 Hello, everyone. It is meeting 13 of the Active Inference Textbook Group, cohort one. It is the final meeting of this section. It's July 28th, 2022. And we are going to be especially thinking about feedback and structuring on this textbook group. On how the first, half of the first cohort went, and also ways that over quite literally the coming minutes and hours and days, we will be able to implement structures to improve our experience for those of us who are continuing on to part two cohort one, as well as those who are joining for a new cohort. In the onboarding page, there are two columns with checkboxes. And clicking these checkboxes will sign you up to continue on in chapter six or ten for cohort one, as well as, or alternatively, to join in chapters one through five with cohort two. So select one or both of these boxes if you'd like to be continuing on. That's the onboarding note. And again, other than that, which I hope those who are here or watching live update according to how they want. Other than that, we're going to be talking about future textbook groups and other things like that. And then also just on the informational briefly on the future textbook group page. Here's the link to share with people to join future textbook cohorts. So to join in and then also please fill out this feedback form. It's really important that even if people think they have limited or non valuable feedback that they provide it, because whether it's on this page directly where everyone can edit. And so every thought that somebody has, like even if right now they can just edit this page, or in this more anonymous and asynchronous mode or through direct communication, but we're experiencing it. And so we have the best insight at this point into how to improve it and the affordances that we're seeing today or the state updates or however you want to use ontology terms to describe it. But what we're seeing today, we can do. So we should have a fun discussion today on some of those topics. And then also look towards project ideas, many of which are related to the educational process and others which are building on education into research and application. And so there will be spaces for those as well. Let's begin though on this future textbook groups page. And anywhere where people want to jump in writing or speaking, how are we going to summarize and distill the experience of the last few months in this textbook group and bring it towards improvements in our coming cohorts. Yeah, I think having worked examples as I've talked about and discussed with Brock like in Jupiter notebooks or like even like simple, like did you understand the material like question and answer like quizzes at the end of each chapter or something like that, I think which just would be useful like to gauge your own understanding like if there is like a definition or something that we can directly refer back to in the text, that'd be useful. Okay, worked examples. Yes. Great. And the questions could also include little tests of knowledge. Not every question needs to be throwing a baseball into the deep unknown. It can be literally clarifying for someone who believes that an answer is in the known unknown space. I believe there's an answer for this I just want to clarify for my understanding, but also once you've clarified that and especially if you did just clarify it, then write that example. Because many of these are opportunities for worked examples and working examples work in certain situations. For formalisms, for example, it might also work for formalisms that aren't mathematical. Like, here's the system this way or this phenomena, how do these ontology terms map? Or if this is the blank, then what is the other thing? So we have, okay, anyone else just go for it or let's just continue to edit and organize and add notes and just have more. Because many people are having many thoughts on directions to go. And I just like, what is the space of some improvements that we can explore? And then how will we be implementing some of these changes? And then in core two when someone's like, it could be this way and they thought of something that's new and not on this list, we add it. If it's on this list, we say, awesome. We agree. Join us and do it. Anyone else with raised hands or we can look more at a few other sections before going to the projects? Like, I see different regimes of attention. I think one way that we can structure that is by clarifying what practices were explicating and where those practices can take place. So then because the reality is and people are here or listening and replay can give their input. Many people may not have known what exactly to do to engage with the material. There's a textbook. There's some regular meetings. There's an information resource. And it's like an infinite game to structure beneficial engagement and where the rubber hits the road is like being clear about what practices people can engage in. So if we have math, like not everything has to be. Practical out into sub sub sub groups. Just including many of the practices. Like if you want to do art about active inference. Here's a practice and then if people are engaging in that practice, it does like create kind of like a group or a guild. Engaging in that performance. So however people think, like what are useful. Minutes and seconds that actually help learning active inference. And we'll be able to have a perspective on that question in years from now that's very different based upon better understandings of competencies and the fundamentals and the approach meant to the area. However, I think the practices table and providing feedback and then sharing and communicating how we're using practices to increase understanding could be very important for those who are learning and learning how to learn. Jessica and then anyone else. Hi, yes. I think to build up on sort of like blue suggestion. I'm thinking that on, like, in some of the courses that I've taken, you have like the other participants giving sort of like feedback as part of the learning process, you give feedback on like the other participants work. And so it could be that like there's an exercise and everybody has to like fill it up and etc. And then the other participants or like it could be like two or three. So not everybody has to do everything. So it's like, yeah, to the people you read like Jessica's like answer and give a feedback and, you know, so on. Because then that way by having to critique or like analyze all the people's work, it helps you think through the material itself. And so that's like another way of also like learning again. There's also missing like you can make you could probably like making a mistake also not applying it correctly. But it's like a tool to help you think through the material by analyzing somebody else's work. Well, also, you know, initially also having to do yourself and then hearing also the feedback that other people are going to be giving you. So I was kind of an idea that came to mind like making reference from like all their experiences. Thank you Jessica. Yes. I think that's a great area to keep in mind and to balance. As our cohorts build and so on, like who's on the supply and who's on the demand side of that. So to speak, like who is actually offering the feedback. And then who is the people who are wanting feedback. Like some of these projects are things where the person was able to be brought into another project. Like the like Noah's and Mike's work, bringing them into synchronous and asynchronous resources that are already happening, provided them with at least a first step or continuing on their journey. Others are there in different areas and so how to keep it open when people have ideas ranging from educational improvements to robotic deployments. What kind of feedback they need and then how to balance like the excitement around applying active inference with what it means to learn. Like there are biochemistry classes. It's not all just biochemistry lab. But who's to say that's even the best way or anything like that. It's just we want to make sure that we keep it all in mind. Jessica. Yeah, and I think this suggestion is not connected to someone's project or net 16 project is more like an exercise. So there's like, you know, short exercise that it's going to be testing, you know, variational inference or something like that. And then, yeah, like every participant in the class go actually does the exercise and then is assigned to other people in the cohort and to analyze that person's submission. And so it's like, you know, three people in the course like analyze like Jessica's, you know, submission and give feedback. And like, you know, it's going to be like for everybody, everyone will be getting three, three feedback on like how they apply the terms or like, you know, there are other ways that they could be done, etc. And I mean, if some of the things are like math related, like, I will have to see and because then the answer is going to be more clear, it's not going to be a subjective. But it's more like, you know, there's a short exercise. Everybody does the exercise. And then some people then get to review that. And, you know, that answer of the person and give feedback on that. So it's like separate from an actual project. Okay, makes sense. It's kind of like a combination of having a good example questions with an organizational structure and then like you said balancing like where there might be a clear measuring stick. Like, was this the right equation or the right calculation or not versus how did this usage of affordances in a new context make sense to you. And also it's not just about getting feedback on the person's idea. It's like the practice of providing constructive ideally written comments is super important. Like the secret side of the questions is the discourse where the kinds of things that questions and ideas that send people on thought excursions and research rabbit holes. We can collaborate on that response together. And it doesn't have to be the end all be all, but sometimes by putting like a seed into the workspace by putting like a prompt or a question. 10 people could write for five minutes and then there would be 50 minutes of writing on that topic from different perspectives. Whereas if there's 10 people also there could be five minutes of discussion, which might add more seeds onto the table and it might continue to invoke whimsy or it might continue to instruct if there was a broad uncertainty that the five minutes of discourse resolved. Yeah. I think like for the questions on it's like somebody has like a question that they're trying to understand and everybody's giving it the crowdsource answer and helping you know, sense make on the meaning of how things are with the exercise. I think similar to I think what blue wants to do with like the short like exercises to test whether people know the concept or not. It's just like adding like by having people critique all the people's work and having to like critically think through the answer is another level of learning. Because you it's not just like oh I think this is the answer to this is like now I really have to understand and want the concept that I'm using to evaluate an answer and then also evaluate that person's answer and kind of compare and contrast. And that like having to make that exercise of critical thinking. I think helps people learn like in a more in depth way because you have to think through your own thinking process and like what things mean what somebody else was saying etc. And yeah and I've seen it applied to like writing. I mean I guess it's maybe like easier to like writing courses and things like that. But yeah it's like another way of. Yeah, like processing through the information and and then also. You know so maybe like reduces the burden or maybe like a person having to like go through everybody's answer and make sure like yeah like I am the teacher correcting everybody's answers like. It's sharing the responsibility around everybody in the group of. You know like kind of answering you know like evaluating people's like responses to an exercise. Thank you Brock and then anyone else with the recent. Just to agree with Jessica and the last part with what I was going to add that but yeah that just systematically created creating a systematic affordance for people to. Review or. Provide some. Validation that it's correct or on the right path or whatever. Will reduce the burden there and. Teaching. It's probably the best way to learn. It's learning how to learn is it's like difficult to do without teaching like learning self learning is about teaching yourself. To get better at teaching you get better at teaching yourself. Which. Just the circle that does kind of require some meta structure to like. Create examples or source them or whatever. And then provide them. So I'm not sure. Outside of the. Project that I put on there and. Maybe it's kind of related to Jacobs as well but outside of those two things I'm not sure we have a clear. Affordance to do that yet so. Or come up with examples of. Yeah like examples to work or whatever yeah like sort either sourcing them and then putting them in a structure that says oh this is for. Section the chapter one or whatever one point one or whatever. This is a not questions that are related to specific parts of the textbook with people to work through but what kind of table or structure. Could be I guess virtually the same but instead of a question a question somebody is asking it's an example. For other people to answer maybe with an answer. That they can see like that they want to reveal it or whatever but these does that make sense. Yes yes. That's that's what it could be an example to work through and discourse it doesn't be in this exact date but just make sure that we have written it down. And everyone can be contributing to like write it down and what the structure is because it's not like this is a impossible affordance. We can spin up the table right now and template it and just start to design it and over the coming weeks we have time to digest all these different inputs. To make it so you know in a few weeks. It will fly by but we might be able to be implementing these things. I wanted to ask what practices did people feel like they engaged in rootfully, whether it's on here or not just when you were participating in the active textbook group. What did you feel like you were doing and how did it help equation translation that was like the most helpful things that I did. And I don't think it's there. Well I guess it's kind of there. It was like to copy the equations down but like putting them into natural language was probably like the thing that I did that was maybe most useful for myself and others. Yep. It was a great practice and it leads to in scaffold so many other why is it this way what does this mean what are the examples it's not like the prerequisite per se. It's possible to ask questions without having the natural language description. But whether it's a basic or a multi line equation. It helps bring conciliance to that discussion. Because there's a non zero probability that then a term is going to be used loosely or it's going to be used one way and then it's going to be conversational use slightly differently. Or people will not agree on which ontology terms are invoked or not. Which can then have every equation start to spiral off into every possible ontology term. And they very well might actually connect that way. But being able to see what is variational free energy. And if there's some term that is being connected to variational free energy that's not being described in the equation. That's fine. Like least squared error isn't in the linear regression y equals MX plus B. So there can be formal and informal concepts that are super important for understanding. That aren't with a symbol. But this was a great way to be clear what the symbols actually are. What other practices did people feel like they benefited from? What would they recommend for a daily schedule for somebody who wanted to learn a lot? I think plus one there on blues. Like writing the equations which is like maybe a super simple but open-ended almost kind of version of Jessica's suggestion. But also in general just adding stuff to the coda. Asking questions. Like if you read and you don't ask a question. Did you really grapple with it at all? I don't know. How do we improve thinking by writing in the actual number of typed words that people add in as a proxy? During a discussion is one thing but even just afterwards or on their own timeframe. How can we have 10 people who do read a chapter just scaling numbers? Each of them could think of several questions. Ranging from simple I wasn't sure about to simple again like more like an example one. There are the three kinds of this and it's totally known and their answers fully listed to research your questions. But how could we have many more inputs in terms of the questions that people are asking? Do people feel like they wrote questions in their own notes but didn't bring it into shared workspace? Or do people feel like for themselves or others that they observed that the questions weren't actually phrased? Jessica, then anyone else? I was going to share more about the student practices. But I guess like the last question you mentioned. There were things that I didn't put in the general textbook thing. Which were more about passing information and which was like helping me connect the dots and synthesize information. Which Ali actually helped me with some of that. But those things were sort of like personal for me to like try to make sense of the material. And in terms of the student practices I like initially I was thinking to do like every day. Like an hour each for like the reading and maybe like an hour for the math to kind of get foundational things. But then I found it was like better for me to maybe dedicate like a longer period of time on one day or two days for the reading. So I was supposed to like a little bit of the reading or like chunk more on two or three hours in one day. And then maybe like follow up another day with like a longer period of reading as well instead of daily reading a little bit. And I think that's more because I guess for memory. I mean they do say to do like repetition for memory. But I think like for me it was like well read longer and you know have like the material fresh in your mind. So like reading as opposed to going every day. And then like for the math because I guess there's less of the you know keeping the material fresh in your mind. Like yeah like daily I was trying to do that. So then it fell off a little bit on that part. But yeah kind of at least one every day like four days a week at least like an hour or so like trying to focus on the math. But yeah that's kind of how my practice you know got you know develop over time through the course. So the textbook is for as the like the research people who want to apply in their research if we remember like from the introduction. Provide readers with the ability to understand and use active inference in their own research. We'll have a better sense after part two with a more applied side. Who might the next cohort of the textbook group be in and communicate to yes. Over many iterations and you know can go many ways but when people are thinking like in the next few weeks. Who blew wrote OK. Can we review which questions and understand. Because I don't quite understand how. Like so if we go to the like if you want to check out the questions check out like chapter two like and a lot of the questions were so if you scroll down. Keep scrolling keep scrolling like I'm trying to think about. Which ones that were skipped let's see. And also there's no skipping there's what we can address in the one hour or two hours. Right but I mean like we can raise our attention to all the important questions. Yeah it was just like and then I realized like most of my questions were like of a largely technical nature. So let me scroll through the code and see if I can find them but it was like in chapter two it was like let's see. Like the general thing to be is how can we among other things. Prevent question self censorship and filtering. The question isn't just like this one like genius meets genius perfectly framed specific answer that will be understood in one sentence. But that question is like a broader space of prompts that don't necessarily have any prior on whether it has to do with an application or it could be a technical question. It was early it was like in chapter two and chapter three and there were like a lot of questions. I'm trying to like look at the specific I'll have to look at my own code instead of watching you beautiful people but it was. And like it just happened early like in those meetings and it's not that there wasn't time like that we didn't address the questions in order it was like let's skip this because it's technical let's skip this because it's technical. So I just stopped putting my questions in because I feel like they're technical like what does this little squiggly symbol mean and that's why like for me the translation of equations into natural language was the most useful thing because I feel like all of my questions are technical like related to the math. So I stopped entering them into the code because they were just skipped over. Okay, thanks. Well, you appreciate it. Again, I personally hope that we can have structures that uncouple the perceived importance of real time speech from the contributions that people make. So that we can be really clear about like which questions we're going to be addressing a conversation, but and use conversation for what it can bring. Many things. And then have 200 questions in chapter two with many upvoters or with whatever mechanism that brings different questions to the for so there isn't even like the plausibility of being skipped it's just a question of like what will we raise up and focus on individually and as a group. From the many, many questions of different kinds that support a beginner with fundamentals and a pre beginner with background and somebody who's asking research questions as well. So I think one simple thing is if you go back to the questions table like you have a interest column here. Interesting. Yes, even better. Maybe there is a what we will address in the net coming live stream or a session kind of table that is taken from the questions table. But the questions table has no such like projection of what is interesting or not in some especially in some one dimensional kind of like yes or no. It's just a separation of random things prompts questions. And then there is some process to get it to what will what will be addressed in the live stream that again there's ways to like try to do things like randomization or trying to like ensure that you know we don't continuously ask. Focus from one person or things like that. The popularity doesn't kind of win the day. There's more complex ways to do this that are kind of almost like decision making processes like. The femoral group processes and things like that, but I think just separating them. It would be helpful there at least. Another thing maybe like zoom you can do breakout rooms. Maybe that's not the format, but again just that that's kind of an ephemeral group processing like perhaps everyone doesn't want to focus on the same questions. Should there be temporary temporary split of regime of attention and that for some short period of time and then we combine. Don't know what that means for a live stream or recording precisely, but that's possible solution. Yeah, thanks. Gather is awesome for these things, but it does require clarity around how it's going to be done. Like mass one on one breakouts mass smaller great breakouts and rearranging people using their spatial location to express different things and assemble into different ephemeral teams. Okay, let's look at some project ideas. Okay, so for those who are here and I hope I'm not misrepresenting anything. Mike brought up some token economies that's something that we've been actively going towards in active block fronts. So I'll just add just a note column in active block fronts as well as as mentioned earlier. Working with Noah on some more some related topics. Okay. Audio book recording of the textbook just as we go through these projects. I've recorded the first chapter and the preface type of materials. Do people think that an audio recording has utility? And if so, is anyone interested to pursue this avenue with blue and I because it's very defined and it involves primarily the equations having natural descriptions and then just reading. And it can go many other directions too but this is like kind of a clear accessibility affordance for different participants that might be interesting as well as just get out there on like YouTube and other channels as having that artifact blue. I do think it's useful obviously if I've elected to I didn't realize I signed up to be the participant lead but I'm so happy to do it. Co-lead, but but I do think it's useful as someone who drives like a considerable amount of time like every day. If it's not like I'm not a very good auditory learner but it does help to like reinforce the concepts and I just wanted to mention. Even if like you're not comfortable perhaps like doing the math or reading the textbook. Daniel thinks he can read it until it's perfect. I would prefer to do some video or audio editing. So that's also an option is to do audio editing like where he's like you know you occasionally like stutter. Daniel doesn't very infrequently but I but there's the occasional tripping over a word or like saying a word twice like when you lose your place or something like that. So video or audio editing. I do it with Windows video editor but like audio editing is also another possible affordance there. And it would involve like you know read along with the textbook to make sure that it's accurate that no sentences are skipped and then to just splice out any like mishaps or just ask him to re-record the you know Markov blanket for precision just so you can place it over the trip trip up if needed. Yeah thanks definitely we would do a multi part. I just did one pass on the first chapter no equations etc draft but okay anyways just so people who are interested here or rewatching can know about this okay resources for conducting research or developing active inference agents ways to facilitate the starting of that work for participants just a seed idea. A lot of this is going to be explored in part two because it's a recipe for designing the models and we'll have a lot of infrastructure around that as well. So I hope whoever wrote this continues in that interest and builds that starting point for conducting research or developing agents based around what they learn and see in part two. Okay. Okay. Also typing just directly in the code is a little better than comments because sometimes it's hard to understand what the comments are like referring to like okay complete mathematical overview of active inference. Derivations of key results maps of equations. Yes. All awesome thank you Yaka for suggesting this. I think the closest related works are in the active ontology and knowledge engineering and other areas to. And I hope that we're on this path. So however, Yaka for anyone else wants to connect on that. Okay, active foundations. So here we added this page. If Brock or anyone else wants to add any comments or describe how some of these comments can be integrated with textbook group. Short answers I'm not entirely sure the precise way to integrate it with the textbook group. But they're just perhaps this is related to questions making questions, removing the stigma or whatever. Like it's just helpful I think to have some affordance or regime of attention or whatever thing that says, hey, if you don't understand this thing about active inference, that's a foundational thing like okay I get entropy but and I get energy like the basic concepts of them but like when you're subtracting them. What the hell are we talking about you know and then maybe there's you know some follow on questions about mathematically what entropy is. Mathematically what is energy or you know what is Shannon entropy specifically what is the difference. These sorts of things. A it's that and affordance for that but B it's also just acknowledging like there's a wealth of knowledge already here. It's not a complete mystery what those background foundational things are it's or what learners might want to get out of some relatively short course to like get up to speed and feel real confident reading the textbook. And so there's a kind of top down affordance there too I think that needs to be made for people that can say hey there's a bunch of equations over or examples over here that are these are the most helpful ones I've found for understanding what Bayesian inference really is how that really works. So I guess it would be those two things but if you look into the project notes there on that project idea. And it's there's a little more detail of them. Yeah just a couple bullet points there of just it's really hard for people I think to get into a new thing if there's not a model. If you do chemistry without Bors model like good luck. If you have to jump especially to something like electron probability clouds or something that's just like why would you do that when you have a perfectly good periodic table and simple discrete you know and it's enough even to do basic physics like Newtonian stuff. Right so need a model like I think a lot of people history is helpful there very minimal amount but just this is where this came from is why it's relevant. It's really hard for people I think to blindly calculate without that sort of simple context. So just kind of made a very rudimentary still kind of scaffolding it. Coda it's already linked in that project row but I'll put it in that sub page there to super work in progress. Yeah I mean the idea is just there's these questions are going to be asked over and over and over again. There will be people coming from a more mathematical more physics a more neuroscience a more from different backgrounds. And so why pretend like that's not happening or reinvent the wheel every time somebody asks a question like just having a central resource where people can get a lot of value but also maybe ask a question. That's a little more they might not want to ask and seem like it's too simple or it's too whatever in the live stream. Okay thank you. Any other comments or thoughts on this otherwise we'll just continue on for the final minutes. Okay. Textbook enrichment. Yes I hope everyone continues to annotate figure out also things on their own and explore and like we're building something really powerful with annotation. So I hope that people can continue to be enriching with the structures that we're building. But yes, so much of what this group has been about has been about textbook enrichment and improvement. Robot modeling. Unless anyone has other comments I think learning from JF and staying in part two will help. And yeah, Jessica and then anyone else. Sorry I think I didn't raise my hand from yours. Yes. Well, any general comments that people want to provide on the textbook group? I think the live streams are super important for the lab and the whole space for maybe obvious reason the textbook group is equally important. These are like the two most important things that can add value to the space and the lab participants. Indeed glad that those who have made it this far see that and over the right time scales in the right way. We see more and more vistas to increase the accessibility and rigor of active inference. So I hope we can hold that frame. And focus our attention. To provide these functionalities. Because if it's stateable, it's doable. In a way. And if no one from the first cohort wants to be an active facilitator or steward for textbook group to some of these changes will be integrated. The more people who want to step in to steward. In our participatory context. Are going to be leverage points for so many more people who come through. So there's like multiple levels of affordances for those who see them. Especially at the end of this section. And it's been quite a few months. Any last comments? Okay, I'm going to stop the recording in three seconds unless somebody wants to say something. Okay.