Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

All right. Hello fellows. We are in by all independent accounts the last of the meetings of cohort 7 in chapter 10. We've already had some interesting sharings and consolidations. So what would be interesting in this time? How how wh where if chapter 10 is the takeoff, where do we go? Or where are we in our own chapter 10? Well, for me, for me, you say Stephen. Yeah. I was just going to say it's model building. It's applying applications is where I'm at. Andrew. Um well, for me, um it's just an achievement to have finished reading the book. I still have to read the appendices, but um uh upon reading the book, I feel like I have a sense of the complete picture, maybe in a murky way. Uh and then what it invites me is a couple of u things. One is to be able to tease out the flesh and the bones I guess or or to say like what is actually active inference saying and then what is kind of like additional ideas that come out from applying active inference because those additional ideas they could be brilliant or they could be just not so brilliant. Uh but that's a separate thing. So um and you were just talking about that if you just look at the bare bones active inference that's actually a common theme across all different types of let's say um I think it was like learning you know large language models or or different types of AI approaches. Um so um what that means is that in order to really appreciate active inference it's important to look at it minimalistically like not what are all people trying to do to apply but really what is it like? It's a little bit for me like calculus like calculus is absolutely important for modern physics. Um and so you wouldn't want to make calculus more physical than it is. it's just a mathematical um way of thinking and then you wouldn't maybe want to say that the calculus is a theory of everything but to be able to really appreciate so what exactly is calculus and what is it contributing you know and how does that relate to linearity let's say or not uh and things like that uh I think that's where I'm at and I think like in order to do that maybe the next step for me is just to uh compute you know so uh to learn how to uh use the Python uh modules I guess and to compute things that would be relevant for me and get some hands-on experience. I think that's probably the next step for me. And then maybe after that like then to say okay take a fresh look at let's say what are baianesian statistics and the related math and stuff and work from there. That's where I'm at. And maybe to say like uh what's attractive for me? It does seem uh to be um very sweet in terms of putting perception and action in a in a unified footing. But that's a very particular case where you're assuming that you're looking at the world with a generative model and then from the point of view of that model you know you have this active look at perception and so you have this active look at action let's say and then you have this ability to quickly switch from one to the other based on I guess free energy um the the free energy uh I guess maximizing or minimizing the free energy So that's all fine, but that's that's assuming a certain, you know, outlook. Uh, and I don't think that every outlook is always based on a generative model. Um, so there's alternative outlooks, let's say. But, um, so just to kind of tease that out, what what what kind of alternative outlooks other than looking at systems in terms of like their input and their output? Well, I just don't think like I don't think necessarily that the physical world is based on a generative model, you know, because a generative model is kind of like a it's a symbolic assumption that like so in the way I would think about like um there's knowing versus not knowing, you know, so like there's you know maybe values and variables and so um the generative model presumes that you have this world of variables of not knowing of symbolic language um But I think that something like a…