization because I don't know the difference between those two nor dissociative identity disorder. So I don't know the differences between those.
Well, it's nice you bring all those three. The phenomenology is very, very closely related, but the way you described it, it was almost perfect. You know, this either a derealization, which people normally described as they're looking, they are sensing,
Something that doesn't quite seem real. The outside world seems like it's a movie. It's lost that tangibility. There's no grip that can be attained that reassures you that stuff is actually out there. The converse, of course, is I'm no longer a person. And that could, I think, be even more frightening. And both of these, I think, speak to
speak to this dissociation that comes with a dissolution of selfhood, or at least one's construction of explanations for the self in a lived world. You used a phrase which I think was quite pertinent here, which is, one should not pursue this kind of thinking and self-exploration and introspection
Unless it is affirmatory and self-affirmatory. And in a sense, that's exactly from the sort of mathematical and dry perspective of the free energy principle that just is maximizing model evidence. This is self-evidencing. It is, you know, if you can affirm
through all the right kind of interactions, be the emotional, social, sexual, intellectual, that you've got the right kind of model of your world, which is constituted by people like your interviewees and your wife and all your colleagues. Then affirming that model simply is securing evidence for that good model of your rich and encultured world.
That just is that sort of affirmatory process. However, there are dangers that lurk in the shadows, which you've clearly contended with. To do that properly, to be able to secure the right kind of evidence that you are living in a world of positive, constructive, nourishing,
enabling human beings, others, other things like you, you have to be able to communicate to be able to communicate, you actually have to entertain the idea that there is there is self and other. And simply, you know, to engage in turn taking, you need to have this, this explicit part of your generative model, that I am a self. In so doing,
You will also have contextualized as part of your generative model itself in a number of different states. And you need to have, remember before we're talking about the good scientist having a portfolio of hypotheses, we're generating a new set of hypotheses so that we can now gather evidence for these hypotheses. So that's part of what you are doing. I mean, in doing these podcasts, you are on a journey
of developing different hypotheses about ways of being. The danger, though, when it comes to different ways of me as a self existing is that you can sometimes lose, entertain the null hypothesis that I am not a self or I do not exist in a sentient fashion.
in the fashion of having particular kinds of qualitative experiences. This is a perfectly viable hypothesis. It can be had in a non-emotional context when you meet people like philosophers, entertain things like the zombie hypothesis or brain and bat thought experiments. These are all really interesting instances of a very high order life form. Think about, well, what would it be like if I was me without mehood?
If I was a zombie, what would it be like if it wasn't actually me talking? What would it be like if there was no reality, it was all just me? All of these alternative ways of being that speak to selfhood or its negation or its nihilism now become plausible hypotheses for which you have to search for evidence.
And of course, if you are just doing this in your head and ruminating, introspecting, there's going to be very little evidence available to you to disaffirm your hypothesis. Perhaps I'm not me. If you don't actually engage with somebody else, there is no evidence at hand for to refute the hypothesis. So on the one hand, that is a remarkable capacity. The very fact you can have this hypothesis in your head, I am not me.
Or it is not me having these thoughts. It is not me having these quantitative experiences is quite remarkable. It's a very, very high level ability, which I would imagine where you and I are the only kinds of
sentient artifacts in the universe that can entertain that kind of counterfactual hypothesis, which is necessary to explore different models. It's necessary indeed to actually just behave and have plans. You have to have counterfactual hypothesis about the future of different ways of being because it comes with it comes with a price. If sometimes you entertain a hypothesis for which there is no evidence on hand,
that could possibly refute it. And certainly, if you adopt a particular behavioral strategy or engagement with the world, that precludes soliciting evidence against the hypothesis that you are, say, you. So examples of that would be things like obsessional compulsive disorder or agoraphobia, or your rumination disorders, where you physically
or possibly even pro-socially, prevent yourself from acquiring any further information that would allow you to change your mind. And of course, once you realize that's a possibility, you can also get into the negatively veiled emotional aspects of entertaining these hypotheses. So if I've got schizophrenia, that now offers
a very plausible hypothesis for these dissociative experiences. Why am I having these dissociative experiences? Well, one very plausible hypothesis, I am having, I have a dissociative order or I have disorganized thought disorder that characterizes schizophrenia. So this now becomes a very plausible hypothesis. How would you go secure evidence against that? Well, it becomes very difficult because the whole point of having these very high order counter
factual hypotheses or models at hand to explain your experienced world is that they're very plausible explanations and it's very difficult. If you have got schizophrenia, you're not going to be able to trust any information anyway because they could be illusionary. So it's one of these wonderful self-fulfilling hypotheses that are very difficult to dismantle.
And you see the same phenomenology in many instances. You mentioned behaviour therapy before in a way that doesn't have this quite the same horrible existential angst that you clearly experienced. You can actually construe many phobias, for example, of the same kind in the sense that if you are very frightened of spiders, you're very unlikely to actually go and solicit
the evidence that would that would confirm the alternate hypothesis that spiders are not frightening because you just don't go near spiders. So this is the kind of hypothesis that maintains itself because it's undermined the way that you act as solicitor evidence for your beliefs. But for you,
Of course, it's not a question of being frightened of spiders, it's a much more existentially deep and worrisome notion that in fact you're not you or you're not even a self and you're not even perceiving reality. These kinds of hypotheses that come along qualified by or associated with plausible explanations, I'm going mad.
are very difficult to refute, especially if you ruminate on them. They will go away as soon as you start reengaging with other people.
That's what I found to be the case, Carl, is that the more I talk about it, the more I'm speaking with you, that just speaking with people and not being so in my head, almost instantaneously, it starts to dissipate. And also, just so you know, there's a danger for the people who watch this channel that it's seen as a spiritual triumphant state to be egoless, to have ego death.
So then, just so you know, when I was feeling that, I was then telling myself, but am I supposed to have ego death? Is this what ego death is like? And now I feel bad. Am I just not strong enough as a person? Which then I feel like I'm less on the social hierarchy because I'm just a coward who cannot deal with, who isn't psychologically strong enough to deal with potential ego death, if that was even ego death.
so that those are there's so many dangers in this and i didn't realize this i had this insight in that episode or experience that kurt kurt kurt kurt something was like i was talking to myself or something was talking to me and telling me kurt you've been thinking way too much this is you're extremely analytical i never would have thought of myself as someone who's too analytical before
It was clear to me that I was too analytical, and there's nothing wrong with being analytical. There's nothing wrong with thinking, but you've been thinking a bit too much. Feel more. Just ground yourself here. Engage in this world more. And there was also this feeling that I don't love myself, which is a strange feeling, Carl, that I'm a bit hard on myself. And I would have never said that before. I would have thought, I'm not hard enough. I need to study harder. I need to prepare for these podcasts harder. I need to release more. I need to learn more.
There's so much self-analyzing that's going on right here, Carl.
Anyway, I felt like, Kurt, you were thinking too much. Just relax, perhaps. Don't engage as deeply as you have been with these ideas. Don't entertain them as much. Maybe you're in a fragile state right now. You can do that later. Right now.
take it back that's what came over me and also this lesson that I don't have enough self-love or self-acceptance last night I looked up I was using a rubber band for my thoughts like self-administered aversive conditioning when you snap yourself if you're performing a bad habit so I thought maybe I could do that with my thoughts so anytime that I feel like that that would encourage my mind to not feel like that I looked it up and it said that works for physical habits but not for mental habits so stop that and what works is act I believe acceptance
and commitment therapy, which is a CBT technique. So then last night I was telling myself, you know what? Try that out. Accept these slots. And I'm not exactly sure what it means to accept them.
I don't precisely think it means entertain them, but at least meet them, not with horror and dread and avoidance, but just say, that's okay, man. You think like that. Some people feel like that sometimes. So I started doing that. And last night was the first time in the three weeks since I've had this, that I felt absolutely like my regular self, almost from five minutes of this quote unquote acceptance therapy. And then I woke up this morning, felt like my regular self, like I'm behind my eyes. I have a, I'm in this world. That was fascinating.
Okay, so I said quite a few statements right there. I'm sure there's plenty you want to comment on. What I'm also interested in, for you to put a pin in your hat, is what the heck does this self-acceptance, self-love, if that seemed the same, mean in terms of the free energy principle? Okay, so I basically just poured myself out there.
There you go. Sorry, Carl, for you to pick up all those pieces. I know it's a huge responsibility. I trust you. You're someone who studies this. You're extremely, extremely bright and you have such a wealth of knowledge. And I do feel like what I'm going through is not something terribly unique. I do feel like this is something that many people who are on this journey of understanding the world and consciousness in their place go through this perhaps at some point. I think I was trying to emphasize that by saying that, you know,
It is a gift that you're able to have these existential crises. There are dangers that lurk.
which you've clearly encountered. But at the end of the day, the very fact that you can entertain these counterfactuals is quite remarkable. To my mind, it would be the highest expression of the human condition just to consider those alternative hypotheses where bits of our humanity are just not there anymore or have a different disposition or relationship to reality or indeed other people. But lots of things you said made a lot of sense.
from simple things like soliciting reassurance from others that you don't have to do any more work in terms of preparing yourself and being a good expounder of ideas or interviewer or book writer or whatever. That's exactly what you should be doing. That's exactly securing evidence for yourself models of an affirmative thought.
And the more you do that, then the more the more that model will be fit for purpose and that you will have the right kind of marginal maximizing your marginal likelihood. But you're in a difficult position because your job is actually to explore other ways of making sense of things and other ways of thinking things. You know, you're compelled to explore alternative hypotheses and, you know, you described
you know, the potential nihilism and that I presume quite nightmarish. I can't now remember my nightmares, but I imagine that all of us do go through so we have nightmares as children. I think you stop having true nightmares in your 30s and 40s, but you know, that sort of, you know, before you've really got that grip and maintain that grip on selfhood and you in your
Very frightened of having one?
Yeah, yes, that is correct. And luckily, I wasn't well, what I imagined what I was going through would feels like how someone would feel on a on a terrible psychedelic trip. Yes. And it's interesting you bring up psychedelics, because of course, you know, there is a move in sort of pharmacologically assisted talking therapists, you know, to use psychedelics, simply
not to aspire to, I don't know what ego death means, but I can have a guess at it from the point of view of meditation and mindfulness. But there's a really interesting connection between the use of psychedelics and the aspiration of many meditation-like practices that would, I think, subsume the internal attention states
of meditation and, you know, current practices of mindfulness, which is really to try and redirect your attention to the sensorium and usually the intercepted parts. It's like breathing, for example. So this is the exact opposite of what you've been doing. It's exactly deploying that attention, all that game control we're talking about in terms of selecting
what kind of information, sensory information, sensory states to engage and determine your belief updating. So putting all that attention out to the sensory side of your deep hierarchical models, your constructs, what you've been doing is the opposite. You've been actually wandering around at the deeper, the highest levels of these hierarchical models that
Sorry, so in the back when we were initially talking, I said there's a black box, the input output, and then there are two black boxes. It's as if you're saying pay attention to the sensory. Forget about the action. We haven't talked much about the action, though, obviously, as embodied creatures, sensory and action are tied.
It's as if what I've been doing was staying within the box and creating my little sensations and actions within there as little loops. And you're saying pay attention to the senses that come from the outside. Is that what you're saying? Yes. Well, I'm not saying you should do, you will do everything in the right way under the free energy principle. What I'm saying is that the skilled practitioners of mindfulness and meditation
I would imagine of the kind that would lead to ego death or I'm not quite sure what that means. It would employ exactly the same kind of internally mediated and sometimes if you're very skilled volitionally called forth mechanisms that are actually targeted by psychedelics.
require you now to do all your belief updating and evidence assimilation and sense making.
much more the input level of the black box, not at the highest level, which you could think of as an output in terms of selecting the plans and what am I going to do next? And what kind of person am I? Narratives that we give ourselves that contextualize the way that we behave and the other things that we do. So, you know, I think it's mechanistically a really interesting connection between the notion of you ruminating, going in circles around your head, exploring ever darker and darker
That's interesting. So I think that, you know,
You might, you may, I guess you probably are skilled in meditation.
But that kind, I think the objective of becoming a skilled practitioner of mindfulness or meditation is simply to get some volitional control over that sort of attention that is paid to these deeper machinations versus attention that is paid to the sensorium that underwrites the psychedelic aspect, the allure of just sensory patterns and
the sensations and textures that you get, you know, when taking psychedelics. So, you know, getting control of the balance. So it's not the fact you're attending to your breathing, which is important. It's the fact that you could volitionally redirect your attention away from selfhood. Extremely interesting. So, but you seem to have got control by this acceptance. And it strikes me that, you know, in the absence of
Feelings of an emotional sort that were articulated in your description in terms of, you know, I'm an awful person. I'm not fit for purpose. I'm not well prepared. I'm not able to man up and own this ego, ego death. All of this sort of personal but quite emotional constructs, affective, valenced.
There are also hypotheses. There are also explanations. So the idea is that, you know, anything that you can talk about has to be part of an explanation for particular ways of being. But, you know, I can be a worthless person in the sense that I am a worthless person. I'm not a worthless person. You know, I'm in a state of anxiety. I'm not in a state of anxiety. We have to recognize when we are anxious. Anxious is just a state of being.
which is necessarily called for in certain situations that rebalance this attention or precision waiting in computational psychology to make your belief updating fit for purpose in this particular context. What I'm trying to get to is why acceptance might have worked, because I imagine that if you were suddenly found yourself ruminating and locked into those ruminations,
then very much akin to someone on a bad trip who thinks, ah, I am going to be locked in this forever. This is going to be my uncertain state of being for eternity, because there is no reality to it. That's how it feels, very heartfelt, yeah. That feeling will, well, if that is true, I must be feeling anxious. You are then going to look for evidence that you're feeling anxious and
because you are anxious, you will experience certain cardiac acceleration, certain interceptive flight or fright like responses that will supply evidence that you are anxious. And then that becomes evidence. Yes, I am right. I am now in a state of nihilism. I do not exist and I should be anxious about that. And yes, I am anxious. So you get into a vicious circle.
So this is a sort of good old fashioned cognitive behavioural explanation for things like panic attacks, that you know, you quite reasonably in an entirely base optimal way, make sense of interceptive bodily gut feelings, literally, you make sense of them with the hypothesis, oh, I must be in an anxious state of being. And that
conclusion hypothesis then generates autonomic actions that are realized reflexively in the way we talked about when we're looking at the Parkinson's idea and that will be reflected in terms of cardiac acceleration, neuroendocrine releases into your body and your body will change and your body will supply signals, intercepted evidence, yes I am anxious. So it's again a self-fulfilling prophecy in exactly the same way that the idea motor theory means that raising my hand
is just a realization of my beliefs, but I can literally raise my levels of anxiety in exactly the same way, but invisibly from the outside, but not when I can sense my own body. So if you got yourself caught up into this joint hypothesis, I am in a nihilistic existential state, and I should be really worried about this and anxious about this, then it is quite understandable
That you are going to use signals of anxiety and angst generated by your body as evidence that, yes, I am now divorced from reality. I'm an unpersoned. I am in a dark and nihilistic place. There could be no other place. There could be no reality. That's a hypothesis. If you can just wait for the evidence that that hypothesis is incorrect,
simply by letting your body calm down so there is no further evidence of any anxiety then you can find secure even from your own body even without talking to somebody you can secure evidence ah the hypothesis that i am in a state of internal nihilism and anxiety is a silly hypothesis because the evidence refutes it i can now feel my
not possibly personally, but you will synthesize all your interceptive feelings in a way that provides evidence against the fact that now I am in a deeply nihilistic, anxious state. I think that just is the motivation for the acceptance. It is pushing through to get to a state of mind and a state of body, literally realizing your good and making sense of your gut feelings.
so that you can now refute the joint hypothesis that I am in a state of nihilism and I should be jolly frightened about this because there's no evidence that you're frightened anymore. So this was just a piece of hypothesis building of a very sophisticated and philosophical sort, the highest level of exploring alternative ways of relating to the world. You now have
I think a very useful insight into the gift of retaining selfhood that I think we all take for granted, but it's quite a fragile thing. Luckily, most of us get through the day, if not hopefully most of our lives,
you know, by retaining that grip and just, you know, continue securing evidence that this is the right hypothesis, I'm a person, I am me, I am functional, this is, you know, this is the way I'm meant to be, and these are the kinds of things I do, and the kinds of people I talk to. But that is such a fragile, self-assembled, and has to be maintained hypothesis that you don't, you know, that fragility, I think, only is revealed occasionally, and only to some people when they have the alternative hypothesis that, you know,
How is it that you prevent yourself from getting into existential crises when it seems, at least to me from the outside, that you study similar, not theories per se, but the free energy principle is about existence?
and is about selfhood. So is your model so strong, Carl? Is your model already so strong that you can entertain these without worry that you'll break yours or without that being a potential possibility? Or is it because you purposefully analyze it dispassionately, analytically, intellectually, so that you're not identifying with what you are writing down on a piece of paper? How is it that you prevent yourself? Or is it constitutional, like by predilection, by personality? You're not neurotic, you're not as neurotic as I am. I think that's probably
truth to all that and by being neurotic you know that's an entirely fun you know neuroses are absolutely essential to get through life you know i have a particular set of neuroses which you know are high treasure and try to admit it in the big five model yes but i think you're also right that some people might call this mentalizing i think what you were doing is that you
In a raw, enthusiastic and creative way, we're just exploring different hypotheses about the way that you yourself, should it exist, relates to the rest of your mind and the outside world. But for you, it got conflated with emotional states like anxiety,
and your self-worth and those evaluative states of being. So if you want to, in your words, be more dispassionate about this kind of mental exploration on the inside, then you have to train yourself not to conflate the two. One way of doing that is just to become to mentalize. So mentalizing is usually taken as a way of, if you like, disentangling
So if you'll say very late at night,
You're normally going to feel fatigued and slightly less robust in a way physically. Now, it doesn't have to be like that, but that's the way that normally the world works, and that's the way that you've learned to model yourself. If you train to be a professional, say like an airline pilot trying to land a plane at night,
or like a psychiatrist who has to deal with these, has to see people going through these emotional or valence or effective episodes, whether it's sort of effective in the sense of, you know, maybe your depression or whether it's
What's that? It's a mixture of abnormal mood with having psychosis of the kind associated with schizophrenia.
So it's having an emotional aspect to saying delusions and hallucinations, for example. Oh, I thought that those those don't go hand in hand. So some people have delusions, but they're OK with it. Yes, yes. Some people have hallucinations and would not would not even seek medical help or their family and friends would not seek medical help. So there are people out there. I guess my point here is that you can
mentalize it and build professional defenses against the emotional aspects. I think that's what I've done. So I think about these things as a doctor or a mathematician, not as a person. And if you want an honest answer, the other trick is to not use psychedelics, but you should try smoking. So smoking and cups of tea are great ways to... Smoking what? Cigarettes.
You were saying the smoking of the pipe does what? Because I'm not the type of person, I don't smoke and nor do I want to smoke. But you were saying that the smoking does
What was the purpose of the smoking? It's just a way of using, um, nuancing your neurotransmitters and control that synaptic gain or the gain and the way to you for different, uh, different parts of your, um, belief updating. So it has the opposite effect of psychedelics basically. So that's why people like tea and, you know, have a cup of tea, calm down,
You know, stay calm and carry on wherever it is. Nice cup of tea, cigarette and cup of coffee. That'll sort you out. And it's because all of these drugs, the caffeine, the nicotine, they all act on the same kind of receptors at different levels in your neuronal hierarchy that are responsible for setting that gain control and gating during the attentional and the attenuation aspects that we're talking about.
I would have thought that the caffeine would have the opposite effect of increasing your anxiety.
that you might be anxious and in fact you're not, then just sensitizing certain interceptive sensations that register the fact in fact you're not terribly, your heart isn't pounding. That'll actually calm you down a little bit. If it goes too far then beta blockers will be another nice example. That's what I used actually. Oh right, oh well there you are. So what I'm trying to say is sometimes with skillful use, the use of a tobacco pipe
can have the same kind of effects as a beta blocker. That's all I would say. All right, professor. Again, thank you so much. It's much safer to stick with beta blockers, then you won't get mouth cancer or anything. So if you could...
I found the beta blockers work and there's also some evidence of fear extinction. Have you heard of that? There's this professor named Meryl Kenditt who I'm going to speak to at some point. She takes people through phobias where she gets them to be extremely frightened. She actually makes them almost manic or panicking at least. And then she gives them a beta blocker and then they still feel fear that day. It's only the next day when there's memory reconsolidation and the emotional tags are removed because of the beta blocker. So you have to have a good night's sleep.
and then the fear seems to dissipate drastically for incurable PTSD. I was trying to do that to myself so when I was feeling some fear I would take a beta blocker and then I would feel the fear less still wouldn't be zero but then thinking okay great maybe this will be like my own intervention of a fear extinction technique.
Yes, that's a very clever idea actually to combine beta blockers with what some people call flooding. So that's exactly the kind of way I was counselling. You would use people who smoke use smoking to moderate in the right context.
Thank you so much. I will talk again. And perhaps the next time we speak, there were some questions that I didn't even get to from Michael Levin, from Professor Norman Wildberger, who's a mathematician, and there's quite a few. Okay. Carl, I kept you on for so long and this is gone. Well, this didn't, I didn't intend for it to go in this direction, but I'm extremely glad it has. And maybe my episode can serve and your advice, episode plus the advice,
can serve as a cautionary tale and as well as some practical advice as to what to do and how to avoid situations like this when exploring topics like this. Thank you so much, Professor. I appreciate it. I look forward to the next time.
The podcast is now finished. If you'd like to support conversations like this, then do consider going to patreon.com slash C-U-R-T-J-A-I-M-U-N-G-A-L. That is Kurt Jaimungal. It's support from the patrons and from the sponsors that allow me to do this full time. Every dollar helps tremendously. Thank you.
▶ View Full JSON Data (Word-Level Timestamps)
{
"source": "transcribe.metaboat.io",
"workspace_id": "AXs1igz",
"job_seq": 10454,
"audio_duration_seconds": 7869.8,
"completed_at": "2025-12-01T01:43:03Z",
"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": 78.114,
"index": 3,
"start_time": 66.203,
"text": " 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 we'll give you a better deal. Now what to do with your unwanted bills? Ever seen an origami version of the Miami Bull?"
},
{
"end_time": 96.34,
"index": 4,
"start_time": 78.626,
"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 Miami Verizon store today and we'll give you a better deal."
},
{
"end_time": 126.135,
"index": 5,
"start_time": 96.34,
"text": " Recall that you can click on the timestamp in the description to skip this intro, as there will be a lengthy introduction. This episode may be the most important Theories of Everything video. Namely, we talk about the perils of attempting to understand consciousness, theories of everything, even the UFO phenomenon. These are aspects which almost no one talks about. Because of this, I've written some notes, so I'm going to proceed to read. You'll find that those who think of themselves as gurus tend to believe that almost all their lessons apply to almost everyone at almost all times,"
},
{
"end_time": 153.746,
"index": 6,
"start_time": 126.135,
"text": " though it doesn't take into account where you may be psychologically. Certain spiritual teachers or people who consider themselves to be enlightened suggest that you should understand their point of view and if you resist their truths, it's because of deleterious fear and they do so without the consideration of the mental state of you, also without a conception of salutary fear slash uneasiness. This episode serves as a cautionary tale for those who are seriously trying to investigate"
},
{
"end_time": 182.773,
"index": 7,
"start_time": 153.746,
"text": " fundamental reality, how to identify when you're on a nourishing track versus being off track, etc. Again, this may be the most important of all the theories of everything content. The push to not feel fear, slash not feel the ego, quote unquote, slash, etc. can be dangerous, not because the lesson is false, though it may be, but because of the social pressure around those words. In fact, there's a universally frowned upon quality called being egotistical ego is embedded in it, and also to be seen as retreating is the same as"
},
{
"end_time": 207.79,
"index": 8,
"start_time": 182.773,
"text": " being seen as being defensive which is the same as being fearful which is seen as being lower on the spiritual hierarchy perhaps it's true that you should feel less disquietude about certain facts though perhaps your disquietude is justified for where you are now and that's okay it's 100% okay it took me about one year to realize that in fact perhaps more than one year it took me over one year to realize that these spiritual lessons"
},
{
"end_time": 237.415,
"index": 9,
"start_time": 207.79,
"text": " tend to not have the physics notion of path dependence. Perhaps you shouldn't uncritically adopt an Eastern practice in a Western world and vice versa, or perhaps certain truths are dependent on the path that is where you are now rather than being independent of anything. Frankly, it's embarrassing to reveal myself as so selfish and malicious and cowardly and to do so so unguardedly to Carl Friston and even to you as I'm a"
},
{
"end_time": 255.196,
"index": 10,
"start_time": 237.619,
"text": " Relatively thin-skinned person, at least currently I am, but hopefully in the end it's a net benefit. The dangers of going on to journeys of investigations into consciousness, the self, computational reality, and even the phenomenon, as it can lead to a spiral down some viciously dark paths,"
},
{
"end_time": 280.179,
"index": 11,
"start_time": 255.196,
"text": " that are at least ostensibly irrevocable while having certainty that your new founded conception of reality is indeed truthful. While I don't like to state personal opinions because I generally don't have many strong views and the views that I do have are fickle, changing week by week sometimes, my current conclusion is that if it's not life-affirming, if it's not loving, if it's not something you can tell others without others being horrified and"
},
{
"end_time": 303.763,
"index": 12,
"start_time": 280.179,
"text": " Frightened for you then use that as a sign that what you're experiencing or what you're feeling or what you think is not truthful Please seek help don't suffer alone You're not alone and you can verify that simply by talking to as many people as you can if you're listening to this podcast Then it's likely because you at least have some trust in it. So trust that statement or at least test that statement out"
},
{
"end_time": 332.654,
"index": 13,
"start_time": 303.763,
"text": " Now onto the guest. Carl Friston is a professor of neuroscience at University College London and is one of the most cited individuals in his field. He's also the inventor of the free energy principle and is someone who's made advances in the field of active inference that's matched almost by no one. He may be one of the most insightful and brilliant people on the planet. I've been lucky enough to speak to Carl before in a substantive four-hour podcast, linked below, which is best watched at least twice to properly understand."
},
{
"end_time": 353.729,
"index": 14,
"start_time": 332.654,
"text": " This episode is an outlier in the Toe Channel because this will serve as an introduction to that episode. As many after watching that were still unclear as to what the free energy principle is. If you feel like you're lost, that's okay. Let that be a theme. Understanding is less like a light switch and more like a stovetop. That is, it's not like all of a sudden you have an insight and then you understand some phenomenon. You went from zero to one."
},
{
"end_time": 375.094,
"index": 15,
"start_time": 353.729,
"text": " It's more like you get a gradual increase in understanding where a modicum here makes more sense and then a smidgen here. As Wheeler says, and I like to quote, the point isn't to drink from the fire hose, but instead to get wet, trusting that your conscious befuddlement is part of the process and unconsciously more is becoming clear. Click on the timestamp in the description if you'd like to skip this intro."
},
{
"end_time": 403.114,
"index": 16,
"start_time": 375.094,
"text": " My name is Kirchheim Ungel. I'm a Torontonian filmmaker with a background in mathematical physics dedicated to the explication of the variegated terrain of theories of everything from a theoretical physics perspective, but as well as analyzing consciousness and seeing its potential connection to fundamental reality, whatever that is. Essentially, this channel is dedicated to exploring the underived nature of reality, the constitutional laws that govern it, provided those laws exist at all and are even knowable to us."
},
{
"end_time": 427.432,
"index": 17,
"start_time": 403.114,
"text": " If you enjoy witnessing and engaging with others on the topics of psychology, consciousness, physics, etc., the channel's themes, then do consider going to the Discord and the subreddit, which are linked in the description. There's also a link to the Patreon, that is patreon.com slash KurtGymungle, if you'd like to support this podcast, as the patrons and the sponsors are the only reasons that I'm able to have podcasts of this quality and this depth"
},
{
"end_time": 447.125,
"index": 18,
"start_time": 427.432,
"text": " Given that I can do this now full-time, thanks to both the patrons and the sponsors' support. Speaking of sponsors, there are two. The first sponsor is Brilliant. During the winter break, I decided to brush up on some of the fundamentals of physics, particularly with regard to information theory, as I'd like to interview Chiara Marletto on constructor theory, which is heavily based in information theory."
},
{
"end_time": 466.186,
"index": 19,
"start_time": 447.125,
"text": " Now, information theory is predicated on entropy, at least there's a fundamental formula for entropy. So, I ended up taking the brilliant course, I challenged myself to do one lesson per day, and I took the courses Random Variable Distributions and Knowledge Slash Uncertainty. What I loved is that despite knowing the formula for entropy, which is essentially hammered into you as an undergraduate,"
},
{
"end_time": 491.476,
"index": 20,
"start_time": 466.357,
"text": " It seems like it comes down from the sky arbitrarily. And with Brilliant, for the first time, I was able to see how the formula for entropy, which you're seeing right now, is actually extremely natural. And it'd be strange to define it in any other manner. There are plenty of courses, and you can even learn group theory, which is what's being referenced when you hear that the standard model is predicated on U1 cross SU2 cross SU3. Those are Lie groups, continuous Lie groups. Visit brilliant.org slash tau, T-O-E."
},
{
"end_time": 518.78,
"index": 21,
"start_time": 491.476,
"text": " to get 20% off an annual subscription, and I recommend that you don't stop before four lessons. I think you'll be greatly surprised at the ease at which you can now comprehend subjects you previously had a difficult time grokking. The second sponsor is Algo. Now, Algo is an end-to-end supply chain optimization software company with software that helps business users optimize sales and operations, planning to avoid stockouts, reduce return and inventory write downs while reducing inventory investment."
},
{
"end_time": 543.729,
"index": 22,
"start_time": 518.78,
"text": " It's a supply chain AI that drives smart ROI headed by Amjad Hussein, who's been a huge supporter of this podcast since near its inception. In fact, Amjad has his own podcast on AI and consciousness and business growth. And if you'd like to support the Toe podcast, then visit the link in the description to see Amjad's podcast because subscribing to him or at least visiting supports the Toe podcast indirectly. Thank you and enjoy."
},
{
"end_time": 568.268,
"index": 23,
"start_time": 543.729,
"text": " Let's do an introduction once more. So what is the free energy principle specifically, specifically its relationship to a theory of everything as that's what this channel is about? Right. So my usual response here is do you want the high road or the low road? Let's take the high road, given that we're talking about theories of everything. So"
},
{
"end_time": 598.37,
"index": 24,
"start_time": 569.002,
"text": " The free energy principle is a principle of least action. So it's a method really. It's put most simply a method that allows you to identify the most likely dynamics, the most likely way a process will unfold. What is the process that we're talking about here? Well, we're talking about things like you and me, particles or people. So the first thing to note is that you can"
},
{
"end_time": 627.961,
"index": 25,
"start_time": 598.882,
"text": " apply standard variational principles of least action to any system, specifically a random dynamical system of the kind that can be described by say a Langevin equation that underwrites most of the physics we know. But the special thing about the free energy principle is it tries to understand the relationship of something in relation to everything else. So you're starting from the premise that something exists and then you have to think carefully"
},
{
"end_time": 656.152,
"index": 26,
"start_time": 628.473,
"text": " What is a thing and how do I distinguish it from something else? So immediately you start, you have to introduce a partition between the states of things that comprise states that are internal to the thing and states that are outside the thing. And thereby in think carefully about how those states are coupled. And what happens is that you have to introduce a further set of states known as blanket states."
},
{
"end_time": 673.848,
"index": 27,
"start_time": 657.022,
"text": " So you've got this notion now of taking the states of any universe, any system that you want to try and understand,"
},
{
"end_time": 704.155,
"index": 28,
"start_time": 674.411,
"text": " and partitioning it or carving it into four subsets, internal, external, and then sensory and active states that constitute the blanket states where the active states mediate the effects of the internal on the external, the inside on the outside, and the sensory states conversely mediate the influences of the external on the internal. So with this particular partition, which you wouldn't need if you wanted to now just move on and develop"
},
{
"end_time": 727.227,
"index": 29,
"start_time": 704.701,
"text": " say quantum mechanics or statistical mechanics or Lagrangian classical mechanics we're now in the special case of dealing with a system that has this partition and in this special case those conventional variational principles of least action translate into or can be used"
},
{
"end_time": 754.94,
"index": 30,
"start_time": 727.585,
"text": " to associate the paths of least action, the most likely paths that the system will take, particularly the states of the particle or person that we're talking about. We can associate those paths of least action with paths that minimize something called variational free energy. So that's basically it. And then the story is, well, what's variational free energy? Well, variational free energy is"
},
{
"end_time": 783.114,
"index": 31,
"start_time": 755.367,
"text": " As with all actions in physics, it's just a path integral or a time integral, a sum, an accumulation of a functional, usually known as a Lagrangian, where in this instance the functional in question, this variation of free energy, can be read as exactly the same quantity that statisticians"
},
{
"end_time": 812.261,
"index": 32,
"start_time": 784.565,
"text": " This Marshawn beast mode lynch prize pick is making sports season even more fun on prize picks whether"
},
{
"end_time": 834.872,
"index": 33,
"start_time": 812.619,
"text": " Football fan, a basketball fan, it always feels good to be ranked. Right now, new users get $50 instantly in lineups when you play your first $5. The app is simple to use. Pick two or more players. Pick more or less on their stat projections. Anything from touchdowns to threes and if you're right, you can win big. Mix and match players from"
},
{
"end_time": 844.735,
"index": 34,
"start_time": 834.872,
"text": " any sport on PrizePix, America's number one daily fantasy sports app. PrizePix is available in 40 plus states including California, Texas,"
},
{
"end_time": 874.138,
"index": 35,
"start_time": 844.974,
"text": " Florida and Georgia. Most importantly, all the transactions on the app are fast, safe and secure. Download the PricePix app today and use code Spotify to get $50 in lineups after you play your first $5 lineup. That's code Spotify to get $50 in lineups after you play your first $5 lineup. PricePix is good to be right. Must be present in certain states. Visit pricepix.com for restrictions and details. The sensory, the path through the sensory states under a particular model that would try to explain"
},
{
"end_time": 903.234,
"index": 36,
"start_time": 874.718,
"text": " hypothesis that would try to explain those data. So you've got this interpretation or certainly the fact that the functional form of this variational free energy that is being minimized by the passive least action of the autonomous the active and the internal states of something a particle or a person now looks as if it is trying to minimize a quantity which provides an approximation or a negative"
},
{
"end_time": 933.336,
"index": 37,
"start_time": 903.592,
"text": " approximation to marginal likelihood or model evidencing. So in other words, it looks as if the most likely paths are trying to maximize model evidence. So one can read that in many different ways. One can read that. And as Jacob Howey has read it, a philosopher and neuroscientist who has a commitment to understanding to this kind of formulation as self-evidencing."
},
{
"end_time": 960.213,
"index": 38,
"start_time": 933.729,
"text": " literally interacting with your environment, actively inferring, engaging with your experienced world, your eco niche in a way that looks as if you're trying to solicit the most evidence for your model of that world. This quantity is also found in machine learning. It's known as an evidence lower bound. I should qualify or just clarify that"
},
{
"end_time": 988.336,
"index": 39,
"start_time": 960.657,
"text": " In machine learning, this free energy is the reverse of the free energy used in physics. So in machine learning, they try to maximize the negative free energy, aka an evidence lower bound or ELBO. While in physics, you're always articulated in terms of minimizing a potential energy or minimizing, in this instance, a variation of free energy. So you'll find that exactly the same functional, the same quantity"
},
{
"end_time": 1016.015,
"index": 40,
"start_time": 990.401,
"text": " of the goodness of some model of how you think your sensations, your sensorium was generated in terms of this variational free energy that leads you to an interpretation of people, particle, persons, inferring, accumulating evidence"
},
{
"end_time": 1044.002,
"index": 41,
"start_time": 1016.271,
"text": " making inferences about the states of the world that are generating their sensations. So that in a nutshell, or perhaps not a nutshell, is a sort of formal explanation of the free energy principle. I did slip in at the beginning, very much like things which you may be familiar with when you were at school, say Hamilton's principle of spatial reaction. It is"
},
{
"end_time": 1074.445,
"index": 42,
"start_time": 1045.35,
"text": " just a method. It allows you to identify, predict, simulate, engineer those trajectories, those paths, those dynamics, those processes that will minimize this action functional or the path integral of the variational free energy. So what that means is that if you now understand or have at hand the functional form"
},
{
"end_time": 1103.797,
"index": 43,
"start_time": 1074.991,
"text": " of your variation of free energy that depends upon posterior beliefs or beliefs about the outside world under a geratin model, quite simply a probability density over the external causes of the consequences that are the sensory states. Then you can use that methodology, use that principle to start to simulate, to build sentient artifacts"
},
{
"end_time": 1132.278,
"index": 44,
"start_time": 1104.104,
"text": " that behave in the way that we assume under the free energy principle that you and I are behaving. Let's see if I can make a nutshell of the nutshell for some people who are more familiar with physics. So let's say you have a ball, you want to know the dynamics of the ball, you can understand this with Newtonian force and that's useful for certain calculations or you can think of it as minimizing a Lagrangian and then that actually provides a more useful for most cases way of calculating the trajectory of the ball. So then"
},
{
"end_time": 1156.493,
"index": 45,
"start_time": 1132.278,
"text": " Right there, when I said ball, we've identified the ball. Now here, you're not just limiting yourself to balls, you're saying let's say cells instead of talking about people, because we can break it down simply, at least for me, for myself, I find it much more easy to think of a cell. Okay, so then we want to know what are the dynamics of the cell? How is the cell, a single cell going to act? Okay, firstly, the first step would be identifying that cell. And when you mentioned the word"
},
{
"end_time": 1180.708,
"index": 46,
"start_time": 1156.817,
"text": " Markov blanket that's a way of identifying what this object is because in order to identify what an object is it's often useful to identify what it isn't at the same time what it is and isn't so that Markov blanket delimits the cell now is that correct so far absolutely yeah so you're adding the cell is a perfect example here you've got you know a cell is just defined by the boundary which would be the surface"
},
{
"end_time": 1215.776,
"index": 47,
"start_time": 1186.032,
"text": " inside the intracellular states would be the internal states, and then the milieu in which that cell lives, which is usually in a bunch or a family of other cells, would constitute all the external states. So you've made the very first step just talking about a cell, a thing. You've defined it in terms of this partition into internal, external, and the sensory states that bound and separate the inside from the outside."
},
{
"end_time": 1242.159,
"index": 48,
"start_time": 1216.118,
"text": " You may be asking, well, where have the active states gone? For a cell, I usually like to associate those with something called actin filaments that just lie underneath the surface, the sensory states, and actively push the cell surface, say, the filia, the mechanisms that cause motion of the cell, should it move around."
},
{
"end_time": 1262.927,
"index": 49,
"start_time": 1242.568,
"text": " those parts, those states that are responsible for cellular motility. So if we're now thinking about a moving cell or a swimming cell, for example, then you need to supplement that partition with active states. So that's absolutely the first step. And then you're going to go on and now translate the ball. What's the ball in this metaphor?"
},
{
"end_time": 1288.677,
"index": 50,
"start_time": 1263.507,
"text": " You also mentioned the word bi-directional earlier. And when I was learning about this, the way that I like to understand this was that in computer science, generally computer scientists think and can only think in terms of you have an input and then there's a black box and you have an output. And that's essentially computer science in a nutshell. It's a bit more complicated than that, but it's in a nutshell. It's you have an input, you perform some calculation and you get an output. The reason why you said bi-directional, at least the way that I understand it,"
},
{
"end_time": 1307.671,
"index": 51,
"start_time": 1289.036,
"text": " So please correct this is it's because of actions in that model where there's input black box output. That's generally if we want to use the term embodied cognition more on the cognition side. So you're just this black box where you're impressed with sensory information. The input is your senses. You're sensing all of what's happening. And then this output now."
},
{
"end_time": 1334.121,
"index": 52,
"start_time": 1308.643,
"text": " is the actions that generally a computer sits there but we're creatures in this world at least we think we're creatures in a physical world the black box has a model and then one way of maximizing or showing that your model is correct is by updating your model or it's by changing what you've sensed in order to match the model now to change what you've sensed is to act so it's useful to think of two black boxes so there's one that's inputting to the other"
},
{
"end_time": 1362.875,
"index": 53,
"start_time": 1334.121,
"text": " and then the output goes into the other, which is external, this is the internal, this is the cell, it senses, and then it acts upon the world, and then that changes the world, which then changes what you sense. So is that the reason that you said bi-directional? Yeah, no, I think that's a really important observation, yeah. So, you know, before we were talking about these states that separate the inside from the outside, and I made a big thing that they had comprised sensory and active states, and that there is a two-way traffic here,"
},
{
"end_time": 1390.213,
"index": 54,
"start_time": 1363.302,
"text": " And that two-way traffic instantiates exactly what you're talking about, which is a circular causality. So, you know, I use the word self-evidencing and described it as soliciting evidence for sensory data that provides evidence for your model of what's going on. But in soliciting, this is something that a computer would not normally do, a passive or a sessile"
},
{
"end_time": 1418.78,
"index": 55,
"start_time": 1391.988,
"text": " would not be in charge of nor would it terribly worry or indeed be equipped with the capacity to select the inputs, to select the data, say it had to analyze. So just having a statistical package, you give it some data and it produces an outcome, say a p-value or some kind of inference. So a lot of computing and a lot of machine learning, you give it data and it provides a prediction or a classification."
},
{
"end_time": 1447.995,
"index": 56,
"start_time": 1419.445,
"text": " That's not the kind of computational process that is implied by the free energy principle. It's much more, as you put it, inactive. It has the aspirations to provide a mathematical image of the action perception cycle, where crucially, you have to go and decide what data to base your inferences on by acting upon the world."
},
{
"end_time": 1477.91,
"index": 57,
"start_time": 1448.404,
"text": " That sort of inactive aspect is the thing that distinguishes this kind of formalism, this calculus, this Bayesian calculus, which is quintessentially inactive. It's not just a good computer. It can be made a good computer just by removing active states if you want to. But that would be a limiting and a rather biophysically unrealizable and unsustainable kind of artifact."
},
{
"end_time": 1505.555,
"index": 58,
"start_time": 1478.66,
"text": " The whole point here is that there is a reciprocal exchange of the world. So your notion of having two black boxes talking to each other, I think, is a really nice notion. And indeed, when applying this principle of least action, the free energy principle, to interesting scenarios, one ends up very often putting two"
},
{
"end_time": 1532.739,
"index": 59,
"start_time": 1505.879,
"text": " sets of internal states each equipped with Markov blankets talking to each other. So the active states of one blanket or one particle or one person now constitute the sensory states of the other and vice versa. So you've got now two systems trying to model each other and if their game is to effectively provide or solicit"
},
{
"end_time": 1562.278,
"index": 60,
"start_time": 1533.097,
"text": " the most evidence for their model of the rest of the world, which is just another artifact very much like you, who's trying to do exactly the same thing. What will inevitably emerge from that joint minimization of free energy or the maximization of marginal likelihood is the kind of active engagement that renders, say you and me as the two black boxes, renders as mutually predictable."
},
{
"end_time": 1585.998,
"index": 61,
"start_time": 1562.483,
"text": " So what we will do is we will come to share the same generative model, the same narrative, so that I can predict exactly what you're going to say next, given that's the kind of thing that I would say, because that's the sort of world that I'm trying to model. So immediately you have this picture of the minimization of the joint variational free energy"
},
{
"end_time": 1611.834,
"index": 62,
"start_time": 1586.271,
"text": " or the maximization of the joint marginal likelihood for shared models. And one gets from that some quite compelling and interesting formulations of multi-agent games, communication, synchronization between conspecifics, that all rest upon trying to resolve uncertainty"
},
{
"end_time": 1641.374,
"index": 63,
"start_time": 1612.295,
"text": " Can any system, let's say it's more than two. So in the example that I gave and that we were talking about, there's one cell and then the external world. But obviously the world is much more complex than that. And let's just imagine slightly more complicated. There's two cells and then an external world."
},
{
"end_time": 1671.578,
"index": 64,
"start_time": 1642.142,
"text": " Is it always the case that from the perspective of one cell, you may as well just treat the external world, even though it comprises an external world plus another cell out there. Can you always treat that as its own singular black box such that we can always understand the dynamics as two black boxes? Or is there ever a need for a third, a fourth, a fifth? I think there probably would be a need if the external states are themselves"
},
{
"end_time": 1701.391,
"index": 65,
"start_time": 1672.056,
"text": " So if my universe is composed of external states that can themselves be carved into natural kinds, you know, like other artifacts and objects and air, planets, liquids, natural kinds, as soon as you talk about a natural kind, you're talking about a kind of thing, as soon as there is a thing that's in play, there's a mark of blanket. So what that tells you is,"
},
{
"end_time": 1708.097,
"index": 66,
"start_time": 1701.578,
"text": " In principle, it would be possible to take all the states that constitute, say, an eco-niche."
},
{
"end_time": 1731.749,
"index": 67,
"start_time": 1708.302,
"text": " find, identify the thing of interest, say me and my internal states and then I would identify my blanket states and then I got the rest of the states. But now I can start again, I can find another set of internal states and identify its blanket states and then take those states off the table and then start again and recursively tile all the external"
},
{
"end_time": 1753.422,
"index": 68,
"start_time": 1733.422,
"text": " Because there's a statistical insulation or separation of internal"
},
{
"end_time": 1779.514,
"index": 69,
"start_time": 1757.329,
"text": " I don't need to know your internal states because everything that is knowable about your internal states is on the surface, on your blanket states. So now the picture that emerges is an ensemble of things, natural kinds, that are coupled to each other via their blanket states."
},
{
"end_time": 1808.37,
"index": 70,
"start_time": 1780.213,
"text": " No individual particle or person or thing will ever have access to the internal states of anything else. But there will be coupling and influences that are mediated by the blanket states, the active states, and possibly the sensory states. Mathematically, you can actually have the sensory states also influencing the outside. But for simplicity, let's assume that sensory states do not influence external states."
},
{
"end_time": 1840.52,
"index": 71,
"start_time": 1811.203,
"text": " You mentioned something extremely interesting, and I want to pick up on this. You said that the sensory states can influence the external states. At least that's not how I understand it. So how can a sensory state influence an external state? So I said that mathematically it is that is possible. So if you if you think about, you know, from a pure from the point of view of somebody who's trying to understand"
},
{
"end_time": 1857.585,
"index": 72,
"start_time": 1840.862,
"text": " the statistical behavior, the probabilistic behavior of a set of dynamics described as differential equations or coupled systems, then the name of the game really is to ask your"
},
{
"end_time": 1885.981,
"index": 73,
"start_time": 1857.773,
"text": " What statistical dependencies that define a Markov blanket, specifically a Markov blanket, renders the internal states conditionally independent of the external states, given the Markov blanket. So that's their role. That's what is implied by a separation of the inside from the outside, that all you need to know is the blanket states, because given the blanket states, the inside and the outside are conditionally independent."
},
{
"end_time": 1913.251,
"index": 74,
"start_time": 1886.681,
"text": " So the question now is, what kind of sparse coupling among all the states at hand give rise to that conditional independence that definitively specifies the Markov blanket? It so happens that all you need is the following rule. Internal states can only influence themselves and active states."
},
{
"end_time": 1926.476,
"index": 75,
"start_time": 1914.241,
"text": " External states can only influence themselves and sensory states. That's all you need. So with that condition in place, you now immediately have"
},
{
"end_time": 1955.401,
"index": 76,
"start_time": 1926.749,
"text": " a Markov blanket or a particular partition. From that follows the conditional independence and therefore the interpretation of the autonomous dynamics, which is a description of the trajectories or the paths of the processes taken by the active and the internal states of a particle. So that means that it is actually mathematically allowable"
},
{
"end_time": 1983.541,
"index": 77,
"start_time": 1956.067,
"text": " in the sense that you do not destroy that Markov blanket or the statistical properties of that Markov blanket. It is allowable for the sensory states to influence active states. And you may be thinking, well, that's a bit silly. There's no way in which my photoreceptors or my cochlear in my ear materially affect the outside. I think that's absolutely true. So for things like you and me,"
},
{
"end_time": 2013.865,
"index": 78,
"start_time": 1983.899,
"text": " then you would not normally expect the sensory states to influence the external states. But that's not the case for much simpler things. When you look at much simpler things where most of the dynamical influences, the coupling that we're talking about that you'd write down in terms of the differential equations, the Langevin equation, most of those influences are basically short range."
},
{
"end_time": 2037.551,
"index": 79,
"start_time": 2014.48,
"text": " So we can preclude action at a distance, for example. If you're a single cell, unless you're next to me, I don't really know what you're up to. I'm having to exclude here all sorts of interesting things about long-range electrochemical communication and electrical gradients and the like, but let's just take a bag of cells that"
},
{
"end_time": 2059.138,
"index": 80,
"start_time": 2037.944,
"text": " They just diffuse stuff locally. They touch each other and that's the only way that they can influence each other is by being proximate in some Euclidean space. So everything's short range. Now in this situation, you have a very different kind of structure. So here, the sensory states are the cell surface."
},
{
"end_time": 2088.131,
"index": 81,
"start_time": 2059.411,
"text": " And when these active states that lie underneath the cell surface change the environment, they do so by pushing the cell surface into the external milieu and cause changes in the sensory states, the surface states of other cells. So being a cell is probably a good example of where your sensory states, which just are your surface states, actually do all the heavy lifting in terms of changing the outside."
},
{
"end_time": 2113.626,
"index": 82,
"start_time": 2088.336,
"text": " the spatial relationship with all the other cells that you're co-inhabiting a particular organ, for example. So it's a little bit counterintuitive, but I think a really interesting sort of thought experiment and actually a practical consideration when it comes to looking at the different ways in which"
},
{
"end_time": 2143.234,
"index": 83,
"start_time": 2114.019,
"text": " insides coupled to outsides and the nature of interactions between the internal states of something and the external states of something and we sort of take it for granted that we can do action at a distance. Literally I can talk to you in Canada while you're talking while I'm in London. That is a beautiful example of this action to distance that means that we have quite a special"
},
{
"end_time": 2163.729,
"index": 84,
"start_time": 2143.473,
"text": " a kind of coupling between our sensory states and our active states that is not a gift of very simple cellular like structures, an automata that can't interpret signals or be subject to or be sensitive to"
},
{
"end_time": 2189.565,
"index": 85,
"start_time": 2164.07,
"text": " Just so you know, or for people who are listening and if this wasn't obvious enough, the free energy principle is famous for, or perhaps infamous, for being somewhat inscrutable. For people who have a physics background or math background, what would be the minimum prerequisites? This is how I like to conceptualize the podcast before I interview someone. I think about, okay, what are all the pieces of knowledge I need to know prior to interviewing them?"
},
{
"end_time": 2215.128,
"index": 86,
"start_time": 2189.565,
"text": " So obvious example is let's say QFT. You need to understand classical mechanics, Lorentz of variance, quantum mechanics, perhaps bundle theory if you want to understand Yang-Mills and so on properly. So what would be the prerequisites for someone to properly understand the free energy principle, let's say if they have a math or physics background? It wouldn't be, it would be probably much less than you'd expect to encounter at university."
},
{
"end_time": 2243.097,
"index": 87,
"start_time": 2215.469,
"text": " I think you'd certainly need to know sort of, you know, the basics of variational calculus and it would be really nice if you understood where a Lagrangian came from and to understand the simplicity of action, you know, basically, you know, exactly Hamilton's principle of least action and how that inherits from patheticals of Lagrangians and how that arises from"
},
{
"end_time": 2264.019,
"index": 88,
"start_time": 2243.49,
"text": " dynamical systems theory of a very simple sort where you can write down any random dynamical system as, you know, a stochastic differential equation or a Langevin equation. I think that's probably quite sufficient, you know, what you would need, I think, to"
},
{
"end_time": 2293.695,
"index": 89,
"start_time": 2264.343,
"text": " appreciate all the interpretational richness afforded by what happens to the paths of least action when you have a Markov blanket, you need to know a little bit of probability theory, you need to know what a conditional distribution is, you need to know what have an intuition as to what is meant by model evidence for example or marginal likelihood. So with I would imagine"
},
{
"end_time": 2318.166,
"index": 90,
"start_time": 2293.933,
"text": " a week's foraging on Wikipedia, you could probably acquire that if you were fluent already in sort of empathological-like formulations of the kind you can find on Wikipedia. So you certainly wouldn't need to know what sort of bundles were, you wouldn't need category theory, you wouldn't need to know anything about gauge theories, you"
},
{
"end_time": 2345.759,
"index": 91,
"start_time": 2320.828,
"text": " Your example of Wikipedia is"
},
{
"end_time": 2373.66,
"index": 92,
"start_time": 2346.203,
"text": " is on point because the way that I conceptualize this interview, this podcast and the way that we spoke about it prior to going on air via emails is it's going to be tendril-like or gossamer-like where you have these words imagining what you're saying is like a Wikipedia page and you say certain keywords which would be if you were a Wikipedia page underlined in blue and worthy of their own investigation in and of themselves. So some of these I'm going to pick out and interrupt you"
},
{
"end_time": 2399.821,
"index": 93,
"start_time": 2374.411,
"text": " sometimes and I apologize for that like I did earlier in order to in order to properly set the prerequisites because we had a four-hour behemoth before and that was meant to serve as an introduction to the free energy principle however there are some people who were still uncertain as to exactly what the free energy principle is so this will serve as an introduction to that introduction okay hear that sound"
},
{
"end_time": 2426.92,
"index": 94,
"start_time": 2400.759,
"text": " That's the sweet sound of success with Shopify. Shopify is the all-encompassing commerce platform that's with you from the first flicker of an idea to the moment you realize you're running a global enterprise. Whether it's handcrafted jewelry or high-tech gadgets, Shopify supports you at every point of sale, both online and in person. They streamline the process with the internet's best converting checkout, making it 36% more effective than other leading platforms."
},
{
"end_time": 2452.978,
"index": 95,
"start_time": 2426.92,
"text": " There's also something called Shopify Magic, your AI-powered assistant that's like an all-star team member working tirelessly behind the scenes. What I find fascinating about Shopify is how it scales with your ambition. No matter how big you want to grow, Shopify gives you everything you need to take control and take your business to the next level. Join the ranks of businesses in 175 countries that have made Shopify the backbone"
},
{
"end_time": 2476.34,
"index": 96,
"start_time": 2452.978,
"text": " of their commerce. Shopify, by the way, powers 10% of all e-commerce in the United States, including huge names like Allbirds, Rothies, and Brooklynin. If you ever need help, their award-winning support is like having a mentor that's just a click away. Now, are you ready to start your own success story? Sign up for a $1 per month trial period at Shopify.com"
},
{
"end_time": 2486.613,
"index": 97,
"start_time": 2476.34,
"text": " Go to Shopify.com slash theories now to grow your business no matter what stage you're in Shopify.com slash theories."
},
{
"end_time": 2507.619,
"index": 98,
"start_time": 2489.838,
"text": " Razor blades are like diving boards. The longer the board, the more the wobble, the more the wobble, the more nicks, cuts, scrapes. A bad shave isn't a blade problem, it's an extension problem. Henson is a family-owned aerospace parts manufacturer that's made parts for the International Space Station and the Mars Rover."
},
{
"end_time": 2536.101,
"index": 99,
"start_time": 2507.619,
"text": " Now they're bringing that precision engineering to your shaving experience. By using aerospace-grade CNC machines, Henson makes razors that extend less than the thickness of a human hair. The razor also has built-in channels that evacuates hair and cream, which make clogging virtually impossible. Henson Shaving wants to produce the best razors, not the best razor business, so that means no plastics, no subscriptions, no proprietary blades, and no planned obsolescence."
},
{
"end_time": 2558.319,
"index": 100,
"start_time": 2536.101,
"text": " It's also extremely affordable. The Henson razor works with the standard dual edge blades that give you that old school shave with the benefits of this new school tech. It's time to say no to subscriptions and yes to a razor that'll last you a lifetime. Visit hensonshaving.com slash everything. If you use that code, you'll get two years worth of blades for free. Just make sure to add them to the cart."
},
{
"end_time": 2582.875,
"index": 101,
"start_time": 2558.319,
"text": " Plus 100 free blades when you head to h-e-n-s-o-n-s-h-a-v-i-n-g dot com slash everything and use the code everything. You mentioned marginal likelihood slash model evidencing and I believe you see those two as the same and I'm unclear as to what any of those two are. So can you give an explanation? What is model evidencing slash marginal likelihood?"
},
{
"end_time": 2613.046,
"index": 102,
"start_time": 2583.968,
"text": " Excellent question. So this is where you would need to be slightly fluent in the concepts and constructs that statisticians would bring to the table. So if I now want to use probability theory to articulate the job of a statistician trying to make sense of data, just note"
},
{
"end_time": 2637.329,
"index": 103,
"start_time": 2616.305,
"text": " You'd start off by understanding the problem in terms of estimating under uncertainty, sort of assimilating data whilst quantifying and accommodating"
},
{
"end_time": 2663.951,
"index": 104,
"start_time": 2637.654,
"text": " uncertainty due to random effects. So, you know, the random effects here can be read as the random fluctuations or the vena fluctuations or the innovations in some stochastic process from the point of view of a physicist, from the point of view of a statistician. That randomness is inherited as random effects. Observation noise, for example, would be one if you were dealing with"
},
{
"end_time": 2689.838,
"index": 105,
"start_time": 2664.258,
"text": " a state space model, which is really a model of a dynamical process that a statistician might use to do data assimilation given some time series data. There may be a distinction between observation noise or sensor noise and the noise or the random fluctuations on the states that can't be directly observed, and that would be known as state noise or system noise,"
},
{
"end_time": 2721.169,
"index": 106,
"start_time": 2694.292,
"text": " But from the point of view of the statistician, this becomes noise or uncertainty or random effects. So how do you deal with this? Well, basically, you're confronted with a problem of some data and you have a hypothesis or a model about how those data were generated by some latent states, sometimes known as hidden states and hidden"
},
{
"end_time": 2749.172,
"index": 107,
"start_time": 2721.459,
"text": " Here, I think, just as to try and connect it to the Markov blanket, they're hidden behind the Markov blanket. So the statistician can, if you can think of it like being in internal states, they're surrounded, they're enshrouded by a Markov blanket. The data is impressing themselves on this sensory veil. There's something on the outside that are latent in the sense that they're hidden behind this veil. So they only have access to the data. They don't know how"
},
{
"end_time": 2778.268,
"index": 108,
"start_time": 2749.531,
"text": " the data were caused so for example they might have a hypothesis that this drug treatment materially affected some pathophysiology that reduced the symptoms that were measurable or recordable by the doctor but it's only the observable data that the doctor can get from the patients that the statistician has at hand now the statistician now has to infer was there a drug effect"
},
{
"end_time": 2806.442,
"index": 109,
"start_time": 2778.712,
"text": " And what she will do, she will create a generative model that if there was a drug effect of a certain size, then I'd expect a separation of the data, for example. Now, that generative model at the end of the day can always be written down as a joint probability density over the causes and consequences where the causes are the latent or the hidden states"
},
{
"end_time": 2836.408,
"index": 110,
"start_time": 2806.698,
"text": " that you are trying to estimate, and the consequences of the observable data that you have at hand. Given that generative model, you can now estimate, given some data, the most likely hidden or latent states or causes of that data, basically the treatment effect that you hypothesized. So that's the posterior distribution. That's the probability of some hidden state, say the external states, the Markov blanket,"
},
{
"end_time": 2863.319,
"index": 111,
"start_time": 2836.817,
"text": " given conditioned upon the data. So that's the holy grail usually of a statistical analysis and you know what your most classical inference for example those based on t-tests or f-tests right through to more sophisticated and informed log-odd ratios such as Bayes factors. These are just scoring"
},
{
"end_time": 2893.063,
"index": 112,
"start_time": 2863.933,
"text": " the evidence for an effective interest expressed in this latent state space in terms of the probability of one model relative to another model and usually in science that would be the altered hypothesis versus a null hypothesis and then you can look at the ratio of the likelihood of this model or more specifically the likelihood of the data given this model divided by"
},
{
"end_time": 2914.36,
"index": 113,
"start_time": 2893.285,
"text": " likelihood of the data given another model, usually the null hypothesis. So to come to our example of the statistician trying to infer whether the drug has had a remedial effect or not, what she would do is compute the probability of her data with"
},
{
"end_time": 2943.78,
"index": 114,
"start_time": 2915.162,
"text": " an effect and then compute the probability of her data if there was no effect of the drug. So the data sets from the treatment and the non-treatment group were treated as from a single group. And then she would use that to make an inference that there was indeed an effect. So that probability, the probability of the data given a particular model, a specific model, be it the alternate hypothesis or the null hypothesis, that's the marginal likelihood"
},
{
"end_time": 2973.899,
"index": 115,
"start_time": 2944.087,
"text": " So why is it called a marginal likelihood? Well, what you've done is you've estimated, given your generative model, the probability of the data having marginalized or averaged away the thing that you don't know, which is the effect size, the treatment effect. So the probability of the data given a model is the average probability of the data"
},
{
"end_time": 3003.558,
"index": 116,
"start_time": 2974.258,
"text": " given that model and the effect size averaged under your beliefs about the effect size. So that's why it's called a marginal, a marginal likelihood. That's a great question because it actually leads one to the genesis of variational free energy on one reading in the context of Richard Feynman's pathological formulation of quantum electrodynamics."
},
{
"end_time": 3033.712,
"index": 117,
"start_time": 3004.138,
"text": " He had this marginalization problem. So he wanted to, or one could read the history as him wanting to evaluate the marginal likelihood of paths of, say, small particles, which would necessarily involve computing or marginalizing"
},
{
"end_time": 3060.162,
"index": 118,
"start_time": 3034.224,
"text": " integrating of extremely high dimensional probability distributions which is intractable. So in order to solve the marginalization or the integration problem to ensure everything sums to one effectively to get your partition function, he introduced the variation free energy and that was a really clever move because he converted what was an intractable"
},
{
"end_time": 3087.261,
"index": 119,
"start_time": 3060.52,
"text": " a non-realizable and incomputable marginalization or integration problem into a tractable optimization problem simply by introducing this variational free energy which was a bound on the marginal likelihood. So that's why it's called an elbow or an evidence lower bound simply because it now becomes a computable object or mathematical functional"
},
{
"end_time": 3114.77,
"index": 120,
"start_time": 3087.619,
"text": " that you can evaluate given some data and a generative model that is always in machine learning lower than the log of the marginal likelihood. In my world, the physics world, it's the other way around, but let's stick with the machine learning world. So the negative variational free energy of the physicist is always smaller than the thing you want to actually evaluate. So if you just optimize by pushing up"
},
{
"end_time": 3142.841,
"index": 121,
"start_time": 3115.077,
"text": " the elbow or the variation free energy until you can't make it any larger. And under the assumption that the bound, the times that the bound is high, you've not got a good approximation to what you were chasing, which is the marginal likelihood. And once the statistician has that marginal likelihood, she can repeat the process for another model. So the null hypothesis with no treatment effect. And then she can take the ratio of the free energies, the variation free energies,"
},
{
"end_time": 3171.869,
"index": 122,
"start_time": 3143.114,
"text": " and start to compare the models using base factors and then go back to the doctor and say, look, I am pretty confident that that drug really did have an effect. And I can say that because the evidence in your data suggests that the marginal likelihood of a hypothesis or a model that included an effective drug was 52.6 times greater than the marginal likelihood if I assume there had been no change."
},
{
"end_time": 3192.449,
"index": 123,
"start_time": 3172.346,
"text": " So that's where the notion of a marginal likelihood comes from. So the marginal likelihood just is the evidence for a model. It just is the probability of some data in the present context, you know, sensory or from the part of the free energy principle, the sensory states."
},
{
"end_time": 3221.749,
"index": 124,
"start_time": 3192.756,
"text": " or the sensory trajectories, say time series, given a particular model marginalizing away all your beliefs about the particulars, effect size, and the parameters, and all the quantities that were responsible for generating those data under the structure afforded by the model. Is that clear? I mean, these are the fundaments of"
},
{
"end_time": 3248.609,
"index": 125,
"start_time": 3222.176,
"text": " machine learning, data simulations, system identification, and you could actually argue this is one way of nuancing the scientific process, not quite in using the rhetoric of Karl Popper, but certainly has this Popperian aspect to it, that every bit of scientific inquiry"
},
{
"end_time": 3277.261,
"index": 126,
"start_time": 3248.933,
"text": " including the way we live our lives, is just about hypothesis testing. And we just generate alternative explanations for the evidence at hand. And then we try to evaluate the evidence for this hypothesis relative to that hypothesis, evaluate the marginal likelihood of the data at hand for this hypothesis versus that hypothesis, this model versus that model. And then we usually commit to the model with the greatest evidence. This just is."
},
{
"end_time": 3300.213,
"index": 127,
"start_time": 3277.449,
"text": " evidence-based scientific progress. You're having committed to or selected, technically known as Bayesian model selection, this hypothesis. You then think about, okay, what's the next space of hypotheses or what's the next portfolio of ideas or models that I want to now go and gather evidence for?"
},
{
"end_time": 3328.814,
"index": 128,
"start_time": 3300.93,
"text": " I think this analogy is quite nice. I'm glad you let me tell this explosion because a good scientist then, the good one, will generate new hypotheses and then the problem contending the scientist who's just chasing these marginal likelihoods or model evidences for her models of her fielding of inquiry is to design a good experiment."
},
{
"end_time": 3352.841,
"index": 129,
"start_time": 3329.701,
"text": " So here we come back to the other side of the coin that takes us beyond machine learning as a black box, give it some inputs, give it an output. We now come back to the design of our experiments, how we actively secure and solicit the evidence that will best disambiguate"
},
{
"end_time": 3381.749,
"index": 130,
"start_time": 3352.841,
"text": " among our hypotheses and I repeat usually when you're writing scientific papers it's a null hypothesis versus an altered hypothesis but in real life and in real scientific thinking people you know normally have eight hypotheses or a hundred hypotheses and they're scoring all of them by evaluating the marginal likelihood of the data that they have carefully solicited by Bayes Optimal Experimental Design"
},
{
"end_time": 3408.097,
"index": 131,
"start_time": 3382.056,
"text": " to maximally disambiguate in terms of the marginal likelihood on the model evidence between all of their hypotheses. So that sort of brings another aspect of probability theory to the table which would be an important complement for somebody wanting to understand Bayesian statistics. The base aspect of course is that"
},
{
"end_time": 3434.309,
"index": 132,
"start_time": 3408.49,
"text": " As soon as you start to talk about conditional distributions of the kind that a marginal likelihood is, it's conditional in the sense that it's a probability of some data conditioned upon a model or given a model, having marginalised away all the parameters of that model. And building a posterior, the probability of my hidden states, the parameters of my model,"
},
{
"end_time": 3458.677,
"index": 133,
"start_time": 3434.735,
"text": " given the data falls out of Bayes rule. So we're talking about Bayesian inference here and just straightforward probability theory that can be articulated in terms in terms of Bayes rule. But there's another really important application of Bayes rule which speaks to this sort of inactive situated statistician where statistician now"
},
{
"end_time": 3487.159,
"index": 134,
"start_time": 3459.189,
"text": " is not simply in the service of the doctor supplying her with data and asking a particular question. Now, the statistician is in charge of designing the perfect experiment that gives her the data that allows her to make the most efficient and statistically powerful inference about whether there was a treatment of bed. And that's the problem of optimal experimental design."
},
{
"end_time": 3512.978,
"index": 135,
"start_time": 3487.602,
"text": " There is an answer to that, it's called the principle of Bayes' optimal design, first articulated by people like Lindley in the 1950s, and then re-articulated by people like David McCain in active learning. And it is, if you like, the other side of the Bayesian coin when it comes to"
},
{
"end_time": 3542.619,
"index": 136,
"start_time": 3514.189,
"text": " Bayesian assimilation of statistical assimilation of data and inference. And one could also argue it's the other side of the coin when it comes to things like Bayesian decision theory as well. But that's a slight distraction. I think we just stick with the big idea here that, you know, there are good. Let's look at it like this. If it is the case that the marginal likelihood is just the probability of some data given a model,"
},
{
"end_time": 3572.056,
"index": 137,
"start_time": 3543.217,
"text": " And your job is to evaluate that quantity and usually optimize the marginal likelihood or maximize the evidence for your model. You can either change the model or you can change the data. And that means that you've got sort of two problems to solve, as it were, if you cast this as an optimality problem, as a maximization of an evidence lower bound,"
},
{
"end_time": 3602.022,
"index": 138,
"start_time": 3572.534,
"text": " I can either change my mind about you or I can try and change you by querying you, asking a question, performing an experiment upon you to get some better"
},
{
"end_time": 3631.34,
"index": 139,
"start_time": 3602.415,
"text": " better data from you that resolves my uncertainty about you. Is there a relationship between changing your evidence versus changing your model and the psychoanalytic repression? So repression would be saying, you know what, that data doesn't exist, let me ignore it, versus the perhaps a cognitive behavioral therapy technique would be to say, no, go out and perhaps update your model of it, view it differently, view the maybe it's our arachnophobia, for example. I think it's a beautiful analogy. And"
},
{
"end_time": 3659.923,
"index": 140,
"start_time": 3631.852,
"text": " Hear that sound? That's the sweet sound of success with Shopify. Shopify is the all-encompassing commerce platform that's with you from the first flicker of an idea to the moment you realize you're running a global enterprise. Whether it's handcrafted jewelry or high-tech gadgets, Shopify supports you at every point of sale, both online and in person. They streamline the process with the internet's best converting checkout, making it 36% more effective than other leading platforms."
},
{
"end_time": 3685.981,
"index": 141,
"start_time": 3659.923,
"text": " There's also something called Shopify Magic, your AI-powered assistant that's like an all-star team member working tirelessly behind the scenes. What I find fascinating about Shopify is how it scales with your ambition. No matter how big you want to grow, Shopify gives you everything you need to take control and take your business to the next level. Join the ranks of businesses in 175 countries that have made Shopify the backbone"
},
{
"end_time": 3709.343,
"index": 142,
"start_time": 3685.981,
"text": " of their commerce. Shopify, by the way, powers 10% of all e-commerce in the United States, including huge names like Allbirds, Rothy's, and Brooklyn. If you ever need help, their award-winning support is like having a mentor that's just a click away. Now, are you ready to start your own success story? Sign up for a $1 per month trial period at Shopify.com"
},
{
"end_time": 3733.148,
"index": 143,
"start_time": 3709.343,
"text": " Probably beyond an analogy, there are people in computational psychiatry who think exactly like that."
},
{
"end_time": 3753.2,
"index": 144,
"start_time": 3733.968,
"text": " You can find articles, trends in cognitive science, reviews of the kind, you know, some that deal with, you know, how do you resolve cognitive dissonance? How do you understand wishful thinking? So examples, I think that's a perfect example of this way of maximizing"
},
{
"end_time": 3766.357,
"index": 145,
"start_time": 3753.456,
"text": " You shouldn't try to change the evidence to match your model, you should update your model."
},
{
"end_time": 3794.94,
"index": 146,
"start_time": 3766.647,
"text": " And in fact, there are some political implications. I'm sure people shout that on the streets across various political divides. So is this an argument to say that updating the model is actually the one that should be primary? Because the way that I understand from how you've explained it is that both are equally valid. However, when I look across the world and I look at how we think just in terms of common sense, psychological advice, at least to us in the 21st century, it sounds like you should update your model. But I think you're right to say that it would be"
},
{
"end_time": 3822.841,
"index": 147,
"start_time": 3795.452,
"text": " disingenuous from a mathematical point of view to privilege the model versus the data. So how does one choose between those mathematically? I don't think you can in the sense you can't split a coin. A coin has two sides to it. The whole point of this inactivist take on sense making"
},
{
"end_time": 3842.978,
"index": 148,
"start_time": 3823.063,
"text": " is that you have to optimally go and listen to the right kind of news, talk to the right kind of people in order to solicit the right kind of data that's going to maximize the evidence for your model, whilst at the same time you are updating your model, sometimes known as basing belief, updating"
},
{
"end_time": 3871.954,
"index": 149,
"start_time": 3843.439,
"text": " or certain basic model selection, for example. So the two things go hand in hand. I think one way of resolving or answering your question is just to say in a political or perhaps even sort of communication science sense or these sort of the social aspects of it. What you're saying is that"
},
{
"end_time": 3901.92,
"index": 150,
"start_time": 3872.722,
"text": " There is now, once you acknowledge that you also are in charge of getting the data, then you also have to contend with the problems of how much weight you afford to that data versus that data and where you get your data from. Which news channels do you forage? Who do you listen to? Who do you trust? Who do you ignore?"
},
{
"end_time": 3926.8,
"index": 151,
"start_time": 3902.568,
"text": " And we do this all the time. We do this in a very skillful way, so personally, in terms of deploying our attention. So if you're an engineer and you were thinking about assimilating visual information using a Kalman filter, for example, then one of the most artful challenges or artful resolutions"
},
{
"end_time": 3952.824,
"index": 152,
"start_time": 3926.954,
"text": " that the Kalman filter brings to the table is to get the Kalman gain right. How much weight do you afford? The prediction errors that supply the upper term of a Bayesian or a Kalman filter. And we do that on the fly all the time, right down to the level of turning on and off the attention or the precision or the gain afforded"
},
{
"end_time": 3981.596,
"index": 153,
"start_time": 3953.08,
"text": " visual information. So my favorite example is, um, saccadic eye movements. Um, so just to demonstrate that, you know, one can do this experiment at home. You can just look, um, from one side of your visual scene to the other side and back and forth and back and forth with saccadic eye movements. Uh, and you will not perceive the world jumping around. However,"
},
{
"end_time": 4007.022,
"index": 154,
"start_time": 3981.988,
"text": " If you reproduce the same visual impressions just by palpating your eye in the spirit of Helmholtz. What's that palpating eye? What does that mean? Well, you can literally do this now. If you just take your finger and place it gently on the outer aspect of your eyeball and just gently press in until you see the world shift around and move around. So you can see visual motion."
},
{
"end_time": 4035.811,
"index": 155,
"start_time": 4007.637,
"text": " that is a consequence of optic flow introduced in this instance by you pushing your retina, translating your retina with the input fixed. And that is basically exactly the same from the point of view of the retina, the back of your eyeball, as moving the eyeball with the world stationary. But notice, you only saw motion of the world"
},
{
"end_time": 4065.708,
"index": 156,
"start_time": 4036.237,
"text": " when you were moving your eyeball. When you move your eyes in the normal way, you don't see that motion. And the reason is that your brain ignores visual input during eye movement. And that's called saccadic suppression. So I think it's a beautiful example of the fact that every moment we are selectively ignoring information that is not newsworthy. We are selecting it. We don't know we're doing it, but we do it all the time."
},
{
"end_time": 4093.729,
"index": 157,
"start_time": 4066.152,
"text": " Of course, at a much more abstract and epistemic level, exactly the same principles apply when we choose which newspapers, which Twitter accounts to follow, which friends to talk to, which universities to go to. Is it going to be a Scholarpedia or a Wikipedia? All of these decisions that we make in terms of securing the right kind of informative, precise data"
},
{
"end_time": 4114.497,
"index": 158,
"start_time": 4094.053,
"text": " that will literally have a salience in the sense that it complies with the principles of optimum Bayesian design in the sense that it will resolve the most uncertainty about what I don't know. So this is absolutely crucial, this sort of being in charge of the data."
},
{
"end_time": 4137.176,
"index": 159,
"start_time": 4115.026,
"text": " even before thinking about the consequences of your belief updating, just knowing that you have to solicit my debt and of course from a sociopolitical point of view what we're talking about now is how to disambiguate between fake news and true news, how to"
},
{
"end_time": 4161.63,
"index": 160,
"start_time": 4137.176,
"text": " solicit data in a way that precludes as being isolationist or precludes as succumbing to cult-like ideologies, you know, precludes fundamentalism of a dysfunctional sort, or why is it dysfunctional? Well, usually because that kind of fundamentalism does not speak to and does not"
},
{
"end_time": 4191.527,
"index": 161,
"start_time": 4161.852,
"text": " So what you will find is that that dialectic between changing your mind on the basis of some evidence and changing the evidence by changing where you look for that evidence, that balance is at the heart of action and perception. And indeed, you could argue that all of action"
},
{
"end_time": 4207.381,
"index": 162,
"start_time": 4192.278,
"text": " is simply in the service of gathering better data. What do you mean by better? Those data that minimise your surprise, minimise your expected surprise or your uncertainty, resolving uncertainty about the way you think"
},
{
"end_time": 4224.787,
"index": 163,
"start_time": 4207.79,
"text": " This sounds like a deeper issue than just a political problem. It sounds like an issue for the process of suggesting that the conclusions that one has come to was from a dispassionate assessment of the evidence."
},
{
"end_time": 4253.524,
"index": 164,
"start_time": 4224.787,
"text": " So you hear people say this, generally people who like to think of themselves as extremely rational people who've come to their conclusions because they have objectively analyzed some situation. It sounds like that is a doomed project because we're all sampling different data without realizing it. Is that the case or no? Yes, no, I think that's absolutely right. I don't think there's anything bad about it. I mean, that's just the way things are. So how does one overcome that? What do we do? How is it that the scientific process seems to work?"
},
{
"end_time": 4273.609,
"index": 165,
"start_time": 4254.77,
"text": " Well, it works because the operational definition of working simply is maximizing the marginal likelihood. So you are a liberty to actually maximize the marginal likelihood by either changing the data or changing the model. So notice that"
},
{
"end_time": 4302.21,
"index": 166,
"start_time": 4273.933,
"text": " in this instance, you're not going to make any false inferences, provided that all the data you solicit is always very consistent with your model. You won't need to change your mind. If you've got the wrong model, the data will provide disconfirmatory evidence and you may well try and get data from another source. And if you can do that and continue engaging until your model is fit for purpose, what you have basically done is find a niche for yourself"
},
{
"end_time": 4331.186,
"index": 167,
"start_time": 4302.449,
"text": " where your ideology, your conceptions, your understanding of the way the world works is perfectly fine. So you've found a cultural eco-niche or possibly a physiological eco-niche or an information eco-niche that suits you and you will have a high adaptive fitness and survive and persist in that eco-niche. If you can't, you move to another one. And so now we have a model of speciation. We have a model of"
},
{
"end_time": 4358.234,
"index": 168,
"start_time": 4331.476,
"text": " shared narratives, shared convictions that can be communicated, so supply and teach each other, that is self-assembling and self-constructing, co-created simply because you are looking for, at the same time you're looking for information that resolves the uncertainty about your models of how the world works,"
},
{
"end_time": 4378.268,
"index": 169,
"start_time": 4358.558,
"text": " which makes it look as if from the outside, you're basically changing the world to make your hypotheses come true. And that, of course, is just another way of describing action. So if we come away from the sort of the exchange of information as a scientist or somebody"
},
{
"end_time": 4406.596,
"index": 170,
"start_time": 4379.838,
"text": " In a sense that is you changing the data to match your model. So if you"
},
{
"end_time": 4436.596,
"index": 171,
"start_time": 4406.92,
"text": " So this falls out of the free energy principle because, you know, if it's the case that the autonomous paths, the dynamics that constitute our active engagement with the world, if they're trying to minimize free energy, and we in this instance minimize free energy as a prediction error or surprise, remembering that it's the negative of the marginal likelihoods, it's something if"
},
{
"end_time": 4463.507,
"index": 172,
"start_time": 4437.022,
"text": " the probability of these data, given my understanding of my construction of the world at this point in time, is very very low. Then it will have a very very small marginal likelihood or have a very very high free energy and therefore it will be surprising given my model of the world. Some people like to articulate that surprise in terms of a prediction. What I predicted was going to happen"
},
{
"end_time": 4489.002,
"index": 173,
"start_time": 4463.746,
"text": " is nothing like what I actually sense the data I actually got. So imagine now you apply that sort of concept that the autonomous states that include the active states are just in the service of minimizing prediction error. How would you understand that in terms of elemental movement and behavior?"
},
{
"end_time": 4518.319,
"index": 174,
"start_time": 4489.48,
"text": " And one way of understanding that is that basically your models of say how you're going to move are actually realized by creating data that is consistent with your predictions. So I have in my head a model, a hypothesis, the notion that I'm going to raise my arm. And all that's happening in that notion is that that's my belief about state of the world."
},
{
"end_time": 4548.524,
"index": 175,
"start_time": 4518.951,
"text": " And I'm telling my spinal cord and my musculature and my muscles, that's how I expect the world to be. And then my reflexes are driven by the prediction errors, driven by the free energy gradients, literally like the ball rolling down the hill, to realize, to change the world so it supplies the kind of data that I predicted. I just wanted to mention that notion of"
},
{
"end_time": 4569.821,
"index": 176,
"start_time": 4548.763,
"text": " movement as simply a realization of predicted endpoints, intentions to move, is physiologically fully licensed by things like equilibrium point hypothesis in motor control and motor physiology,"
},
{
"end_time": 4598.695,
"index": 177,
"start_time": 4570.179,
"text": " But it goes back even further to idea motor theory. So this was in the Victorian age, how people understood movement, that basically it was a physical manifestation of beliefs about the way my body should be posed or what my body should be doing. So to me, that is, if you like, an elemental and quite fundamental example of the importance of changing the data, not the model."
},
{
"end_time": 4625.845,
"index": 178,
"start_time": 4599.138,
"text": " So if you changed your model in the context of, I'm going to stand up now, and you looked at the data and nothing is happening, you would immediately change your mind and say, oh, I'm not standing up. What would that look like? That would look like Parkinson's disease. That would be a failure to properly attenuate the evidence that I'm not moving when I strongly believe that I am about to move."
},
{
"end_time": 4655.93,
"index": 179,
"start_time": 4627.142,
"text": " And of course you can understand now the pathophysiology of Parkinson's disease that's due to a failure of neuronal message passing and particularly the weight or the gain afforded proprioceptive information in that light. It's a failure of active imprints, it's a failure of the active states to properly respond to"
},
{
"end_time": 4669.258,
"index": 180,
"start_time": 4656.152,
"text": " The evidence that or change the evidence at hand. So this is a lovely example of ignoring selective ignoring just like the second suppression with the eyes."
},
{
"end_time": 4690.111,
"index": 181,
"start_time": 4669.462,
"text": " That enables movement. If you didn't do that, we couldn't move. So what I'm trying to say is that, you know, right from the inception of these kinds of this kind of thinking in the inactive domain in terms of idea motor theory, the there is a there is I think a fundamental importance of, you know,"
},
{
"end_time": 4720.503,
"index": 182,
"start_time": 4690.776,
"text": " of the other side of the coin from changing your mind, which is basically changing the data which causes you to change your mind through sense making. And without that, we wouldn't have action. So you can almost say that action and perception are the two sides of the same coin. The perception is changing your mind by changing your model to maximize the marginal likelihood given a particular model."
},
{
"end_time": 4751.084,
"index": 183,
"start_time": 4721.135,
"text": " Action just is changing the data by acting upon the world to maximize the marginal likelihood, basically maximizing changing the data. So the probability of the data given the model is so active inference. So that coupling of action perception is essentially trying to realize your beliefs about or your models of your lived world. And if you're successful in doing that, you've got the right eco niche. You will have a high marginal likelihood."
},
{
"end_time": 4780.555,
"index": 184,
"start_time": 4751.34,
"text": " you know, from an evolutionary perspective, you will have a high adaptive fitness from a social perspective, you will be, people will listen to you and teach your job into these podcasts, suggesting you do indeed solicit that kind of information that causes people to, you know, to go to your channel, to attend to your stuff, simply because they think they're going to get good stuff and they're going to get uncertainty resolving stuff."
},
{
"end_time": 4809.599,
"index": 185,
"start_time": 4780.828,
"text": " Okay, well, perhaps let me undermine my own credibility here. So a few weeks ago, and I think about a year ago or so we talked, and I had some experience about, and I was worried and you said, Kurt, don't be worried. I had some experience where I thought I heard a voice. It was in a hypnagogic state. So I was just about to fall asleep when"
},
{
"end_time": 4811.391,
"index": 186,
"start_time": 4810.623,
"text": " Many people"
},
{
"end_time": 4842.056,
"index": 187,
"start_time": 4812.244,
"text": " that shouldn't be a state that I attribute much credence to but I remember being afraid because I heard my wife either say okay or just some one syllable word and then I was so worried and I worked myself up into anxiety thinking do I have schizophrenia oh my gosh there was no other voices since then but I was extremely anxious and then you told me don't worry and even Kurt even if there is there's medication and many people go through psychosis and or psychotic breaks and so on but luckily I didn't luckily there was nothing wrong with me"
},
{
"end_time": 4868.148,
"index": 188,
"start_time": 4842.056,
"text": " And your words did calm me, so thank you for that. However, and coincidentally, perhaps not, but coincidentally, a few weeks ago, I had a strange experience, Carl, and I wasn't sure if I was going to bring this up on the podcast for one reason, because I don't want you to think that I'm someone who's beset with psychological issues. But on the other side, I am driven by a filmmaking adage, which says that the more personal the pain, the more widely applicable it is."
},
{
"end_time": 4882.892,
"index": 189,
"start_time": 4868.422,
"text": " And the opposite is also false. The opposite is false. So if you try to speak generally, you end up speaking to no one. What happened a couple of weeks ago was I had this sense and it was so and even to talk about it now is a bit frightening."
},
{
"end_time": 4906.766,
"index": 190,
"start_time": 4883.2,
"text": " I had this sense that I was like, oh shoot, solipsism. Is everything in my head just that thought? And I've, because I'm on this podcast and I've entertained many ideas, it's extremely taxing to do so because I try to take different people's points of views seriously. And perhaps I shouldn't, but I feel like in order to give them a fair shake, I need to hear that sound."
},
{
"end_time": 4933.729,
"index": 191,
"start_time": 4907.654,
"text": " That's the sweet sound of success with Shopify. Shopify is the all-encompassing commerce platform that's with you from the first flicker of an idea to the moment you realize you're running a global enterprise. Whether it's handcrafted jewelry or high-tech gadgets, Shopify supports you at every point of sale, both online and in person. They streamline the process with the Internet's best converting checkout, making it 36% more effective than other leading platforms."
},
{
"end_time": 4959.872,
"index": 192,
"start_time": 4933.729,
"text": " There's also something called Shopify Magic, your AI-powered assistant that's like an all-star team member working tirelessly behind the scenes. What I find fascinating about Shopify is how it scales with your ambition. No matter how big you want to grow, Shopify gives you everything you need to take control and take your business to the next level. Join the ranks of businesses in 175 countries that have made Shopify the backbone."
},
{
"end_time": 4985.674,
"index": 193,
"start_time": 4959.872,
"text": " of their commerce. Shopify, by the way, powers 10% of all e-commerce in the United States, including huge names like Allbirds, Rothy's, and Brooklynin. If you ever need help, their award-winning support is like having a mentor that's just a click away. Now, are you ready to start your own success story? Sign up for a $1 per month trial period at Shopify.com slash theories, all lowercase."
},
{
"end_time": 5002.79,
"index": 194,
"start_time": 4985.674,
"text": " Go to Shopify.com slash theories now to grow your business no matter what stage you're in Shopify.com slash theories. Embody them in some way, shape or form. So if they say, well, all is mind, for example, that's the idealist."
},
{
"end_time": 5032.483,
"index": 195,
"start_time": 5003.558,
"text": " I'm trying to imagine what would that be like and if someone's a materialist well what does it mean that this was dead and it somehow emerged that we have consciousness either way the point of that is to say I've been feeling like for almost a year and a half now just in a void not sure what is true and what's not true that's fine I can deal with that except that that led me to this spiraling of such it was probably the most terrifying experience I've ever had in my life and it took me weeks and I'm still recovering from it where I felt like shoot"
},
{
"end_time": 5055.316,
"index": 196,
"start_time": 5032.961,
"text": " Am I all that exists? And then just that thought alone? I'm like, I don't want to believe that that could be the case. And I remember looking at my wife and thinking, is my wife even real? How much of an imbecile and how ungrateful and how foolish is it? And how that I would even think that I feel bad that I would even entertain that thought. Oh my gosh, I'm scared. I don't want to go to the"
},
{
"end_time": 5077.142,
"index": 197,
"start_time": 5056.084,
"text": " Hospital because I felt like if I was to truly accept that thought I could I was on the brink of just losing it into an insensate spiral. I became extremely scared of myself and I think that I'm the type of person I am an anxious person and I am the type of person that tends to obsess over thoughts."
},
{
"end_time": 5099.155,
"index": 198,
"start_time": 5078.456,
"text": " So then I started to obsess over this and for the past couple of weeks, I was so scared, so scared of my own mind. So almost like a hypochondriac for mental disorders where I didn't want to even look up what's the definition of schizophrenia or psychosis was because I was afraid that I would see the signs everywhere. For example,"
},
{
"end_time": 5123.012,
"index": 199,
"start_time": 5099.804,
"text": " A few days ago, or about a week ago, remember, generally we all speak to ourselves in our own heads. And sometimes I use the phrase, so I don't like that, or I like that. I use the word I in my own internal monologue, but sometimes the word, yeah, you can do this, you can do this. So sometimes they use I, and sometimes they use you. And I imagine we all do that. It's colloquially okay to exchange those two. However, then I thought, oh, shoot,"
},
{
"end_time": 5152.261,
"index": 200,
"start_time": 5123.985,
"text": " What if I felt distanced from my thought? What if the one that's saying you is a thought speaking to me like a schizophrenic may hear? Oh my gosh. Oh my gosh. I don't want to hear this. And I was so afraid and I felt derealized and depersonalized. When you mentioned Parkinson's earlier, that the difference between the model and then the data, that didn't strike me as a symptom of Parkinson's. The way you were describing it struck me as a symptom of derealization. Anyway, I'm bringing this all up and actually feel"
},
{
"end_time": 5173.695,
"index": 201,
"start_time": 5153.814,
"text": " Good about speaking about it. It somehow feels Curative to even voice these concerns. I Well, actually, I don't know I don't I don't know why I'm I'm voicing this mainly to Mainly, okay. How about this? I just verbally cascaded incoherently perhaps"
},
{
"end_time": 5198.114,
"index": 202,
"start_time": 5174.241,
"text": " What are your thoughts on what I just said? The reason I'm saying this one more time is that people in the comment section, when they view different theories of what consciousness is, what reality is, that's the theories of everything channel, in a nutshell, I see many of them in a similar place where they feel lost. For example, I've heard that some people who view Yoshabok's ideas that this is all an illusion or this all simulated consciousness, some people become suicidal."
},
{
"end_time": 5225.265,
"index": 203,
"start_time": 5198.387,
"text": " That's actually extremely terrifying. Luckily, I'm not suicidal, though I'm scared that I could because I've heard that other people could. I'm not sure if my problem is just obsessive thoughts or if it was indicative of something else and I was so afraid. And it's such a scary feeling, Carl, to be afraid of one's self, to not even be able to sleep at night because you're afraid of your own thoughts and what that may lead to."
},
{
"end_time": 5251.408,
"index": 204,
"start_time": 5225.811,
"text": " Does any of that make sense? And I know that sounds like it's completely off field. Perhaps it's not. No, no, I think I think it follows on very gracefully. But you've taken it right now to some of the most important aspects of modeling, which we all contend with, which, of course, is I mentioned before, you know, getting models for our"
},
{
"end_time": 5280.896,
"index": 205,
"start_time": 5251.886,
"text": " our lived worlds. But of course, the most important part when we're a baby of the world is our body. So the first thing we have to do is to build a model of our body and work out what things that we are able to move, hence motor babbling and rattling. Probably more important than that, of course, is exchanges with the mother"
},
{
"end_time": 5308.899,
"index": 206,
"start_time": 5281.169,
"text": " developing a sense that mother is an object and crucially an object that is separate from me. And that's quite a skillful move and it takes a lot of my belief updating and changing your models and building a coherent explanation for all your sensations, both the political touch and the physiological intercepted consequences of being suckled, for example, and being cared for."
},
{
"end_time": 5336.732,
"index": 207,
"start_time": 5309.701,
"text": " And I'm taking this route to the key thing, which is basically models of self and selfhood. So if we are coming back to this picture of some mathematical universe that I've now tiled with multiple Markov blankets encapsulating or defining lots of particles, and basically we"
},
{
"end_time": 5366.186,
"index": 208,
"start_time": 5336.954,
"text": " inhabit a universe that is constituted by things like us and initially you know mum and dad and brothers and sisters and then you know peers at school. That means that our models of our lived or experienced world have to have models of other in it which means you need to be able to disambiguate between sensations caused by others and sensations"
},
{
"end_time": 5391.988,
"index": 209,
"start_time": 5368.217,
"text": " That brings with it something quite special, which is a model that encompasses selfhood."
},
{
"end_time": 5419.206,
"index": 210,
"start_time": 5392.944,
"text": " So you wouldn't need this if you lived in a world that did not involve an exchange with other conspecifics. But because we do live, our world is constituted by other things like us, we need to be able to contextualize and make the inference, I did that, or you did that, or it's my turn, or it's your turn, or you've got that intentional stance, and I've got this intentional stance."
},
{
"end_time": 5446.886,
"index": 211,
"start_time": 5419.991,
"text": " So just having as part of the generative model of the hypothesis, you know, the hypothesis that I am me and you and you is something which characterizes one would imagine very high life forms. Now we go even higher to basically philosophers and people like you. So if you've got, if you have, if you spend your life worrying about"
},
{
"end_time": 5474.445,
"index": 212,
"start_time": 5447.09,
"text": " What is the self? Does the self even exist? Well, precisely. That seems to be the essence of your existential angst and all its attendant, interceptive and emotional consequences. Do I exist? I think really usefully you brought to the table the notions of depersonalization and derealization"
},
{
"end_time": 5490.794,
"index": 213,
"start_time": 5474.718,
"text": " If this is for physicists, then can you explain to your audience what those things are? Because I think that it would be really useful for people who do not know what... Well, to be quite frank, Carl, I didn't want to look up this exact definition because I know that I'm a valedictorian and I would just..."
},
{
"end_time": 5515.691,
"index": 214,
"start_time": 5491.118,
"text": " obsess over myself having these issues. So I can give an explanation of my experience and then perhaps you can delineate what the psychopathology is. Okay, I'll make an aside first. Some people say that what we need to do is realize that the self is an illusion. You can hear this in the more eastern end. And I know that many people who watch this channel also feel like, oh, they take these people seriously, like I took them seriously. And I don't know if that's exactly true."
},
{
"end_time": 5537.483,
"index": 215,
"start_time": 5516.049,
"text": " Not whether or not the truth of the self being an illusion is true, but whether or not that is indeed useful for everyone. It may not be. It may be. At least for me, it was extremely harmful and perhaps still is. And it may be dependent upon where you are in psychological development, spiritual development. And sometimes these lessons that are told to us by philosophers or spiritual gurus"
},
{
"end_time": 5557.176,
"index": 216,
"start_time": 5537.483,
"text": " or people who seem like spiritual gurus, perhaps they should be taken with such a grain of salt. And I've come to the conclusion that if it's not life-affirming, then perhaps don't follow that. Perhaps use that as a clue that that's a pathological path. And don't do anything that can be drastic. Don't give up. Because I felt like I was in a state where I could"
},
{
"end_time": 5581.237,
"index": 217,
"start_time": 5557.722,
"text": " snap and feel like, well, this is all a dream. I remember feeling, looking at my arms and feeling like, am I even behind, like right now I'm feeling great and sorry that I don't seem like it, but I'm speaking with you. I feel safe. I love my wife. I've actually whatever I'm feeling happy right now. And I feel like I would a month ago, which is I'm behind my eyes and this is a person speaking a week, two weeks ago, three weeks ago, I was feeling like,"
},
{
"end_time": 5604.838,
"index": 218,
"start_time": 5582.329,
"text": " All of that could be a model. I could emulate anyone else's head. I could be anyone else. This could all be in my head. So debilitating, destabilizing, walking around feeling like, is this a dream? Constantly checking my fingers, because that's one of the signs you can check for lucid dreaming. How many fingers do you have? And then also feeling bad, like, why the heck am I thinking this is a dream? Shoot, what if it is a dream?"
},
{
"end_time": 5634.292,
"index": 219,
"start_time": 5605.589,
"text": " Spiraling, spiraling, tumbling, tumbling thoughts. Okay, that's what I had slash still have to some degree, though not right now. I wouldn't say that I'm over it because it was such a recent experience. I'm not going to say that I've solved it. I feel like it's tapering off. Luckily. Anyway, that's what I had. I would characterize that as derealization in the sense that I didn't feel like it was real. I don't know if it's depersonalization because I don't know the difference between those two nor dissociative identity disorder. So I don't know the differences between those."
},
{
"end_time": 5657.858,
"index": 220,
"start_time": 5634.906,
"text": " Well, it's nice you bring all those three. The phenomenology is very, very closely related, but the way you described it, it was almost perfect. You know, this either a derealization, which people normally described as they're looking, they are sensing,"
},
{
"end_time": 5685.384,
"index": 221,
"start_time": 5658.353,
"text": " Something that doesn't quite seem real. The outside world seems like it's a movie. It's lost that tangibility. There's no grip that can be attained that reassures you that stuff is actually out there. The converse, of course, is I'm no longer a person. And that could, I think, be even more frightening. And both of these, I think, speak to"
},
{
"end_time": 5714.189,
"index": 222,
"start_time": 5685.794,
"text": " speak to this dissociation that comes with a dissolution of selfhood, or at least one's construction of explanations for the self in a lived world. You used a phrase which I think was quite pertinent here, which is, one should not pursue this kind of thinking and self-exploration and introspection"
},
{
"end_time": 5732.995,
"index": 223,
"start_time": 5714.599,
"text": " Unless it is affirmatory and self-affirmatory. And in a sense, that's exactly from the sort of mathematical and dry perspective of the free energy principle that just is maximizing model evidence. This is self-evidencing. It is, you know, if you can affirm"
},
{
"end_time": 5761.869,
"index": 224,
"start_time": 5733.387,
"text": " through all the right kind of interactions, be the emotional, social, sexual, intellectual, that you've got the right kind of model of your world, which is constituted by people like your interviewees and your wife and all your colleagues. Then affirming that model simply is securing evidence for that good model of your rich and encultured world."
},
{
"end_time": 5790.469,
"index": 225,
"start_time": 5762.159,
"text": " That just is that sort of affirmatory process. However, there are dangers that lurk in the shadows, which you've clearly contended with. To do that properly, to be able to secure the right kind of evidence that you are living in a world of positive, constructive, nourishing,"
},
{
"end_time": 5820.708,
"index": 226,
"start_time": 5790.998,
"text": " enabling human beings, others, other things like you, you have to be able to communicate to be able to communicate, you actually have to entertain the idea that there is there is self and other. And simply, you know, to engage in turn taking, you need to have this, this explicit part of your generative model, that I am a self. In so doing,"
},
{
"end_time": 5847.619,
"index": 227,
"start_time": 5821.22,
"text": " You will also have contextualized as part of your generative model itself in a number of different states. And you need to have, remember before we're talking about the good scientist having a portfolio of hypotheses, we're generating a new set of hypotheses so that we can now gather evidence for these hypotheses. So that's part of what you are doing. I mean, in doing these podcasts, you are on a journey"
},
{
"end_time": 5870.401,
"index": 228,
"start_time": 5848.114,
"text": " of developing different hypotheses about ways of being. The danger, though, when it comes to different ways of me as a self existing is that you can sometimes lose, entertain the null hypothesis that I am not a self or I do not exist in a sentient fashion."
},
{
"end_time": 5898.046,
"index": 229,
"start_time": 5870.555,
"text": " in the fashion of having particular kinds of qualitative experiences. This is a perfectly viable hypothesis. It can be had in a non-emotional context when you meet people like philosophers, entertain things like the zombie hypothesis or brain and bat thought experiments. These are all really interesting instances of a very high order life form. Think about, well, what would it be like if I was me without mehood?"
},
{
"end_time": 5918.746,
"index": 230,
"start_time": 5898.166,
"text": " If I was a zombie, what would it be like if it wasn't actually me talking? What would it be like if there was no reality, it was all just me? All of these alternative ways of being that speak to selfhood or its negation or its nihilism now become plausible hypotheses for which you have to search for evidence."
},
{
"end_time": 5949.019,
"index": 231,
"start_time": 5919.172,
"text": " And of course, if you are just doing this in your head and ruminating, introspecting, there's going to be very little evidence available to you to disaffirm your hypothesis. Perhaps I'm not me. If you don't actually engage with somebody else, there is no evidence at hand for to refute the hypothesis. So on the one hand, that is a remarkable capacity. The very fact you can have this hypothesis in your head, I am not me."
},
{
"end_time": 5963.797,
"index": 232,
"start_time": 5949.514,
"text": " Or it is not me having these thoughts. It is not me having these quantitative experiences is quite remarkable. It's a very, very high level ability, which I would imagine where you and I are the only kinds of"
},
{
"end_time": 5993.677,
"index": 233,
"start_time": 5964.275,
"text": " sentient artifacts in the universe that can entertain that kind of counterfactual hypothesis, which is necessary to explore different models. It's necessary indeed to actually just behave and have plans. You have to have counterfactual hypothesis about the future of different ways of being because it comes with it comes with a price. If sometimes you entertain a hypothesis for which there is no evidence on hand,"
},
{
"end_time": 6021.766,
"index": 234,
"start_time": 5993.968,
"text": " that could possibly refute it. And certainly, if you adopt a particular behavioral strategy or engagement with the world, that precludes soliciting evidence against the hypothesis that you are, say, you. So examples of that would be things like obsessional compulsive disorder or agoraphobia, or your rumination disorders, where you physically"
},
{
"end_time": 6048.473,
"index": 235,
"start_time": 6022.005,
"text": " or possibly even pro-socially, prevent yourself from acquiring any further information that would allow you to change your mind. And of course, once you realize that's a possibility, you can also get into the negatively veiled emotional aspects of entertaining these hypotheses. So if I've got schizophrenia, that now offers"
},
{
"end_time": 6077.961,
"index": 236,
"start_time": 6048.797,
"text": " a very plausible hypothesis for these dissociative experiences. Why am I having these dissociative experiences? Well, one very plausible hypothesis, I am having, I have a dissociative order or I have disorganized thought disorder that characterizes schizophrenia. So this now becomes a very plausible hypothesis. How would you go secure evidence against that? Well, it becomes very difficult because the whole point of having these very high order counter"
},
{
"end_time": 6107.363,
"index": 237,
"start_time": 6078.217,
"text": " factual hypotheses or models at hand to explain your experienced world is that they're very plausible explanations and it's very difficult. If you have got schizophrenia, you're not going to be able to trust any information anyway because they could be illusionary. So it's one of these wonderful self-fulfilling hypotheses that are very difficult to dismantle."
},
{
"end_time": 6136.391,
"index": 238,
"start_time": 6107.602,
"text": " And you see the same phenomenology in many instances. You mentioned behaviour therapy before in a way that doesn't have this quite the same horrible existential angst that you clearly experienced. You can actually construe many phobias, for example, of the same kind in the sense that if you are very frightened of spiders, you're very unlikely to actually go and solicit"
},
{
"end_time": 6155.947,
"index": 239,
"start_time": 6136.578,
"text": " the evidence that would that would confirm the alternate hypothesis that spiders are not frightening because you just don't go near spiders. So this is the kind of hypothesis that maintains itself because it's undermined the way that you act as solicitor evidence for your beliefs. But for you,"
},
{
"end_time": 6181.067,
"index": 240,
"start_time": 6156.254,
"text": " Of course, it's not a question of being frightened of spiders, it's a much more existentially deep and worrisome notion that in fact you're not you or you're not even a self and you're not even perceiving reality. These kinds of hypotheses that come along qualified by or associated with plausible explanations, I'm going mad."
},
{
"end_time": 6191.527,
"index": 241,
"start_time": 6182.21,
"text": " are very difficult to refute, especially if you ruminate on them. They will go away as soon as you start reengaging with other people."
},
{
"end_time": 6213.575,
"index": 242,
"start_time": 6191.8,
"text": " That's what I found to be the case, Carl, is that the more I talk about it, the more I'm speaking with you, that just speaking with people and not being so in my head, almost instantaneously, it starts to dissipate. And also, just so you know, there's a danger for the people who watch this channel that it's seen as a spiritual triumphant state to be egoless, to have ego death."
},
{
"end_time": 6235.043,
"index": 243,
"start_time": 6213.968,
"text": " So then, just so you know, when I was feeling that, I was then telling myself, but am I supposed to have ego death? Is this what ego death is like? And now I feel bad. Am I just not strong enough as a person? Which then I feel like I'm less on the social hierarchy because I'm just a coward who cannot deal with, who isn't psychologically strong enough to deal with potential ego death, if that was even ego death."
},
{
"end_time": 6255.811,
"index": 244,
"start_time": 6236.049,
"text": " so that those are there's so many dangers in this and i didn't realize this i had this insight in that episode or experience that kurt kurt kurt kurt something was like i was talking to myself or something was talking to me and telling me kurt you've been thinking way too much this is you're extremely analytical i never would have thought of myself as someone who's too analytical before"
},
{
"end_time": 6285.657,
"index": 245,
"start_time": 6255.811,
"text": " It was clear to me that I was too analytical, and there's nothing wrong with being analytical. There's nothing wrong with thinking, but you've been thinking a bit too much. Feel more. Just ground yourself here. Engage in this world more. And there was also this feeling that I don't love myself, which is a strange feeling, Carl, that I'm a bit hard on myself. And I would have never said that before. I would have thought, I'm not hard enough. I need to study harder. I need to prepare for these podcasts harder. I need to release more. I need to learn more."
},
{
"end_time": 6316.22,
"index": 246,
"start_time": 6286.22,
"text": " There's so much self-analyzing that's going on right here, Carl."
},
{
"end_time": 6332.551,
"index": 247,
"start_time": 6317.551,
"text": " Anyway, I felt like, Kurt, you were thinking too much. Just relax, perhaps. Don't engage as deeply as you have been with these ideas. Don't entertain them as much. Maybe you're in a fragile state right now. You can do that later. Right now."
},
{
"end_time": 6359.206,
"index": 248,
"start_time": 6332.551,
"text": " take it back that's what came over me and also this lesson that I don't have enough self-love or self-acceptance last night I looked up I was using a rubber band for my thoughts like self-administered aversive conditioning when you snap yourself if you're performing a bad habit so I thought maybe I could do that with my thoughts so anytime that I feel like that that would encourage my mind to not feel like that I looked it up and it said that works for physical habits but not for mental habits so stop that and what works is act I believe acceptance"
},
{
"end_time": 6371.305,
"index": 249,
"start_time": 6359.735,
"text": " and commitment therapy, which is a CBT technique. So then last night I was telling myself, you know what? Try that out. Accept these slots. And I'm not exactly sure what it means to accept them."
},
{
"end_time": 6401.493,
"index": 250,
"start_time": 6371.681,
"text": " I don't precisely think it means entertain them, but at least meet them, not with horror and dread and avoidance, but just say, that's okay, man. You think like that. Some people feel like that sometimes. So I started doing that. And last night was the first time in the three weeks since I've had this, that I felt absolutely like my regular self, almost from five minutes of this quote unquote acceptance therapy. And then I woke up this morning, felt like my regular self, like I'm behind my eyes. I have a, I'm in this world. That was fascinating."
},
{
"end_time": 6418.797,
"index": 251,
"start_time": 6402.346,
"text": " Okay, so I said quite a few statements right there. I'm sure there's plenty you want to comment on. What I'm also interested in, for you to put a pin in your hat, is what the heck does this self-acceptance, self-love, if that seemed the same, mean in terms of the free energy principle? Okay, so I basically just poured myself out there."
},
{
"end_time": 6446.664,
"index": 252,
"start_time": 6420.623,
"text": " There you go. Sorry, Carl, for you to pick up all those pieces. I know it's a huge responsibility. I trust you. You're someone who studies this. You're extremely, extremely bright and you have such a wealth of knowledge. And I do feel like what I'm going through is not something terribly unique. I do feel like this is something that many people who are on this journey of understanding the world and consciousness in their place go through this perhaps at some point. I think I was trying to emphasize that by saying that, you know,"
},
{
"end_time": 6453.882,
"index": 253,
"start_time": 6447.039,
"text": " It is a gift that you're able to have these existential crises. There are dangers that lurk."
},
{
"end_time": 6483.183,
"index": 254,
"start_time": 6454.206,
"text": " which you've clearly encountered. But at the end of the day, the very fact that you can entertain these counterfactuals is quite remarkable. To my mind, it would be the highest expression of the human condition just to consider those alternative hypotheses where bits of our humanity are just not there anymore or have a different disposition or relationship to reality or indeed other people. But lots of things you said made a lot of sense."
},
{
"end_time": 6510.845,
"index": 255,
"start_time": 6483.183,
"text": " from simple things like soliciting reassurance from others that you don't have to do any more work in terms of preparing yourself and being a good expounder of ideas or interviewer or book writer or whatever. That's exactly what you should be doing. That's exactly securing evidence for yourself models of an affirmative thought."
},
{
"end_time": 6537.227,
"index": 256,
"start_time": 6511.186,
"text": " And the more you do that, then the more the more that model will be fit for purpose and that you will have the right kind of marginal maximizing your marginal likelihood. But you're in a difficult position because your job is actually to explore other ways of making sense of things and other ways of thinking things. You know, you're compelled to explore alternative hypotheses and, you know, you described"
},
{
"end_time": 6566.323,
"index": 257,
"start_time": 6538.302,
"text": " you know, the potential nihilism and that I presume quite nightmarish. I can't now remember my nightmares, but I imagine that all of us do go through so we have nightmares as children. I think you stop having true nightmares in your 30s and 40s, but you know, that sort of, you know, before you've really got that grip and maintain that grip on selfhood and you in your"
},
{
"end_time": 6595.316,
"index": 258,
"start_time": 6568.507,
"text": " Very frightened of having one?"
},
{
"end_time": 6620.486,
"index": 259,
"start_time": 6595.794,
"text": " Yeah, yes, that is correct. And luckily, I wasn't well, what I imagined what I was going through would feels like how someone would feel on a on a terrible psychedelic trip. Yes. And it's interesting you bring up psychedelics, because of course, you know, there is a move in sort of pharmacologically assisted talking therapists, you know, to use psychedelics, simply"
},
{
"end_time": 6649.735,
"index": 260,
"start_time": 6620.794,
"text": " not to aspire to, I don't know what ego death means, but I can have a guess at it from the point of view of meditation and mindfulness. But there's a really interesting connection between the use of psychedelics and the aspiration of many meditation-like practices that would, I think, subsume the internal attention states"
},
{
"end_time": 6678.677,
"index": 261,
"start_time": 6650.009,
"text": " of meditation and, you know, current practices of mindfulness, which is really to try and redirect your attention to the sensorium and usually the intercepted parts. It's like breathing, for example. So this is the exact opposite of what you've been doing. It's exactly deploying that attention, all that game control we're talking about in terms of selecting"
},
{
"end_time": 6708.063,
"index": 262,
"start_time": 6679.224,
"text": " what kind of information, sensory information, sensory states to engage and determine your belief updating. So putting all that attention out to the sensory side of your deep hierarchical models, your constructs, what you've been doing is the opposite. You've been actually wandering around at the deeper, the highest levels of these hierarchical models that"
},
{
"end_time": 6722.551,
"index": 263,
"start_time": 6708.712,
"text": " Sorry, so in the back when we were initially talking, I said there's a black box, the input output, and then there are two black boxes. It's as if you're saying pay attention to the sensory. Forget about the action. We haven't talked much about the action, though, obviously, as embodied creatures, sensory and action are tied."
},
{
"end_time": 6749.411,
"index": 264,
"start_time": 6723.131,
"text": " It's as if what I've been doing was staying within the box and creating my little sensations and actions within there as little loops. And you're saying pay attention to the senses that come from the outside. Is that what you're saying? Yes. Well, I'm not saying you should do, you will do everything in the right way under the free energy principle. What I'm saying is that the skilled practitioners of mindfulness and meditation"
},
{
"end_time": 6779.121,
"index": 265,
"start_time": 6749.94,
"text": " I would imagine of the kind that would lead to ego death or I'm not quite sure what that means. It would employ exactly the same kind of internally mediated and sometimes if you're very skilled volitionally called forth mechanisms that are actually targeted by psychedelics."
},
{
"end_time": 6786.783,
"index": 266,
"start_time": 6780.043,
"text": " require you now to do all your belief updating and evidence assimilation and sense making."
},
{
"end_time": 6816.715,
"index": 267,
"start_time": 6787.073,
"text": " much more the input level of the black box, not at the highest level, which you could think of as an output in terms of selecting the plans and what am I going to do next? And what kind of person am I? Narratives that we give ourselves that contextualize the way that we behave and the other things that we do. So, you know, I think it's mechanistically a really interesting connection between the notion of you ruminating, going in circles around your head, exploring ever darker and darker"
},
{
"end_time": 6844.224,
"index": 268,
"start_time": 6817.022,
"text": " That's interesting. So I think that, you know,"
},
{
"end_time": 6850.469,
"index": 269,
"start_time": 6844.462,
"text": " You might, you may, I guess you probably are skilled in meditation."
},
{
"end_time": 6880.367,
"index": 270,
"start_time": 6850.776,
"text": " But that kind, I think the objective of becoming a skilled practitioner of mindfulness or meditation is simply to get some volitional control over that sort of attention that is paid to these deeper machinations versus attention that is paid to the sensorium that underwrites the psychedelic aspect, the allure of just sensory patterns and"
},
{
"end_time": 6910.708,
"index": 271,
"start_time": 6880.947,
"text": " the sensations and textures that you get, you know, when taking psychedelics. So, you know, getting control of the balance. So it's not the fact you're attending to your breathing, which is important. It's the fact that you could volitionally redirect your attention away from selfhood. Extremely interesting. So, but you seem to have got control by this acceptance. And it strikes me that, you know, in the absence of"
},
{
"end_time": 6936.084,
"index": 272,
"start_time": 6911.51,
"text": " Feelings of an emotional sort that were articulated in your description in terms of, you know, I'm an awful person. I'm not fit for purpose. I'm not well prepared. I'm not able to man up and own this ego, ego death. All of this sort of personal but quite emotional constructs, affective, valenced."
},
{
"end_time": 6964.138,
"index": 273,
"start_time": 6936.271,
"text": " There are also hypotheses. There are also explanations. So the idea is that, you know, anything that you can talk about has to be part of an explanation for particular ways of being. But, you know, I can be a worthless person in the sense that I am a worthless person. I'm not a worthless person. You know, I'm in a state of anxiety. I'm not in a state of anxiety. We have to recognize when we are anxious. Anxious is just a state of being."
},
{
"end_time": 6993.473,
"index": 274,
"start_time": 6964.36,
"text": " which is necessarily called for in certain situations that rebalance this attention or precision waiting in computational psychology to make your belief updating fit for purpose in this particular context. What I'm trying to get to is why acceptance might have worked, because I imagine that if you were suddenly found yourself ruminating and locked into those ruminations,"
},
{
"end_time": 7020.623,
"index": 275,
"start_time": 6993.814,
"text": " then very much akin to someone on a bad trip who thinks, ah, I am going to be locked in this forever. This is going to be my uncertain state of being for eternity, because there is no reality to it. That's how it feels, very heartfelt, yeah. That feeling will, well, if that is true, I must be feeling anxious. You are then going to look for evidence that you're feeling anxious and"
},
{
"end_time": 7050.23,
"index": 276,
"start_time": 7020.896,
"text": " because you are anxious, you will experience certain cardiac acceleration, certain interceptive flight or fright like responses that will supply evidence that you are anxious. And then that becomes evidence. Yes, I am right. I am now in a state of nihilism. I do not exist and I should be anxious about that. And yes, I am anxious. So you get into a vicious circle."
},
{
"end_time": 7073.66,
"index": 277,
"start_time": 7050.435,
"text": " So this is a sort of good old fashioned cognitive behavioural explanation for things like panic attacks, that you know, you quite reasonably in an entirely base optimal way, make sense of interceptive bodily gut feelings, literally, you make sense of them with the hypothesis, oh, I must be in an anxious state of being. And that"
},
{
"end_time": 7103.609,
"index": 278,
"start_time": 7074.36,
"text": " conclusion hypothesis then generates autonomic actions that are realized reflexively in the way we talked about when we're looking at the Parkinson's idea and that will be reflected in terms of cardiac acceleration, neuroendocrine releases into your body and your body will change and your body will supply signals, intercepted evidence, yes I am anxious. So it's again a self-fulfilling prophecy in exactly the same way that the idea motor theory means that raising my hand"
},
{
"end_time": 7130.35,
"index": 279,
"start_time": 7103.865,
"text": " is just a realization of my beliefs, but I can literally raise my levels of anxiety in exactly the same way, but invisibly from the outside, but not when I can sense my own body. So if you got yourself caught up into this joint hypothesis, I am in a nihilistic existential state, and I should be really worried about this and anxious about this, then it is quite understandable"
},
{
"end_time": 7160.657,
"index": 280,
"start_time": 7130.862,
"text": " That you are going to use signals of anxiety and angst generated by your body as evidence that, yes, I am now divorced from reality. I'm an unpersoned. I am in a dark and nihilistic place. There could be no other place. There could be no reality. That's a hypothesis. If you can just wait for the evidence that that hypothesis is incorrect,"
},
{
"end_time": 7184.667,
"index": 281,
"start_time": 7161.186,
"text": " simply by letting your body calm down so there is no further evidence of any anxiety then you can find secure even from your own body even without talking to somebody you can secure evidence ah the hypothesis that i am in a state of internal nihilism and anxiety is a silly hypothesis because the evidence refutes it i can now feel my"
},
{
"end_time": 7214.053,
"index": 282,
"start_time": 7184.94,
"text": " not possibly personally, but you will synthesize all your interceptive feelings in a way that provides evidence against the fact that now I am in a deeply nihilistic, anxious state. I think that just is the motivation for the acceptance. It is pushing through to get to a state of mind and a state of body, literally realizing your good and making sense of your gut feelings."
},
{
"end_time": 7244.172,
"index": 283,
"start_time": 7214.394,
"text": " so that you can now refute the joint hypothesis that I am in a state of nihilism and I should be jolly frightened about this because there's no evidence that you're frightened anymore. So this was just a piece of hypothesis building of a very sophisticated and philosophical sort, the highest level of exploring alternative ways of relating to the world. You now have"
},
{
"end_time": 7264.394,
"index": 284,
"start_time": 7244.497,
"text": " I think a very useful insight into the gift of retaining selfhood that I think we all take for granted, but it's quite a fragile thing. Luckily, most of us get through the day, if not hopefully most of our lives,"
},
{
"end_time": 7294.394,
"index": 285,
"start_time": 7264.394,
"text": " you know, by retaining that grip and just, you know, continue securing evidence that this is the right hypothesis, I'm a person, I am me, I am functional, this is, you know, this is the way I'm meant to be, and these are the kinds of things I do, and the kinds of people I talk to. But that is such a fragile, self-assembled, and has to be maintained hypothesis that you don't, you know, that fragility, I think, only is revealed occasionally, and only to some people when they have the alternative hypothesis that, you know,"
},
{
"end_time": 7319.275,
"index": 286,
"start_time": 7294.394,
"text": " How is it that you prevent yourself from getting into existential crises when it seems, at least to me from the outside, that you study similar, not theories per se, but the free energy principle is about existence?"
},
{
"end_time": 7348.063,
"index": 287,
"start_time": 7319.599,
"text": " and is about selfhood. So is your model so strong, Carl? Is your model already so strong that you can entertain these without worry that you'll break yours or without that being a potential possibility? Or is it because you purposefully analyze it dispassionately, analytically, intellectually, so that you're not identifying with what you are writing down on a piece of paper? How is it that you prevent yourself? Or is it constitutional, like by predilection, by personality? You're not neurotic, you're not as neurotic as I am. I think that's probably"
},
{
"end_time": 7378.234,
"index": 288,
"start_time": 7348.422,
"text": " truth to all that and by being neurotic you know that's an entirely fun you know neuroses are absolutely essential to get through life you know i have a particular set of neuroses which you know are high treasure and try to admit it in the big five model yes but i think you're also right that some people might call this mentalizing i think what you were doing is that you"
},
{
"end_time": 7405.589,
"index": 289,
"start_time": 7378.609,
"text": " In a raw, enthusiastic and creative way, we're just exploring different hypotheses about the way that you yourself, should it exist, relates to the rest of your mind and the outside world. But for you, it got conflated with emotional states like anxiety,"
},
{
"end_time": 7434.872,
"index": 290,
"start_time": 7405.828,
"text": " and your self-worth and those evaluative states of being. So if you want to, in your words, be more dispassionate about this kind of mental exploration on the inside, then you have to train yourself not to conflate the two. One way of doing that is just to become to mentalize. So mentalizing is usually taken as a way of, if you like, disentangling"
},
{
"end_time": 7459.65,
"index": 291,
"start_time": 7436.732,
"text": " So if you'll say very late at night,"
},
{
"end_time": 7482.227,
"index": 292,
"start_time": 7460.06,
"text": " You're normally going to feel fatigued and slightly less robust in a way physically. Now, it doesn't have to be like that, but that's the way that normally the world works, and that's the way that you've learned to model yourself. If you train to be a professional, say like an airline pilot trying to land a plane at night,"
},
{
"end_time": 7504.241,
"index": 293,
"start_time": 7482.602,
"text": " or like a psychiatrist who has to deal with these, has to see people going through these emotional or valence or effective episodes, whether it's sort of effective in the sense of, you know, maybe your depression or whether it's"
},
{
"end_time": 7526.817,
"index": 294,
"start_time": 7506.118,
"text": " What's that? It's a mixture of abnormal mood with having psychosis of the kind associated with schizophrenia."
},
{
"end_time": 7552.142,
"index": 295,
"start_time": 7527.056,
"text": " So it's having an emotional aspect to saying delusions and hallucinations, for example. Oh, I thought that those those don't go hand in hand. So some people have delusions, but they're OK with it. Yes, yes. Some people have hallucinations and would not would not even seek medical help or their family and friends would not seek medical help. So there are people out there. I guess my point here is that you can"
},
{
"end_time": 7582.125,
"index": 296,
"start_time": 7552.79,
"text": " mentalize it and build professional defenses against the emotional aspects. I think that's what I've done. So I think about these things as a doctor or a mathematician, not as a person. And if you want an honest answer, the other trick is to not use psychedelics, but you should try smoking. So smoking and cups of tea are great ways to... Smoking what? Cigarettes."
},
{
"end_time": 7610.759,
"index": 297,
"start_time": 7582.91,
"text": " You were saying the smoking of the pipe does what? Because I'm not the type of person, I don't smoke and nor do I want to smoke. But you were saying that the smoking does"
},
{
"end_time": 7633.882,
"index": 298,
"start_time": 7611.22,
"text": " What was the purpose of the smoking? It's just a way of using, um, nuancing your neurotransmitters and control that synaptic gain or the gain and the way to you for different, uh, different parts of your, um, belief updating. So it has the opposite effect of psychedelics basically. So that's why people like tea and, you know, have a cup of tea, calm down,"
},
{
"end_time": 7662.432,
"index": 299,
"start_time": 7634.155,
"text": " You know, stay calm and carry on wherever it is. Nice cup of tea, cigarette and cup of coffee. That'll sort you out. And it's because all of these drugs, the caffeine, the nicotine, they all act on the same kind of receptors at different levels in your neuronal hierarchy that are responsible for setting that gain control and gating during the attentional and the attenuation aspects that we're talking about."
},
{
"end_time": 7692.637,
"index": 300,
"start_time": 7662.637,
"text": " I would have thought that the caffeine would have the opposite effect of increasing your anxiety."
},
{
"end_time": 7720.828,
"index": 301,
"start_time": 7692.841,
"text": " that you might be anxious and in fact you're not, then just sensitizing certain interceptive sensations that register the fact in fact you're not terribly, your heart isn't pounding. That'll actually calm you down a little bit. If it goes too far then beta blockers will be another nice example. That's what I used actually. Oh right, oh well there you are. So what I'm trying to say is sometimes with skillful use, the use of a tobacco pipe"
},
{
"end_time": 7734.48,
"index": 302,
"start_time": 7721.118,
"text": " can have the same kind of effects as a beta blocker. That's all I would say. All right, professor. Again, thank you so much. It's much safer to stick with beta blockers, then you won't get mouth cancer or anything. So if you could..."
},
{
"end_time": 7763.302,
"index": 303,
"start_time": 7735.026,
"text": " I found the beta blockers work and there's also some evidence of fear extinction. Have you heard of that? There's this professor named Meryl Kenditt who I'm going to speak to at some point. She takes people through phobias where she gets them to be extremely frightened. She actually makes them almost manic or panicking at least. And then she gives them a beta blocker and then they still feel fear that day. It's only the next day when there's memory reconsolidation and the emotional tags are removed because of the beta blocker. So you have to have a good night's sleep."
},
{
"end_time": 7781.732,
"index": 304,
"start_time": 7763.302,
"text": " and then the fear seems to dissipate drastically for incurable PTSD. I was trying to do that to myself so when I was feeling some fear I would take a beta blocker and then I would feel the fear less still wouldn't be zero but then thinking okay great maybe this will be like my own intervention of a fear extinction technique."
},
{
"end_time": 7806.613,
"index": 305,
"start_time": 7782.944,
"text": " Yes, that's a very clever idea actually to combine beta blockers with what some people call flooding. So that's exactly the kind of way I was counselling. You would use people who smoke use smoking to moderate in the right context."
},
{
"end_time": 7833.336,
"index": 306,
"start_time": 7807.21,
"text": " Thank you so much. I will talk again. And perhaps the next time we speak, there were some questions that I didn't even get to from Michael Levin, from Professor Norman Wildberger, who's a mathematician, and there's quite a few. Okay. Carl, I kept you on for so long and this is gone. Well, this didn't, I didn't intend for it to go in this direction, but I'm extremely glad it has. And maybe my episode can serve and your advice, episode plus the advice,"
},
{
"end_time": 7847.056,
"index": 307,
"start_time": 7833.643,
"text": " can serve as a cautionary tale and as well as some practical advice as to what to do and how to avoid situations like this when exploring topics like this. Thank you so much, Professor. I appreciate it. I look forward to the next time."
},
{
"end_time": 7869.804,
"index": 308,
"start_time": 7850.623,
"text": " The podcast is now finished. If you'd like to support conversations like this, then do consider going to patreon.com slash C-U-R-T-J-A-I-M-U-N-G-A-L. That is Kurt Jaimungal. It's support from the patrons and from the sponsors that allow me to do this full time. Every dollar helps tremendously. Thank you."
}
]
}No transcript available.