Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

hello everyone it's October 14 2022 we're in meeting 20. we're in our second discussion of Chapter eight we have a few topics and points in chapter eight and then maybe we can preview some chapter nine and have a lot of space for any other thoughts or questions people are having so we can jump to the questions and look at the um written chapter 8 questions or does anyone want to suggest a point for us to go to last time we know that some of the areas we wanted to pick up on involved learning and hierarchical modeling and hybrid modeling one question that came up I think here but maybe we can revisit is the Birdsong model is widely pointed at as an example of how communication is modeled in active inference or for active inference entities however it's quite literally singing from the same hymn sheet and the turn taking is based upon uh joint expectation of that song being sung that way and then the actions that the birds can take the kind of turn taking that they enact is against related to whether it's like well somebody has to say it is it going to be me or is it going to be that so how do you think we would broaden that scope of communication to account for situations where there isn't a pre-scripted song how do other Frameworks for communication cybernetic ish or otherwise deal with the semantics of conversation improvisation open-endedness in conversation uh actually I had a somewhat more general question about uh the bird songs and its relation to uh this particular systems that they've developed in the textbook uh you see some researchers believe that I'm sure you're much uh qualified to talk about uh this than me but there are lots of research going going around about the territorialization uh function of bird songs so if we take that as uh if we take a birth song as uh as its territorializing mechanism at least one of uh it's territorializing mechanism uh can we somehow put the dynamic I mean the dynamic boundary of the markup blanket uh I think would be affected uh from this point of view and then I'm not sure how that would translate into this specific model if we take that Dynamic nature of the mark of blanket boundary yeah I guess I would just build on that and say that um the communication serves a purpose and so um I think a model shouldn't incorporate what that purpose is and how it's served by the communication and then whether it happens to be you know syncing up with some predefined pattern or or whatever is is a is a follow-on consideration the generative model in the singing from the same hymn sheet is like is very much like the sheet music example provided in the textbook there's a known sheet music script such that even when there's like a error in the playing of the music still the sheet music can be heard so just coarsely what generative models um I think it's pretty straightforward to see how a generative model could have the notation or the sheet music and then use the sheet music as its prior to kind of converge reality to what extent is communication that way is it totally disparate and differently structured or is it a little bit like a hybrid like you're doing inference on the sheet music that they're about to say and then you use your inference about what they're about to say to converge what they do say well you ask a good question which is well what are some alternative Frameworks and um I I know I I know I know some or should know some and I they're not coming to mind what comes to mind uh most immediately in this example is a phase lock loop in electronics which is simply a way of synchronizing one circuit's oscillation with another and um this is like a phase lock loop with a bigger vocabulary so figure out where we are in the in the pattern and then um synchronize with it and um but to the answer the question of what other Frameworks um I mean there's this whole field of um conversation analysis um which um well it feels actually called CA conversation analysis and they have identified…