Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

Hello, it's March 1st, 2023. We're in Cohort 2 of the Paradol textbook group. We're in Meeting 18, our first discussion on Chapter 8. First, Michael, if you wanted to just raise the context or question that you wanted to begin with, and we'll just pick up there. You're on mute. I'm not sure if you're asking. Sorry, my microphone was off. Yes. Yeah. So the map territory fallacy, I was wondering what this is. And whether you could perhaps elaborate a little bit on that fallacy. All right. I'm moving it to Chapter 9, even though we're in Chapter 8, because we're going to bring in a figure from Chapter 9. Oli, you want to give a first thought on this? Yeah, sure. Actually, map territory fallacy is one of the fallacies that is widely used to somehow criticize a free energy principle, because as the name of the concept may suggest, some believe that it's a kind of fallacy to somehow conflate the map with the territory. So by this, I mean that free energy principle claims that, well, we have, so very informally, we have a map of the principle or the theory or theory or whatever, and the territory that it applies to. So the system or the agent that we're trying to model. But on the other hand, by claiming that the system or the agent actually does the kind of inference that recovers the map itself, some people believe that this kind of conflation that map converts into territory or the vice versa is a fallacious claim. So we need to separate map from the territory. But in this recent paper on the map territory fallacy fallacy, Ramsted et al argued that actually, this is not a fallacy at all. It is the structure of these of the the FEP theory. So in other words, we're well aware that there is a kind of inference as if epistemology happening here. So yeah, but it's a bit more extensive. They have lined up some number of arguments to defeats this map territory fallacy fallacy fallacy, and why it's not fallacious at all. So yeah, that's basically the gist of it, in my opinion. Does it have to do with a distinction between model and process? And the model is the, maybe the model is the map and the, no, the other way, the process is the map, and the model in the external world, I understand. No, the external world is the, is the, what is the process is the external world and the model is the, I mean, the generative model of the agent itself. Yeah, it definitely relates to that. But on the other hand, it also relates to this concept of self-evidencing, which is a way to describe the mechanism by which the model in the environment, or let's say the agent in the environment, models itself according to the generative model. So it's a kind of, as the name suggests, a self-referential way of modeling the system and the environment. Yes, it's definitely a roller coaster, and not a resolved one. But still, we can look to some of these key recent works. So, the map territory fallacy would be just like it's been since the classical Borges writing. This realization, like, the map is not the territory, kind of related, like all models are useful, or some models are useful, but none are true. Like, correlation is not causation. This sort of broad space of aphorisms that's basically like, instrumentalism isn't realism, as we might say it. So that's the map territory fallacy. We'd be making the fallacy that the map is the territory. I made a linear regression between cancer and heart disease, ergo they are actually causal that way. And when we're using active inference framework, one might be slightly more tempted than the linear regression to say, well, because I've generated a model that has a structure that seems to articulate certain natural things, this is like an explanation. It might even have predictive power. And so that might lead somebody to make this kind of a fallacy that the map is the territory. Then Ramsted et al have kind of agreed and amplified, which is saying, actually, when you use the map territory fallacy to criticize the kind of…