Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

Hi, it's February 8th, 2023. We're in meeting 15 of the ACT-INF textbook group. We're in our second discussion in cohort two on chapter six. So last week, we looked at a few questions and in this modeling chapter, there's many questions and topics we can come to. So does anybody want to just raise a question or thought on six? Otherwise, we will look through questions and see what we can do. Yeah, well, I've put a few questions there that came to my mind. Great. Which one would you like to start with? So we already talked about this last time, this figure 6.1 on page 110, I believe it is. And where is it? Page 108. And my question was this vertical connection there between active states and sensory states. And I was not able to really understand why we would need to model this. And I've put this into the question there. And that chapter six, should I read out that question? So how is the mutual interaction between active states and sensory states meant? Can they mutually change their states without impacting neither internal nor external states? So that means they cycle around. If yes, what is it supposed to model? For instance, could they cycle around a thousand times, mutually modifying their states, and then eventually arrive at the state where they change the external or internal states somehow? So that means my question is, if they mutually are able to change inside these Markov blankets the states without any trace in the external or internal states, then they can catapult or manipulate themselves to come into a situation where suddenly they produce a trace. So they produce a trace. And if this is allowed, then I'm wondering what are we supposed to model with such processes. So because, I mean, you can explain any arbitrary behavior and is it Turing complete, this kind of machine that we are then trying to implement there? And basically, doesn't it undermine the explainability of behavior? um so I guess you got my uh my question but maybe I'm understanding something wrong there anyone else want to give yes please Ali and then go on I believe those uh bi-directional paths between uh active states and sensory states are uh simply there for the model to be uh more comprehensive and uh they basically refers to uh they basically refer to uh something like reflex arcs uh which don't necessarily need uh to pass through cortical paths uh so that's something uh much more directly manipulated uh but of course those um direct paths between active states and sensory states could be removed from the diagram and uh we we can talk about bypassing internal states or external states when the reflex arcs happen but I believe those direct paths are there to account for those kinds of uh intrinsic or some somehow uh those um uh the behaviors akin to reflex arcs thanks Ali but but isn't this isn't this then uh the question of what you define to be internal states if you say the internal states only start up from a certain point above the spine and below it's no more internal then it's an arbitrary decision right so you could just also say that everything that happens in the hypothalamus is uh not part of internal states and only what is in the prefrontal cortex is and I mean how um I mean this is arbitrary you know and if what is the reason to have such kind of arbitrary decisions there you could then put everything into into this direct uh interaction or depending on where you yeah yes actually uh it it is uh to some degree uh an arbitrary uh boundary because uh as we also read in chapter six uh defining the markov blanket boundary depends on the situation we're trying to model uh so uh it's not necessarily something predefined and it depends what behavior we're trying to model so uh yeah in sense in some sense uh it depends on the context uh and yeah but but see my point is if we are trying to model reflexes we would put this into the internal states but if we are not trying to uh i mean i mean if you allow this direct interaction between…