Your Brain is a Beautiful Lie | Building AI That Understands Humanity

Where does a consciousness end and the rest of the world begin? Where's the line between inside and outside, between life and not life? Between the parts of the universe that are conscious and those that are not? Between you and not you?

We don't respond to like necessarily truth in the world information. We respond to information in the world that affects our homeostatic relationships.

The first time your brain ever lied to you was the second time you opened your eyes.

We look at the boundary of consciousness. Do we have to include all the parasites that are in us also?

Hey everybody. Welcome to the ABCs for building the Future podcast where we bring on researchers, founders, scientists, entrepreneurs, people doing big things with big stories to tell. People will build in the future. Today I have on with me my co founder and co host with Epistemic Meet and we are doing an extension to the previous podcast where we Talked about the 19 Ways of Looking at consciousness by Patrick House, who Jonathan met. He's a researcher in the consciousness space at Stanford. We ran into him at the Vitals Bay conference and Jonathan actually had a conversation with him and that's how we heard about his book. And so we started reading the book as our next Epistemic Me book club just randomly and we are on a consciousness kick. So we figured let's just keep rolling with that. We went through chapters one through three last time and this time we're going to deep dive into another chapter which we really like that ties into what we're doing with Epistemic re really well. We will riff about it a little bit from a product meets academia standpoint on Markov blankets and high level how those work. There is a chapter in Patrick House's book, chapter five called A Secondhand Markov Blanket, which has a little bit of a funny backstory to even the title and the editing process that Jonathan could tell too when he talked to Patrick House. We found a little Easter egg that was really cool. So we'll do that and then we're going to go into a build in public update where Jonathan had a really great build cycle over the weekend. I had some evolved thinking on our marketing and positioning and how we'll do our go to market and we'll probably just riff on that a little bit, just out in the open. Perfect. So let's get into it. I think where we'll start is Jonathan, what the hell is a Markov blanket?

Markov blanket in Carl Friston's free energy principle, also called active inference. It's a mathematical construct that can define the boundary system of A living being. And so it says like, what is external and what is internal and what is the boundary? And does this mathematically by creating a statistical model of internal external states. And then the Markov blanket is a. From the same person of the Markov kernel and Markov chains. It's a statistical model that basically says like, given the change dynamics and internally external, this is the mathematical boundary of the system. You can think of that being like a cell wall, for instance, or you could think of it being the boundary between a human and their environment. So yeah, as in the 19 ways to look at consciousness, I think it's chapter five. They introduced that humanity was given a challenge, which was to define like what consciousness was or how. How this person Anna was experiencing consciousness, or they could define free will. And in either, either way they needed to define the boundary of what consciousness was for Anna, the person in the story. And so like Markov blanket was brought in as a tool to say like, what's the physical boundary? And so you're usually thinking about three, the three dimensional space and like, and what's the internal, external and boundary system of this. Usually thinking about the sphere, like a spell wall. And then they're also talking about the Russian dolls of Markov blankets. Our structure for observation context is basically a Russian doll of Markov blankets. That is how we. That's how observation contexts are defined. It was definitely influenced by these ideas of what is a belief in active inference? And like what. How does a belief interact with the world? There's. Yeah, we could go into that more detail. That's a can of words to recap a little bit.

Markov blanket just explains the boundaries between high level. Let's just call them agents for now. And Patrick House wrote this chapter, a secondhand Markov blanket. And let me just read the first few sentences which give a great foundation for the discussion. And then we'll go back to the observation context concept that was influenced by all of these topics in the science of consciousness into our SDK. So by Patrick House, where does a consciousness end and the rest of the world begin? Where is the line between inside and outside, between life and not life, between the parts of the universe that are conscious and. And those that are not, between you and not you. I think that's really beautiful. He's a really great writer. It's very provocative. Maybe we even start this by talking about those questions and then we could tie into more of our philosophies around Epidemic me and the implementation that last one between you and not you. I think that is so important when it comes to the free energy principle, as Jonathan said. And I think what I want to double click on now is that Russian dolls concept, paralleling, observation, context. And I'd love to hand it to Jonathan if you could speak more on that and how we were influenced by this and how it kind of works in the grand scheme of our product and we could mold it around how we're thinking about AI. Brian, I think that could be a good anchor there, a use case to explain this. Yeah.

Where to start? One thing I was curious if this relates to your own experience because you studied chemical engineering and if you remember, like in the atom atomic structure, you have these kind of different valence bands that each have their own, like, you know, electrical characteristics, electron characteristics. And so like there, there's kind of like, you know, atoms can get pretty complex, right. In terms of like the electron shell structure. And you probably remember these like equations like sp2, so on. Does that ring a bell? And does that conceptualize the nested Russian doll idea from a physical standpoint?

You mean, for me, or do I think the whole audience would grok that?

I think chemical engineers, I guess for you, does that seem like a, like a relevant sort of analog?

I'll speak for me and the other chemical engineers from. And particularly for me as a former chemical engineering student. Like, I was trained in it, but I worked in technology all my career. Yeah, I like that. That totally makes sense. And it is the relationship between these entities, the valence bonds that you were speaking of. Like, I remember like studying these things and looking at how much energy does it take to break these. And especially in the biochemistry context as well, it's. It is interesting, like how we all walk around doing normal human things like eat food and stuff like that. And then also you could just really deep dive into the atom microscope and then start calculating how much energy it takes to break bond between a couple atoms.

Cool. So I think, yeah, it's kind of like an. There's an intuitive thing that like, I think a lot of us introduced you in school, which is the. This rusted, nested Russian doll structure in the atomic structure. And I think when we think about nested Russian doll structure of Markov blankets, we're kind of pulling up like a cell or, you know, or complex something that we can identify perception actuation loops in. I'm not saying that there aren't perception reactuation loops in a molecule, a chemical molecule. I'm just saying that I don't have any observational evidence about that. And like, nor do scientists. So I think the Markov like is most easily thought of, like thinking about cells and thinking about things that are. That are built of cells that are more complex. And so when you think about our own, like, perception, actuation, loop with the world and our own sense making, there's a lot of things that we're simultaneously managing from, like, a homeostatic sense. We could be managing temperature, we could be looking for food, we could be avoiding a threat or predation or, you know, whatever. In the nest and structure of these dolls, I think each of them is sort of managing its own homeostatic process. And in fact, chapter six, Secondhand Markov Blanket, talks about simulations and how this idea that we respond to is actually, I think, very similar to Donald Hoffman's work, which is the idea that we don't respond to, necessarily, truth in the world, information. We respond to information in the world that affects our homeostatic relationships. So if you think about yourself as like a set of nested person dolls that are simultaneously perceiving many different homeostatic signals at a time, each sort of like the boundary system of each one of those dolls to, like, figure out, like, what the threat is and then like, to, to route it through the, the nervous system infrastructure and the various, like, you know, adaptive homeostatic processes that we have built up from our evolution and respond to it. And I have. I have a couple different artifacts that to share to help visualize.

Yeah, let's go.

Yeah. First, there's a great video of a monkey beating a human in a memory test. Have you ever seen this before?

I was about to die at. If you stopped that. There's a great video of a monkey beating a human. And I was about to be like, what. How does this relate to consciousness, Jonathan?

Okay, so watch this chimpanzee in a memory test. So he's being flashed on the screen eight or 10 numbers. He's going to memorize all 10 numbers, the location of them, and then tap on them. Do you sort of get the idea what's going on there?

Yeah, yeah, yeah.

So you put a human next to them, which they did. I just initially saw this on a BBC documentary. And the human, like a smart guy, educated, like us, who runs a lab, there was like, maybe I can remember six numbers. And the. There's a theory that, like, why can't a chimpanzee remember, like, so much more information? And I think they showed chimpanzees memories up to 16 digits at a time where they just flash the numbers, remember the Exact location, the order and tap them off and they get like a little treat afterwards. This wasn't like the Michael Jordan of monkeys, heavy memory, right? This was just like one sample from a random place in Japan that they trained him to do this and the humans can't do it. And so there's an idea that there's a cognitive trade off between language, being able to have linguistic development and like the, the pace at which like you, you memorize, you can memorize things in the environment as a primate. And chimpanzees and great apes actually have like a more evolved neuron, neuronal system or brain than, than monkeys themselves. So you, you have monkeys, then there's a layer of brain on top of that, which is for ch. Chimps and gorillas and the great apes. And then there's a layer of brain on top of that for humans, the prefrontal cortex, which now I'll share like a way that, that sort of like cognitive trade off how I think about it from a frequency domain perspective. So this is just like an analysis I did about the frequency response at different homeostatic functions. And you can see from the paleo reptilian brain, the pal, mammalian brain, primate brain, ape brain, insular cortex and neocortex. And basically the more complex information you get from your environment, the slower your interaction with the environment is. And like, you know, the difference between say an ape brain at say 311 milliseconds versus a human brain at 721 milliseconds tells you like both though we're slower at perceiving the world than our primate ancestors. And like, we sort of like have a different memory relationship with it. So I had, I visualize this to like just take a frequency response sampling of the nervous system. I built this up a couple years ago and you know, so what Patrick says in the book is that like we, the secondhand Markov, like it is because we interact with the world at the frequency of one second. Like as in we can respond to the world, we can move our whole body, we can actuate onto the world in that frequency and it's actually about a 721 milliseconds, so almost a second. Whereas like our fight or flight responses, it can be as fast as 1/7 of a second and then some of our startle reflexes can go as fast as 14 milliseconds. And like we basically have this situation where the more complex information that we're making sense of, the slower it is. And so we're optimized to learn through shared language, through our interactions in the world, and we lose the ability to do some, like, faster processing. Like we just saw that the chimpanzee able to memorize information. And I thought that was kind of an interesting example. I'm curious if that ties together for you.

The AI agrees that if this is impossible, it will settle for a statistical distribution of probable or highly likely Hannah thoughts instead of an exacting list of them all. If both of these prove impossible, because free will is truly free, the AI adds an allowable success condition. It will settle for an exact atomic description of where Anna's consciousness ends during her surgery in 1996. As long as it accurately defines the line between Anna and not Anna during the experimental trials. This was a great chapter. I friggin loved this one. And so that was what humanity had to go through in this story, which is. And thought experiment, which is. What is that answer? And to give some more context for the audience who might not have heard the previous podcast where we talk about Anna's brain surgery. She had a brain surgery. This was in. I think it was in chapter three or two, I can't remember. And we talked about this in the last podcast, but she was getting brain surgery. Her head was cut open. And during the surgeries, when certain electrical pulses were in certain parts of the brain, she would laugh or say the most random things or do the most random things. And so Patrick, in this essay and chapter in this book, is essentially posing this thought experiment of like, well, what if AI was saying, hey, we're going to kill you all unless you tell us exactly what happened there? She logged in, read, go. What is consciousness? What is it? What does it mean to be in that moment? What was happening to her? What was the difference between her and everything else? Jonathan? Yeah. Do you have anything to add to that narrative tying to the Markov blankets and epistemic B?

Yeah. So I think the. We talked a bit about the spatial definition of like a boundary system. And I think what Patrick introduced in the book was the go through. I think it was like a middle schooler asking a question on an Internet forum. The case of a caterpillar becoming a butterfly. Like, how do you. Like, where's the boundary? First you had a boundary of a caterpillar, now you have a boundary of a butterfly.

Like, what.

What is going on? How do we describe this? And that's when he introduced the idea that like really need a boundary to define like from second to second. The difference in the organism. Even to answer the question Accurately. And so the chapter does a really good job of describing the complexity that we're living in. And the punchline of the chapter at the end of it was they just, they defined that Anna was everything all of the time or something like that. And she was meant to say that her consciousness was just part of everything to some degree, or if maybe specifically part of everything, but. And they had chosen that answer just to confuse the AI so that it could maybe on an off chance that it would take forever to like invalidate the question, they could like just hold it at bay. The overall thesis of the chapter was somewhat like, like, hey, we don't really have the tools to define the boundary of consciousness, which is interesting. But I can, I can give my own takes on, on these different tools real quick.

I'll just give humanity's answer so we get to the punchline for the audience and then we'll back into some discussion between us on this topic. Related topics, Humanity said, dear AI Anna, is all of it forever moving forward and backward in time to everything. Warmly yours, Perth.

Yeah, yeah, it's very thoughtful. Illustrative figurative chapter. And one of my favorite things that happened this last week was I asked chatgpt03 to like give me its take on the chapter and it interpreted the second dash hand literally and gave me like, it gave me like an in depth metaphor explanation of how this was a secondhand blanket being passed on as opposed to the like second hand on the clock blanket. So it actually got confused by that typo. Like I thought that was hilarious.

To give more context on the typo, I had mentioned we had this Easter egg moment from Patrick House himself, but we didn't give, we didn't talk about it in this pod. Could you tell that quick story?

Yeah, I, I mentioned, I talked to Patrick like five minutes at a, at a dinner for the Vitalist Bay concert conference. And I had seen him give, he gave like one blurb of, of speech during Brian Johnson's presentation where he talked about like the metabolic function and evolution and my ears just perked up. It was like, there's not enough people talking about metabolic functions and evolution. I gotta check out this guy. And when I found, I went to his website, his book is listed and so I just bought it right then and started listening to it that night. And then I met him the next day. I was like, hey, I downloaded your book, I'm listening to it. And, and I also just listened to Annika Harris's Lights on and he said, I'm on chapter four of Annika Harris's lights on. There's the one about parasites and plants. And Patrick's work definitely has this, like, this embedded, like, question about, like, if we look at the boundary of consciousness, do we have to include all the parasites that are in us also? Is that like, part of the equation? Because it affects your, your interactions with the world.

Right?

It's so. And it's, it's a really interesting point. But anyway, I talked with him and he said, because I, I mentioned what we were doing and what we were working with, epistemic me. And in this nerdy group, I would talk about, like, how we were building belief systems based off of some of Carl Friston's work and the free energy principle. And he said, you know what? I've actually got a chapter in my book called the Second Hand Markov blanket. And he said this is the only typo that he knows about in the book. And his editor changed it from second space hand to second dash hand, which is the second hand means like, it's like goodwill. You're giving like, you know, a used piece of clothing to somebody else.

Right.

But he meant secondhand is in the second, because that's how long it takes for the collective sensorium, or sensory information of a human being to collect and for us to make a decision and to act and to make an embodied motion of the world. And so he wanted a second by second Markov blanket to describe the border of consciousness from like, as. As we sort of are being in the world.

I love that. That's really awesome that we had that Easter egg. So thank you for telling that story again. I already know about it, but it's really awesome that you heard this guy talk, was interested, bought the book that night, went again, met him and talked about this. And now we're talking about on the pod. Maybe that's the free energy principle at work. Who knows, folks?

Yeah, and like I said, my favorite thing this week was that he actually confused chatgpt03 in the same way it confused his editor. And they made up a whole story about, like, what it meant. And I said, no, it's actually this kind of secondhand. And then it reinterpreted the chapter correctly. And that was, like, really funny.

Well, one thing back to when we talk about the difference between us and our primate brethren and how we have this different. I'm just going to abbreviate it and just say we have a different refresh rate for processing data, like how much time it takes our brain and different parts of the brain for that information to be made sense. This is called a free refresh rate. Very easy. If you're a gamer or have a monitor, you might have seen 60Hz or 144Hz or whatever. That's just the refresh rate typically in gaming. And I only know about because I used to play a lot of games and semi competitively as well. Like especially in first person shooters, you want the refresh rate to be really high so that you get the information signal as fast as possible. You know, know fast. The faster the better. Right. So that you can compete against the competition. Like if you see somebody before somebody sees you, obviously that is better. And that can be said for probably just predator and prey as well. So going back to the primate where they have to do this berry picking and they have this higher refresh rate than we do and maybe this is not the right way to phrase the question, but one of the things I was thinking is this second by second Markov blanket would just be a very different. We'd have to take time into account for that particular agent too, like the primate and their particular refresh rate versus us as an agent and our refresh rate.

Yeah, I think it's got a different set of homeostatic functions it's performing. It's got a different interaction with its environment, it's got a different refresh rate for how it interacts with the environment.

That's so interesting. So it's sensory perception loop for evidencing beliefs is different. Right. Then for us what that means is we'd have to take that into account for epidemic me and modeling belief systems. And because we have this vision that we should have a data model that is flexible enough for every agent and organism on the planet so that we could do collective belief modeling. So could you talk to us a little bit about how we're thinking about that with the SDK, Jonathan?

Yes, let me just give an overview. I actually have a diagram that I built a few months ago that might be helpful to walk through. But so in the epistemity self model there is a belief system tree that sits underneath it. And what we call the, what we've been calling these nested Russian doll models is just, you know, like a, a tree with a, with an edge. Tree with an edge. Tree with an edge. And then a set of states that are the leaves that can perceive the world. And so the, the belief system is built in this way where you could take a snapshot of that at a given second or given single inference. And I Think about that. By the way, from the philosophical sense, I think of that as being like being in the world. So it's a metaphysical construct, the relationship between self and world, a single interaction. Sure. Say it's about a second long and then it captures like a belief expression during that time. And you know, it could include the actuation that you had with the environment. And I use a tool called the observation context to define the set of information that's being that we're using our, our conscious attention towards both in terms of receiving information and also, also manipulating the environment. So if I was going to. We talk about one second intervals, we're really talking about movement, you know. So let's say that maybe I'm at the gym and I'm doing some squats and I am like perceiving the weight, you know, that I'm holding on my back. Maybe I'm actuating so movement, getting feedback in terms of how my body's reacting. My body's in this loop and it's expressing beliefs as I'm, as I'm like squatting up and down and doing my reps and I'm feeling, I'm tired, you know, and my beliefs are like reinforced that I'll get tired on rep 6 and then I'll, I'll be finalized on rep 8 or something like that. Those are all like, that's, that's my beliefs expressing themselves and me actuating on a second by second basis through a dynamic interaction. It gets more interesting when you start thinking about like the process of learning and how beliefs, if you look at, if you sample them every second. Sure. What we're really trying to do is identify like the change in beliefs and the updates to beliefs where they're reinforced, where we see surprises and curiosities, where we have to create a whole new edge on the system because we had confusion and we're just on the wrong track. And we can't understand this without a whole new sort of belief structure. If you think about the time lapsed version of that data as we take it over time, that's really a way that you can learn about how your users are interacting with your user experience and with your conversational agent or whatever, you have a second by second snapshot. But you can see how the beliefs are changing over time. So you have like a meta learning annotation on it. We call that like one, at least one of the aspects of that meta learning annotation. We call it epistemic emotions which modify how beliefs are updated from cycle to cycle. Yeah, Diagram that, yeah, might be helpful. And this was a point I was actually looking at like, you know, all the different sensor loops, like this is how light is perceived in our organism and different like clocks that we have in our system, endocrine dynamics through a different day through like your. And so these are all different like phenomenological interactions that we have with the world. Now something, something like circadian rhythm. When he talks about a nested Russian doll, like in a sense of like space, there's also a nested Russian doll in sense of time.

Right.

And so a circadian rhythm happening in a 24 hour cycle nested within that is an ultradian rhythm with an hour 90 minute cycle nested within that is like a whole bunch of, of sort of second by second motion interactions. Right. And you can see what I mean by the nested doll and that those are just like very high level.

And then we triple or million click down through. And then it's quantum.

Like it's exactly. This is all happening different frequency domains, as we say in electrical engineering. So yeah, when I initially modeled the belief system, this was one of the tools I used. And this is the observation context, observing the world. This word ngram comes from Michael Levin. It's a way to describe how memory is being stored as in like a substrate. So you can just think of like every cycle that we have of like perceiving the world can encode a memory at a certain like frequency. And I, if you remember I talked about earlier how they started measuring a baby's memories through FMRI and how like they're a lot smaller than adult memories, like they take up a different footprint.

Right.

And so they're probably perceiving the world at a different frequency than an adult and also storing memories. And I think like on a second by second basis you could think of like, what the limiter is like is actually can I store memory and react to that memory during that time frame. So yeah, I have some different sequences here. But then to model active inference, I have basically created this observation context. We have a last observed state of the world, expected observed state of the world. Both of these are stored in memory and stored as a belief. And then we use evidence to predict based on that what the future state of the world's going to be. And then we have an epistemic motion, like confirmed belief that stores back into the belief system. So that's that one second type we're talking about. This is kind of description of like how it's working in the memory structure of our Epistemic me code. And yeah, I was showing some different examples here, like collecting evidence to confirm a belief. Collecting evidence to update a belief. If you just confirm a belief, we use your beliefs to predict the world was exactly what you observed. Again, you can store belief confirmation. If you predict something that's slightly different than what you expected, that's called surprise. And that's going to update your belief in a way that recalibrates the probability distribution of what you're going to see in that observation context the next time.

I have a question, Jonathan. People lie to themselves. We all lie to ourselves. You're. There's a great quote from Patrick's book which is the first time your brain ever lied to you was the second time you opened your eyes. And in particular this a lot with people that lie to themselves about adopting healthy habits where they say I'm going to do it tomorrow. That's a very common lie. Right. And so there's that belief that you'll tell yourself about yourself and is it ground truth or not? And so I'm now thinking about the use case where how does our SDK know In a software interaction or with an agent, how can it discern that this is a false belief that this user has told themselves? Like it's very meta in health, it makes a lot of sense. It's like if you are saying you believe health sleep is really good for you, I should be seeing your sleep improve over time. Or you should be doing actions to improve your sleep. You should be trying to improve it if it's so important. So we could see perception or the belief that one has of oneself and then we could see the wearable data and we could see the exercise data and we could see the tracking data and we could see does that map to behavior and habit? But what about use cases where we like don't have the behavior habit data? Like is there ever a way where we could eliminate that use case that I just talked about?

Like, what's an example that you're thinking about?

I'm more thinking in generalities. But now to restate sort of the question, it would be, are there use cases out there where a software system could use our product to learn more about the user but there is no behavior loop to close out evidencing of beliefs and therefore that like corner case I just talked about. I suppose that's the question.

Like you have no, no history of this person at all. Like you just. Or trying to model their beliefs on the fly.

Oh, okay, here's one. Personality tests they're one way. You don't get any. You don't get any say into it.

It just happens.

I take the Myers Briggs test and there's no agent or human I'm talking to or there's no, like, oh, I can confirm this. Right. The test is there. It's a static result. So. Yeah. So that there's no behavior involved. And at the point in time, I'm sure there's. There's gotta be people that have studied this. I'll look this up with Perp Perplexity Pro after this. But think about the amount of people that would take a personality test and based on if they're in a stress state, would they say certain things about themselves or if they're in a really good state. And I think your perception and your beliefs may change based on that. That's the observation context.

Right.

But like, then in that personality test powered by epistemic B, how would they. They wouldn't ever have that loop. They couldn't actually confirm it because there's no actual loop.

So you're saying if you took a Myers Briggs test results and like, something that doesn't have an experiential memory to it and tried to integrate into our system or.

Yeah.

Or like, it would just be okay.

It would just be wrong. Like, we could just never get to. We feel pretty accurate about this.

Well, I mean. I mean. Okay, so just think about your experience of taking the personality test, you know, and I'll speak to myself because I think when I first took the Myers Briggs, maybe I was in college and I thought, like, oh, this. You know, I'm wording. I'm reading these words on a page. It's kind of like a horoscope, right? It's like, they don't seem wrong. It kind of seems like me. It sounds like me. And then I take those hypotheses about whether I'm extroverted and the next time I'm in a social situation, I'm testing that. Say, like, yeah, you know, and it might even be, like, help helping to reinforce what the test came out with, which was that I'm an introvert, by the way. And so. But think about what's happening there is that I've gotten, like. I've gotten a piece of data from this test that, like, filled out likert bubbles and it gives me a formula back for, like, how I might behave in the future. And I can use that for. As my own evidence to say this is how I behave in certain situations. This is how I don't and I can weight it with extreme authority. I can say this thing is 100% right and I'm actually going to, it knows me so well that I'm actually going to behave in the way that it says that I should behave because it's so right. Or I could take it as like, like a very low evidence that, that I. It's like it's just like one input into of many and I'm going to compare it against other experiences I have and other observational studies that I do and it's going to be a weighted average and that that's how it's going to evidence my beliefs in the future. So I think of it like we just, I would, I would categorize it as a piece of evidence and then some of the personalization we talked about are meant to like unfold. Like, you know, why do you believe this thing that you believe? And I think there's some cases where we would, we would interview users to the point where we would say like, oh, they believe, they believe this because one of the evidence is this personality they test, they took and it's wading into their beliefs and that's a valid piece of evidence for them and it's useful. And then there are sense making in the world. So anyway, I put it as a piece of evidence as opposed to like an observational experiential model or behavioral model. Interesting.

Okay, okay, that makes a lot of sense. Thanks for sharing. I suppose one of the things that I take away from this from a go to market lens, if we landed with a personality quiz provider, by the way, shout out any developers of personality quiz providers. If you want to make your shit better, hit us up. But if you want to make your shit better too, having a way for your users to come back to you and give you more data or updates or say, hey, this part of this result actually I don't agree with then powered by our thing, I think then it can be better. But okay, great. So we went on a decent tangent there. That was fun. We were talking about secondhand Markov blankets. We talked about Patrick House's episode, his essay on that, his short story on it. We talked about the narrative where humanity is basically under threat by AI in this thought experiment story to go figure out consciousness, how to define the boundaries. And so to summarize, in that story, humanity kind of gave a cop out answer, which is Anna, the subject in question and her consciousness at point of brain surgery is everything and everything all at once. It's like everything before and everything that comes after. And then for us we have this observation, context, concept that Jonathan came up with where we could store those different lenses or refresh rates, processing speed in which an agent or organism perceives the world and gains information and therefore the rate at which their belief systems update. And that parallels a lot of that concept. Patrick House writes about the second by second Markov blanket, the Markov blanket over time for this particular agent. That is really like how you would define the boundaries of consciousness between agents, at least per that view on the world.

Yeah, a few things to say there. I mean, I think one being the philosophic, the philosophical distinction between being and becoming and what that is. You know, if like we're, if we're putting a boundary between each interaction that we have, each cycle of interaction we have with the world, I'm usually looking at that as called, as what we call being like is. It's a metaphysical. But then there, there's a place where the underlying belief structures that you have in the world change enough, where you can say I'm becoming something, maybe through the process of learning. Like through the process of learning I'm updating my belief systems and become something different than I what I was before. And that's not going to happen necessarily on a second by second basis. It may happen over some number of days, weeks, months. You know, sure, you can learn things in a shorter period of time. But you know, I thought, I think it was like micro becomings.

Micro becomings. That's such a great concept. I love that. Yeah.

Great content. Yeah. And that's where these nested Russian dolls come, you know, like in a frequency domain. The nested Russian dolls give you change that you. Of how you're changing over periods of time, like short intervals and long intervals. And so something like Piaget's model of embryonic development where he necessarily give you change of how you're changing over periods of time, like short intervals and long intervals. And so something like Piaget's model of embryonic development, where he's a psychiatrist who studied his own children and their phases of development and wrote a paper about embryonic development that was like appropriate for the time, maybe it was like 1930s or 40s. And you know, he, he described these phases, like, you know, adolescence. You know, he's. He's trying to find these points in time where the caterpillar becomes the butterfly. Where do things fundamentally change? And there's I think a famous paper in perspective taking by Selman S E L M A N where he identifies there there's like a point in a boy's life where between like age of 7 or 8 where they can perspective take other people and read their emotions and read their points of view. And it's just like before he's a psychiatrist who studied his own children and their phases of development and wrote a paper about embryo at that point. And they can't do it. And after that they can. And there's, there's like a period of like cognitive development that happens that enables a new set of like, homeostatic functions the, the human embryo child is able to exhibit. You know, and this is measurable thing. So. And we have a number of those, like I think, I think I mentioned before, like our, our prefrontal cortex doesn't fully form in our brain until we're 27 years old. So we still, even as adults, we still go through these changes. And then like, you know, there's these new like aging studies that we have, like these aging cliffs where you really like severely change epigenetically maybe around the age of like 45 and 63 or something like that. So there's this like, there's this process of impregnic becoming that's constantly going on. There's this process of like us changing our habits as an individual that might happen in periods of like six weeks or six months. There's a period of like, you know, these, these micro becomings are learning. So like how we're like learning in the short term. And I think what Patrick's describing, development that was like appropriate for the time, maybe it was like 1930s or 40s. And you know, he described these phases like, you know, adolescent. The second hand markup blanket is like just what I normally think of as being, which is like maybe the smallest amount of change that you can have. But it doesn't really meet the definition of philosophical becoming because it's like you're the same structure or form of belief system that more or less from one.

Second to the next, it's like enough code for a meaningful PR Commit.

Yeah, a diff.

Right? Have you done that? Is it enough?

Yeah, it's like the difference between a commit and a release. There's a change code, but like, does it actually change the functionalities for the user or like. That's a good point. Change the system as a whole. I love that since, you know, he's, he's trying to find these points in time where the caterpillar becomes the butterfly, where do things fundamentally change? And there's, I think, a famous paper in Perspective Taking by Selman Sel in.

Visualization of what comes up to mind. It's just a provocative visualization. The question of when does the caterpillar becomes the butterfly? And in the book for the audience, Jonathan touched on this a little bit, but in the book they kind of talked about that as, so some little girl brought up in this five year race to figure out consciousness. In that narrative that Patrick House writes, he wrote like some little girl posed, hey, what happens when a caterpillar becomes a butterfly? And everybody was like, oh, oh my gosh, like, okay, this must mean something. What does this mean? Like, and so they brought in that becoming concept and that evolved in that storyline into, oh, there's a goo across all consciousness, some sort of consciousness goo that glues us all together. That's the observation context, y'. All. Let's shut it down. Pods over, Mic drop, get out of here. Yeah, but anyway, that is how we think about that with epistemic me. It is the glue that will hold things together. And we hope that one day it will be holding collective intelligence together too. That would be really cool. So we'll be able to go back to this pod and reference this someday.

Yeah. So speaking of the caterpillars, the butterfly and that transition and what the glue is. Right. I have to bring up Michael Levin's work on improvisational memory.

And please.

He based it on his observation in a lab of like, they trained a caterpillar to do something and then it metamorphosized into a butterfly and it retained the memory, but it used it in sort of like a different way that was like more butterfly specific. And so it led him to this, like, philosophical theory of he calls self improv, improvising selves, where, like, you think of the actual structure of memory is a communication to your future self. So there's no more like storing data on a register type of memory. Like, our brains aren't holding. Holding these memory in the way that our computers do. It's just sitting there for a future version of yourself. So that, like, when your observation context kind of makes it relevant again, it pulls up that that information, it completes the reception of the communication to your future self. I'd say you can imagine this way if we have these secondhand slices of who we are as people. Michael Levin sees that as each one of them is just sending a message to its future self. And so you're both sending and receiving messages from your future self. And the illusion of that is what we call memory. But and by the way, like, it doesn't, like, the memory doesn't have a, an ultimate like, meaning or purpose. Like, so if a caterpillar killer discovers that, like this leaf tastes good, right, that might mean to a butterfly that like, that chemical signature is a good place to lay its eggs or something like that. It uses the information that it has in that context from its own experiential memory and it applies it through its, its intelligence mechanisms to like, what, whatever its problem is at that time, which is also just like a fascinating thing to think about. And if you think about, like how you reuse some of your own own memories, like, if you're like, you know, certainly, like, I don't pull these, like, I don't just have facts in my brain. I pull out these experiences that, you know, may be relevant to a certain situation in a way that I don't even understand. Why am I remembering this right now? What is this? What is the bearing on the world? How am I going to interpret it, by the way? And what, what mean am I going to make out of it? That's more the human condition than storing vocabulary from flashcards and having to recapitulate it. Well, I can attest.

Definitely true.

I definitely am where he identifies. There's like a point in a boy's life where between like age of 7 or 8 where they can perspective, take other people and read their emotions and read their points of view. And it's just like before that point they can't do it, and after that they can. And there's, there's like a, a period of like, cognitive development that happens that enables a new set of like, homeostatic functions that the human embryo child is able to exhibit. You know, and this is measurable thing.

When I am encountered with situations, any situation, my brain's architecture and visualization is not somebody takes out a note card. It definitely is more evocatively emotional. There's a sense of, yeah, there's sensory connection there to a memory. As you bring this up, I almost think of it in two different analogies. Like, it's like a letter you write and throw into a bottle in the ocean. Someday it meets something. Another example is it's sort of like Kafka events. You don't know what's going to subscribe. It maybe doesn't even exist yet. I don't know.

You mean Kafka the, the framework or Kafka the event?

Event framework. Okay, yeah, yeah, yeah. Like the whole pub sub thing. It's like you publish no idea who's going to sub. No Idea might not even exist, so. Interesting.

And we have a number of those, like I think, I think I mentioned before, like, our. Our prefrontal cortex isn't fully formed just.

Yet, but like, you just do it and it's there. And there's no like in. In that example. There's no purpose for why you put it there, but, like, it's there and maybe you'll have.

Yeah, that's exactly how Dean and I envisioned the memory structure. It's called event sourcing as a pattern. And it just, you know, you keep. You keep version logs of a specific object in the memory, and you emit signals that may be received or may not by their counterparts, but if they are, they mutate the object. And so event sourcing as a pattern seemed like a really good way to kind of model this.

Makes a lot of sense. Yeah. Makes a lot of sense. Sweet. Okay, so looking at the clock, just because it's getting down near my wind down period. So we'll have to do a quick build in public update. But this actually connects really well.

So HEPA's in our brain until we're 27 years old. So we still, even as adults, we still go through these changes. And then like, you know, there's these new, like, aging studies that we have, like these aging cliffs where you really like, severely change epigenetically, um, maybe around the age of like 45.

And me models the self and we've built out all this processing logic and.

Engine so that 63 or something like that. So there's this, like, there's this process of impregnic becoming that's constantly going on. There's this.

We can empower evolving belief systems and all these concepts that we just kind of talked through. That secondhand Markov blanket or second by second Markov blanket concept, the free energy principle, all of those things have influenced our philosophy in our implementation as we described. So we really loved this chapter. And actually, Jonathan, you should tell people about how you literally have a blanket with the word Markov on it. Could you talk about. Oh, yeah, that's like a quick little awesome Easter egg. Tell. Tell us how you got there.

Hold on, I'll sit. You want to see it?

Yeah, I want to see it. Yeah.

So I saw this. I just had to order it. And you see that?

Yeah, with the nodes, can't see you. I'm under my blanket. You're outside my boundary, bro. Great.

Yeah, it's just like a nerdy little purchase. Like, I was like, who would make that? But if somebody made It, I'll buy it. And that's what happened.

Now I have it.

And it literally creates like a boundary underneath my bed between my wife's clothes and my computer stuff.

That's so great. I love that there's some teleology in the Markov blanket. Cool.

So one more example of this that I think is pretty illustrative. Yeah, this would be a pretty quick one. This is also coming from Levin's work. It's a paper he wrote with Daniel Dennett, recently deceased philosopher, who's just like amazing sort of consciousness philosopher. And this was, this is when they were trying to explain, I guess, like the how a biofilm. So they have different cells. These are each cells this process of like us changing our habits as an individual. That might happen in periods of like six weeks or six months. There's the period of like the cells communicate to each other through ion channels. And they have this example of a bacterial biofilm, multicellular cooperative bioelectrical networks. So we have this. So I always look for analogies to like think about how multicellular life formed. And in this example we have the cells all communicating to each other through ion channels. And they're getting the big picture awareness. You know, these, these microbe comings are learning. So like how we're like learning in the short term. And I think what Patrick's describing in the second hand markup blanket is like just what I normally think of as B of the outside world that they can all share and be coherent together. So very just like concrete example of where you start to see a boundary between this group of cells in the world and sort of like a multicellular biofilm. And in my own work there's a. Which is like maybe the smallest amount of change that you can have, but it doesn't really meet the definition of philosophical becoming because it's specific phenomena. I think if you take the speed of the communication channel between each ion channel and the refresh rate on the world. So let's say this biofilm updates like one time every 100 or 100 times a second or something. And the speed of, of the communication is like one thousandth of a second. How many cells wide can this biofilm be before it's not coherent with itself? You know, and that's something that like electrical engineers are sort of talking in terms of Nyquist sampling theorem. But I think that it can really inform this type of work. And I haven't seen it used anywhere yet. But just like, what is the. How many sort of like cells can we have within this boundary, the world before they're incoherent. What happens when we add a cell that it can no longer get the information fast enough within the frequency domain that it needs it, that it can behave right. Maybe it becomes cancerous or.

Right.

I think that's. And I'm not the first person is just that the root of cancer is cells that can't communicate. Matter of fact, I've heard Levin sell that multiple times. But even just like if you think about tissues like you're the same, you're in the same structure or form of belief system that like, you know, over more or less from one second to the next. It's like an organs that are built on animals on this planet. Like there's a definite like sort of limit to how many ion channels they can connect with each other and share information about their environment. And that's just a cellular example. I think that same phenomenon is true if we talk about like groups of coherent people, you know, say like coordinating on a battlefield or something like that. Or in a factory number.

Yeah, yeah.

There's a relationship between the internal communication of beliefs and the external ability to respond.

Yeah. It really does come down to that when you put on a systems lens to enough code for a meaningful PR commit.

Yeah, a diff.

Right. Have you done that? Is it enough?

Yeah, it's like the difference between a commit and a release. There's a change code. But like, does it actually change the functionalities for the user or like. That's a good point. Change the system as a whole.

I love that. It as like think of every agent as a node. There's information shared between nodes. Collectives of nodes get together. They have to share information between the nodes. What is that rate? What's the rate of understanding? What's the refresh rate, I should say of that outside world with that boundary. So you see that everywhere, like social circles. Who gets invited to what party generally just comes down to oh shit, like somebody you remember. Like did somebody have that person on Facebook to add to this? You know, like it's like that or. Or in a meeting at work at. In big tech. A lot of times. A lot of times there's politics because somebody got. Didn't get invited to the meeting because of Conway's Law. Because we're bad at Conway's law.

Yeah.

You see that all the time. So it's funny to see it down to the things that make us. Us have the same problem.

Yeah, totally. And so cool. It's funny to see how that dynamic changes Too. Once we have, you know, self representations, once we have AI that can communicate on our behalf, I mean, we're already so much more connected than we were for all of humanity, right?

Yeah, yeah, yeah. Like just the evolution of if you just abstract it to information processing, right? Like how fast can we process information. Like before it was pigeon courier, right? Or let's get runners to go run to a village to freaking like, oh, actually we're at war now. And then it might be two day time delay. Now we get it in a microsecond visualization of what comes up to mind. It's just a provocative visualization. The question of when does the caterpillar become on Twitter? Like we're at war now. And then thinking about that evolution too. We got telephones that rich people could buy and then that became mass market and the audio sucked for a while, signal sucked. Or when we had dial up at the Internet, like mom picked up the phone, Internet cuts out your gib. Video game drops. Multiplayer is single player now. And then now we're in where when we're recording a pod virtually. Yeah, yeah, it's wild. And then there's the butterfly. And in the book for the audience, Jonathan touched on this a little bit, but in the book they kind of talk. Now the next step is I have a little. We have AI agents making sense of the world, right. It's the most powerful microscope ever built to process all the information on our behalf. So next step would be, I suppose, you know, you see like the ray Ban Facebook meta glasses. I suppose that's even another step of augmenting humans with a about that as so some little girl brought up in this five year race to figure out consciousness in that narrative. AI sense making tools situationally. Next next step would be the neuralink stuff and becoming that. Patrick House writes, he wrote like some little girl posed, hey, what happens when cyborg. I'm all for it. I'm all for it.

Yeah, yeah, I think that was. That's actually where my. So this whole idea of, of like how we change over time is called teleology. And I heard this great podcast, the Mind Body Solution, he's got some great guests on there and he had Michael Levin talking to Terence Deakin. They both had their own sort of theories about teleology, dynamic teleonomy or I think, yeah, Levin calls it teleonomy and Nikon calls it like teleological dynamism or something like that. But they are, they're literally trying to understand like how did dynamics shape life, how did. How does how does the embryo like progress? How does it form? And you can kind of simplify the thought experiment down to like how an organism knows to make the next cell. Kind of like how an LLM needs to come to the next word. Like we need to know the next cell and how it's going to differentiate and what's going to help us become and in the future. And I thought I heard Deacon give examples of fungi, animals and plants and how they all have a different strategy for like differentiating the next cell and dealing with the selfishness of cells inherently. And I thought where you just left off like this.

A caterpillar becomes a butterfly and everybody was like, oh, oh my gosh. Like, okay, this must mean something. What does this mean?

Neuralink. I think the ability for us to tell ourselves to be able to like create external connections to silicon that can then like interact with our body. And within the 1 second, the 1 second Markov time blanket that we've been talking about is. Yeah, that's how I see like the human like.

And so they brought in that becoming concept and that evolved in that storyline.

Into a condition co evolving with technologies is that we can actually like through pharmaceutical interventions or through technology feedback loops, actually differentiate ourselves in way ourselves in ways that can start communicating with silicon technology.

Oh, there's a goo across all consciousness, some sort of consciousness goo that glues us all together. That's the observation context, y'. All. Let's shut it down. Pods over mic drop. Let's get out of here.

Yeah, but we can sort of evolve into them so that we can then have ways to communicate with each other so that we have the ability to synchronize ideas within this one second loop also. And then you can imagine the types of organizations we can form, the type of soft problems we can solve.

Yeah, dude, we. Everything would be solved.

Yeah, this new monstrosity of intelligence.

But yeah, we mean, I mean we'd be this global hive mind. So. Yeah, that's interesting. Maybe at that point the rest of the universe's intelligences then show themselves to us. You know that whole theory, right?

Yeah.

A civilization gets to a certain point of intelligence and then it's on Star.

Trek starting the Next Generation.

Yeah. Anyway, that, that is how we think about that with epistemic me. It is the glue that will hold things together. And we hope that one day. It came up recently, I think. Did it come up in the Bratton talk? I don't remember, but this came up computational. Yeah. Something we, we read or watched or consumed. But Anyway, we'll get there someday, fam. Yes, Future mind fam. I hope I'm still around and helping out the fam. You'll be holding collective intelligence together too. That'll be really cool. So we'll be able to go back to this pod and reference this. Okay, so I think, let's see. Do we want to do a quick build in public update? Because it's getting closer to my wind down and I'm trying to stick to someday.

Yeah. So speaking of the caterpillars to butterfly and that transition and what the glue is. Right. I have to bring up Michael levin's.

Work like 8:30 ish.

Cool. Yeah. Now if you need to wind down, we can do that another time or.

Okay, yeah, that could be helpful because I got to wrap some deliverable kind of stuff, like emails and comms and stuff. And then I gotta wrap my day. So maybe we just wrap it there. I think that was a really wonderful discussion. So that. That is a pretty good point. Is it hard there? I guess. Check. Are there other things you want to bring up before the wrap?

Funny, I pulled up a bunch of links. I think we got to most of them.

Hell yeah, that's great.

There's other stuff we could go deeper on, but no, it is like a nice surface of a lot of complex topics that I thought on improvisational memory. He based it on his observation in the lab of like, it fit together nicely from that book.

I really liked your suggestion to focus on the Markov blanket one because we do have so much to say. There's. And I think that would really resonate with the philosophical builder icp. And as well as the topics we talked about. We could like make clips from this and send them to people we admire to be like, hey, this is what we're thinking, Michael Levin, like, come on the pod. So I think it was very worthwhile to talk to these things. The Excalidraw you had of the observation.

Con, they trained a caterpillar to do something and then it metamorphosized into a butterfly text.

I don't know if I ever saw that or. Yeah, maybe. Maybe you did send it to me. And I never actually deep dived on it.

I think I didn't send that because I didn't want it to overwhelm anybody. And it retained the memory, but it used it in sort of like a different way that was like more butterfly specific. And so it led him to this like philosophical theory of he calls self improvising. Makes sense. But now that chance to Introduce it, of course. But I think one more thing we could add that would, is like the use of data for training agents on that sort of customer back layer that Sequoia was talking about.

Oh yeah, yeah.

I wanted to talk a bit about like the self model and how I think it creates like the perfect training data for, you know, an LLM wrapper application, a, an agent. And I think as entrepreneurs we're kind of in the space where we have AI as a tool, but the apps that we build and the experiences that we build really don't have any kind of like mode or meaning to them unless you can create a unique data set. And I think that's one of the, one of the punchlines of what I'm seeing with selves where like you think of the actual structure of memory is a communication to your future self. So there's this whole discussion that we had about the data set. We talked about the self model that can model beliefs over time, say on a second by second basis, but actually on an interaction by interaction basis and how valuable that could be as like a training tool. And I think a training tool for AI and we're specifically working on in the health space. And so if you think about what's the most valuable data that we could create, I think it gets to like making a mirror model of each of the people that interact with it. No, more like storing data on a register type of memory. Like our brains aren't holding, holding these memory in the way that our computers do. It's just sitting there for a future version of yourself. So that like having a second by second description of their nuanced beliefs, be able to, to like upload data, ask questions of the user, learn more and more about them. So we get like, like sort of this infinitely complex belief system structure that describes who they are. We can also look about it over at how it changes over time. And if you think about thousands of people, hundreds of thousands, millions of people, and we have a data footprint of them interacting with AI agents in that way, then we can use that data for training purposes to make more and more nuanced evaluations. And I think one example of this is like if you think about taking a sample of 300 people that all interact, say over like some exercise regimen, and let's just say it's like maybe an athlete level exercise regimen, we're really going to like push people in a particular direction during the VO2 max training or something like that. When your observation context kind of makes it relevant again, it Pulls up that, that information, it completes the reception of the communication to your future self. I'd say you can imagine this way if we have these secondhand slices of who we are as people. Michael Levin sees that as each one of them is just sending a message to its future self. And so you're both sending and receiving messages. If we can take a, if we can take slices throughout each of the development of the cycle, we could see the things that people struggle to learn, the things that people struggled with during that period. We can understand how to adapt the intervention to make it sort of like more feasible. And we can split, test things between different parts of the group. And we can, we can use all of that to train an agent on the outcomes that were most proficient. Not just for a person in a specific, like demographic group, like male, female, of an edge group, but you're talking about like something that's a hundred times more nuanced. Being able to place people together, be able to say this plan worked for people that were in this exact time sliced part of their teleological journey in their learning of this particular field. And these are, these are the plans, interventions that help people get along with that the best and just feature self. And the illusion of that is what we call memory. But and by the way, like, it doesn't, like, the memory doesn't have an ultimate like meaning or purpose. Like, so if a caterpillar teller discovers that like this leaf tastes good, right, that might mean to a butterfly that like that chemical signature is a good place to lay its eggs or something like that. It uses the information that it has in that context from its own experiential memory and it applies it through its. Imagine being able to do that with a team at precision and be able to run these kind of experiments and really just elevate people's learning. And these are like these little micro becomings, as we called them earlier, to push them along. I think that's the kind of data set that we as entrepreneurs can create as a moat. That's one of the reasons epistemic me is so meaningful right now, is that it creates this counterpart to the objective data sets, the analytic data sets that already exist in a domain. And it maps the experience of the humans that use them. And by doing that, you can create optimal user experiences and have people work with AI in ways that help them on their journey of becoming whoever they are.

Yeah, I love all of that and I totally agree and I had gotten to that thought too, which is just layer on another lens to Add to the discussion, Jonathan. The Jobs to be hon framework, which I've been recently getting intimately aware of, as you know and have showed me so obviously so jobs to be done. The framework by Bob Moesta, whereby the philosophy is that users, customers hire a product or service to do a job they need to be done. And the best way, the best products, the best products that convert and retain the their users and increase their clv is to show them their progress on their jobs to be done as fast as possible and get their jobs done in a great way. So that's just like best in class user experiences. What they talk about in that framework is how do you build the best product that solves the jobs to be done in the best way for the users? And the lens in which they look at that is from the question of what are the common causal structures that led this user to this precipice of not being able to do their job and then you gotta hire something to do that job and what's every step after that. So I love this what you say.

About how hey, intelligence mechanisms to like whatever its problem is at that time, which is also just like a fascinating thing to think about. And if you think about like how you reuse some of your own memories, like if you're like, you know, certainly.

Somebody uses epistemic me and maps out these causal structures, you know, thinking about normal product analytics, you're looking at user cohorts by clicks to figure out how do they get to this event, firing this action that I want them to do that converts or retains or grows, right? And so now we're saying if we take a step back, this is no longer about just optimizing for the right click. It is fundamentally as entrepreneurs, we're solving problems, we are building a solution that gets you your job.

Like I don't pull these, like I don't just have facts in my brain. I pull out these experiences that, you know, may be relevant to a certain situation in ways that I don't even understand. Why am I remembering this right now?

What is this to be done as a user in a target market for an underserved need and you want to see progress on it. And we do that in a better way than anybody else in that target market for that underserved need. Right? So we're saying with epistemic me is use us and we help you help your users transform to who they want to become. Right? We help them do their job to be done by helping you map out all the common causal structures that Got them there and how they evolved along their journey of doing that job or their jobs to be done. So I just wanted to add in that layer that I kind of started thinking about as you were saying it because it makes so much sense, right? Because the goal is to make money. Making money is an engine, it's a system. What are the levers? Generate leads, convert leads. What are the levers under that? Micro levers. It's all these things, right? Like how do we actually capture attention, convert that attention? How do we grow more users? How do we retain the users that we upsell them? All of that comes down to identifying these causal structures for the reason why they're even there in the first place. What do you think of that?

Yeah, it's a great metaphor, I think, talking about Bob Mwesta's version of jobs to be done and that customer journey and he's got like a really great picture that I was visualizing as you were talking to describe that.

Yeah, totally. I'll link the book in the show notes as well for the audience. One of the red hidden product and entrepreneur port. Only regret I had was I didn't read it at the beginning of my career. Fully regret. So if you're at the beginning of.

Your career, don't be like just awesome.

Read that book was after you were born. That's next Best time is today. Anything else to wrap up the discussion, Jonathan, before I end the pod?

No, that was a great like bringing it all down to earth in terms of like that. What is the bearing on the world? How am I going to interpret it by the way? And what mean am I going to make out of it? That's more the human condition than storing vocabulary from flashcards and having to read the evolution of a customer using, you know, an AI powered web service. That's pretty cool.

Going from multicellular agent networks with bioelectricity all the way back up to how do we make money as entrepreneurs and help other entrepreneurs make money?

That's how we do it.

That's how we do it. That's how we roll, y'. All. All right everybody, thanks a lot for listening to the ABCs for building the Future podcast. This has been such an awesome time. Thank you so much, Jonathan, my co host, for facilitating this with me and having these discussions and debates. As always, drop comments in YouTube and everywhere else you could find us. Hit us up. Don't our website. If you're a developer that wants to build the best in class personalized agent that helps your users get recommendations, they actually Want to be who they want to be and get their jobs done. Hit us up. That's what we do. All right, y' all take care. Have a great day. Bye. Bye.

Recapitulate it. Well, I can adjust.

Definitely true. I definitely. When I'm encountered with situations, any situation, my brain's architecture and visualization is not somebody takes out a note card. It definitely is more evocatively emotional. There's a sense of. Yeah, there's sensory connection there to a memory. As you bring this up, I almost think of it in two different analogies. Like, it's like a letter you write and throw into a bottle in the ocean. Someday it meets something. Another example is it's sort of like Kafka events. You don't know what's going to subscribe. It maybe doesn't even exist yet. I don't know.

You mean Kafka the framework or Kafka the event?

Event framework.

Okay.

Yeah, yeah, yeah. The whole pub sub thing. It's like you publish no idea who's going to start. No idea. Might not even exist yet. But like, you just do it and it's there and there's no. Like in. In that example. There's no purpose for why you put it there, but like it's there and maybe you'll have.

Yeah, that's exactly how Dean and I envisioned the memory structure. It's. It's called event sourcing as a pattern. And it just, you know, you keep. You keep version logs of a specific object in the memory and you emit signals that may be received or may not by their counterparts, but if they are, they mutate the object. And so event sourcing is a pattern. Seems like a really good way to kind of model this.

Makes a lot of sense. Yeah, makes a lot of sense. Sweet. Okay, so looking at the clock, just because it's getting down near my wind down period. So we'll have to do a quick build in public update. But this actually connects really well. So epistemic me models the self. And we've built out all this processing, logic and engineering so that we can empower evolving belief systems and all these concepts that we just kind of talked through. That secondhand Markov blanket or second by second Markov blanket concept, the free energy principle, all of those things have influenced our philosophy in our implementation as we described. So we really loved this chapter. And actually, Jonathan, you should tell people about how you literally have a blanket with the word Markov on it. Could you talk about. Oh, yeah, just I. That's like a quick little awesome Easter egg. Tell. Tell us how you got this.

Hold on, I'll sit. You want to see it?

Yeah, I want to see it. Yeah.

So I saw this. I just had to order it. And.

Yeah, with the nodes, can't see you. I'm under my blanket. You're outside my boundary, bro. Great.

Yeah, it was just like a nerdy little purchase. Like, I was like, who would make that? But if somebody made it, I'll buy it. And that's what happened.

Now I have it.

And it literally creates like a boundary underneath my bed between my wife's clothes and my computer stuff.

That's so great. I love that. Yeah, there's some teleology in the Markov blanket. Cool.

So one more example of this that I think is pretty illustrative. Yeah, this would be a pretty quick one. This is also coming from Levin's work. It's a paper he wrote with Daniel Dennett, recently deceased philosopher, who's just like, amazing sort of consciousness philosopher. And this is when they were trying to explain, I guess, like the how a biofilm. So they have different cells. These are each cells, the cells communicate to each other through ion channels. And they have this example of a bacterial biofilm, multicellular cooperative bioelectrical networks. So we have this. So I always look for analogies to like, think about how multicellular life formed. And in this example, we have the cells all communicating to each other through ion channels and they're getting a big picture awareness of the outside world that they can all share and be coherent together. So very just like concrete example of where you start to see a boundary between this group of cells in the world and sort of like a multicellular biofilm. And in my own work.

There'S a.

Specific phenomena, I think if you take the speed of the communication channel between each ion channel and the refresh rate on the world. So let's say this biofilm updates like one time every 100 times a second or something. And the speed of the communication is like one thousandth of a second. How many cells wide can this biofilm be before it's not coherent with itself? That's something that electrical engineers are sort of talking in terms of Nyquist sampling theorem. But I think that it can really inform this type of work. And I haven't seen it used anywhere yet. But just like, what is the. How many sort of like cells can we have within this, this boundary of the world before they're incoherent? What happens when we add a cell that it can no longer get the information fast?