 Hello and welcome. This is active inference or extreme 6.1. It's January 22nd, 2024. And we're here with the d20 governance team and project. So I will let them all introduce themselves. This should be a very fun and interactive session. So thank you all for joining and looking forward to your presentation and the discussion. Cool. Yeah. Thank you, Daniel. So I'm set hosting and I am a community lead at the meta governance project and one of the developers and researchers with the d20 governance project. I'm here with Hazel and Janita and I'll let them introduce themselves. Yeah, maybe Hazel and then Janita. Hi. Thank you for coming to the live stream. My name is Hazel. I'm affiliated. I'm kind of in a medical community environment atmosphere. I am political theorist. I'm a researcher with the 20 Before I pass to Janita also want to highlight another of our team member who isn't here. Well, it's her birthday. Happy birthday. Well, Hi, I'm Janita. I'm a software developer. Have been involved in the doubt ecosystem working on some projects around dows in the past and this project is my main sort of interaction with the medical community but I'm generally pretty interested in all the work that they're doing and Yeah, research and governance, etc. Cool. So with that, we're planning to do a presentation today and then the presentation will also have some interactive demos for people who are joining us in the stream. And yeah, so without we get started, I should also say before we get started that I want to thank Daniel for inviting us to come and speak at this. I've known Daniel for as long as I've been doing the community lead work, medical and have had a lot of really interesting discussions with him about cognitive security, Stigmargy and governance in general, less familiar with the kind of resources happening at the active and friends Institute. So really glad that Daniel and their community was interested in having us come and join so that we can present our work and learn from the insights of this group of researchers. So with that, let's go ahead and get into the presentation. Great. So before starting, there is this discord link and actually it's not in my clipboard so I'm going to generate a new invite. And then I'll share it in our chat doing zoom and Daniel can share it. And then you think whatever way is applicable. And then, yeah, so this is going to be a link to our discord server where the bot for this project is hosted. It's the D20 governance server. And we're going to do some demos today. And so if you want to participate, this is the place to join for that. If you join, come in and say hi. And yeah, so with that, I'm going to pass it over to Hazel who's going to give us a bit of an overview for today's presentation. Yeah, so the flow of this presentation today will be some demonstration of the functionalities we built sprinkled with some conceptual thinking. So we're going to start with giving you guys a background on where this how this project came to be and we will introduce our first functionality. That's D20, where we're going to be interacting and we will demo the cultural modules and talk a little bit about community of rule, which is where and how this was inspired this project. And we'll talk about two more functionalities that's value group voice, actually three more and continuous input. And then we'll end with a Q&A and discussion. So moving on to the next slide. A really brief overview of this project is D20 is a Discord bot that allows communities to play governance games in a AI large language model mediated environment. So groups of people can come together to embark on a governance, what we call governance quests and experience a diversity of governance structures, decision making processes, power distributions and cultural dynamics. Through building the Discord bot, D20 aims to help lay the collective decision making foundations necessary for online communities to conduct their own governance experiments. Do governance education, do future scenario planning and make real life decisions while bonding communities through play and a sense of playfulness. I'll pass it to Janita. Yeah. So actually sent if you don't mind going to the next slide, I think there's a little gift. I wanted to reference. Okay. So the original inspiration for this project was a tool called community rule, which this gift on the screen depicts community rule is basically a templating tool that allows you to define a modular stream. It's a structure for your community's governance. So there are four types of modules that you may select and combine culture modules, decision modules, process modules and structure modules. And community rule basically allows you to combine and create sort of nested hierarchies of these modules that you can then export into a template format. So our original vision was to basically take this, allow someone to export this template and then import it into our Discord bot and have it just work for sort of your community's governance. But we also at the same time wanted to kind of have this like motion of gameplay or allow people to experiment with these modules in a like very low stakes manner that you could then, you know, move forward and take into higher stakes situations like actually using these modules to govern your community. So the idea of low stakes is kind of where the governance game idea came from. So we started out, you know, developing a more fully fledged kind of game experience and we were trying to design a narrative that took people through a series of different decisions that they had to make as a community. And we're trying to be pretty honest to the original like structure of community rule with respect to these four different types of modules. But something we realized as we went along was that like game design and like making a really kind of interesting and exciting gameplay experience is like a hard problem in and of itself. And we're kind of trying to accomplish a lot of things at once we realized. So we decided to pair back the game structure a bit and sort of focus it around what we call like these mini quests or governance experiences where people in the community can kind of like still experiment but in a in a sort of like on a shorter timeframe and with less narrative overhead they can still experiment with these governance modules. So we'll get a bit more specific about the actual structure of the game later on but that was kind of the origin story of the D20 bot. One other thing we'll mention is that as we were like developing this and doing play tests with different communities. We got a lot of feedback around the sort of this LM mediated communication feature. So one of our original ideas was to implement these like implement the idea of culture models but in a more wacky manner by using LLMs to modulate how people are communicating. So, you know, for example, we have like an eloquence on types in the discord is translated into Shakespearean English. And we found that like people were finding a lot of delight in these culture modules. And so we started thinking more about, you know, how this game can allow a lot of people to experiment with communication and LLM mediated information landscape or like how even the frustration that you experience when trying to communicate amongst sort of this like chaos of information can be interesting or can can teach us something about governance. Yeah, next slide please. Actually, I'm going to just go back to the other side really quick because there is something I wanted to add on to what you were just saying there, Jeanina. Something that's also kind of interesting about, you know, modeling the kind of communication system based around community rule is that the kind of implementation of the different kinds of culture modules can vary quite a bit. So if you look at this, actually, well, you can't see because it's not on culture right now in the GIF, but in culture there's a values module and we'll talk about that a little more. But one of the interesting things is that that's a quite broad category for like structuring communication as like mediated and filtered through a language model. So some of the modules are very narrowly defined and some of them are more abstract. Another thing that's sort of interesting is that this tool is kind of originally designed to basically help communities like define the way that their governance actually is rather than necessarily the way that they want it to be. It's more of a retrospective tool rather than a prospective tool. And part of the reason that it's like that is because the community rule itself is not actually programmatic in the sense that setting up these structures doesn't do anything. It's not connected to any APIs or the web app doesn't even do anything other than just represent. And so kind of thinking through how to translate this into something that's programmatic was actually a really interesting exercise. And it's worth saying that too because one of the things that kind of was also part of the impetus for this work was some of the past user research that we had done with communities who were interested in incorporating more governance tooling into their suite of online tools. We validated in particular, did a lot of user research with community participants and leaders while we were doing some work on both MediGov's gateway project and also policy kit and now collective voice projects. And one of the things that came up a lot in those conversations was the need for some kind of primer or some kind of like accessible way to even develop the kind of governance muscles or impulses that an online community would need to really make use of the tools. And community rule is actually, at least in my opinion, a really great tool for introducing communities to what a very simple kind of representation for what governance could look like. It could look like the drag and drop web app and the kind of simplicity of like seeing your structure represented even though like you can actually get these very strange kinds of like recursive structures. But you could have values inside of a board inside of values inside of a board inside of values inside of a board just kind of infinitely recursing. But it's a really accessible and approachable way of actually seeing what it means to have a structure to make something that is often implicit. We heard that there is a lot of this need for making governance more understandable and tangible. So that's part of the reason that we were taking community rule as a foundation. And then we also heard that there is this desire to something that is more explicitly interactive. A lot of tooling can sometimes be set up a poll and then let it run. And you don't really get any kind of interactive feedback as a result of participating in that process. It can feel like a lot of online interactions where you do something, maybe you see like a react number go up or down. But largely speaking, you do something and you kind of don't even know if what you did had any impact. And so this idea of kind of creating a way of introducing people to online governance through play seem to be something that was coming up in our research. Janina talked about this a little bit, this sort of in terms of the kind of conceptual development of the project moving from a kind of general governance tool. And even a game to thinking of it more as a kind of surface for governance experiences. And I think this is maybe important to lean around just a little bit because there's something about having a good experience with governance or a meaningful experience with governance that can really change your whole relationship to why you would even care about the topic. And so even though like as we'll kind of get into our game is our kind of governance experiences are designed in some ways to kind of be reduced confusion. They're also designed to kind of stimulate the imagination for what a kind of governance could look like and still have some fun in the way in order to basically translate the experience of playing and to thinking about implementing these kinds of systems in your communities. The other thing is that for any one of the hard things about designing governance scenarios, games, environments, etc. is determining what the resources are that are actually being governed at all. It's very nice to kind of imagine this very abstract governance scenario. But if there are no stakes in terms of resources, then there's really entirely true, but it's much harder to kind of devise a governance interaction space. And so, you know, I think we started and maybe we'll pick up a little bit on this and some of the kind of prompting discussion later on, but we started with this idea of kind of living in a kind of like post communication collapse type environment where people for whatever reason still needed to communicate over the internet, but the ability to do that had degraded so much that communication itself was this kind of work. Clarity of meaning was a scarce resource. And the other thing that we were interested in was thinking about how to kind of like produce that kind of scarcity or that kind of fleeting sense of meaning. And the way that we use culture modules was one way of kind of artificially placing people in these kind of abstract communication environments. So yeah, let's go to the next slide. Janita, we'll talk a little bit about the D20 Agora. Yeah, so what you see going on here on the right is the D20 Agora. So the game is basically broken up into two modes. There's the Agora, or sorry, the bot rather is broken up into two modes. The Agora is what we're looking at. It's kind of this asynchronous environment where all of the culture decision, et cetera, modules are available to experiment with, but there's no kind of active game or narrative going on that requires people's synchronous attention. So we intended the Agora to be an experimentation zone and the jump off point for the actual games, experiences, whatever you want to call them. And we'll talk about that second piece in a bit. This is basically just a kind of gif, but actually because we're doing the stream this way, it's basically redundant. So let's go to the next slide. So we're going to go into a little bit more detail about the way the culture modules are implemented and designed. We'll start with Hazel and then I'll pick up a little bit after you. Yeah, I think culture module is one of the really interesting functionalities that we build and it's the way through which we mediate kind of all messages and communication through the switching on and off of culture modules. So to give an introduction, I think culture module is one of our most notable features. And it works to influence the group's communication norms with what we call, we picked up the term culture modules from the community rule, where basically we use large language models to modify the content and the input of users' messages and temporarily place the users in the artificially constructed culture communication environment. And a lot of these culture modules have pre-generated large language model prompts that the AI use to transform the text input of the user's message on Discord and give an output. And we see a lot of the transformations happening here. If you type something, your language will be mediated and changed by a bot. And with reference to whatever culture module that's active. So these are the six culture modules that we've built so far. And to just, yeah, we have Amplify, we have Eloquence, Obscurity, Ritual, Values, and Wildcard. And Amplify just intensifies the message sentiment. Like if I say that, you know, today's fine. It's going to amplify it to be something like today's great or today's awesome. The best day ever. Eloquence makes messages more persuasive. And in our formulation, it kind of turns the message into like a Shakespearean, like poetry kind of which is really, really interesting to read out loud. In the environments, and Obscurity kind of makes the messages more obscure by like a scrambling of words. But we actually found out like really surprisingly that it's still kind of readable if you stare at it for a little while, but it just makes it a little bit harder to read. And Ritual is interesting because it actually changes the meaning of your input. It will align, if we turn on the Ritual module, it will like align your message and to like with the previous message that's being sent. That if the previous person say like, you know, I love like tea or something and you say I hate tea or changes will I like tea to have this like really interesting mediated group consent that's very constructed. And we have values. We have a list of values as defined in the Orgore channel. You can propose revision to it and you can check values with the check values function. And the last one is a little bit special, the wild card module. It transforms messages into the voice of the culture that's being defined by the previous group. So this is like a special one we built in. It's designed to construct like a new cultural module every time a group successfully completes a build a community quest. So the transformation prompts constructed using the text content of the decisions made through governance mechanisms made during the quest and made by the group. And the questions the players are tasked with making a decision are related to the voice that's going to formulate and inform what the wild card culture module is. And we will talk a little bit more about the build a community quest later. And I'll pass to send to talk a little bit about the conceptual thinking behind using AI generated prompts. Yeah. I talked about it actually a little bit already kind of maybe before it was scheduled in our presentation. But I think one thing that maybe to kind of come out from a different perspective is, you know, we have people at D web camp when we did an in person play test of this. And they when the eloquence mode was turned on, they ended up just like reading these posts out loud. And afterwards we had people sort of talking about how there was this you can imagine a scenario where they're in a discord or some kind of communication platform and people are basically talking past each other. They're on their own individual tracks. And there's the idea of kind of coordinating or responding to what the other person was saying is basically impossible. And they said it would be interesting if they had the ability to go into a kind of eloquence mode or step into a channel where all the communication was structured in this way. And I think this kind of points to the something that I think we want to come back to in the discretion, but this kind of question of producing artificial stability in the environment. And like, I think often it's maybe speaking for myself, I think often when I'm posting, I become very attached to what it is that I'm posting for a couple reasons. The content is, I can really only express myself through text, usually. And so a lot of the nuances, of course, of how I might actually express something are lost in the communication platform or the medium. But also, there's the kind of sense that what I am saying is also kind of part of a long archival history and part of the kind of reputation that I've either explicitly or implicitly developed within a community. Well, the chat and being able to kind of create a situation where everyone who's posting is kind of conditioned by the same logic of communication, I think helps to serve a coordinating function. Even if it's a frustrating coordinating function, but it also serves to give a little bit of distance between what it is that you're saying, what it is that you're communicating, and then how you continue to communicate as a result of those transformations. People including myself also notice that the way that you type changes because you're in some ways trying to kind of imagine how will your text be transformed by the language model. And I think this kind of introduction of a kind of flexibility in this kind of artificially stable system allows for a slightly different approach to thinking about how you make decisions with other people online. I think it increases the kind of complexity of making decisions because the ability to accurately communicate is made more challenging, but it opens you up to new ways of relating to your peers and that environment. And so I think one thing that we're sort of interested in is how do these cultural modules kind of influence the decision making patterns within online communities. So at this point we were planning on doing a demo, but the discord is already quite active and the cultural modules have been pretty thoroughly engaged with. But one thing that we were interested in is sort of right now the discord bot and the project is hosted on our discord server. We still need to go through the process of setting it up as a bot that other discord servers can install. So if there's interest, the code is open source and you can set the environment variables and actually just put it in your own discord. But the reason that I'm saying this is because right now it's hosted up here and this is a kind of decontextualized space on the internet in some ways. People from your community are here right now and you're missing all of the context of the channels, the kind of flare that you might have set up with your account, the ability to interface with the kind of specific environment that you operate in. And we're actually really curious to see what that is like because right now it feels kind of distanced and also there's no real incentive or magnetism to this particular discord server. The only reason you would come here is to like play with this, but it's not integrated into the place where you are in community with others. And so to kind of stimulate that experience a little bit, we thought it would be interesting to have alongside the rest of the presentation people having a conversation about something to do with active inference. We were thinking that it would be interesting to have just a kind of general account from the active inference community of what it is that they're interested in with active inference and how they understand the term. And then to maybe kind of just carry on attempting to have a conversation in the way that you might normally do it in your own discord. Just to kind of give a little bit of that sense of the kind of transformations happening in a contextually meaningful environment. So I'm not going to pause this, we'll just go to the next slide. And this is going to be Janita talking about one of the specific cultural modules, which is values. So did you intend for there to be a few moments for discussion? For discussion, like in the discord? Yeah. I'll read a question from the chat and also we can talk a little bit about active inference or continue on. But I'll read a question first. Upcycle Club writes, can you combine two or more modules into a new fused one? Yeah, I think that's actually what we were experiencing a few minutes ago. So you can have any number of cultural modules running at the same time. Right now I think there aren't any cultural modules running, but if for example we wanted to turn on amplify as well as eloquence, anything anyone pulled going to be run through both of these prompts. So the sentiment of your message will both be amplified and transformed into Shakespeare in English. One of the interesting things to point out from a technical perspective, just how the code is actually implemented is the order in which the cultural modules are activated as an impact on the way that the text is transformed. The cultural modules are stored in a list and that list is being read through in order. And the order of the list determines the sequence of the process. So you have your input text, let's say we have amplified and eloquence. So your text goes to the amplify transformation and then the output of that goes to the eloquence transformation. So there's even kind of an ability to basically amplify followed by eloquence and eloquence followed by amplify or two kind of different functions. And I mean that's like in a technical sense, but there's another way in which, in my experience anyways, the eloquence module kind of has this way of overriding more or less all of the other modules. It's so distinctive in terms of the way that it's characterized that I often find that if I want to get a sense of the flavor for the other modules, I have to deactivate the eloquence module. There's also, you know, I can't remember, but I think in the Agora you're not even the ways of changing the cultural modules isn't exposed or maybe it's not exposed in the quests, but to also kind of answer what I think the person in the chat might have been asking is, there's no way to kind of make, like, to specify a module that isn't already specified. So you kind of have what you're given. The wild card module changes as a result of the quest and it's only temporary. So after another group plays the quest, then the wild card module changes. We would like to explore the idea of more or less being able to kind of specify the way in which the large language module is transforming the text. I think there's a lot of ways that the kind of system was architected. It was very kind of organic and not predetermined. Maybe at some point you need to be interested in talking about that, but the way that it was developed didn't kind of have the path dependencies that led us to a system where you could just create your own module on the fly. Awesome. And just one more question from Pablo FM. He wrote, if in our role-playing game, we have our set of values and our lingo and lore, can it be adapted? So one thing that we kind of mocked up a very simple like web app that would allow people to kind of basically define their own quests and it never really is finished being developed, but we're interested in definitely being able to let people construct their own role-playing scenarios, define their values in advance, maybe even have them stored in a database. Every time that the server where the bot is hosted is restarted, all of the data is wiped because it's just stored in the way that governance is conducted in that community. So you can imagine joining a Discord server and then in order to participate in some collection of channels, you first need to go through a governance quest. And part of that would involve communities being able to define their governance in advance in the way that their quest is structured so that when people join, they have a stable, reproducible experience. And this is also the idea of kind of working with community rule. We explored this idea of like actually exporting the... When you play a quest, you have these different opportunities to change which culture module you're interfacing with and what decision-making method you're using. And we had this idea of kind of showing people how their governance stack or their governance structure changed over time and then allow them to basically export any slice of that and then use it as a way of generating any quest. So these are all like any future development ideas, but the short answer to Pablo's question is, yeah. Awesome. Continue on. This is really fun, really fun. Definitely bringing a smile to many people. And there's a lot of seeds that you planted, like the contextless, the decontextualized or the recontextualizing of the dialogue in a space that's new to us. It's in your lobby and there's all these different formal, informal structures. So we'll return to some of this, but let's see a little more. Awesome. Cool. Okay. Yes. So Janita to talk about values. Yeah. So another one of the culture modules we have available is called the Values module. So I think Hazel or a sense of touched on this earlier in the presentation. It's basically a way that you can sort of softly enforce the values of your community. So anyone can propose. So there's kind of a fixed number of values that we have and anyone can propose a revision to anyone of these values to replace it with a new value. And then what you can do is when someone posts a message, you can see if that message is in alignment with the values of the community of chosen by running this command called check values. So I'll just demonstrate that on a message. So we see at the, I think someone ran the values command. So the community defined values we have our inclusivity, collaboration, trust, no respect and intentionality. So I'm going to go ahead and run values on upcycle clubs message. And the values that analysis appears to be that this message does indeed align with their values. And then there's a description for, for why the bot thinks that it aligns with the values. Yeah, so just another building block that we've included to sort of allow people to experiment with like how things can be made more things that are generally implicit can be more explicit. And, okay, yeah, I guess we can try if Daniel, if you want to try the proposed value revision command, just propose underscore value underscore revision and then it'll take you through the flow of replacing the no respect value. Maybe go go through it on your side, just because that's what screens captured. But this is epic. It's very flexible and I'll just read one comment just while you're setting it up. I'll read one comment from Dave. He wrote, here's a very practical application of the kind of transformations you're demoing. Glossing a discourse that's expressed in one vocabulary into a discourse expressed in a different analogous vocabulary. Identifying corresponding terms and concepts in two fields or into approaches to a given field is expensive. Deploying the result of such discovery will be useful on its own and a spur to the effort to develop more correspondences. This would involve a lot of under the hood activities required by full machine translation, e.g. German to French, but could be driven by much simpler rules, a.k.a. patterns. Thank you, Dave. All right, so you're typing up respect and justifying it. Yeah, so you'll see that we all have the option to object. It's kind of like the pass by default unless there are any objections to the value within a certain time period. So provided that no one, yeah, lazy consensus is the term for that decision making process. Provided that no one objects to this value, it'll be added to our list. Cool. How long is the time like you could set different amounts of time? Yeah, I think the default is like a minute. Yeah, yeah, you could set the time like you could program that, but we haven't partially because one the someone could set it for like 60 minutes, or someone could set it for like one second. I mean, you could like specify the like the boundedness of what's allowed. But usually because anyone who objects to one of these values is going to be able to pretty quickly realize the object. Our sense was that you really don't need that much time for this decision. Oh, it totally makes sense. It just reflects how that could be a one month review process like in a DSI scientific review setting with the right interfaces or it could be really rapid like feedback from the individuals and their time bound activity. And then that feeds back into the game like new rule, here's how we propose new rules or here are the options for rules, or here's how we're going to apply which filter. And that just adds so many plot twists, even before active inference or without any kind of cognitive modeling at all, or systems modeling just approaching it from the pure gameplay. It's really fun. And people in the chat are like, yeah, already pointing out a lot of ways that can be used to play if not be more useful. Yeah, one thing to point out that's kind of hidden at the moment is, so let's go to this value analysis of upcycle club. Where are they? So they were aligned. And if you go to their profile. Oh, no. There should be something that I thought that there was a misaligned tag on friendlies profile. Yeah, yeah, I don't know if it's doing it right now, but I think we might have taken it out at some point because it wasn't part of the full features that we wanted to ship. But this function can assign an aligned or misaligned tag based on the results of the analysis. And then you could do what you want with that from a programmatic perspective. One idea would be that like, if you as a community are aligned, any user that you're aligned with, you are either not affected by the disruptions and language that are caused by the culture modules, or the effects are less intense. And so there's an incentive to remain aligned if you want to be able to effectively coordinate with your peers. So the demo thing that the demo idea that we were going to explore was asking people in the chat to answer the question of what do you value in communication environments, perhaps from the perspective of active inference. And then we can do some value analysis, even propose new values. We can see this kind of iterative feedback layout. So maybe we'll just leave that again, like just keep playing with that in the background in the chat. It feels, you know, something about just stopping and watching the chat and not saying anything feels too strange. So, yeah, we'll go to the building a community mode. I know that I was just to talk about this, but I feel like I've been talking for a really long time, even though we like structured the speaking setup. Hazel, did you want to talk about this at all? Or do you want me to still pick it up? Go for it. Okay. Yeah, so building community. We have a build a community quest. It's we actually explored a bunch of different quests, this kind of like post communication collapse quest. We explored one that might still be in the repo somewhere called the Josh game, which was a kind of. Actually, Hazel, can you talk about that just very briefly, like what the Josh game was? Yeah, I think the Josh game is like, it's like intending to be an a game that's placed on individual identity. It's kind of inspired by I think one of us saw like a news online about the like IRL Josh game that a bunch of people are named Josh. And they kind of had like a just like a fake like noodle sword fight for the name Josh where the last person standing who like one like via rock paper scissors is the last person who's crowned Josh. And the rest of the people like lost their ability or lost their right to be Josh. And I think like through we kind of made some version of that where we all entered the game with our like avatar and name replaced with like a random Josh like can be like jostling Josh or like Josh or tree Josh or whatever else. And then we kind of like compete we kind of give spielson speech in this like kind of word based contest where we kind of compete to be who's the most worthy Josh or who's the last Josh. And yeah it's just like like a playful way to experiment with who's who. How do you have like what's in the name and how do you like kind of like establish or spin out like a little playful identity, and we imagine like a lot of different ways we can potentially play with that. But but that's the that's the general that's the general like playful context of the Josh game. Cool. Thank you. Another funny thing about that game is Joshua Tan was the executive director of medical at that time and is still currently at medical as a resource director. And we actually never played this with him, but it's his GitHub handle is the last Josh. And so I don't know there's something very funny about spending once time, making a game that is not even directly referencing someone, but is yet nevertheless connected somehow like organizationally. So a little bit of, you know, meta for for the extreme posterity. So to the build a community quest. Yeah, 25 minute narrative blood game. And you can prompt it in the chat. I mean, you can try it. It's a little while since we've actually done it. The bot might break. So and also it takes a long time to do, but basically, you can set up the parameters of a quest. You can decide how many players need to join and whether or not images are generated as part of the quest, and then you go and you play. And the goal in the game, the quest is to basically kind of as a group collectively establish yourself as a new group to give yourself an identity of some sorts. You have to answer questions like what's your name, what's the main topic of interest that your group talks about, and in what way does your group can speak or communicate to each other. And for each of these questions, there are proposal functions. You can revise proposals. There are voting decision making processes. There are some deliberation questions that are set. And there are also different methods for making decision. At the moment, I think we've implemented two or three, like a very small amount of decision making methods. And then there's a fourth kind of implicit feudalism method, which is if the Google players are unable to make a decision after three tries, the bot basically just picks a decision at random. Partially because I don't know. It's hard to know how deep we want to go into the implicit feudalism thing. But basically one kind of background thing that's also connected with all of this is the idea of thinking about the ways in which the platforms we use to communicate with each other are structured. And Nathan Schneider has written a really nice piece called the implicit feudalism of social media. And it is looking at the history of the server client architecture from as far back as the BBS era up to now. And his argument, which I won't go into a lot of detail about, is basically this architecture and the ways that we communicate in general are basically representative of a kind of feudalism. It's not explicit, but it's just implicit in the way that we communicate. And one of the thesis is that we could give communities online a different ways of governing or setting up governance systems that are not inherently feudalistic. And this is kind of where the modular governance and the meta governance angle of this project comes into play. And a lot of the game is about kind of introducing people to the idea that they could change on the fly how they make decisions. They can have agency over the way their communication happens in a way that is circumscribed by the developers of platforms. And to be honest, it's also circumscribed by us. It's very difficult to create a system that doesn't have any kind of predetermined structures in place. But for example, in the game, if you fail to make a decision about any of these topics, then you're given the ability to change the way you make decisions. And then if none of that works, then a decision is just made for you. And this is in some ways meant to kind of bring to the surface for players the ways in which they can and cannot exercise agency. When communicating. And the other thing to say about this question, why you go through this process is that all of the answers to these questions are then fed into a template for an LLM transformation module. This is the wild card module. And what that means is that one thing that I think is there are a lot of experiments with kind of multi user input interfaces into LLMs, but not so many. And I think one thing that's sort of interesting to think about is how can we open up an imaginary about this two things. One is having more collective inputs into the way that the large language model is interfacing with communications, but also kind of being clear through experience, the ways in which the way that we communicate and the decisions that we make have an impact on the ways in which our environments change around us. So this is kind of maybe like the stigma G kind of thing a little bit like you have these like interactions, these like pheromones that people are leaving behind, which are their decisions. And then you kind of experience the result of that in the Agora and that also changes the way that you interact in the Agora and maybe the way that you interact in the build a community quest. So I mentioned that if a failure, if a decision fails, then you have the ability to change it. And the way that that happens is through a kind of continuous input mechanism. There's a kind of multi-user input as a way of inputting your preferences for a decision. And then once a certain level of preference collectively has been expressed, then a decision is made and the mechanism is triggered. It's also a display of community conviction. And it also just kind of acknowledges the fact that there are governance failures. I think one thing that kind of comes up sometimes and discourses around governance and also the perceptions around governance is that governance is kind of a, when we talk about it as a system, it can sometimes, or a model, it can sometimes be perceived as this like static set and fix and don't think about it again, experience. And sort of a lot of the literature that kind of thinks about, especially the emergence of medical governance and governance in general as a kind of development of government is basically like hyper focused on the ways in which failures manifest and decision making. So we wanted to make sure that when this happened, there was a way of exercising agency and kind of giving players the experience of steering or responding to the environment. Another thing to maybe say just quickly about this is that this kind of mechanism is also one of the like most popular, often cited ways of references for this is Twitch plays Pokemon, which is basically a emulator of Pokemon set up on Twitch and then some Python binding so that anytime someone types up in the chat, then an up command is sent to the emulator and the character moves up. And so you can have a chat that's basically just collectively distributed the controlling the game. And this is cool and extremely chaotic. But there are maybe like two scenarios in which place Pokemon are like in Pokemon where you really have to press the same button, maybe like six times in a row in order to even move to the next part of the map. And so you can't stay in the kind of pure chaos boat or else you never complete the game. And so the developer of the Python binding set up this way for people to oscillate or move between democracy and anarchy. And when you're in anarchy mode, it's exactly like we just described. It's all continuous just taking the input. And then when you're in democracy mode, there's a kind of time buffer in which all of the responses are collected, whichever response gets the most input in that period of time is the response that's triggered. So it sort of slows down the leading of input and allows for groups to coordinate and actually get past these situations where they couldn't otherwise. And then it's a super interesting project. It's very old at this point as well, but it's highly recommendable or not familiar with it already to go and look at it. Another great reference in this area is Trusts Moving Castles project, which is also looking at community control games through Twitch and is kind of a more contemporary example. Yeah, so we wanted to demo this and you can see that it's already happening in the Discord chat a little bit. People are typing a culture module and then plus or minus one. And then once it passes a certain threshold, then that module is activated. So we didn't have any prompt for this one. So maybe we'll just kind of move to the next section unless you want to pause and discuss something since that seems to be a pattern here demo discussion. I think one quick thing I would add is we told we built the cool down time because I think when we were playing it we ourselves find ways to break it and sheet on the system. I think that's also part of the fun of building the game of thinking about not not just thinking about ways that we can play but also to think about to play through the disruption by disrupting the disruption, which was really interesting, which was why we built the cool down so nobody can spam and just get what they want. At least not immediately. Yeah, that's a great point. Yeah, yeah. Thank you. Anything you want to add maybe, Janita? So progress. We've done a play through of this demo camp. We talked about that a little bit. We did a kind of annotated play through. I can annotate a version of that play through. I'm going to show it because it's not set up. I can't remember where the link is actually, but anyways, that was a really interesting process and we should make a slide for that next time. We also did a presentation as part of a symposium in Geneva called the sea shore in those worlds. And we've also presented this at a one of the weekly meta governance seminars. So, and now we're here, which is very exciting and really looking forward to this stream existing because I think this is a very fun engagement with the project. So, yeah, with that, with that, maybe we would move into a little bit of discussion, even though we've had this nice kind of interjection discussion so far. And we were thinking of, like, in what ways is this kind of project? How does the active inference institute community see relevance in it? I guess in terms of both movies, setting up experiments or kind of using it as a compliment to the subject of research. And then another thing that we're interested in is kind of thinking about this kind of thing that we alluded to earlier, this kind of like communication collapse. Like a kind of post LLM or whatever kind of technology were posed at this point. Communication environment where the like, this is idea as far as we understand we're not experts in active inference that one of the kind of reasons to study and model this in this field is to kind of be able to anticipate and respond to unpredictable environments. Somehow like the idea of active inference, and I'm curious if this is correct, is to kind of produce kind of more stable system than we might otherwise design. And this game in many ways kind of purposefully introduces instability into the communication, but it's also those disruptions are themselves kind of metastable environments. And so there's this kind of strange interplay between stability and instability and the ways in which this kind of experience of expanding the kind of governance imaginary for how we might interface in a more ambiguous, epistemic context. So we'd love to discuss that a little bit, and then to maybe leave everyone with some information about us here are our names and our respective roles in the project. Some of our contact information, and then some links to our projects. And she also mentioned that the project was created and supported by that governance project. So, yeah, that's our presentation. Thank you so much for having us. And I'll leave the discussion concept and we'd love to see where the conversation goes. Awesome. Well, thank you for presenting this. Like, there's a lot to say, a lot of places to jump in it feels so natural, despite the tech in some ways being so new in practice. And in our hands, but even though you all have made this it feels like people just slotted into it. And I think it does touch on a lot of active inference ideas like on a more technical but also on a more informal level, like just starting with the informal, the idea that there's multiple agents or multiple entities with varying blends, new mixtures, new filters and overlays of bodied and digital, all these different modifications and layerings and identities. So this kind of like open cyber complex identity engagement space. And the way that still there are these features of ecological psychology, and that those fundamental features, like relating to sense making on the inbound and decision selection on the outbound. We can understand whether it's like leaving an emoji or the words that are produced. All of that is behavior, and it's behavior that's different, but with our body. So there's all these kind of like rich dynamics. So the field provides more questions than answers on those fronts but shows like how far the gap is between the general unified approach and how deep the last mile is in application. But this really like connects it from what otherwise could be super abstract about like multi agent online interactions, which is like what the research papers in MetaGov or in active inference, how they'll describe it. And this like really connects it and it's like in this discord. So it kind of becomes embodied in the real chat. So that was one thought. Another thought was like, it shows that words matter, and that our speech acts online matter by representing them in this kind of fun and unique way. And there's other ways to re-present people's speech online like going back and curating older posts. That was something we had done or seen in the MetaGov Slack. And then here with some new shiny affordances from the language models and all of that to have words that have informal import with the sentiment that we convey, filtered or not. And then also to bring in acts that have kind of quasi formal import, like actually changing the rules of a game at some level of analysis and being able to hold the fun and playful and comical and that whole aspect. And also seeing that that's even paradoxically or not like what's important for high reliability distributed settings or for establishing trust or new ways of communicating when there's change. Thank you for letting us speak. I think to be fully transparent. I think we're still very actively, this is not a finished project and we don't present it as a finished project. I think we're still very much in the process of imagining use cases, not just use cases, but places where this can be immediately useful or playful or make sense. Because I think today we're really fortunate to demonstrate it in a way that feels that has the split screen to demonstrate in a way that feels very calm. But we've also had like playthroughs and demonstrations where when it needs to be like put into like say like 10 minutes is really confusing and it's just like a bunch of messages blowing up and everybody speaking. And I think, yeah, some users think it's really kind of chaotic. Yeah, I think, yeah, I think like we, yeah, we really appreciate speaking today, but would really still love to hear from you guys, what are some of the use cases and even like modules that you find that's like most accessible. Or any research potential that could potentially be interesting stemming from this, or other ways that we can, I think, play together like for the purpose of like governance, spirit, spiritiveness. Cool. If someone writes something in the chat, we can read it, but just one thought that's kind of a general approach in active inference modeling, which is analogous to some of the work we did like in 2022 with DSI, distributed science, decentralized science is take a look at systems that are often engineered from the top down. And then think about it from the bottom up in terms of the sensory inputs and then the capacity for action, the affordances of different agents in the system. And so take that kind of view from the inside. So one kind of scaffolding of an approach that is pretty general across active inference settings is like connect up that constraints or like the modifications that you're putting on to the environment, and then interface that with schemes for the kinds of input that people receive and actions that they receive. So like different actions, maybe people have different roles so they can only do emojis or people can type in chat, and then different inputs, people could see different channels, or maybe some people are seeing it visually, and some people are having it read. So make a kind of scheme for who are the players, and then with the players plus the board on that's kind of the board game setup. And it's a different board and different players, but I think even with the title of the project you kind of allude to that role playing. Yeah, that's really interesting. Yeah, the title is also kind of a funny one, because there are no dice involved, but it's catchy. Yeah, I think having like more persistent consequences is an interesting avenue. I mean the kind of aligned, misaligned role tagging that we showed is kind of pointing a little bit in that direction. But yeah, like imagining like a kind of blank server, the first thing you do is interact with this and everything that you learn about the server as a result of the decisions and interactions that you have through this would be like a really interesting way of kind of producing a somewhat like labyrinth and non stable, constantly evolving environment that people would interact in. And like, yeah, that's a great idea. It's interesting. I mean, that is newly on the table. And so it's exciting to think about digital environments, whether they're personalized or shared that have a defined construction to them. So it's legible. Here's how we're modifying the incoming posts or something. So there could be situations where that's exposed, or like transparent or open source, and then there could be other situations where for one reason or another they don't. And they would be accepting the trade offs with or without sharing that kind of information. And it brings like new nuance to all the filters that we do receive information through. And as that becomes more posts or multi lingual, then there's there's so many new ways that this can come into play because people are hashing these things out. And they are having a model of who's aligned and not and so making some of it visible. Can maybe that's a disconcerting look in the mirror. So not every representation is probably right for every setting, but in like different settings, there will be people who love this mod facilitator, meta governor role, and you're producing tools that help bring that to the discords where people really are. Yeah, there's there's two things that are interesting somehow there. One is actually working with the discord interfaces and also the API is actually kind of a nightmare. Like there's a there's a function that we have that has a lot of text streaming. So it's taking a message and then like sending a couple words chunks at a time to produce this effect that the message is being written by some external agent. But there's a pretty aggressive rate limit on how quickly you can edit messages in discord. So there are some kind of like constraints there. And I think the there's something about this. I think I can I want to that I can sort of see is that there this this kind of model that we're talking about of like a kind of really a margin and flexible and some what kind of. Unpredictable structure for and and a structure where no one person really has the same view on the system. There's no holistic view. Maybe even as an admin then like you're opening up so many kind of unpredictable moderation questions. I mean we already have enough challenges or at least the discourse will kind of point to the kind of numerous moderation challenges that we have. And kind of maintaining our disc our discursive spaces aligned with some kind of sense of community value. But like the ability to kind of have an environment where no one has the same view kind of opens up this like huge like how do you even conceive of moderation in that way. There's also the like there's also this this problem of information overload. So we explored like being the the generated response actually show you the information like the the original input using like a spoiler tag. But it was just too just too much like it was visually cluttered. And another thing that actually happens here is like I want to figure out how to like have it match token length so that the input and the output are kind of like correlated to each other in length. Right now there's this kind of it will elaborate so much that it breaks the natural flow of communication. And it introduces a kind of temporal friction and so like more moderation to kind of expose what's happening in the background. And it naturally puts a kind of like break in some way on the kind of way in which we're communicating which could be beneficial or not. And then yeah like this dystopian thing. Yeah I mean there are a lot of ways this could be applied that are not benevolent very like authoritarian in nature. So it's it is there is also this kind of yeah yeah the moderation surface just becomes so difficult to figure out how you actually approach. Yeah well one interesting approach there is that now with these LLMs the space of moderation is in the semantic not just the syntactic. So a lot of discord bots or trending lists or all these things are based upon the syntax of like how something is spelled or actually like how word is used with all the kind of false positives and false negatives. Now we're in a different space or mode of false positives and false negatives with the semantics. And that's a different and a more latent space and one that really gets at the cognitive diversity and different people's understanding of different words and decisions and events but that's kind of the real space as well. And so making the maps of that cognitive semantic territory instead of saying OK no name calling here's the list of names. Now that is going to be elaborated on from something and maybe and also just so new vistas for how people want to communicate and engage or not. And like you brought up massive opportunities and challenges for attention and overload and just the pace of all this augmented text is extremely high. And so it's very intense also though people do a lot of intense digital experiences. Yeah it's so true. I think there's like some relationship between the intensity of digital experience that you're comfortable or have previously experienced and how quickly you pick up this spot. We've done this in some other contexts where people are just like I have no idea what is happening. And I think if you are if you tend to be like extremely online this feels very natural. And if you're not it's it's somehow just way it's too much. So that's also it's also interesting from a governance perspective as well. I mean I can't make any real extrapolations but I wonder if in terms of the kind of the viability of online community based self governance if there is a kind of correlation between the kind of like intensity of digital experiences that the community is game for. And how broadly effective they will be or adaptive to online governance they will be. Whereas communities that are less active or less digitally intense would generally find online community self governance harder to adopt. That's super interesting and communities that have variation amongst their participants in that as well. Well this is an epic part one and kind of play through I guess with this like maybe as we kind of close where do you want to go from here or what would be something cool in 2000. To get to or try. I don't know if we can each kind of try to give an answer to this. I think one thing that would be really cool is like cleaning up the code base so that it's like easier for other developers to actually extend the work that we've been doing and contribute meaningfully. I and also getting this set up so that communities can actually experience it in their own contexts rather than just in the surface. So if there's anyone who is part of this community or who's watching who's interested in kind of sifting through the code base and figuring out how to contribute that would be really cool. Yeah that's kind of my thought. I don't know Hazel or Janita anything from you. Yeah I'm on a similar page. I really think of this bot and this tool right now as like just a set of building blocks that communities can ideally use to like configure their own experiences. And so I think yeah I think it'd be great to figure out ways to extend this to make it more accessible for for people to continue to experiment and build upon what we've done. Yeah I feel like I really wish that more people watching this live stream or not will just come talk to us and say hi and tell us how we can make this more useful or practical for you. And that would make us really happy. And I think another thing that a topic that I'm really interested in is what kind of thing is randomness in governance. How does like being random work and having a function that assigns like a random value work. I think there are quite a bit of we build like a kind of random overlord function in build a community quest. But I feel like that's something I really want to explore more especially with regard to what's the relationship between randomness and agency. Like what's the setup that gives user like the agency the feeling of agency of governance. That's something I'm interested in thinking through more and hopefully with this project. Cool super exciting I hope people in the Institute discord and elsewhere tune in to this kind of cool way to think about our shared services as governance services really converting a space and taking everyone on this journey. That's so in the moment that it's kind of like intense and flies right by. So it's been an epic stream and maybe we can play around with it as you all continue to develop the framework. So thank you all till next time. Yeah, thank you so much for having us. Cool. Bye.