0:00 Today I will speak about my recent work which is called tser theory of evolving self representations and the centralis of this theory is that living organism is a self-regulating system that maintain heostasis and uh uh self-representation is a is a is a property of self-regulation systems and I will show it uh a bit later and so while regulation is evolved moving pushed by evolution. self- representations are evolving accordingly also and uh they are in fact that is a subjective world uh that that we have uh this uh um psyche and consciousness and
0:47 intelligence and so on and this is the graded property um and and so I I'm in a constructive position I I believe that consciousness can be understood uh because it's a result of evolution it's not something mysterical that some guys say uh but so this the series is about uh is about how it is possible [snorts] uh and uh I will mention that I'm theoretical physicist by my training uh from Moscow State University uh and um my my main researches is in biohysics and and brain modeling and I'll pass that so um so a couple of theoretical foundations uh so uh we will uh refer to the system uh as a set of
1:34 elements which are located in some region of space and uh in a living system it's a set of molecules of course and uh heostasis is the ability of open system to maintain a stable and balanced uh internal environment. It's not a it's not a full equilibrium which is stable but it's a dynamical equilibrium which is always needs to be maintained and for that uh goal uh the system needs to have some uh regulation some regulators. So usually regulators define the control device that monitors the state of the control system and uh generates some control signals uh and and trying actually to to keep the system in some range of parameters and uh in ' 70s um
2:20 there was a nice theory which was proved by Kan and which is called good regulator theorem which says that every good regulator of the system must contain a model of that system. Uh it is a very simple and mathematical theorem. uh generally uh if if the system has a a lot of uh um states potential states that but the regulator wants to keep it in a in a in a smaller subset let's say G uh then this model of the regulator will be like reverse mapping of this G to the initial uh parameter uh space uh so it's a mathematical theorem which was proved And uh actually
3:07 it is it works for every regulator uh and also for some simplest ones uh like error based. So it's uh nice nice work and from that point we denote uh the self-representation as a complete set of all models of all regulators of the system. So if the system has a lot of regulation regulators and each regulator has its uh model then the self-representation is a full set of these models they can be independent this can be dependent uh that that does not matter this is a uh self-representation so selfresentation is a model of the system that system has for for regulation purposes and uh I I will I will note that there there is a
3:55 different thing which I which I Um for me it is very different. So I can make a model of the system. Uh let's say I look to the cell and make a mathematical model of how the cell is working and this will be a representation of the system. This is my model model of a external observer who looks at the system and tries to understand how it works. But uh this is not self-representation. Self-representation is when the model appears inside the system for the purposes that system for the better regulation of the system and system use this model for the better regulation. So this will be the self-representation and uh we'll talk how this will actually appear and another thing which I use in
4:42 in in my work is uh just some um ideas from neural networks uh very simple idea that uh neuron is processing information and in uh generally it sums the the inputs uh and and gives some some output and uh If we see to if we look to the actually every neural network finally at the end of the day it's like a big system of equations like arithmetical equations and we we will use this notation just to uh to to have some ideas about how this soft representations evolved and so uh and I will go through this evolution of self representations and uh we will
5:28 start with the emergence of life uh for uh let's say for for the better understanding of the self-representation is and uh first of all I will mention that uh there is a idea that many people think that uh uh life actually starts with the replicators but I think it's a wrong idea because replication is a uh is necessary but not u sufficient condition for for life you can have a millions of molecules of DNA which will replicate but there will be no life at all. So what is life and uh when life actually appears. Uh I think that uh generally life appears uh the very first
6:15 step to the life appears with a boundary. So uh evidence says that life appeared in in a sea water and uh what we know about every uh living organism that there is some boundary and uh uh the organism tries to keep something useful inside and there is some other things outside. And so this difference between what is inside and what is outside actually creates a thermodynamical problem because uh uh when uh the system has something uh that is not uh going outside this is this violates the second law of thermodynamics and uh actually the nature uh the thermodyamical nature tries to to to to break it tries to to
7:02 to uh destroy this membrane and it's it is known uh as osmotic pressure. It is very well known thing. Uh if you have a membrane and put something inside let's say uh around 30 30 mm of proteins DNA and so on which is usually in a cell which is not uh which which is not uh in a in a uh outside environment. uh uh so osmotic pressure will immediately try to break this uh membrane apart. Actually it will destroy it like this. It will uh because the water will go inside uh this membrane to to dissolve to diffuse all these uh uh [snorts]
7:48 non-equilibrium state to a fully equilibrium. And to deal with that uh each cell needs to to deal with with pressure and uh we know that the easiest way to do this is actually some pump pump uh some difference uh outside. So every living cell of uh of uh on the earth have these small uh pumps which actually pump out some yons uh actually sodium because it's a main uh main um major ion in in in in water in the sea water pumps out this um sodium just to make the concentrations of yons uh equal and this helps to maintain this osmotic
8:34 pressure. So that actually means that to to keep going this boundary with something useful uh in the water needs some regulation uh needs some uh let's say some pumping some some process to to to to deal with actually to to to to fight against this osmotic pressure. And in human cells we know that uh these pumps works uh even much more interesting. They they they pump a lot of sodium outside and they pump a lot of uh potassium inside uh to to create a bigger uh difference and and that's it have some purpose. But interesting thing that if we make some different differential equations of this uh process here is only a half of this
9:21 equations devoted to to to sodium. Uh if we if we solve this equations finally we will see that actually all this pumping system all this fighting with as oasmotic pressure is regulated by only one parameter uh intracellular uh sodium concentration. So that means that for for this for this boundary with the pumps a big system actually appeared only one parameter very simple model of the system of the system uh uh concentration of sodium which acts as a model of the system for the purposes of osmotic regulation. So we see that in this in this case uh a very simple model of the system appears for the regulation
10:08 purposes which actually acts to help the system to maintain hemoasis uh in terms of osmotic pressure at least and uh I call this um this type of uh very simple u regulation and very simple model uh like a self-representation of level zero So I mean that uh I understand the big gap between the boundary with the pumps which pump out some yons and between the unicellular organism which is much more complicated and have a lot of uh metabolism and replicating and so on. uh for for for my purposes for tips purposes this
10:53 is does not important this is not important because uh generally uh the point is that uh to to make a system which which which will be alive uh we need to have a system which maintains heistasis it always requires some energy and heostasis is needed to fight against some environmental um uh uh changes and and so on. And so I think that this is the way how the very first uh unicellular organisms um appeared on the earth. It was around 4 billion years ago. uh of course um uh cell has a very um complicated and very well reggulated uh
11:39 and coordinated regulation but I will not go further about unisellular organisms because there is no uh there's not not the point which I will talk about and we will switch we will go to the multisellular organisms which appear uh around 600 millions year years ago u and uh the main idea A about multisellularity is uh when you have many many cells and they are doing different things different tasks you need to coordinate them in a way uh and uh multisellular organism can have a very different kind of uh cells some sensors motors uh are different but finally they need to exchange
12:24 information or if you need to coordinate them you need to coordinate them in a in a say in a unified space of signal So uh and and we we we see that uh with this uh multisellular organisms um the first uh nervous system appeared uh it was a very simple diffused uh nervous system without any nodes without any um uh so just just like network but it was built on a on the first cells that were uh specializ ized on on some signal transmission. It was not neurons as you know it was like a fiber cells and uh but I will call this neurons uh because
13:10 of for my purposes there is no uh there is no big difference I mean uh no big difference and so this uh this nervous system first of all was needed to coordinate some uh different cells and some uh some of um some of different parts of the of of the uh of the body of this multisellular body and uh of course this coordinated articulation provides heostasis in a wider range and the the idea about self-representation in this uh in this level is that that um a network of uh these neurons uh appear
13:58 and uh the signals appear in in them and you can imagine in millions of different organisms which happen to to to appear uh in different variations. But in some of them uh uh this this um the signals the signaling in this network happened to produce something like uh stable uh dynamic temporal patterns that just um that just exist in this in this network and uh act back to each neuron. I mean uh this this um this self-impersonation is uh this uh this this temporal pattern are not um present uh are not produced
14:44 by a single neuron they produced by by by all the system but when they produced they are let's say pushing back to uh each individual neuron and this is uh uh the the small video from one of the works of Yugini Jakovich uh this is a spike in neurons uh the the network with spiking neurons and uh this is just to to give an idea how it is uh working. I I mean that in this network some temporal dynamical patterns appear and you see that they are like a uh like a recurrent patterns that are that live in this system. I will say and uh and my point is that uh I think that um among millions of systems uh with a
15:32 millions of variations of a simple nervous system, some of them were just more efficient because uh uh this type of dynamical patterns that happened to to to appear in this nervous system was more efficient to coordinate this regulation uh in in in this in in in this organism and that's why just this uh this type of creatures just happen to to to to live longer but the problem with that kind of uh diffusenet system is that every neuron is equal I mean uh uh if if you write an
16:17 equation uh it will be like uh X1 plus X2 X3 they all are equal because we created this uh not we evolution created this unified space but uh of course different signals are have different importance uh let's say signal of u body which is destroying it is much more important than I don't know smell of something and uh of course it is much more uh efficient in in terms of mathematics to sum this like uh with with some co coefficients. If you can uh sum some one signal with coefficient uh which is bigger another one with coefficient which is smaller.
17:04 But to do this you need to differentiate between between these signals. Uh when when I write this equation I know exactly what is x1 what is x2. But when these signals happen to appear in the system uh and these signals are actually signals from neurons they are equal. I mean you cannot in a real life you can take a wires and you can paint these wires but uh you cannot paint the neurons you can make this colors and this is uh but we need to to differentiate them and in fact uh this differentiation is quite simple is a is a special one when I read when I write such an equation to the computer uh let's say in a python language the compiler actually translate it to a
17:51 different physical locations. So x1 is a physically is a um one memory place and x2 is a number in another memory place. This actually special special differentiation and and and so I think that this is this need for differentiation was a point uh to to to to to to appear uh such thing as a gang. So Gangli is a nodes as you know is a nodes for processing signals but I think that uh the main purpose of ganglia was uh this special signal differentiation. So it was like a micro scheme in a way uh where have a lot of inputs but these inputs come to uh concrete places and
18:37 when you differentiated these signals you can uh do something with them and actually we know this uh as a labelled line principle. uh we know this exactly about human. This is was a work of Penfield in the 60s in in last century. Uh when we uh we know this picture we show that different signals from different sensors and models of of our body come just to different parts of of a brain and that is how the brain recognize them. We know this exactly about the our neo cortex. But my idea is that I think that this actually happened 600 millions years ago with Ganglia. And Ganglia was the was the first thing uh
19:23 that actually was invented by evolution to make this special uh differentiation and make this line on on that simple omazoan organisms. And this was of course for choice of action because they need to to to to to make some choice of action. And uh another thing at that point we see another interesting thing that uh when when I have a signal which is uh which is going into this u self-representation but it has let's say a mark uh which is a a source mark uh from where it is signal it is actually it is a sensation uh it is a a colored let's say colored
20:09 signal uh and uh this is a signal with a preserving information about this data source. And uh another thing is uh emotion. Uh emotion acts like a like a sum of these sensations uh in a way some in a way not not directly sum but like like a uh complete state of this self-representation at at one point of of time. And so uh my idea is that at that point at that uh second level of self-representation with this differentiated signals actually uh we can see a proa psyche. It is a not a a full um
20:55 riches psyche as we have as a human as as we uh used to to to to to say about this word. In a way, these simple creatures had some sensations, some differentiated uh feelings and had some emotions which are like a complete and uh I I can say that C elegance is a as an example actually it it has uh the neuron very hot and the neuron is very cold which uh it's like a real emotion. I know that in in current biology it is called like a behavioral states. Nobody says that it's emotions or um or uh psyche but uh my point is actually
21:42 there is no big difference between these things. uh and uh so so so the idea behind that is that this uh signal differentiation and this protocycle appeared quite early uh around 600 millions years ago and I think that maybe this was a big uh one of the reasons to abalone explosion at that period uh which gives a a birth to a lot of species and the next explosion was a cumbrian explosion uh it was 540 millions year ago when first botarians appeared and also uh it is thought that this is the time of emergence of a predator prey behavior model. Uh but what is a predator prey?
22:31 Someone is going after uh if if I try to catch someone and uh uh he tries to to run from me. This means that uh I need to localalize myself and localize that creature in a space in a way. And so I think that this is a point where uh we can speak about the appear of a subject. This is I use word subject. This is a philosophical term. It's a big term of course but uh actually I divide it in several categories. And at this level I think that uh at that point uh we can talk about appearance of a subject here. Subject here is like localization of oneself in a in in in a space. It's not
23:19 a reflection. Uh it's just it's just a feeling that uh here I am and at a big world outside. And for that purpose actually you need only one thing. You need stereo senses because each stereo sense uh a couple of ears or a couple of eyes gives you an ability to localize yourself u uh uh comparing to to some other external objects and uh this ability actually appeared uh in in a billitarians around that way. So I think that this is the very moment uh not a moment it was million years but uh at this stage uh
24:06 this self-loization appeared and uh it is not a reflection of self not understanding it's not consciousness it's just an it's just a feeling that uh that the I'm here and and the world around and in terms of and and I know that you uh work with a with with a uh in many of your works you use a term agency which I I didn't actually used in in this in in this work but uh we can say that at this point some some lower level of agency appears uh well because agency in in from the point of tips uh agency is like a graded property which is not binary uh there
24:53 was no agency And this just it just happened to be like layers. So this is just just a very first layer of this agency which appeared and we know this was a very major breakthrough because 99 of% of animals today is a billarian uh and I use this uh Minkovski space um permalism to to show some interesting idea about this uh self- representation. So first at first stages uh self-representation was only about the system itself. It is like a point in in a space only small point. But when we talk about localization oneself in space actually it's not uh
25:40 self-representation begin uh uh begin um include not only uh uh self uh in but also some some area around some some area in space uh obviously because you need to localize yourself among other things you need also some simple at least some simple representation of this. This is not objects. This is done more simple things but the objects will appear later but but generally this that means that self representation begins to represent not only the system itself but some some area around and uh next level which appears later
26:27 and uh it seems it's also uh around billetarians but I'm I'm I'm still not sure about this but but but let's The current research says that uh that yes this is like a simplest forms of learning. uh and I mean that learning appeared much uh ahead of intelligence and intellect in a very simple way like conditioned reflexes uh and uh but uh it is not that central point I will keep it quite quite fast and so uh and we'll go further and evolution moves further and speeds increase and uh and actually when uh
27:12 when you When we say about predator prey model, someone is catching someone else and I'm running from a predator. Uh the information needs to be prioritized highly because when I run I have something ahead and I have something on the sides and actually that means different prioritization and uh I think that this was a major next major breakthrough was attention. Uh so it was next level like a dynamical prioritization of sensation depending on situation. In terms of uh this kind of equations we can uh see this like uh this coefficients just can can change its value. Uh in one situation one
27:58 coefficient is is is bigger in another situation one ceicial is bigger. Uh and this uh option seems to be uh appeared in the vertebbrates. um around 500 millions year ago because uh what we know about vertebraes that all of them has uh um three uh parts of the of the brain when the brain develops uh hing brain, midbrain and forbrain and uh what we are interested now in in at this stage is a midbrain because midbrain has this motor nerves and actually midbrain controls the eye movement. starting from fish and ending to mammals. uh uh our uh our um accolometer
28:45 nerves also controlled by by midbrain generally of of course we have some tuning by by neocortex but generally it is a the midbrain and uh so uh it it was an attention and it was also a big uh major breakthrough and uh so so and and but we will move further uh uh and uh So I think that uh the next paradox appears that um uh the evolution pressures uh the speeds increase you have to make decisions much more faster. But uh from another point of view the perception becomes richer. Uh the vision improves attention is
29:32 there's a much more um resolution of the of the picture. So and that means that there is actually a paradox you have a lot more information to to uh work on uh but you make uh you need to make a faster decisions. So this need needs to be solved in a way and uh I think that this is uh the point uh why uh the object differentiation appeared uh uh and uh actually so uh what what what our uh what this level do is uh simplifying the world. So we have a very complic complicated raw picture and and a lot of
30:17 raw pixels which is coming uh to uh let's say to to the vision to to to the to the hearing to any any feelings but finally uh they are compressed they are recognized and compressed into very simple objects which you can um proceed very very quickly with them. And uh what what we know is that uh from the from the uh current research that this is done by a sensory motor cortex. Uh and uh seems that first this cortex appeared in the earlier amphibian around 370 millions year ago. It was called archipalium. It was only three layers network uh not six as as we have.
31:06 But uh generally what we know from convolutional neural networks that this three layers is uh is is is okay to uh to distinguish between different smells between different objects and so on. And so this three layers was actually a first um convolutional network if you could say. So um which um uh differentiated the different um different objects. And another thing which uh appeared at this moment is memory because uh differentiated objects needs to be memorized from the one hand. On the other hand, you cannot memory
31:52 a a millions of row pixels. You can memory some compressed some some objects. And what we know exactly now from the artificial neural networks that differentiation of objects and memory about these objects actually form in the same brain area. This convolutional neural networks they they they have both pieces at the same place and uh at the sensory motor cortex and so the memory appears and when I when I talk about memory I don't mean um long-term memory about the lifetime things. I I think that at that point there was a very shortterm uh memory just maybe some short trajectory of a of an agent maybe some last uh seconds or something like
32:39 that just just short pieces but this already creates something that I call subject now um let's do another microphone uh uh just to confirm uh do you hear me well Okay. >> And so sub. >> Okay. And so subject now uh this is this is something on the previous stage it was a subject here localization one in in a space and subject now means localization one in in a time. So uh the agent appears uh that is not only a point let's say at space but he has a some trajectory. He has a and uh he he has a memory of at least
33:27 some small trajectory but it it's already appears uh um this trajectory and in terms of u when we uh make this like a picture on the Minkovski light cone it means that self-representation at this point includes some past events. So uh this subject now uh begin to include some some past events. Um and I think that this is actually a point when we can talk about uh consciousness appearance of a consciousness in a way that we humans uh uh understand in a in a in a routine life in a in a in everyday life. Now this is not
34:14 self-consciousness. I mean uh these creatures did not reflect the on their own creation on their own existence. Uh there there is not uh self-consciousness that point point but uh this agent has a complete phenomenal world. It has a sensations and emotions like a u some of these sensations. It has an attention which is uh and and some and some objects in in in its attention and it localize itself in space and uh and has at least localiz itself in time. So I think that uh we can say yeah and in terms of uh when I say no no no self-consciousness just to to to give an
34:59 example uh when you when you do like a mountain skiing let's say and go from the from the mountain at a high speed you are very conscious you see every every every ski every every person every tree you see how road is going you're very conscious you see a lot of things and you notice a lot of things but there There's no you at that moment. There is no self. I mean you don't have a time uh or to to have this self to reflect that you exist. The world exists to you but you don't exist. I mean uh this is a the consciousness without self-consciousness. And um uh I'm not sure uh that we have I'll pass this qualia for the moment but we
35:46 can return back. So I have a nice uh explanation of aqualia at this moment. Um and now we see that uh this evolution of a subject uh there was a subject here then subject now and self- representation includes some past events and it is obvious now that next things will be self-representation will uh will include some future things. It is a natural evolution of the self-representation. And by the way, I will mention that if um if I say uh when I say that uh self-representation includes the area around and some past events and some future events uh uh you can say that it is uh not in line with
36:32 my first definition that self-representation is a complete set of models of regulation for homeostasis. uh but uh actually uh I can have a fantastic regulation of inner heistasis but if I cannot see the predator which comes and can eat me then uh my heistthesis will be finished in in a moment. So I mean it is very natural um extension of self-representation of of model of self to include not only the model of the body but the model of some area around and what is what was going and what will go what will be next. So uh and we come to the intelligence. So for for for me intelligence is like a predicting some
37:18 future uh that is that allowing you to prepare for some potential actions for to to to improve homeostasis and we know exactly now how future can be predicted. This is a cause and effect relations and uh we observe them. If uh there is event A and event B and one follows another then we understand that uh it is we can predict that B will be after A and this prediction is like a forecasting and my point that uh here is that if individual neurons or maybe okay group of neurons represent some events or objects and connection between neurons uh uh obviously represents some causal
38:03 relations then um uh Hebian rule which is fire together wire together means that it is just like memorizing of this cause and effect relations and it it it makes this uh happen. So this narrow connections just a way to uh to memorize this um posatric relations. And uh I use this special term like a behavioral intelligence because my point is that intelligence appeared much uh earlier than language. Many people connect now in these days especially with with the large language models. Uh many people think that int
38:48 intelligence is is about language but but not actually at that point um we will see that uh behavioral intelligence is is about some predicting of a future which allows you to prepare for possible actions but uh without any language and the first step of that was uh reinforcement learning try to to to move faster reinforcement learning which was uh learning through trial and error on uh your own experience And it was uh it was for early reptiles around 300 millions year ago because they uh we know that uh the emergence of basal ganglia and stratum uh which control this uh type of learning emerged uh at
39:37 at that time. Actually this is a um but uh what is more interesting is what is called hierarchical learning uh which uh next level uh it's a eight level in in in in in terms of objective cell when you can learn not on your own experience but you can learn on simulations when you can build the model the world and potential um potential uh [snorts] trajectories of what what can go on and and and and uh choose uh the one which is more uh uh interesting or promising just in the simulation and for that kind of uh simulation you need uh some some some some model some hierarchical model of intellect. there is no um there is no
40:23 common uh understanding of how it works but I personally I I I I uh support the hierarchical predictive model of which is proposed by Carl Friston uh and uh this is uh higher levels make this model uh of the world what will be next top levels no top levels lower levels report what is uh actually occurs and uh there's a two ways uh uh coding I think you know about this uh but uh what I will uh mention that there actually there are two ways of doing uh two ways of updating the situation uh in this hierarchical predictive coding model. So if reality diverge from what I expect I
41:10 expect the world is x and I see that it is uh so I have actually two options first I can update my model of world I say okay the world is not okay it's I made a a note but other thing uh other way I can say okay world is but I want it to be x so what I can do to make it X and I can act to the world actively to to change it and this is a active influence concept uh by by Kristen in a very simple words but uh my point is that in terms of tips it's also uh it is just a virtual regulator so goal is like a
41:55 model of a system and realization plan is a like a regulation mechanism and uh that means that at this level system uh self-representation just uh have an ability to create this virtual regulation uh and and there is also a Russian theory about this uh which is called theory of functional systems made by Petrain but uh it was a very old theory which is uh and uh we we know it quite well uh the the kind of uh species that have this kind of int intelligence because you you need a a hierarchical neural network uh let's say deep neural network uh and we
42:43 know about this kind of networks from artificial from machine learning then when you have this network uh it is much longer to be trained and um and when it is not trained it is useful. So it is not useful. So the the agent with un untrained network is uh is is is in a dangerous position. So he needs a very safe um very safe period for training. And that that points to mammals uh that uh this kind of uh mammals feed their feed their kids with the milk for a long time giving them a long period of time to let's say to to learn and that means that this heroical intelligence already uh is presented in
43:32 mammals. So this behavioral type of intelligence and uh the rest uh three uh parts of the of the uh self- representation are connected not with the actually the two slides not with the filiginesis but the ontoises not with the uh physiological changes but with some uh changes in a in a development of individ through through his life. And so uh this this is evolution in society. You can survive solitarily but uh it's much more efficient to survive uh in in a groups. So uh I think that the next level which appeared which I denote like uh omega 9
44:18 is so when you differentiate the objects uh you can differentiate that there is a objects any other objects and there are some species of your of your kind and generally there are two different strategies uh with the species of your kind and the and all other species and uh seems that this kind of differentiation appeared in in in evolution and uh a lot of a lot of species uh u demonstrate this group behavior and I call it mirroring uh maybe because of mirror neurons but uh we know that in in in primats uh there is a prefrontal cortex that actually doing this social things but uh in in in
45:04 a simpler organism is maybe maybe there's mirror mirrors uh but at that point we come to um self-consciousness uh from my understanding and from perspective and from actually it's not my understanding it's it's a lot of lot of philosoph psychologist thinks that self-consciousness is not something that built in into human it evolves through social contexts when someone addresses I begin to pay attention to myself and this reveals my my existence to me and uh so I think that uh this
45:50 self-consciousness appeared uh appears as a as a result of this social contacts and we know that young that the kid uh before before his age of three he does not use the word I he does not seems that he he he does not have this concept of of uh self-consciousness. Uh and uh and the final layer which is uh you know like omega 11 is abstract intellect which is uh uh which is not language uh which not only language language of course boosts this uh catalyzes this process. It is uh like um it erases this object differentiation because when you
46:35 communicate to other subjects you need to this is a boost for creating much more abstract concepts of u categories of objects and events. But uh generally um this um abstract uh language uh and abstract intellect which was built on the language. This symbol symbol separation in a symbolic space appeared very late in the evolution just around 100 or maybe 200 thousands years ago. There's a lot of versions of that you know no one knows exactly how it is. My point is that language is a only a top off of of of this of of this big uh
47:22 of this big uh cake with a with with different self representations and interesting thing about this abstract intellect that there it's uh let's say in in in a way it's a self-consistent level. So this is a symbols in a symbol space and uh of course it is based on uh differentiated objects which were differentiated on a level six and these objects are related and with a cause effect relations which are were um recognized and me memorized on a level seven and eight. But uh but but generally uh this this level is self-consistent. So uh it can be it can
48:10 exist without uh anything else. I mean uh what we now see about LLMs uh is this is this symbols in symbolic space which are like um uh like u they are separated from from from a real world which is uh which is other parts. uh I think that uh I I have some portion about some general uh some some theories uh I will not talk about this what what I'll mention finally is that um I know from your works that uh you uh uh also think about this a agency as a graded property is a not discrete levels
48:59 uh I'm sure that in evolution uh these self-representations als also uh evolved uh not discreetly like like very smoothly. there was a lot of variance and there was maybe there was a lot of overlappingings but uh in terms of understanding in purposes of understanding and uh words we use and terms we use uh I use this let's say discrete levels to to to explain this um here are of self- representations uh but but of course that can be very smoothly Uh so uh and uh there is also a big part
49:46 about uh some um some uh AI agents but uh uh but uh I because we were out of time I will make a conclusion. So the idea of ter is that uh uh introducing this concept of self-representation uh we try to to to to to decomposition to make a decomposition of a uh of of a big word like consciousness or subject or intelligence to a smaller um pieces which are more let's say functional and uh more understandable and uh and and this this is a way to talk about let's
50:33 say consciousness uh in animals in a different way just not just not saying is conscious do they have consciousness though they do they not have consciousness but uh there are some levels are present some levels are not present and uh this is the way I think this really works so at that at that moment I I really will stop and you know that the work is published in uh you know frontiers in human neuroscience recently. So uh I made it very briefly and very fast. Uh sorry for that but uh the big paper is this is published. So, thank you very much for your attention and I I hope to to be faster,
