Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

all right it's our second discussion in cohort 4 of chapter one uh does anyone at the beginning want to just bring up any comment or any specific question about chapter one or otherwise we have we can look at one uh recent stream transcript okay so Terry um you asked a question about the generative model the generative process so maybe just summarize that question and then and then we'll see how it played out recently uh my question was when we were trying to create a model we create the generative model we sort of Define it in four over four sort of domains and and and then at then you define the generative process but my if if we you know think of the Markov blanket as separating B internal state from the underable external State well if we're creating a model and that model depends on us creating something that is in fact on mobile or is there not a corruption within that thought process so it creates an impression that within the context of if I take my Markov blanket is bad my skin then um I uh I cannot know what is the generative process so there's got to be some flaw in our system of modeling if it depends on creating something we can't do all right awesome and uh when you when you asked that it made me think of the recent um stream with Maxwell ramstead now operation and Ali and others and uh this highlighted a uh inconsistency in the generative model generative process uh topic Ollie do you want to summarize um what you what you asked and and what they wrote or said yeah of course uh well uh we were talking about uh the I mean discrepancy between Maxwell's presentation and also the recent paper because uh in figure one in inner screen paper uh they've indicated generative model as encompassing the whole shebang I mean encompassing the internal States the external States and of course the blanket States so um nowhere generative process was visible in this diagram so we were discussing with Daniel Sanjeev and others where exactly uh General to process can be manifested uh in this diagram so I basically asked this this question where is the generative process in this diagram and uh surprisingly Maxwell uh somehow disowned the earlier literature and generative process and basically uh I mean uh he he he just I mean interpreted the generative model in a kind of new more sophisticated way that wouldn't require any separate entity as generative process because General to model basically includes the joint probability of every possible of uh every everything that markup blankets um uh I mean partitions so it it is a joint probability of internal States blanket States and the external States uh but on the other hand when we want to uh model a situation and not just I mean describe a physical phenomena or a physical system then of course we can Define our own generative model as something or let's put it this way as kind of dynamics that we expect the system or the situation the situation we have to follow those Dynamics and so for instance if we want to model a very simple linear behavior of a I don't know mechanical system then of course the Newtonian mechanics or you know the terms the lagrangian mechanics of the system can constitute the generative process of the system so uh yeah that's uh basically the uh gist of Maxwell's argument uh at least as far as I understand it quite a discussion we can kind of parse it out better but the transcripts out there like so um as with everything in the textbook it's sort of like there's a there's some internal ontology and then um it's somehow aligned with the broader active Gestalt and for that you will still it helps to understand the generating process and the particles of the agents and so um like to to the to to the earlier question how can you model the unknown because you're modeling something into existence in the act of modeling so you can have the real temperature of the room do statistics on an unobserved that maps to the observables so in practice there's no issue but then…