Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

Hi, everyone. It's Cohort 3, and August 1st, we're in our second discussion on Chapter 9. So, if either of you would like, before we jump into some specific questions, are there any general comments or any other topics that people want to address? Otherwise, we will probably look through some answers and curate and improve some discourse. Yeah, nothing from me. All right. I added some questions at the end, but we can go through the other stuff first. Perfect. Okay. And Esmail, thank you for your question. What does this text mean about epistemic value? So, let me know if it's okay to modify the question to what is epistemic value? If that's what your question was, or let me know. Okay. Let us... You added some of the new questions at the bottom here, Neil? Yeah, that's right. All right. It's one sort of big question, really. Which one? Let's do... Well, in a non-Metanasiian, so that one, and then... All right. Awesome. Yeah. All right. Let's start here. All right. Okay. So, my take on the chapter is trying to fit active inference into a framework of what science is. The new thing, I think, the big idea for me was the Metabasian inference, which I understand as trying to make... Someone's trying to make inferences about something that has itself got beliefs. So, maybe I don't know if I've got that right. So, the first part of that question is, what does this half chapter have to contribute when that isn't the case? If we're just trying to use it to understand something that can be modelled? And then, the second part is what the chapter talks about, is thinking about it in the context of brains, so psychiatry. But it presumably can be applied to anything in which the system that's being modelled can in any way be said to be of having beliefs. And I've mentioned Dennett's intentional stance there, where we're not necessarily believing, you know, we don't think there's any sentience there. So, you could think of... We could talk about a thermostat thinking it's too hot, so without actually saying anything about any sort of mental processes going on there. So, it seems to me that the Meta-Bayesian aspect could potentially be useful for this, beyond psychiatry, psychology. Yeah, and so, skipping over the third question, which was the question is, is this actually being applied outside of neuroscience and psychology? It seems particularly... Well, I always thought it would be useful to be including it in behavioural economics, and maybe in systems engineering. I don't know. Great, great questions. We'll start with the first one. Oh, there we go. Oh, okay. I see. The second one was here. Also welcome, Molly. Okay. Thanks. And I put the link, but this upload of the textbook, that's a little... It's a good starting point. All right. So, the first thing that... Well... Let's just start by recalling the Meta-Bayesian approach in 9.1. One. So, I think you summarized it perfectly, Neil. We're doing Bayesian inference about something that has beliefs. So, the aboutness of the subjective model is the environment, which may be the laboratory or the niche. And thus, the aboutness of the scientific model is the rat in the maze, which has a model that has an aboutness of the maze. So, that's this kind of like nested or Meta-Bayesian approach. We're using Bayesian approaches from the outside, but then when we get to the good stuff, it is also a Bayesian model with an aboutness of an environment. So, I think this is a great way to put it. Trying to make inferences about something that has beliefs or trying to make inference about something that does inference. So, your first question was, what does the chapter contribute when that is not the case? Here, I believe we can return to our favorite path integrals paper. And point to the continuum where on the sophisticated end of the spectrum, we have a situation like that in this figure. Where, you know, here's us as scientists. Here at the blanket states are the solid line. And then we're inferring a…