Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

Hello, it's February 8th, 2023. We're in Cohort 3 and we're having our second discussion on Chapter 1. So just a few side notes before we jump to the questions. First, I'll place this in the chat, but we can... Hey, Maria. Are you recording the meeting? Yes, I'm recording it with OBS, like on my local computer. Oh, okay, I'm sorry. Oh, no, thank you for checking. I didn't see the icon. Okay. Thanks for checking. Yes, it doesn't mention it. It's like a yes. So last week we talked about NEATS and SCRUFFYS and those terms are due to Roger Schenck and it was actually last week that Roger Schenck had passed away, which I have just placed in the chat. So auspicious though tragic and people may find a lot of in this article by Harold Jarch, a lot of interesting thoughts on learning by doing. Teaching others to do this is the next step in the education revolution. And also there's a student's bill of rights that I had never seen from Schenck before. So there's a lot of interesting pieces here. So for those who are interested, take a look and maybe see how we can bring some of that back to the textbook group. And then also... Awesome. Thanks. Yeah. And then I don't see Yana here, but yesterday Yana and I went to the step-by-step active inference paper, which I'll also put in the chat. It's one that we reference all the time because it has a step-by-step building of like static perception, dynamic perception. These are passive inference and then action comes into play. And we used the ontology to describe what was happening and used LaTeX in the platform to write the equation and then made some tables that allowed us to flip between different ways of talking about what was given and what we're calculating. Wrote the pseudocode, made tables for every variable and interpreted every column. So this was just for the static example, but it was a lot of fun. And I'll also put that... Thanks, Jonathan. That sounds cool too. Yeah. So this was a lot of fun that we had and just there's definitely some walls that we run up against, like doing vector and matrix multiplication in the platform, but it's a fun direction for making really interactive and multimedia systems. All right. So let's go to chapter one. Does anyone want to give any thought or comment at all on chapter one? Any thought or question at all? And then we'll jump into the questions that the cohort has had on chapter one. It feels like a long while now since reading it. Could have been last week, could have been last year. Yeah. Feel free to raise your hand or write in the chat. So let's go to the chat. Any comment or question that's presenting itself to you today, or we will look into the questions that the cohort wrote. Okay. It looks like we may have done this last time. What is the relationship between policy and environment? Maybe anyone can just give a summary or thought on that, and then we'll move it up. Okay. Yeah. Very informative comment thread. Brock, go for it. And then anyone else? Um, policies are, um, selections on affordances in, uh, like latent affordances in the, um, environment. Yeah. Yeah. And then let's look at the comment thread. Generative process. Generative process. Stomach is empty. Stomach emits a rumble. That's a great example. Because the generative process helps frame it in terms of what it is, which is the process that passes observations to the blanket. This doesn't have to be outside quote of the organism. So the true status of the stomach is an external state with respect to the blanket states being interoception and the internal states being cognitive modeling of interoception. Generative process is just whatever passes observations. So it doesn't have to be the niche, but in the case of exteroception, vision, hearing, et cetera, packets coming into a server, then it is outside of the agents, but it doesn't have to be spatially outside of the agents. Jonathan? Yeah. You said that the generative process is the thing that passes observations. Um, presumably it…