Simulating the World & Remodeling Philosophy (Session 7)

Hello everyone and welcome to the now seventh session already of simulating the world and remodeling philosophy. I'm going to pass the mic to the course instructor now, Reza Negrestani. Thank you very much, Theo. Thank you everyone and hello. Okay, we are a little bit behind. I mean, In case if you have any kind of question from last session, please pose it. Otherwise, I will start talking a little bit about those variables of similarity that I promised. And then I move towards a discussion about fictionalism and false models.

And that would lead eventually to a broader discussion about toy models, generally understood. Anything, anyone, questions? Okay then. So, if you remember, not the previous session, the session before it, I talked about that Geyer and cataracts concept of similarity seems to be better than, it kind of put us

on the right track. and similarity for them seems to be the relation that holds between models and the world because it comes in degrees, can be used to compare idealized models to targets, can relate qualitative features of models to targets and so forth. And similarity, again, however, you know, obviously, is a context dependent concept. You can't just think about there is a canonical concept of similarity when we are talking

about models. Of course, not only it is context sensitive, but also it is in a sense a pragmatic concept as well to the extent that it always pertains to a degree to the scientists or modelers intention. For example, what kinds of parameters should I exclude when I'm doing the modeling? What kinds of properties should I cover and not others? All of these questions. Now these questions are in fact important for how similarities to be determined with

regard to the correspondence between model another model or a physical phenomenon. Yet with regard to Cat Wright and Geier's concept of similarity it's just that it doesn't really give us too many details as what similarity actually is, even in the sense of context sensitivity, pragmatism, modeler's pragmatic intentions, so on and so forth. So I'm going to talk about a few of these details as what similarity might be.

The first account that I would like to talk about is Amos Tversky. So around 1970s, Tversky's, which is basically his idea of similarity as constant account, developed a theoretic account of similarity, with which he attempted to capture the everyday judgment of similarity and dissimilarity made by his experimental subjects. At that time, the most sophisticated theory of similarity judgment had been developed by people like Roger Shepard, drawing on some of Quine's ideas.

In Shepard's, for example, geometrical account of similarity, objects are assigned to a location in a multidimensional space on the basis of values assigned to their features. Sorry, Renfra. Yes. But it's really hard to see your face because of the light behind you. Oh, the light, yes. Yes, yes, yes. Thanks a great lot. How about this? Yes, yes. Now I have a problem reading the text. Don't worry. So basically, I was saying that Tversky actually and Shepard, Shepard actually, developed a set theoretic account of similarity. When we are talking about set theoretic accounts of similarity, we are talking that, for example,

you are covering a set of properties, okay, with regard to a physical phenomenon. Like, imagine we are talking in a collision mechanics. So, momentum, mass function, energy, points, what else, you know, velocity, these all are are spatiotemporally defined, right? And the thing is that you can decompose this spatiotemporal account of, you know, such properties of a particle in set theoretic terms. For example, assigning real valued numbers to a space and other real valued numbers to, for example,

the time with regard to those properties. Now this allows you to construct very detailed satiritic accounts and if you remember those of you who took my last course we talked about this with regard to the work of Joseph S. Snead. That is exactly what Joseph Sny does. It's a set theoretic account of the logical structure of theories where real valued numbers are assigned to properties, spatial temporal properties. So this is essentially again not really an

innovation it is something counts a theoretic accounts of properties and of course it has weaknesses and points of the strength now as I mentioned In Shepard's accounts of similarity, objects are assigned to a location in a multi-dimensional space on the basis of values assigned to their features. Similarity then is just the distance between representing objects in this space.

For example, colors might be represented as coordinates in a three-dimensional space. corresponding to their lightness, hue, and saturation. Right? Kind of like any kind of Photoshop color palettes you get. So two colors could then be compared to each other by measuring the distance between them in terms of these real number values. The closer two objects are in this feature space, the more similar they are one to another. So this is a kind of a theoretic account of similarity. Now, Taberski thought that this was not a fully general account of similarity

for a number of reasons. For one thing, he actually believed that not all properties irrelevant to similarity judgment can be mapped onto a dimension of a property space. Some features are in fact qualitative. and cannot be quantitatively captured. He also believed that not all similarity judgments were symmetric, such that you can simply measure the distance between two colors in terms of their lightness, hue, and saturation. for example you know when we are talking about

similarity between China and North Korea North Korea can be said to be similar to China but China is not similar to North Korea okay Okay, so Tversky wanted the account that was more flexible and general than the geometric account, but could also generate the results of the geometric account when they applied. To a first approximation, Tversky's constant account of similarity says that similarity of objects A and B depends on the features they share and the features that they do not share.

So, I'm not going to, you know, details with regard to his equation for this account of similarity, which is actually quite, you know, it's a logical formulation. Essentially, you know, his account developed in a following way. We begin with some set of features, delta, called the feature set. These can be quantitative and qualitative predicates. Like for example, color, intensity of light, intensity of pain, whatever you might call it.

and might include elements such as is read or is left, y is left of x, you know, kind of like a, basically these kinds of primitive predicates to which the majority of our features or attributes can be decomposed. So there can be quantitative and qualitative predicates and might include elements such as is red or is to the left of x. We land on heads rather than tails with probability of 0.5

and just about anything else. Now for two objects A and B, we will define A as a set of features in delta, a feature set, possessed by a small a, and B, the set of features in delta, possessed by a small b. Now this allows us to define a function for any set of features delta. It allows us to define a function from mathematical standpoint which is called a waiting function. F of point. F open parenthesis a point.

Paranthesis closed. Now the term weights, for example, alpha, beta, gamma being the terms of the weight function, allows us to come up with this equation which gives us a similarity, a score that can be used in comparative judgments of similarity. So, essentially the whole point is that this kind of set theoretic approach attributed to both quantitative and qualitative features allows us to define first a weighting function.

And this weighting in a probabilistic sense, okay, or a statistical sense. And the terms of this weight allows us to arrive at a new equations where we can in fact define similarity in terms of scores of correspondence between the features of the models and the features of the target phenomenon. The scores in a probabilistic sense. Now this equation says that the similarity of A to B is a function of the features they share penalized by the features that they do not share.

So if they have too many features that they don't share, the probability of score goes beyond a certain, for example, agreed upon by the modeler score. And hence, they can not be called as establishing a similarity. OK? So this is a kind of a very rudimentary introduction to the first account of similarity, Tversky's account, which is essentially approximation, similarity in terms of approximation, and probability of score. Where the features that they don't share always reduce the score,

the probability of score of comparison and similarity, the weighting function. Any question here? Justin, Meredith, Adam, Jean-Pierre, Joven, Marie, anyone, Mikey. Is it clear or am I just pushing the stuff? It's too soon, I think. Just too soon for questions.

I'm trying to grasp it a little bit more. Okay. It's not that it's crystal clear, but it's too soon. Yes, okay, okay, good. Sorry. Okay. I need another cigarette. We are getting to the nitty gritty of stuff. Sorry. So, ultimately in the sense of Tversky's concept of similarity, a model is similar to its target

or to a mathematical representation of its target when it shares certain highly valued features and does not have many highly valued missing or penalizing the score and when the target does not have many significant features that the model lacks, relevant features are identified in a natural or formal language. of what kind of basically again you remember the Stegmuller idea those of you take my took my class last time when we are talking about natural and formal

language here essentially we are referring to not the ordinary language and formal language per se, we are referring to a set theoretic account couched in terms of natural predicates, predicates of a natural language, or a set theoretic account applied to the predicates fundamentally couched in terms of formal language. So when I am saying natural and formal language, don't think that we are talking about ordinary natural language at this point. We are not. We are simply talking about a naive, basically, axiomatization system in which you apply said theory to the predicates of the natural language.

or we are talking about applying said theory to the predicates of a purely formal system. Any kind of artificial language. So, yes. So relevant features are distinguished or identified and determined in a natural formal language in the sense that I mentioned. and their importance is weighted relative to the goals of the scientific community.

In order to transform this basic idea into an account of the model world relation, we need to consider in more detail where f, the weighting function, and delta, the feature set or set of features and the waiting coefficients come from. Of course we also need to look into how the equation of the waiting function has been put together. And so this was Tversky's account. Another thing that we can talk about is similarity from the perspective of attributes and mechanisms.

You know, in scientific paradigm, in the scientific paradigm of inquiry, it is typical usually to distinguish the properties and patterns of a system from the underlying mechanisms that generate these properties. So for those of you who again didn't take, weren't present in my previous class, so you You see, in contemporary philosophy of science, the idea of explanation, in a sense, at least in the mechanistic term, which is quite a very dominant paradigm right now, the idea

of explanation is defined in terms of mechanisms. So how does this actually go? So you see, of course mechanisms can also be laid out in terms of statistics, statistical terms. So what are mechanisms? Mechanisms what you might call to be these components, these informational packages, information packages which have some relation among themselves at a very specific level and then these at that for example a lower level your for

example from a reductionist aspect the relation that is obtained or derived among these mechanisms explains a phenomenon at a higher level. You can think about genetics as a good example. Mechanisms among genetic properties. Now of course so mechanism is essentially tied to the idea of explanation scientific explanation to explain a phenomenon at a different level. Of course to have that we have we need to have at least three things one.

Three things or four things, sorry. Okay, four things, four things. So think about a level A, a lower level, and a level B, a higher level. For example, any kind of phenomenon that at the level of B, the higher level you see, is observable phenomenon. Okay, you can describe it with naked eye. describe it. One of the tasks of science is to explain why is that we are actually seeing what we are seeing at this level.

okay so you have to go to a different level a different scale an explanatory scale and a start to describe certain mechanisms or information packages whose relations with one another might from, of course, that is completely an open question, from an ascetic point of view or a mechanistic point of view, explain what we have already observed and described on a higher level.

Exactly, like think of like that old example that I have been making with regard to Boltzmann gas theory. So at the level of B, we see that whenever we open a sealed bottle filled with colored gas, the colored gas, as soon as we open the seal, escape from the bottle and never comes back into it. right an irreversible process this irreversible process however is only a property of observable phenomena okay we are essentially describing it we are not explaining the phenomenon what we want

to explain is that what kind of mechanisms are responsible for generating what this phenomenon that we have thus and so observed. Okay? So we go to a lower level. And this lower level, for example, certain kinds of mechanisms involved in the collision of particles of a system of gas. Okay? Now, first, so first I mentioned four. First, the describing of what we are going to actually explain. This is our explicandum, that which is yet to be explained.

So we want to go to the realm of explainants, those which explain, the mechanisms which explain the explicandum. So the second phase that we should do is to describe these mechanisms. Of course, the description of such mechanisms again requires theorization and modeling, so on and so forth. So we describe these mechanisms. Then the second stage, to initiate a process of explanation, to say that given such and such mechanisms at the lower level,

at the level of the particle, whatever happens within a statistical threshold, the so-called nomological expectancy, will yield a result that we had seen at the upper level. Cats having fights here. gas leaving the chamber, the bottle. So this is the third stage. But of course, to make this explanation happen, not only we need the descriptions, two fundamentally different descriptions on two different levels,

not only we need the interrelations between these mechanisms at the lower level, but also we need to somehow move from one level back again to the lower level back again to the upper level to connect the mechanism responsible for the irreversible phenomenon that we have observed to level. So far we have just been working at interval level descriptions and interval singling out of mechanisms.

But of course the task of explanation ultimately culminates in an interlevel move, moving from a lower level, the level of mechanisms, to a level of the kind of phenomenon that we have observed. So these are the four stages. So this is coming back. I just wanted to talk about this for those of you who are not familiar with why mechanisms are important. And a very good book that I really, really suggest if you are to actually a good book. One is William Bechtel called complexity William Bechtel let me just William Bechtel and the other one

is Carl Craver and Carl Craver let me this is you can you can in fact start with this one which is quite accessible and very very lucid. Sorry. Sorry, I'm just trying to...

So this is the paradigm of mechanistic explanation in science. It is what you might call to be a refinement of a bunch of different paradigms that have come earlier in philosophy or science like functionalism, statistical analysis, so on and so forth. It incorporates a good majority of them. Any question here?

The Bechtel book is called Discovering Complexities, is that right? yes that is that is and there is another one which is fantastic a really great book one second it's called mental mechanisms again by William Bechtel philosophical perspectives and cognitive neuroscience mental mechanisms the book opens earlier on and says that you see the mechanistic view many people in continental philosophy when you say mechanisms they have kind of a fit they

think that oh you you're trying to compare us to you know mannequins and automata and William Bechtel actually tries to what he tries to do is based on what he says he says that you know there is no discrepancy between having for example rational autonomy and ideas of rational freedom and being basically comprised and determined by mechanisms and he says that we are humans with dignity and freedom not because we are not made of mechanisms but we are precisely because we are made or consist of the right kinds of mechanisms

this is in the second book you were talking about you told you may get the book actual yes that is that's the mental mechanisms yes that is okay yes that is introduction to the mental mechanism yeah Okay, just clarifying that. Meredith, would you be able to elaborate? I have a problem now that the light is off. I can't see anything. Okay, so, you know, this, so, so I feel like there are two different ways,

you know, so like talking about mechanisms itself, is like is tricky right because what is the mechanism you know so so they are usually defined by invariances of course these invariances can also be basically laid out in various ways that you can, for example, single out an invariance. Predictive invariance, causal invariance, purely a statistical invariance. There are so many ways that you can in fact detect an invariance. So mechanisms are in that sense basically what you might call to

be invariants a kind of you know for example again coming back to an old example in James Woodward making things happen you see a very simple mechanism would be the shadow on the wall cast by a wooden pole okay when we are trying to explain first of all if you are simply looking at the shadow on the wall we are describing the phenomenon we are not explaining it okay to explain it we actually go back and talk about

how the wooden pole within a certain kind of threshold of light and some other elements However, we basically shorten the wooden pole or eleanigate it, it is still cast within this threshold a shadow like this on the wall. Go on Meredith, sorry I interrupted you. No, I'm listening. I just said K that I was muted, but continue. Yeah, no, that's just what I wanted to say. So invariance can be actually detected by various ways. Again, coming back, there is no such a thing as a context-independent idea

of invariance or mechanism. Mechanisms are completely context-sensitive. So would you say... Okay, so I have two questions, or three... I have a bunch of questions. My first question is would you say that the toy model for example from this week's reading is that a mechanism? No, absolutely not. No, no. In fact, toy models try to broaden the scope of how you can possibly using different variables, different theories and and different models can arrive at new kinds of mechanisms, new kinds of singling out invariants. So toy models are a machine for creating mechanisms in a way.

Not essentially. They are basically, you see, they are like these self-consciously, they have reached the Hegelian self-consciousness, that they are absolutely, explicitly false. They are fictions, okay? But the thing is that does not actually, what you might call, relinquish their importance. All models are false. But toy models are in fact saying why they are false, making it expensive. These kind of target system,

Then, in conjunction with different models that have been applied, then you can see that there might be, for example, different kinds of paradigms, different kinds of parameters and variables at work here, which have not been taken into account. and they should be taken into account precisely because they can lead to a paradigm shift, both in the model and in the respective theory or the corresponding theory. It's a kind of almost like, so if you think that models are kinds of, have this kind of

abductive feature, kind of a hypothetical feature, toy models are actually one step beyond that. Precisely because they can. postulate or posit various hypotheses from what appears from the perspective of the current model and the corresponding theories to be impossible or improbable. Like imagine sex, sorry, three sex biology as a way to understand two sex biology or

Arthur Eddington's idea that the idea of the improbable or possible can enrich the idea of the actual. So would you say that this is in some case, you know, to mix metaphors, like going outside of the set to understand the set? Yes, they are absolutely meta-theoretic in a very specific technical sense of meta. Moving from the hierarchy of the set of available data theories to a meta-set and actually systematize this meta-domain. Of course, the meta set does not have any kind of empirical traction at this point.

It is in fact very likely that it is purely false with regard to how we know or talk about the world. But nevertheless, it allows us to see some other alternatives, some other variables and parameters which we have missed. Of course, this brings up the idea that in that case, what actually differentiates a toy model from say that let's talk about, you know, devils, demons, gods, so on and so forth, they are all useful fictions, but why not those and why toy models?

that's something that I will come back to it later. So you know one thing that I kind of find interesting and this might be the like a wrong the wrong perspective but I sort of see an analogy between toy models in particular... between toy models and sort of like a jig or some sort of of scaffolding you would create, you know, if you're creating a piece of art or a, or a, or like a, or a craft, like you're making the tools to make your piece. It's almost like a, a, an epistemology of tool making in a way or a metaphysics of tool.

Like, you know, your toy models are your tool are, are in a sense your tools for, you know, they're, they're, they're your new, they're your new measurement instruments. Yes. Yes. It is, it is, kind of like a, it is a kind of, yes, it's a kind of epistemological scaffolding or epistemological metaconstruction in the sense that you remember that Witkane Stein said, you know, the limits of our world are the limits of our language, right? But of course this statement really

doesn't make sense. When we are talking about a limit, it means that we understand or somehow recognize the limit. Hence, there is no limit, a priori limit to language. That's Carnap's idea and you can get the same thing about toy models these limits in fact for us to determine them such that we can rate rise above these limits we have to construct a different meta hierarchy and that's and this comes back to a kind of i shouldn't think that we should probably this is you need to take it with a very grain of salt. I'm not one of those people who are making these

lavish conclusions from Goodell's incompleteness theorem. But yes, it is kind of like a Goodell's incompleteness theorem, that in order to prove and verify the truth of your axioms as system one, you can't do it. You have to make a system two. And you cannot do the same thing about the system 2, you have to make system 3, so on and so forth, a whole hierarchy of the meta. Yeah, meta, I think that this allegory of a kind of, again, with a grain of salt please, coming back to Godel, that you see, you cannot verify the truth of your statements or prove

them consistently and coherently while you are inhabiting that system, right? It's just a self-verificationist loop. Remember Mikey's, I'm talking to Jovan, sorry. You remember, Jovan, that Mikey gave that presentation when he talked about that these models usually have the habit of self-verifying themselves. You see, that is exactly kind of like a Godelian problem. So in order for you to in fact be capable of verifying, determining these statements, you can't just simply inhabit in your system. You have to create a new artificial system, a new language, a new theory, a new model, which is called a toy model.

A toy model is essentially this meta, this meta level from which you can do a good, better job at talking about, you know, certain kinds of limits that have been going on within your first system. but again those kinds of stuff that you go to your first meta hierarchy like system 2 they cannot also be verified by themselves you have to make create another one another one it's a kind of an incompleteness of the levels of the levels you know of what you might call to be meta insight

Yes, and these are essentially called big toy models. These are called big toy models and there are in fact formalizations to talk clearly about what big toy models are. However, we are not going to that length. That would be just a little bit of that is actually the intelligence and spirit talking about what big toy models are. Okay, let's have here questions, if not, let's have a break and then come back. Adam, do you want to talk a little bit?

Sorry, this is more of a throwaway comment, but it was just a nice... Yes, I didn't get the reference to Jiggs. I just didn't want to interrupt Meredith. As well as a type of like tool, it's a type of dance. Okay, we're working. So you can sort of play with your toys that you made. Yes. That was... Yeah, and again, coming back to what we were talking about last time,

that toy models, essentially, the main idea behind toy models and this kind of meta-hierarchy is this idea that, so if models are representational systems of sorts, our resources for world representation cannot be coherently accounted by exactly those methods of representation. Essentially, our resources of world representation are beholden to our resources for world construction. And this world construction is the domain of toy models.

Even though toy models have no empirical application whatsoever, they are explicitly false models. Yeah, I thought that came across pretty well in the Neuburg piece, or Neuburg piece, the universe. Even though I'm not familiar enough with string theory to really follow the physics of it beyond about page three, I was able to see what he was doing in terms of, okay, well, let's use this simplifying assumption and this simplifying assumption and these radically reduce things like the whole universe is actually just one dimensional circle. It all exists in one dimension, which then circles back on itself.

But we're still the force of gravity and we're still going to use exactly the same equations that we use in our fully, you know, trying to model the full complexity of the universe physics. but we're going to use it on this radically shrunk universe or radically simplified universe to see how these things interact more clearly. Yes, yes, absolutely. And the thing is that when I start to talk about false models, in fact toy models not only just simplify but they also introduce fictional components, entirely fictional properties and entities and sets of events and processes

which are not existent I mean we don't know anything or they are completely imaginary it's a make deliberately fictional right so you sort of say like you get like half the pieces of the Lego set from some model that's been built around some real-world phenomenon and then take it over here and say but what if I add this piece as well, which I never did. Yes, yes, absolutely. Yes, yes. What can I build with that? And then how can that, then eventually that might reflect back on what I was doing. Absolutely, yes. That's a superb example. Yes, coming back to the Lego stuff. Yes, absolutely. And it's not just, you see, as I mentioned, all models are false. I will talk about this.

It's just that toy models are self-consciously false and that gives them an advantage to to think more determinately about the scope the limitations of the models in use okay let's have a break for five minutes and then we come back how about that sounds good excellent

Thank you. Thank you.

Thank you. Thank you.

Thank you. Thank you.

Thank you. Thank you.

Thank you. Thank you.

Sorry. Theo actually asked an interesting question. Can you elaborate? I have an answer for you, but would you be able to elaborate a little bit your question

on the sidebar with regard to wondering if there are any relationship between toy modeling and norms? um well i'm it's kind of a sloppy thought right now but i was just wondering so i'll give like a cons a kantian type of answer where you're saying uh you imagine the the moral world in which like virtue and happiness are proportionately dished out which is something like a counterfactual I guess but it's yeah I'm not sure exactly how to relate

it to the toy modeling right now but yes you see well the thing is that I wouldn't say that yes sure there are counterfactuals but in the sense that counterfactuals that are not you see so basically when we are talking about counterfactuals and philosophy we can talk about it in terms of two things you see when counterfactuals like literally it does not really abide by what is we take as facts okay a different kind of words where different kinds of mechanism different kinds of claims can be put forward right now this is a very rudimentary, perhaps even distorted definition of a counterfactual universe.

Now, there are two ways that in, and of course, counterfactualism is closely related to the so-called modal revolution in philosophy and in science. If we would have had horns, what could we, for example, do? Of course we don't have horns. Some people actually transplant them, but so be it. I'm talking about like as a kind of a massive fact. There are two ways to actually talk about counterfactuals.

One in terms of modal realism and the other one about more of a kind of nominalistic sense. So the modal realist is David Lewis. the kind of a more nominalistic sense of approach to counterfactuals is Nelson Goodman. Now, the modal realist thinks that everything that can be said possibly is real. whereas a nominalist someone like Goodman thinks that the counterfactual scenarios such as having horn

or having such something called apricot instead of electron as a physical entity essentially need to be responsive to the facts at hand, not disconnected, not fully disconnected from them, right? So these are two different kinds of major scenarios of how we can think about a counterfactual. Now, yes, from this perspective, the latter perspective, Nelson Goodman's, We can think about toy models as norms precisely because they are constructible.

But their constructability needs always to be responsive to that which is at hand, namely the existing facts, body of facts or set of facts, set of features within the current theoretical structure. Just like allegory that Adam made. So it is not that we are coming with new Lego pieces completely we are just using some vintage Lego pieces that have been made into creating a Darth Vader in one of those 1980s Lego sets. And then we add a few

other pieces which absolutely have no place in the Darth Vader scenario a Star Wars scenario like a homemaker like you know the kitchen scenario 1960s Lego sets and then we play with them and see what how things can fit together from this point onward yes and in In that sense, yes, there are in fact norms with the understanding that ought implies constructability.

constructability that is nevertheless constrained by certain sets of criteria. One, of course, this also, now this is a fundamentally, I don't want to get into this, I don't want to make more problem for myself. One of the greatest advocates of this view is Hugo Dingler. If you can find his book, that is absolutely one of the most astonishing pieces of philosophy. He was a philosopher of physics.

Let me just type his name here. So, the thing is that Howard Dingler had It's kind of more like a third scenario rather than as a nominalistic idea of counterfactual whereas the model realistic idea of counterfactual. He was a conventionalist. Does any of you know what conventionalism means in philosophy?

Jean-Pierre? Yeah, I have some... there's a whole debate about conventionalism, so I'm not sure what... In a very naive sense... In a clear-cut sense, well, I'm more familiar with the conventionalism as proposed as read to Wittgenstein's philosophy so the idea would be to that the rules that orient our thinking is are they are basically conventions so they are not answerable to fact in a sense they are not absolutely absolutely this is this is this is here so he was a conventionalist in the sense that he took side with

Poincaré. So Poincaré and Helmholtz, you know, the greatest, one of the great debates at the turn, at the end of the 19th century in the philosophy, in science. So Helmholtz is a kind of person who believes in the kind of the empirical facts and sensory processing as the germ line for science. Whereas Poincaré believes in linguistic innovation like you know let's introduce the delirium of Riemannian geometry in this kind of new entities can we discover how can we renegotiate the

bounds of the Euclidean geometry as applied to physics you see now Dingler takes the Poincaré in a sense in philosophy of science to a different degree he says that to get rid of the contingency of our empirical observations and so for majority of our facts are at some level are bottomed out in precisely the contingency of our empirical descriptions and limits of empirical description then we need to institute rules only in order to play and decide whether we can arrive at new basically empirical descriptions of physical

phenomena is actually representing a very very particular conventionalist stance with regard to the toy model but yes so essentially you can now see that the idea of toy model can be thought in terms of some sorts of a bridge or kind of connection between conventionalism, between fictionalism and you know what you might call to be constructivism, logical constructivism. Yeah, I think I'm having a bit of difficulty figuring out what would keep the conventionalist

from lapsing into full global fictionalism or something like that. Yes, yes, that is exactly, you see, this is what I wanted to say that I actually prefer Goodmanian nominalist or irrealism rather than this kind of full-on conventionalism. In fact, Carnap, when he read the famous essay by Dingler, he, despite Dingler conventionalism, which was completely opposite in contrast to Carnap's view, he endorsed it. But he also reminded in his review of Dingler's work that this absolutely, if conventionalism

does not have some sort of connection with the existing facts or the empirical facts, literally as you say the collapse or relapse back into global conventionalism free-floating rules is completely uh you know uh inevitable and precisely goodman the ways of world making is in fact a kind of a tongue-in-cheek to some extent answer to this debate between carnapp and Dingler where he says that we can only make new universes new worlds new toy

models out of the ruins detritus and facts of the existing world we cannot just simply posits a new world with new rules which are basically in a kind of a platonic heaven nevertheless I think that then you are then you will have to debate the realist because if you are fashioning counterfactuals out of the stuff I would say the stuff office of this will be existing well the model realist wouldn't have a problem with this would he he would simply say that

what you are fashioning is as real as this world in a sense yes yes but that is not exactly yes this is one of the premises of the model realists but there is also a stronger friends that any possible world regardless of whether it is connected to this world or not can be said to be any possible that's right sorry we are going to get into kind of but so what I'm I mean you are into trouble because you know I know I wasn't talking about the rest you see

everyone so when we are talking about all of this stuff you should understand that all of this stuff are built from so many other kinds of philosophical assumptions, philosophical theses. And one of them is absolutely the idea of modality and counterfactual. And to really get into the most depths of these ideas of how models work, toy models work. There should be, you should also investigate like a stuff about modality, modal realism, counterfactualism, nominalism, irrealism,

so on and so forth. So this is, we are just simply trying to kind of bracket this stuff to the task at hand, just giving the most modest introduction to scientific modeling and what models actually do mean. But nevertheless, if you really want to do the actual research, there are whole sets of premises behind and beneath these kinds of discussions. And they are not just philosophy, as I mentioned, these are in fact coming from the um you know uh debates between scientists and philosophers uh starting as late as or as early

as you know uh late 19th century so can i ask a question if it's not detouring us too much okay go on i was just wanting to hear your thoughts more on the relationship between toy models and norms because i my thoughts are really underdeveloped on it so um but i was wondering if if you could go into just kind of more detail about yeah exactly maybe it's that i need a little bit better picture of what my models are in which case

case we could just move on. So if by norms we mean just rules, are we talking about norms as rules? Right. OK. Then that is essentially the whole point, that the introduction of new rules, new calculi of rules, can actually not only determine the relations between the previous rules but how those rules were applied to an empirical level of observation. And yes, that is exactly what toy models are. And these rules

are not defined any longer in pure conformity, pure in italic, pure conformity with the empirical facts that we have managed to get by or derive from the organon of current rules. But these are rules which can actually be counterfactual, literally have no application at this time to what you might call to be the empirical dimension. Okay, actually that really cleared things for me a little bit.

Oh, by the way, those of you who are certificate programs, a friend of mine, Thomas Monahan, is excellent, brilliant, brilliant. And of course, Jean-Pierre is also teaching, by the way, next semester. So Thomas is actually going to talk about the revolution of modality in a very kind of comprehensive sense. How what we understand as a risk and rationality is a fruit of the revolution of modality.

So yeah, this stuff will be covered, if not by me, by someone else. Okay, so coming back to the attributes and mechanisms. So as I mentioned in scientific inquiry, it is typical to distinguish the properties and patterns of a system from the underlying mechanisms that generates these properties. When a model is constructed, we can also distinguish, okay, we can also distinguish among static

properties, static patterns, dynamic patterns and processes that generate these patterns. Now of course we can make a major division, a major point of distinguishing between properties and patterns on the one hand and underlying generating processes on the other. Those processes which are responsible, namely mechanisms responsible for making such patterns. You can call this category attributes and the second category mechanisms.

So you see, majority of people who are thinking about patterns, they are thinking about attributes rather than mechanisms. When, you know, for example, we say that a pattern can actually explain certain kinds of regularity, you know, in cognition, in the real, in the physical universe, so on and so forth. The concept of pattern here is quite vague, this way that usually people talk about patterns. You see patterns are usually in philosophy of science considered to be as categories of attributes. Those mechanisms which are responsible for generating these patterns are called, sorry,

components which are responsible for generating these patterns are called mechanisms. And of course, mechanisms themselves can be distinguished in terms of invariant patterns and hence they require further explanation, further explanation, so on and so forth. Now, a different way, two more abstract way to think about the difference between attributes and mechanism is that attributes are states and state transitions and mechanisms are transition rules.

Like for example, using tiering, a classical church tiering paradigm as an example. So we have states and then we have state transitions within a tape, an infinite tape, a computer. The rules which basically elaborate how these states having such and such transitions are called mechanisms. You can think about this, for example, in terms of a sequence of RNA in biology as well.

Any question here? Jovan has been so silent. Can you elaborate on the RNA thing? I'm not sure I get it. Like, right, you, okay, so you have a sequence, the sequence of, you know, RNA, or for example, DNA, doesn't matter, a sequence of certain kinds of combinations, and how, for example, these kinds of components combined with one another and creating a different combination, these are attributes.

Whereas the rules of how and why such and such components can only fit together to make such and such combinations are called mechanisms. Thank you. Absolutely. Sorry. So, as an example of the distinction between attributes and mechanism, for example, consider equilibrium states of Schelling's segregation model.

When the model comes to equilibrium, it contains racially segregated clusters. And it approaches this state with a pattern of contagion. You know, remember, it was like unfolding like an epidemic. Where small clusters lead to bigger clusters, such that it reaches a segregation state. What drives these patterns are the agents' utility functions and rules of movement. You know, coming back to our Schelling's model of segregation. Attributes such as degrees of clustering are states of the model.

and mechanisms such as agents' movements, rules, or the transition rules of the model. Now, insofar as Schelling's model explains segregation in actual cities, then there has to be some relation between the model attributes and the city's attributes. And there has to be some relation between the model's transition rules and the actual mechanisms that derive segregation in the city. So this was a very brief introduction to how another way that we can think about similarity in terms of attributes, our attributes correspond to mechanisms responsible for generating them.

Sorry, can we just to come back to a really super simple example, can we just go over mechanism and attributes on the example of Lego? Okay, give me a moment to think about this. So okay, let's say that Legos have these studs and these studs needs to fit in such and such ways into the hollow, the dense of another Lego, right?

This is what you might call to be a transition rule. Essentially, every fucking Lego model requires a rule of fitting blocks to one another, right? So this is the mechanism. This is the mechanism. But now, of course, the mechanism can actually be put parameterized by new constraints. For example, in terms of not only the transition rules of how you put one block to another block such that they fit, but also in terms of, for example, color. That certain colors should fit together.

and then the kind of constructions like death star the homemaker the doctor office I don't know the the car these these kinds of models which require you know certain kinds of states certain kinds of blocks to be put on top of other blocks such that it produces that end product at end kit these are attributes is it clear yes okay good yeah how these things these blocks fit together is the

mechanism and of course you can constraint constraint parametrize the criteria of fitting together so if I say that where it's usually only yellow blocks that comprise let's say a model of a house if I'm changing the rules to say that you do not fit the same color together and you must only use contrast no the same color should ever be on top of one another then I'm working within the mechanism parameters or am I still dealing with attributes? No, you are actually, absolutely, you are trying to diversify the mechanisms. And that creates a fundamentally new sets of attributes, right? A black and white cow instead of a yellow cow or a brown cow.

And of course, again, I mean, you know, Lego has always been expensive. I didn't have money to buy Lego. But nevertheless, we know that, for example, when you put a head of a cow on top of a regular Lego block, it just doesn't fit. You need to have a bridge point here. And these bridge points are those kinds of parameters that can actually change very drastically how pieces fit together.

my apologies. So after mechanism attributes we can talk about one more criteria of similarity and of course there are many many more criteria of similarity but I just don't want to talk about them. I will try to come up with some sort of reading list so you can study them. So we can go to fictionalism, false models and toy models discussion.

So there is another way that we can talk about similarity and it's basically in terms of feature sets. Again coming back to Tversky's work you see remember that we talked about delta and what was delta? sets, I mean features set or set of features. Now of course Tversky, the way that he defined a feature set, namely delta, is quite liberal. So in that sense, you can in fact decompose

the sets of contents with regard to such and such features and properties into attributes and mechanisms. Some of this liberal deliberate, some of this liberalness is deliberate in diversity's account of Delta or feature sets because you know the the myriad comparison comparisons theories are required to make you know there are different ways that you can establish comparisons and diverse his accounts and basically

arrive at the waiting function and calculate what you might call to be the scope of the comparison penalize it or basically reward the similarity between two features in Tversky's account. Now, for example, the elements of delta can be qualitative. Interpreted mathematical features such as oscillation. oscillation with amplitude A. The population is getting bigger and smaller.

You know, these are all qualitative and so on. Now, they can be a strictly mathematical term such as is a Lyapunov function or they can be physically interpreted terms such as equilibrium or average abundance. which of these kinds of terms should go in Delta and which shouldn't well the whole point as I mentioned to you similarity is not a context free thing so correspondingly there is no context free answer to the question of which of these kinds of terms should go into the set of features, delta. A combination of

context and prior practice will determine how both model and system are broken up into parts and properties. This is conceptually before the establishment of the similarity metric and is part of how the target and model are conceptualized. Sorry, I used a very esoteric word. How many of you took my computation class other than Adam? I don't think there are anyone. So, basically, I mentioned Lyopunov. So, Lyopunov is like the god of the ultimate system dynamics.

He was a Russian mathematician. He came with an idea. Oh, one second, let me... There are different spellings but I think this is one of the canonical ones. Alexei Lioponov. So essentially Lioponov is still his work is considered to be on the forefront of system dynamics. Lioponov came up with a quantity which is called Lioponov exponent.

and this exponent is a measurement of how given such and such perturbations within a dynamic system in what ways such a system diverges from its initial conditions you know that the dynamic system at least classically speaking is is defined by the threshold of perturbations. That even if the initial and boundary conditions of a dynamic systems are this and that,

if we inject the smallest perturbations, when I'm saying the smallest perturbations, They can even mean beneath the threshold of existing measurement metrics. If we give these perturbations, inject them into the system, the system evolution can fundamentally explode toward fundamentally new courses of evolution. Now, the thing, however, is that this is just a mathematical idea. This really cannot be applied to an actual physical system. It's an idealization. Essentially, Lyoponov's idea of exponents,

these kind of perturbative exponents, is based on one idealized assumption. Given an infinite time, the system trajectory will explode, will diverge from, fundamentally diverge from its initial condition, where it has come from, how it looked like at the beginning. So these are mathematical stuff. So this is just, I wanted to kind of, I made a very esoteric reference. I just wanted to talk a little bit about what actually Lyoponov stands for, and what its significance is for dynamic systems.

Of course, the thing is that, as I mentioned, Leopold of exponents can be also thought, first of all, there are idealized mathematical models. not only idealized mathematical models in the sense that we have been talking about, but there are also mathematical idealizations. Mathematical idealizations. It is actually, you have heard this stuff in folklore of complexity. I mean, you should trash them. for example they say that you know for

example given such and such changes in the system this system can be technology can be human can be the planet earth can be the universe so on so forth there is absolutely no way to guess or estimate even within a probability threshold how this universe how this human how this technology evolves within a time limit no this is absolutely a very very naive interpretation of lyoponov exponent and it's in fact responsible for the majority of misinterpretations of the chaos dynamics within complexity theory

You know, you cannot simply use this in order to talk about physical systems. This is simply an idealized model whose mathematics, as I mentioned to you, is also idealized. But nevertheless, the whole point that I was trying to say is that the changes in the systems creates new rules of evolution, new tendencies. And these new tendencies are kind of like mechanisms. The output of the system, however it might diverge from its initial conditions, are new attributes. Okay.

Okay. So as I mentioned that which of these kinds of terms should go in the feature set delta? As I again as I mentioned there is no context free answer to this question. The combination of context and prior practice will determine how both model and system are broken up or decomposed into parts and properties. This conceptually before the establishment of the similarity metric and is part of how the target model are conceptualized.

With the conceptualization of the target and model into properties in hand, then the scientist can add elements, new elements, to delta feature set. For example, an ecologist might include terms like equilibrium abundance and for example maximum population size. For Schelling's model, relevant terms might include segregated clusters of size n, racial exposure index r, spatial layouts of cities and neighborhoods and descriptions of various movements, rules, and utility functions. As science progresses and more is known about a

model's target, the content of delta or feature sets may fundamentally change. Modelers might initially deem some elements of models and targets important, but these might fall by the wayside as the science progresses. Similarly, new properties of the target might come to be recognized as especially important. These changes in practice and interest will occasion a change in delta and consequently a re-evaluation of the model-world relationship. These changes alone can have the effect of rendering the model more or less similar to a target.

At first, this might seem like a disadvantage of the account, suggesting that the account's flexibility or malleability precludes it giving a good analysis of the model relationship. relationship. However, there are two reasons why this is not a disadvantage. First, the theory of similarity supervenes on properties of the model, properties of the target and the modeler's intention. When context or scientific goals change, these intentions will change and aspects of the relation will change as well.

Second, there are cases in which the perceived quality of a model changes without any new information about a model or a target. This can happen when a better model is created, but the old one continues to have heuristic values. So this is another idea of similarity in the sense that we begin with a kind of almost a very modest way how we define our systems. what we knew kind of epistemological criteria with regard to context sensitivity with regard to what features did we choose and what features we

excluded and then as we toy around with these features with these intentions we We see that the concept of similarity as it has been espoused in our previous paradigm can actually change, become more refined or become falsified in fact. Questions? Okay, I'll throw a question out there. Yes.

How is this not, and I don't think this is a bad thing, but how is this not sort of well-organized question begging? In other words, you know, we say the validity of the model is, maybe validity is not the right term, but we're basing something in similarity. how good the model is. But if the model's predictive, then whatever elements are working there become parts of the similarity function. Is that making sense? Yes, but would you elaborate your last point? Because I think that's an extremely, just a little bit, not just for me, but for everyone. Oh, yeah.

I mean, as I'm understanding it, you know, taking sort of relating the attributes of the algorithm to the city in the shelling case, to the extent that you run multiple models or the quality of the results match of the results of the algorithm match the results in the city, let's say. Maybe this is a bad example. And that you tweak the algorithm and now they don't match as well. Well, you now know something about what to put in the similarity function. So is this process driven by the results? Yes, I think that you are absolutely right.

It essentially comes back to that idea that we said the models end up to be inevitably self-verificationist. in the sense that if I'm, please do correct me if I am misinterpreting what you were saying. So we start with a humble account of similarity, right? Which is a complete fundamental idealization. And then we deploy the model with regard to a target system like a city. And then we add, as how the model works with regard to the world, we come across possible new similarity contexts.

Basically, relation between mechanism and attributes. Which we then plug them back into the similarity. But that also begs the question that essentially the very reason that we came across such similarity points was precisely because the frame of reference that we had already deployed, namely the model, idealized similarity in the first place. And hence, who is to say that such similarity points are actually not any more fictions than any other kind of fiction that you come up with? Yeah, yeah.

I think I might have been in a slightly different angle in the sense that we're looking for something to justify the relationship between the model and the world, and we're calling that similarity. But ultimately, similarity is just the match between the model and the world. So we're... I mean, we are breaking something out conceptually. Yes. But it's not really anything new. No. Well, it is new precisely because, you see, there are two ways to go and think about this. And again, this is something that I would like that, for example, Theo, Adam, Marie, Joven, or Jean-Pierre talk about it,

or any person. Essentially, one of the first things, it is new precisely because the model is deployed within the ambit of a theoretical structure, a theoretical paradigm. Of course, the theory is that which determines under basically which criteria and constraints certain kinds of structures can be applied to certain kinds of observations. This is the whole task of theory. So the model works in the ambit of this.

And of course, theory always gives us something new. Like, for example, the equations of motion. With regard to, you know, if you are simply thinking about pure observations and pure thoughts, we could have been always in the realm of the Ptolemaic system. It's just that Newton comes up with a new kind of theory in which only certain kinds of structures, couched in terms of differential equations, can be applied to the observable motion of celestial bodies. And within that, we can create certain kinds of models.

And these models are new to the extent that they are committed to that theoretical revolution, to how this theory actually gives us fundamentally new facts with regard to the motion of the celestial body. So, in that sense, model is new. But model is not new in the sense that essentially, as I mentioned, the criteria of similarity with model, we yet don't exactly know how much of it is part of a theoretical

paradigm to which it belongs and how much of it is actually an arbitrary decision made by a modeler or a scientist? Now this is absolutely yes, yes, and that's why I said that models are false. All models are false, but that just does not basically undercut their significance. So yes, I probably haven't answered your question precisely because this question is quite a very very thorny question. Hopefully when we go to the toy models we can think about

this problem more coherently but I want other people actually respond to Ian at this point because this is a really fantastic question. It's absolutely one of the best questions. Theo, a response to Ian? I'm thinking, I'm still actually digesting it. I was wondering if you could clarify though the... Me or Ian? Reza. Yeah, I'm still digesting what Ian said actually, but I might have something to say in a little bit. What do you want me to clarify?

Or are you saying that you can't determine the theoretical assumptions of a model from its arbitrary assumptions? Is that, or that becomes a problem? So you have some sort of theoretical assumptions which are already encapsulated within the structure of the model itself. Yes, still within the practice of modeling, we just cannot fully determine how much of what we call by similarity belongs to what theory talks about how certain structures can correspond or should be applied

to certain facts, okay, that's a kind of a theoretical component of a theory structure, from the idea that similarity might actually be fundamentally an arbitrary decision of a modeler with regard to the world. Precisely because models are not theories. Models are subclass of theories. in the sense that the intention of the modeler plays a significant role in what kind of model, with what kind of interpretation, scope, assignment, and fidelity criteria

should be applied to what kind of phenomenon with what kind of features. You see, there is a kind of a bifurcation here. between those parts which are fruits of the theory similarity what kind of a structure corresponds to what kind of features of the world part of the theory and what kind of features of the model should correspond to what kinds of features of physical or worldly phenomenon which can be fundamentally the fruit of a

modeler intention and for that matter they can be arbitrary. Now to determine how these two are actually connected to one another is something that modelers usually don't care nevertheless should be discussed. Similarity in the sense, in the general sense of how theory can allow us to think about certain kinds of mathematical structures as applied to certain features of the world and similarity in terms of different features of the models

apply to different features of the world any person any person who can basically a little bit you know go through that question that Ian posed which I absolutely think that is really very very good question and hopefully we can begin a little bit shed light on this question as we move forward I'm not sure that if I can answer it we can reframe it can you repeat the question yeah please do repeat your question okie doke now you

see you one time you ask a question everyone wants to see this is I was saving it up. You got to accumulate the anarchy. I think sort of the question is, is similarity just circular? A particular principle, in a human sense, begs the question, as you said. It begs the question, because you want to validate the model by saying the model is similar in an X, Y, Z way to the world in ABC, you know, X, Y, Z matches up to ABC or whatever it is. And you say, this matters, they're similar, it's valid. But, you know, if your results are sort of

mismatch, the results are mismatching, not the attributes, but the results, then you say, you know, maybe X, Y, Z doesn't matter. And you find other elements that maybe do matter. so to answer the question between the relationship the model and the world to establish the validity I'm going to say validity of the model you really need to know the validity of the model so you can determine the similarity I think this is really interesting because this is like the bootstrap problem like when you bootstrap a yield curve self-grounding problem yes 10,000 trilemma yeah answer that question that's a whole other branch of you could start your own branch of philosophy

anyone anyone this is really good good good debate is another way of uh thinking about this problem just thinking about it in terms of the difficulties that representational epistemology representational philosophy runs into when it's trying to somehow um absolutely no don't worry don't worry this is just like the order of the day We have cats, cat fights, children fights, whatever. It's fine. We're not going to tweet about it.

I mean, I think the way, I think the whole question is how do you reframe this? Like there's something about this, something about the framing of this problem. No, I think that Theo actually came up with a general problem. It is really the problem of representation. How do you verify your representation without falling in trap of petite principi, circularity? Well, just to give up, just a clue for everyone. I think the answer is you don't. I am happy with that answer, but I don't think that's productive for debate. I do think that this debate, you see, I will talk about this a little bit later, probably in the next few sessions.

So there is this whole debate about two forms of representationalism, okay? And these basically correspond to two different forms of realism in philosophy. One is called global realism and the other one is called local realism, which is essentially heuristic way. A heuristic one is a modest one, which I absolutely am fine with it. These circularities at the level of local are purely heuristic. And of course, but when we inflate them to the level of global representation, global

fidelity or global representation, they become fundamentally, basically they cut you at the kneecap. They literally doesn't allow you to move forward. They create actually a skepticism, fundamental skepticism. But yes, this is really one of the best questions ever asked, the whole series. Isn't this like similar to the question that Goodman raises in his article, the problem of counterfactual conditionals? trying to construe a counterfactual as something that happens like his example

is a match that is lit it it gets lit because of standing conditions okay so if the match is how do you say that in English then you risk the match like this Okay, then the match lids. This is the counterfactual. This is a conditional phrase, not a counterfactual. Condition. Yeah, yeah, condition. But you can... Actually it is counterfactual, modal counterfactual robustness as Brandon says. Yeah, thank you Adam. Yeah, if the match would have been strike, he would have lids. Okay, this is the counterfactual. Then he construed this as in standing conditions S, if the match has been striken, striked,

he would have lit. So this transforms this counterfactual into an embedded counterfactual within this counterfactual. You have to prove the standing conditions, the validity of the standing conditions. If there was standing conditions S, then if I would have stricken the match, it would have lit. So you can never prove a counterfactual unless you prove another counterfactual. Essentially, you cannot prove a counterfactual such as a striking match, a safety match against

a frictional surface if you cannot find a defeaser. A defeaser essentially, for example, a defeaser would be like this. I mentioned this I think a few sessions ago. So a safety match. A safety match is very different from the old matches. The old matches were like basically napalm bombs. The new safety match actually has an oxidizing agent and sulfur and a very particular combination of the match stick. Now the thing is that A diffuser like this, when the match, a model of a match, is stricken against a frictional

surface in the terrestrial atmosphere, it lights up, it ignites, okay? And the matchstick burns, right? Now this is what you might call to be based on certain ideas of similarity in a canonical sense. But of course you cannot prove it by any means whatsoever from a kind of a modal sense. You have to find the defensor and the defensor is a different context for lighting a match. You say that I take the safety match beyond the orbit into the vacuum.

And I strike it against the fence precisely because the match head is a combination of sulfur and oxidizing agent. The match head ignites but the match stick does not burn. because for the matchstick to burn, it requires oxygen. Whereas that oxidizing agent within the match head actually allows it to get ignited. Now, Robert Brandom has a different example of this with regard to a lion. So essentially, we are talking about

a lion inside a cage now this line is being transported from one zoo to another zoo of course at some point you might change the cage okay the lion gets wet is a still a lion you see there is a range of counterfactual robustness Whatever within this range, and this is not something that can be logically or empirically done, it is essentially to find defeasers. Within the range of these defeasers, whatever we do to this lion, it remains a lion.

But imagine that at some point someone instills this lion and replaces it with a zebra. Then the whole thing will become a different phenomenon. But still, I would like you to, while I understand why you are actually posing the question of counterfactual robustness with regard to Jan's question, would you be able to a little bit elaborate to Yen how you can see what you were talking about in relation to Yen's question? I'm mentioning this Goodman article because the way he puts it, it begs the question.

thinks that it is basically impossible to prove or to get a final validity of a for a counterfactual in somewhat an analogous sense to what Ian saying about about models perhaps because you will have, you will need, you will still need, it's the same structure that is reiterated at the level of the standing conditions and then once you get one standing conditions it will beg another counterfactual. This is like a vicious circle

I think this was the structure of his question, if I'm not mistaken. So you are not anytime attending to the world. You are not at any time getting knowledge about the world. I think this was the way he phrased it. You are not getting new knowledge by this. And I think I don't really get the gist, actually. This is, I have to confess, because I don't really get the gist of one answer to this problem, which I get from Sellers, in which he says that the process of devising counterfactuals is the same process of refining the language in which you are constructing them.

So maybe if we can extend the analogy, I'm not sure where you're getting at, Reza, with your course, I mean, I'm not sure what's your goal here, what's the thesis, but perhaps this idea of explication would amount to something like that, like the, what do you say, the elaboration of language itself which is the same process, it's not a different process, you don't have to recognize the rules that are there and then you get to do science.

The process of refining the language you do science is the process by which you assume counterfactuals, I mean assume models to be in some sense valid. Yes, I do generally take side with what you say. But I do think that Ian was not talking in fact about the refinement. He's actually talking about within the ambit, the strict ambit of the deployment of a model with regard to a physical phenomenon or a target system. Now, within the ambits of the model, literally there is no way to get out of the circularity problem.

I really don't think. Whereas what you are talking about is not within the ambits of the model. It requires a refinement of the theoretical structure. structure and also the language within which the theoretical structure of which a model is a part can be reframed. Yeah, yeah, yes, absolutely. But isn't JP's response necessary in some way to speak about getting out of the vicious circularity. Yeah, I don't... Yes, yes, it is. But the thing is that you cannot do this at the level of the model itself.

The model is always going to be circular. That is the whole point. Yes, I see. This is actually a very, very... some of you know these are we are coming across new questions uh which are not even addressed by weisberg or any kind of you know philosopher of modeling uh what jp actually is talking about is not really about uh the model itself the model itself is just simply this kind of Petito Principe. Which of course... Maybe just... Go on, go on, please. No, no, please go on. I was talking about first there was this Goodmanian account of counterfactual

which I think captures the issue because he's not placing this counterfactual into a larger picture of like an ecosystem of different counterfactuals which is something he does in a later text but I will this this this little article the problem counterfactual conditionals is really just a logical analysis of this counterfactual this match counterfactual and his conclusion is exactly this one it is it is circular you can't get out of it I mean it he doesn't phrase it like that he says I don't know how to get out of this I think this is the ending of his article and then I I was I was referring to the cellars in way of

getting out of it which is actually to understand counterfactuals as encoding the rules of constitution of declarative sentences absolutely yes very car not the instance yeah I'm agreeing with Reza because the cellars account is already already is encapsulating all these so-called models inside something which is bigger than the models so the the responsibility the owners to provide the results is not only in the model anymore but it's in the model inside a an ecosystem of expressions I don't know I would say

Yes, yes, yes, absolutely. The theoretical edifice and also the metatheoretical assumptions of that theory. Such metatheoretical assumptions are fundamentally required for us to investigate them properly to refine the system of language. and this is a very Szilardian way to say it, but ultimately it's extremely also a Carnapian thesis, the task of explication. But yes, I agree with both Ian and Jean-Pierre that now the point is that, okay, at the level of the model, it's just...

from a scientific epistemological perspective, we cannot verify the idea of similarity without, in one way or another, endorse a petitoprencipe, a circularity of some sort. but the thing is that that is why a true modeler should not abide at the level of model it should go to the theoretical edifice of which the model is a part not only uh not not not even not even stopping at that phase at that level it should go even beyond

more to revise the very language the very structure in which the theory is constructed and also the model respectively and that's how this problem can be averted but not in a canonical sense you just simply cannot I would say Fault a specific model while you're abiding in that model. Go to 101. Go to 101. Meredith, go on.

I feel like I'm having a parallel conversation with Adam. I don't know. I do feel that this whole circularity, it is a misstate. We are somehow mis... Okay, my children are fighting each other with swords. Don't worry about it. I do think that we are somehow misconstruing the problem of similarity, that it's is creating this paradox. I don't have an answer for what that is, but that is my intuition.

Ian, would you be kind enough to all of us to please, please come up with a kind of crystal clear version of your question and put it on the Google Classroom. Yeah, I actually can. And it's, yeah, yeah. And that was to me, right? I'm sorry? No, no, this is absolutely, you shouldn't be sorry. Absolutely. I think that what you have done is that you have perturbed the entire class, which I'm

so proud of you for that. It's truly, you know, essentially we are in the business of philosophy. We do not want to sell on anything. Any kind of perturbation should be seen as an opportunity. Okay, I will give my best shot at it. I in fact have a real life hypothetical in mind. But yeah, okay, yeah, I can sketch it out. Yes, please. Does it does it actually has some sort of crime case in court? Well, given your background? If you would like it, I can twist it around to have that.

Yes, yes. It's a little more generic than that. We would love to. I can throw in some details. Sure. Absolutely. Yes, definitely. This is exactly what we are trying to do here. Essentially, Ian, you see, so many of us coming from different backgrounds. Philosophers, I love philosophers. I'm a philosopher. philosophers for the most part really don't have any good grasp of you see philosophers are great at basically coming up with this kind of umbrella questions right but when you actually do talk to a philosopher their examples are

never straightforward they're kind of weasley wishy-washy examples I am so sorry Jean Pierre for saying this but as a philosopher I should act actually accept this I really want to hear examples that even might be allegorical to what we are saying but nevertheless they shed an extra light on what we are in fact talking about at this point Thank you so much Ian. Thank you very much. We are reaching the end of our session.

I think I need to have a cigarette. I need to have a little bit of inorganic intake, namely food. So come up with your last heckling. Adam, oh by the way let's ask those people who haven't talked. Marie, Justin, Sean, Cepide, Barrett, Artemis and Alberto. Artemis, go on. No, I was just commenting about philosophers and I thought that me, I mean, we from the

other fields might kind of act as a toy model that has a freedom to be false and accidentally say something new. I was just commenting on that saying. To be honest with you, the thing is that when you see philosophy is a task. Philosophy is not exclusive to philosophers, literally, absolutely. Any person who thinks, who mistake philosophers as the idea of philosophy is a charlatan. You can do philosophy in whatever you do, nevertheless. This doesn't mean that any person who thinks is also a philosopher. To do philosophy is to abide by certain standards, certain constraints, certain kinds of way

of interrogating yourself and the other. Mari, what are you talking about? Go on. Oh, I'm buffering today. I've had enough to process. I will be buffing for another 24 hours. I will also have some food and then I will start talking. Well, Sean, would you be able to talk a little bit about this, what we have been talking about? knowing where you are coming from with regard to the kind of problems that you are tackling in your own field. Yeah, absolutely. So I'm thinking that the whole differentiation between the model and the theory,

or rather like how theory can coordinate with the model has been super helpful for me thinking about like language particularly. And so this kind of like, I mean, because I mostly study like poetry and literature and especially like with poetry, like there's this kind of like, you know, less, I mean, it's interesting. I'm still working this out, but like somebody like Holderine or Mallarmé or some of the Sufi poets that I'm interested in, or yeah, there's this like extreme attention given to individual words and how like the scaffolding of these kinds of like structures. And if you just remove like one word, the entire thing can kind of fall down. And yet at the same time, there's this kind of like yielding to the multiplicity of language.

But the model has been super helpful because like writing on that, it's like, you know, I'm not writing poetry, right? Like I want to write a systematic kind of analysis of what's happening in individual poems in relationship to a kind of larger project. So I'm still working it out. I know that's kind of a rambling response. But, yeah, I don't know. Superb, superb. Excellent. That's. Oh, yeah, I'm still just trying to process kind of some of the things that we're gone over today. I have some sort of like larger, broader questions, but I don't see how they really apply to today's discussion.

So I think I'm going to work out those thoughts and figure that out perhaps for next session. Next presentation. A responses class. Don't say no. What was that? I said that how about you giving the next presentation about today's class. Okay. Next session. Okay. Okay. There we go. There we go. And Ian, how about you working, crystallizing that fantastic question that you posed and pose it as a presentation for the next class?

Don't say no to me. I don't want to say hear no at this point. You came up. You opened up. I don't want to hear no at this point. Literally, you have to put yourself in that situation. Now you have to take it to the ultimate conclusion. Look, if you would stop talking, then I would say no. No, just to be clear, you want the write-up and then a presentation in the class. Yes, kind of in a way that incorporates your ultimate question. Sure, sure, sure, sure. Superb. Excellent, excellent, fantastic, excellent. All problem solved. Okay. I think that, so next session, again, as I mentioned to you, whatever we are, I owe you, you know, right now I have to keep in, I have kept the tally.

I owe you one hour and nevertheless beyond that we are still whatever those of the major discussions that we haven't talked I will cover them at the end of this of our seminar next session we are going to talk about false models and and fictionalism in models. And that opens up an entryway to what toy models are. From the toy models, then we move to different kinds of thought experiments which are not exactly models in the way

that Weisberg has been talked about, but I would say that there are models in the strict sense of toy model building. And so I'm just giving you a little bit of flash forward of how we are moving forward with regard to our discussions. Thank you so much, everyone. Fantastic, fantastic class. Thank you for all of your contributions. All right. I'm going to go ahead and end the broadcast now.