Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

all right welcome thank you everyone for joining it is the first early session for cohort 7 in this second phase of activity so anybody who uh wants to sort of just check in with uh that I guess we could have done before starting recording but just want to check in where are we at in this cohort as we head into the second half of this textbook in people's view or what would they like to do for this phase of activity for the kind of second half for this cohort and then we will turn to chapter six um maybe ali uh and then anyone else who wants to raise your hand or just go for it uh okay so uh chapter six actually marks the um uh beginning of the second part of the textbook uh as we probably already uh know the book organized into two different uh parts so part one is mainly theoretical and part one part two uh basically deals with more practical applications of active inference uh so as as we saw in the first part of the uh textbook um the theoretical Foundation of active inference is actually work in progress so uh it's not something uh that that has been uh uh completed uh in order to applied uh as a kind of uh comprehensive uh modeling tool but at the same time uh the second part of the uh textbook provides very promising road maps uh to um towards how we can apply even these uh tentative foundational work um in order to build uh some real World models of the intelligent agents or cognitive agents uh so I think that's important to keep that in mind because uh in many textbook groups uh I've seen people uh somehow uh rais this issue as to well it's not as comprehensive uh as we thought as as a modeling tool so uh since the public a of this book U obviously active influence research has progressed a lot uh but even the uh the theoretical foundations provided in the uh first part of the book uh I believe provides enough material to um at least build some uh Toy models uh that can capture some of the real world element of a real um a real world agents or at least cognitive agents thank you Ali great anyone else want to share with their ad as in this uh as we head into this part of the textbook yes so uh I think I would like to do that um I missed the first half of this um group uh so I'm I'm asking a question which belongs to chapter two I think or chapter three I don't know um and so my apologies for that and I don't want to hijack the discussion here uh and I've put my question also in the in the chat but basically comes down to um having tood the role of PR in um active inference as having a dual role which is one kind of the intuitive role for me at least being vely familiar with bence inference is you know encoding prior beliefs and also um having a kind of indicator role of encoding uh preferences and I think this comes out of uh discussion of um ocular motor Behavior as well as other things and I was just wondering if you have discussed this at all and uh if I've yeah if a brief comment on this would be great but if you say well you should have you know attended the previous sessions and I'm fully happy with that as well and then we'll just tag along for the right for now yes um there's a slide in this the point raised in this paper um that points to Prior being used as a um preference to drive pragmatic behavior in lie of for example what a reward function does for reinforcement learning because when things are framed in terms of likelihood directly rather than projecting onto a explicit rewarder utility uh function only just working with a base joint distribution um then the prior can be seen as preference constraint so it's like the preference for blood sugar has a different connotation and and um function and everything than one's felt psychological preference for something or revealed or stated preference but um at least um from a modeling perspective it's more similar than not okay so I'll I'll have a look at uh those reses thanks so okay Jerry yeah I think as as we're going into at least my area of interest and my…