Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

Hello, it is April 5th, 2023, and we're in our second discussion of Chapter 5 in Cohort 3. So this is going to be our final recorded session for the first half of the textbook, Chapters 1-5. And then next week, we'll have an unrecorded session. We will have time to fill out the feedback form for those who haven't yet. We'll discuss project ideas, which people can also add ideas to. We'll just talk a little bit more broadly and informally about the textbook and about the textbook group. And then we will have an inter-semester break, or it's not a semester, but an intermission. And we'll pick up when Cohort 4 begins, we'll pick up with the second half of Cohort 3 going through, just as previous cohorts did, Chapter 6, 7, 8, 9, 10, and so on. So, we're in our second discussion of Chapter 5. I'll head over there now, and we'll begin just with anybody in the chat, or via raising their hand, or just unmuting. How does Chapter 5 sit with you? What are any thoughts or questions about Chapter 5 that you want to just raise at the outset? Chapter 5 Jonathan? Yeah, I sort of feel that while, in some sense, it says that you don't need to have digested Chapter 4 to understand Chapter 5, it still feels like there's a lot of background there, which is really important. And the lack of technical clarity from Chapter 4, I sort of feel there's a bit of a hangover from that into Chapter 5. The technical debt hangover. Yep. Yes. Okay. Thank you, Maria. I was a bit frustrated with this chapter because I thought it would be more insightful, but it seemed to me it was just more of the same. Bunch of equations narrated and plotted. Yes. I guess the apologist's view would be Chapter 4 was equations and plots in general. And Chapter 5, in contrast, is applying it to the specific mammalian nervous system examples and to empirical studies. However, broadly, I totally agree with that. Any other thoughts or primary reflections on Chapter 5? Any aspect of it or some part of the discussion or even any other active inference learnings over the last week or two that come to your attention when you're in this setting? Okay. So, again, raise your hand or write in the chat, Terry, you wrote, I was struck by the similarity to neural networks. You could describe a little more if you want to unmute, or we can see how that might come into play. So, in the previous chapter 10 discussion from cohort three, we were playing with the idea of reading chapter 10 first to provide a lot of the conceptual scope and the historical connections and similarities and dissimilarities of active inference with other topics. And so, it's interesting to note, neural network is described in the citations and twice in chapter 10 in this paragraph. But there's a lot of context that is kind of held out till chapter 10. But the intuitions that you're having along the way are very good. Any other just general thought or a question on five? Otherwise, we will look to the questions and address those which we haven't addressed previously and then continue on through. Okay. So, continue on this neural network just while we're here. Let's use the figure 5.5. So, here we have 5.5. Oh. The cortical system. The limbic or the midbrain system associated with policy selection. And the spinal motor execution arc using this differential between perceptions and set points. And so, then, Terry, you wrote, the series of nodes passing information along a chain with inhibitory mechanisms alongside. Yet, neural networks use transformers while active inference uses free energy. Well, great topic. To go deeper into the relationship between neural networks and free energy principle and active inference models. In livestream 51. I'm just going to the slides now. The core claim is that here on the right, we see the Markov blanket and the internal and the external states. Sense coming in. Perception. Action selection resulting in embodied action. And an effect upon the generative process, the niche, which then…