Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

All right. Hello everyone. Welcome back to the active inference uh institutees textbook reading group. Uh today this is cohort 7 wrapping up with chapter 10. It's been quite a journey through uh the first part of the book which as we know is a bit more focused on formalisms on theory on anything from marov blankets to wrapping our heads around hierarchical uh variational message passing in microcortical circuits as well as looking at sort of a basian framework for doing updating and then linking that to action and perception and action perception loops. uh from the active inference framework perspective. And then with the second part of the textbook, we saw a lot more in terms of application models. We had a recipe for active inference ranging from uh defining our generative models, our generative processes, considering different kinds of components for models, likelihood and state transition matrices, looking at the discrete case versus the continuous case and all the rest. And then our last chapter for this cohort prior to chapter 10, chapter nine, we were looking at what what happens whenever we want to actually fit models to empirical data. Whenever we want to do group analysis, uh, and and kind of wrapping our heads around the idea of composing a subjective model for our agents, then having our own kind of model of the agents model. So something like metabasian modeling or analysis. Now the chapter 10 which kind of sticks out a little bit uh a bit of an outlier in the the second part of the book. It's almost as if they could have written a third part of the book that would have been labeled something like you know active inference and its connections. We saw in this chapter that we're actually no longer directly talking about kind of the mechanical or or sort of of of uh engineering aspects of creating or testing or fitting models. Now we're just kind of wrapping up with looking at the inter relationships between active inference and a variety of other fields. Both fields from which active inference kind of grew as well as fields to which it can contribute and uh and those which fall within both of those prior categories. So in uh the introduction to chapter 10 we're uh described uh it's described to us active inference is a complete solution to sentient organisms adaptive problems which attempts to provide a unified perspective on action selection perception attention and even emotion regulation. Uh that list could continue on and on actually for those of us who have kept up throughout the book. Um and it's kind of reiterating you know we know that active inference is essentially a unified it's a normative framework operating from first principles brings back the old uh neats and scruffies uh argument or debate from the first chapter of the book where we are in the realm of the neats although we certainly would not uh deny a scruffy or two uh especially to welcome them into the fold if they so chose um in In section 10.3 there's a pretty inf interesting quote uh from Daniel Dennett the late Daniel Dennett of course and uh his suggestion to kind of model the whole iguana uh with active inference uh we we we do something like that in an attempt to create proxies of reality whether it be creating a model to simulate in silicone or it be to fit a model to a a real world organism or phenomena or otherwise where here free energy minimization acts as a kind of first a first principle by which we can derive implications about specific cognitive processes in their neuronal underpinnings. So active inference takes a kind of inactive integrative approach to things like preference realization, self-evidencing, self-preservation. And we're reminded of homeostasis or or as well as alostostasis or a kind of homeostasis of the future. Uh the idea that you know we have bound there's bounded rationality that that organisms and agents and so on kind of exist within their immediate context. They come with you know their their own histories…