00:02[Applause] [Music] It's gonna be alright. [Music] All right everyone, welcome to the Stoa. Today's session is called Emerging Projects in Public Sensemaking with our friend Daniel Schmachtenberger. So Daniel reached out to me and wanted to do a session at the Stoa to shine a spotlight on projects and people he thinks are doing good work in the domain of public sensemaking. So I was like, yeah let's do it, um, and so we're here. And how today's gonna work, I'm gonna hand the keys to the Stoa to Daniel in a moment, and he's gonna interview three people for around 15 minutes, um, and then they'll share the project that they're working on.

00:51And then hopefully we'll have some time for Q&A, um, before the session is over, and I think we're here for around 90 minutes. So if you have questions anytime, pop them in the chat and I'll call you, you can unmute yourself and ask them to Daniel or any of the three people he'll be interviewing. And um, yeah, and I'll be taken near, uh, back at the end to close the session. So that being said, I'm gonna hand the keys over to you Daniel. Welcome back to the Stoa my friend. Very happy to be here, it's good to see you Peter, and I am happy to see everybody else.

01:21It's been a minute since being here at the Stoa. Like Peter said, the aim of today is there's a few projects that I think are really good projects that I wanted to just be able to kind of introduce to the community for a few reasons. We've got three people joining us today. Jamie Joyce from the Society Library, we're gonna talk first about what she's doing, and then Speaker John Ash from Cognicism, and then, uh, I just got my view moved I'm going to move it back so I can see everybody, here we go, and then Brad de Graf from, uh, NewNet.

02:00And they're, they're conversations I've had in the last few months, uh, Jamie I've actually known longer than that, with those projects I just think are really great projects, and I would like more people to know about them because I'd like the projects to succeed. But I also wanted this Stoa as a community and everybody's here to get a sense of what does an individual who is thinking about these topics of the state of the world and how do we better collectively make sense of what is needed and go about doing it together, what can you do if you're a person who doesn't have a lot of capital support and you don't have institutional support.

02:33And I wanted to give examples of some people who just started doing amazing on their own, and it's actually turning into or has turned into real projects, because it's, um, it's inspiring, and they're very different in type in ways that draw on their own unique backgrounds. So uh, without further ado, I'd love to dive in. I would like to know, we've got three because that's what we have time to do here and these were three great ones to start. There are definitely more people in the Stoa and Stoa community working on pretty exceptional projects, and maybe we'll get an opportunity to highlight those in the future, uh, I think there might even be some people in this audience today running some of them, so I might mention names briefly at the end when we do Q&A.

03:04So dive in for basically these short 15 to 20 minute, um, overviews of each, and again they can't begin to cover what they're doing well in 15 to 20 minutes because they're actually thoughtful, deep, nuanced, big projects. So they get to cover maybe a main thread, and I hope that's enough for you to just get a taste to see what you're interested to follow up more with, and then there's opportunity for that to occur. So Jamie, welcome, can you hear and can we hear you? I can hear you, hopefully you can hear me. Indeed. Uh, you at your Berkeley office? I am, yes.

03:34So Jamie started the Society Library, and uh, there's a lot of people, and she will do a good job as the leader does of saying we more than I most of the time, but she really started the thing by herself, and like a huge force of will to make it happen, and um, a huge force of will is one of the things that it requires to make anything meaningful happen. The Society Library is still small as an institution but has already done very real stuff, and like getting to help policy for, um, grid security and nuclear energy and like real things with big players.

04:12So just to have that sense as you're hearing about her. Jamie, um, you have, uh, you have something to share with us as an intro for what some of the main things Society Library is working on are? Yeah um, so I don't know Daniel, do you want me to set the stage about the problems we're looking to solve or just dive right into what we do? Set the stage in like 45 seconds of the problem you're trying to solve, because I think each of you are trying to solve them from a slightly different angle, so the way you set it is context setting.

04:46Yep okay so like Daniel said, uh, many of the people who are in the audience but also my co-panelists are working on solving a number of different problems related to not only the information environment but also our internal ability to be able to process information. And I think it's accurate to say that each one of us, Brad, John, and I, are working on ways to extend our cognitive senses to essentially give new sight and vision and lenses to see the nature of problems through the information that's accessible to us.

05:16So the way the Society Library is working on these issues is to deal with the fact that there is so much information to process, a lot of it is obfuscated, a lot of it is not high quality, links get broken, sources aren't cited in news articles and scholarly works and all across the web and world, so there's this mass distribution of content across so many different platforms. Not only like TikToks and tweets and social media but also government agency documentation and scholarly works and books and textbooks that have been generated through time, and it's an impossible task to ask people to sort through all of this information or to find the relevant bits to construct in order to understand a social and political problem.

05:44So like Daniel said, the Society Library has worked on a number of different projects. We do political decision making models, work on educational curricula, we've written and deconstructed legislation, but today um, I would actually like to talk to you about the main mission of the Society Library, which is creating a new kind of digital library system to serve society, and it is uh also one of the most complex, technically difficult, and expensive things to accomplish. But if Daniel's interested I'm happy to actually show you exactly what that looks like today as we are about to launch our first collection, and also talk about the process of how we go about gathering all this knowledge from different forms of media, extracting arguments and claims, and structuring that content.

06:13For you again, maybe 45 seconds worth of, uh, what a library is relative to an open society, because I think that's not necessarily, uh, obvious for everybody, and why we need a digital form of library to do that now. Got it. So libraries have been companions to advanced civilizations for a very long time, um. If you are going to participate in an open society or even a democracy, it's expected that you have, you are to some degree informed, um, and you have like an inclusive understanding of the different perspectives at the table in order for you to make a free choice about where to vote or where to spend your money or who to advocate for to represent you.

06:41Um, but if information is just completely out of reach because you're not in the areas of the web and world where that information is, is going to just constrict and confine your ability to make informed decisions. So um, a lot of, uh, you know, information and the way that we interact with it can either build walls where there are none in our minds or also open doors where we think they are but they don't actually exist. So it's really important that we start developing the lenses, the tools, the techniques to start interacting with information a little bit more rationally so we can make more informed, inclusive, and less biased decisions.

07:12Yeah, so please, uh, take us in. But just, if I think a lot of people here saw some of the war on sensemaking stuff, uh, this aspect of the Society Library is trying to create an institution to solve a lot of that. Says hey you as a person might have a hard time if somebody on the right backed with the RNC support or a media institution support or the left of the institution behind it says something and they configure statistics in a particular way and like off-frame them a particular way, really being able to ascertain that.

07:40Then you see hugely disagreeing views and mostly all you get to do is decide cams or decide to be a non-actor. Is there a way to actually parse all of that, remove the like off-framing, find the facts, see what's omitted, and be able to inform the public with something that doesn't have vested interests? Obviously we don't trust our institutions that much now, and we had made some institutions that tried to do this but there's a need to be able to rebuild them. They can process all of the content that is here and that has more oversight into how they came to that, so that people can actually trust what is the process.

08:09So that's really fundamentally what the Society Library is trying to do, without which something like democracy or open society just isn't a thing. Thank you Daniel, um. So one of the, uh, as I mentioned we are now getting to the point where we're releasing a library collection. We've worked on many different things but these collections, this research that has demanded of us is incredibly extensive. So I'm actually going to show you the database and talk a little bit about what it looks like in terms of breadth and depth of what it means to go about collecting knowledge to make sense of something.

08:42And so just again another overview in creating our library collections, what the Society Library does is extract arguments, claims and evidence from over 12 forms of media in order to build databases that articulate the reasoning, and not just logical and fact-based reasoning but inclusive of representing people's opinions also. We are modeling the reasoning from all points of view on complex social political issues. So we worked on climate change, covid, a number of big national and international issues. We've never released these because they've never met a level of comprehensiveness like the collection I'm about to show you now, which is a state level issue having to do with a nuclear reactor in California, the last remaining one.

09:11And this is the state and to some degree national debate about whether or not it should remain open. So I'm going to share my screen and then also obviously recommend going and checking out our website to see what content does exist, uh, on our other topics. So this is a database everyone. This is not our front end, I will show you our front end shortly, and actually we have a variety of different frontends, but I'm going to walk you through just the level of depth we're talking about.

09:38So here we have a question: what should happen to the Diablo Canyon nuclear power plant, the last remaining nuclear power plant in California? Based on us scraping, um, I think we have over 4,000 references from which we've derived over 4,000 arguments in this particular database. So we've taken this from TikToks and interacting with the NRC and confirming facts with them, pieces of legislation, television clips, books, documentaries, YouTube videos. We've just went through and collected a bunch of multimedia relevant knowledge about this particular issue and extracted the arguments.

10:05And what we found was there's about nine different positions that people take on this issue, so it's not just should it remain open or closed, but open or closed with certain conditions. As we unpack position one, we see there's a number of different categories of types of arguments that people are expressing. So just to open up Economic alone, you can see there's a lot of economic arguments here, and these are like, we're not splitting hairs, these are very high level arguments that are being made that only if you desire to unpack them we can start splitting hairs.

10:32And I'm kind of going to zoom through it, I don't think we have time to like read through. So can you give an example of a couple arguments in a couple sections just so people get a sense, because I so often, the issue makes it to the news that people see polarized around a single part of it, so they get left or right they don't get nine different views, and those nine different views don't have thousands of competing sub elements, and I think it's important to just get a little sense of that. Sure, okay um.

11:01So backtracking a little bit, I actually think I, I mean I can read some of these and then I can also zoom out and kind of show the complexity, if that's gonna, I think the zoom out would be a little bit better. And luckily I prepared for that. So this is position one, highest level arguments. And when we talk about high-level arguments we're talking about things as vague as uh, market forces, financial incentives, and policy decisions have made the Diablo Canyon nuclear power plant redundant, non-competitive, and undesirable. That's very high level, that's big, there's a lot of information packed into that.

11:31So like, our specific ontology of deliberation starts with questions, then we have positions, and then we've got the categories: economic, environmental, safety, and then we start breaking it down into like the gist of the arguments in each one of these. So if we back up just position one, this is all of the very high level vague compound economic arguments. These ones I believe are the environmental ones, these ones are the safety and well-being arguments, these ones are the energy-related arguments, these ones are the ethical and education-related arguments like public sensemaking issues, and these are all the political arguments.

11:59So really what the Society Library is working on is how do we enable not only a conceptualization of the complexity of these issues but make it so that people actually have the opportunity to interface and interact with it if they choose. They can either comprehend, wow this is amazingly complex, I had no idea that we were beyond this polarization and this debate was so nuanced. But beyond just giving you that like Google Maps-like Earth view, actually also having the street level view as well.

12:32And this is more along the lines of the street level view, which is this pathway I was taking you down before. So we're talking about market forces, we're going to break it down specifically to consumer demand, and then here we have a uh multi-premise argument that breaks down an assessment of the economic um, you know, the supply and demand that would render Diablo Canyon not competitive. And then each one of the premises in this particular piece is also something that can be debated further.

13:00And what's cool about the Society Library's work is that like, we go a little bit beyond pro and con. It's not just true, not true, we are modeled and reflecting the language and dialogue of society, so there's some more nuance to the pro and kind of argumentation that some people, when they're doing argument mapping deploy. Um, we've got like I think 15 different relationships between claims, so you know it's a little bit more complex. But anyway it's a, this is kind of like the street level view of you really being able to get down into the details of what people are arguing, paper by paper and claim by claim derived from paper by paper argumentation about the methodology, all these different things.

13:31And each one of these nodes actually contains a lot of knowledge now. Um, this is not the thing that we normally show, uh, people. We're going to make the database traversable like this, that's completely fine, but the question we've been asked is how do you make an entire database worth of knowledge something that people actually want to see? So we started from kind of, I don't know if it's first principles, but we just started thinking about, well what is the most familiar format in which people experience knowledge, and we came up with the piece of paper.

13:59So you're looking at this piece of paper here and this piece of paper actually contains all of the knowledge of that very complex database, but it's an interactive engaging piece of paper. There's many ways in which the Society Library is probably going to present knowledge, this is our first foray in building the library one page at a time. So let's go ahead and like, unpack here. Here we have that first position, the Diablo Canyon nuclear power plant's license should expire, it should be decommissioned, because of a number of different reasons. These are the categories you saw before, now you can unpack each one of these.

14:32We're actually having a feature where you can opt in to different versions of explaining the same thing which is pretty cool. We also have notes, so you should know that this particular high level argument is extremely popular all about like marine life being impacted by the OTC system. So you'll see notes and tags from us just giving you little clues, and this is something that users told us that they wanted. And what's cool is that, you can read this paper section by section, and you can either unpack it into the arguments that support it, so no evidence provided for this one, kind of important, or you can actually open up every single claim.

15:00Here's the claim about the marine life and see a whole Wikipedia page kind of style, uh, page for it. So here all the quotes from which that claim was derived, so we didn't just make up these claims, these are where claims are found. You can see the different locations, an online publication, a government agency document, um, organizations, blogs, and because we're a library we're essentially providing you know reference material for you. So you can click on different links you can go to the original source yourself.

15:29We work hard to back it up in the Internet Archive, so even if links decay over time there's a backup in the Internet Archive. So all of our work is just to give you the opportunity to see the bird's eye view of a deliberation, dive into the details as much as possible, give you enough context with the things that we've learned like, we've not been able to find any evidence for this, or you should know this is a very popular argument, and then just give it to you with no intention of trying to coerce you, help you make a specific decision, just organizing knowledge and saving you tens of thousands of hours, trust me, finding all and finding all this knowledge and organizing it yourself, and this includes interfacing with government agencies and confirming facts with them which can be very time consuming.

15:57There are other ways in which we're going to be visualizing data, for example this is just like a scatter plot, and you can kind of just like randomly explore little nodes at a time, and this actually clicks into the debate map itself. We will actually be using our decision making model and digitizing that and rendering all of our, uh, data through that as well. So in case you want to make a decision, we actually give you a methodical process where you go step by step, section by section, argument by argument, and it'll keep track of essentially what you decide you believe to be, you know, more likely to be true, more important, etc, and then mirror that back to you at the end.

16:24So it's just the beginning for us, we've worked on a number of things, but this is just the beginning for the library, um, and I think this is a good stopping place so I will. All right, so even though that's just the tiniest little intro to what they're doing, isn't it awesome? Like you get a sense right, like beyond being able to really get into it, you get a sense of like, yes it makes sense that the decisions are that complex, it makes sense that the high level arguments are very high level and then the details are very far out in the tree.

16:53And so then what you get when it's time to vote is yes or no on proposition 12. Yeah, what the, yes or no, there's environmental arguments and so then you end up getting something where okay, it'll ruin the environment but if you don't do it it'll destroy the economy, you're like why am I stuck with this decision, what about the better proposition? So this doesn't just inform how you vote, it also informs what would a good proposition be.

17:20And if you have bad propositions you can't not polarize the society and mess the world up right, because the people who really care about one thing are forced to be in a trade-off with the people who care about the thing that's going to be damaged by it. But if you get to see all those elements you can say well, a good proposition would tend to all these things as best as it can. And yet for a world as complex as ours, it takes some work to do that. And why she's saying a digital library, that's obviously a knowledge management system because there are a lot of claims, and if you were trying to do your own research you couldn't get all of those books and government files and etc, and you don't want them and you wouldn't know what the to do with them.

17:45So let's break down the argumentation that brings us to different ideas. Now you might be like, all right well that's actually still not all that useful because what citizen is going to go through all that, and how could a group of people process that much information? So for those who are asking the question, Society Library is already working on AI applications to be able to do this using the same format where it's a human-done format, so humans can adjudicate it, so it doesn't just live inside of a black box where the AI said do this magical thing, but it can help process the data, humans can actually oversight it.

18:14And somebody mentioned, I think Scott Nelson was in here, that's a great example why I wanted to do this because Scott and Jamie, you two should talk. He was asking, have you webbed three this yet? Because could you do a system where you can actually see the provenance of all the information and then also do something like with crypto incentives where if anything is wrong or missing, people are incentivized to find info and add it, and if they do there's a whole incentive system for civic participation? Yeah that stuff can totally happen.

18:47So um, Jamie's done stuff like this, or Society Library has, on a bunch of other issues, and there's a lot of applications that are really just to inform representatives and institutional like, government institutions, some that are for the public. It would be really worth having her back into a whole session just on Society Library types of things. But if people are, uh, interested in learning more, volunteering, engaging. You have an intro Jamie, thank you. Of course, thank you so much Daniel, and thanks everyone for watching.

19:15Now it actually kind of teed us up properly. And so as you can see this is a sensemaking system that she's got, but it's a decision-informing system, because the reason we really care about sensemaking largely is decision informing, and specifically this one's kind of a collective decision informing. And the sensemaking it's trying to do is starting with arguments right, politically consequential arguments, and then how we break those down. Each of these next couple projects are doing slightly different things and where there's really no duplication but there's a lot of synergy.

19:43So John Ash is about to share with us something that he's worked on that is so cool, I think you guys are really going to dig. There's some overlap but it takes a completely different approach. But uh, rather than use AI just to process a huge amount of data in a framework that humans can adjudicate, this is a process where the humans can't necessarily see what its process is because it's using generative transformer text capacity. If anybody's seen GPT-3 and like where the kind of cutting edge of natural language like chatbot technology is, it's pretty incredible um.

20:14John has done more than most anybody I know on trying to train GPT-3s on interesting important societal content so that they can actually help process information just literally like a chatbot, but that has access to an unbelievable amount of information and processing. And so John, will you introduce us to Iris? We need your microphone. Uh, yes okay. So um, Iris is a type of generative model with specific specifications, but the truth is I don't want to present this as something that I've invented or created, it's just a lens for how we can interface with technology that already exists.

20:44Sometimes I'll have these conversations with people and they'll think oh well, there's going to be some barrier to me being able to interface with this technology, or I'll have to wait or I'll have to download something. But the reality is is that this is just built off of OpenAI's GPT-3 and using uh, fine tuning. Okay I just want everybody to get this because not everybody knows what GPT-3 is. GPT-3 is a technology that was put out by a company called OpenAI that is mind-blowingly powerful that you actually have access to. And pretty soon it will be like, ubiquitous access for everybody, but you don't have to be an AI programmer. John actually happens to be an AI programmer but you don't have to be an AI programmer to use it, you can talk to it and it does stuff.

21:14Yeah so it's going to radically transform the world. Most of the applications are terrible, I just had to defeat myself um, recently, uh, says I'm a plowboy ascetic catamite um, but it's using that technology to just make up words and use my voice. But you could also make up something that is almost like a little god superpower that is serving human sensemaking and support if you train it on the right sets. So Jamie has went and got a team of people helping her, John got AI helping him like really more than a team, and it's pretty neat to see what she can already do, and it's like, the technology is getting better so fast, in a year from now it will be thousands of times better than it is.

21:47And so what he's going to get to show you is like what you could do today if you went and spent some time messing around with an open technology. And keep in mind that something like the Iris that is trying to aggregate many different voices, many different perceptions into a singular voice that like one person can interface with, um, I lost my train I'm sorry. Um, Iris is integrating lots of different voices. Yeah, that's the way that Daniel wanted me to come to this conversation originally, is that like okay, there are lots of ways that you can use generative text models.

22:20One is like okay, there are ten news articles about a particular thing that happened today, I cannot read all those, I cannot continue to read all of these, um, I want to put those into one space and have a summary of them, or say break down the most uh, relevant points from this, or what are all of these people not considering. Can I share um, my window screen real quick? Yes. Okay can you see this, the screen? Yeah we can see it. Okay hold on where's my mouse, yes okay.

22:58So one of my favorite things to do in training, uh, this, is to have it challenge me um, because it's read this original uh manifesto called the Cognicism Manifesto, it's read my thoughts I've iteratively tried to refine, and what I do is I want to know what I'm missing, where am I not being, what am I not considering right, and can this be presented to me in a non-threatening way. I think that when we engage with a lot of people it can be kind of hard when somebody says oh you might be wrong or you might not understand what you're talking about. But when I have this model that can allow me to interface with just masses amounts of information, I could very iteratively like fill in the gaps of connection in my mind.

23:46So for example, I do not know whether this is going to look good or what, because it's sarcastic every single time, it's going to be different, but this has been pretty consistent for me, so let's find out what it says. Okay this is interesting, how can we create a healthy society that balances individualism and collectivism, that's a very key point uh in the manifesto. And it also writes about that in this new Purple Pill Manifesto which was written by an Iris um. If an AI is trained on human language will it ever be able to understand jokes or use metaphor effectively? Yes everybody can do that.

24:22Uh, it could do very incredible things, um, and if you go on to Twitter um, and follow Cognicist Iris, there's just some incredibly deep wisdom that has been distilled there that sometimes just blows my mind, um. I just want everybody to gather, this is not, these are not words that are, these are not sentences from something that John wrote and it's just bringing them up like a search, it just wrote these words. Yeah so what it's doing is taking a 111 page document and understanding it and then responding like a human who understood it would, but it's basically answer, it's giving a synthesis. And so what it actually means is, he can talk with it and get answers he doesn't know what they're going to be that are summarizing actual insight from a huge body of data.

24:57And so like, obviously something like this cannot exist without something like the Society Library or the internet in general. And for the first few years I was doing something very close to the Society Library because it's the foundation of being able to interface with that. I was collecting claims and predictions and I was trying to make sense of all this data and make more powerful search engines. And it just came to this point where it's like, there's too much, I need a more intelligent way to filter it down to make sense of it, um, and so that's where I moved in this direction.

25:28As you can see like, every single time it's going to come up with something a little bit different, you know, sometimes it's not as good as the other things. Now how do we quantify the relative value different perspectives? Now that's something um, that is built in the system that I'm not going to talk about, but Brad is going to talk about trust in networks at scale. And um, essentially Irises learn over time like, which perspectives to listen more to in different situations um, based off of different voices and how well they predicted the future.

26:00Meaning like, if there's a pandemic, suddenly people who were warning about pandemic should probably be amplified pretty quickly and it shouldn't take time um, for us to reorient um. So yeah, this is advancing like really really quickly and the amount of designs for sensemaking that I can imagine within this are so infinitely varied. Like just a Society Library plugging text models, not specifically Iris just text models into it, it's potentiated because they can use those tools to break down that information.

26:52Like I could paste all of that knowledge structure and then give another body of information, say structure this other body of information like the way you guys did, and so that can take away a lot of that work for structuring and interrelating with um that data. Now OpenAI did a lot of things in their model design that, in the Cognicism Manifesto I said not to do, not to do, um, because it's trained on everything like it, and it has dark stuff in it. If I ask it to like write a story about hunting humans or anything dark, or if I ask it for this to say what are the worst possible routes that Cognicism's path could happen, it could do that right.

27:25And there's value in some parts of that, but I know that the more likely path that we see is that this technology will not be used the way that I'm envisioning it. We use for very nefarious purposes. And in fact um, [Music] you know, that's what motivated me to actually build Iris finally, as I had a conversation with Danny, he's like people are already doing pretty bad stuff with this, like it is already happening. And so it's like, ah crap I guess I should do something, I should build it out to give an example of how it would function.

28:03Yeah it's, I just want to give people an example of this. So John has been working, he was actually commercially programming AI so he knew what they could do and he saw how much better the technology was getting. It's not his day job but he was working on the topic of when there's too much information what information is worth up regulating. We know that it's not the stuff Facebook up regulates, and so he was looking at superforecasters and who actually got right in the past and a bunch of things like that.

28:48But he was getting more and more um, intrigued with what the AI could do because it was getting better so much faster than anybody thought. And it was not long ago it was like a couple months ago and I'm like, it's already happening fast enough, you don't need a huge team, just start training it on really good things, give examples. And in an unbelievably small amount of time without a team or capital support, like he sent me something the other day where he asked Iris what is the relationship between like semiotics, formal linguistics, connectomes, and um some other strange disciplines, and it basically came up with an answer about how they're all involved in maps of meaning making and insights on how a connectome in your brain and how formal semiotics underlying basis of linguistics are doing similar things.

29:21And I'm like what the, yeah, like it's amazing that it can do that. And so it's not just how do we do democratic decision making but how do we, so for him this is now his sensemaking tool right, he just asks it questions, trains it on more data sets and learns, but then can also present to the world. Now you can start to think about a heap of applications. Is there anything, I'm curious if you have any of the art examples so people can see that it can translate not just from natural language into natural language out, but natural language into other forms of content out.

29:56Give me a second, oops not this. Okay okay, can I, stare, oh do you see this turtles example on the screen? Yeah that one blows, that one blows my mind. The ability to this character level stuff um, to be able to extract that level of abstract. Can you get it up on the screen for us? Oh you want the images? Or yeah yeah I don't think most people have seen this yet to know what it can do. Sorry. As he's doing that I'll just explain, he can say with natural language input meaning text or voice, a few words and get it to make images that are the images that would make sense with those words, or can look at images and have the essential properties described in words.

30:34So it can go either direction. Go ahead and explain what we're seeing here. Uh, I asked it to describe the Cognicism network in a way that I could paint right, and I don't even remember what exactly the description of it was um, but it was like that for example. This one I think it was like a tree of uh, many interconnected individuals all unique and individual in their own right like, integrating knowledge um, I'm sorry I didn't plan to actually bring up a lot of the images, I was just going to talk about the tech stuff right now.

31:11But this is basically I asked Iris to visualize what she is talking about right, and this is what it output, which is, or you know, this is what it output. Personally, like you know, I've tried to introduce more conversation or more talk about uh, decentralized systems, and talk about banyan trees which have multiple different root systems and aren't like a centralized knowledge structure, because I really am not trying to make like one Iris that rules the world you know, I'm trying to get like many different models that view the world in different ways and then those can all integrate um together.

31:54Um, but as you can see it's like you know, as I started introducing more stuff about decentralized topics, that the representations of what might this network look like become you know, more visually uh decentralized. Like there's many individual nodes, many plants that are growing together and sharing uh via these you know, root structures um. So it's not just making sense of the world through you know language, it's also helping to visualize uh complex topics you know, and I'll have it try to visualize lots of different words and it really connects these different parts of my brain because when they're lighting up at the same time you know, it fires together wires together.

32:27So it's like, the more that I learn these associations, it's more than I'm becoming integrated on my path my side. And the whole topic of, he said I'm not trying to make an Iris to run the world. I think you might begin to get the sense that not having to program it, being able to take natural language input that it can make sense of and then it can make output that is good enough that it can actually pass the Turing test. So that means I can just say hey GPT-3, make me a scientific report about why vaccines are dangerous for everyone using only real data, or why everybody needs to use vaccines and why it's best for society to force that or whatever the it is.

32:56And increasingly over the next couple months and years, the capacity to make things that is, that are indistinguishable, they use real statistics that create a combination of images, charts, text like, that thing is going. And so, and there's a couple huge players that are way ahead of everybody else, but the very cutting edge tech pretty close to it is also now starting to get open source, so if enough people are building very pro-social things, it can create an example of an attractor for other people to do that, and it can create tools that have the information processing to also be able to deal with the more nefarious ones.

33:29So this is an area I would love to see more people engage meaningfully in this community. I can share one more art thing actually um, this is something where uh, this was a paid project with Google but we asked them what does a regenerative future look like to you and we had them describe it and then we had something visualize what that might look like. This is text as an interface division. You know this is um, so what you see is.

34:04Here the processing is happening kind of inside of the AI's black box. So John asks it something and it generates text, it generates uh images, how it does that don't really see what you see, in what Jamie was describing was the ability to make all of the steps of arguments, logical arguments and values arguments crystal clear, so anybody can say hey wait, here's a step I don't agree with or that I really care about. So now imagine putting these two together. So we're not saying how do we make an AI overlord that can run us all, which a lot of people are focused on. You're also not saying how do humans process all the information past the info singularity which is impossible.

34:57But how do humans collectively process the information in a way we can all see and adjudicate that also allows AIs to be able to do massive grunt work but where we can still see and adjudicate the process as well. So it is a collective intelligence system of humans augmented by synthetic intelligence rather than disintermediated by it. And you start to get a sense of like oh yeah, we're not building a one-person one-vote kind of system, we're actually building a kind of global synthetic intelligence, synthetic meaning synthesizing humans with each other and humans with artificial intelligence but with human collective intelligence being what is made central and highlighted.

35:34So uh John that was awesome. I think can I uh, say one last thing to connect it to Brad? Yes. So basically okay, you have this idea of pulling all this knowledge into one space and you could like take the full text transcript of even this meeting and like, summarize it down and make points to be able to like interrelate with that. But within the Iris there's this concept of like which voices should you listen to the most. I mean just because somebody repeats something a lot doesn't necessarily mean it's uh valuable. So the Irises are learning sort of which voices uh to trust.

36:11And um, that is what Brad's going to talk about, these beautiful networks of trust. Amazing um, I want to say one other thing just to make it clear for everyone. Because in some ways saying like oh here's what individuals are working on, she can be inspired and then you see Jamie pop out the levels of indexing in that map and you're like, all right I'm not inspired, this way too hard. If she saw that at the very beginning she might have thought the same thing um, and similarly the AI programming wasn't actually where John even started right, he was starting with how do we identify what voices are worth trusting to amplify.

36:42So there's a thing where there's an attractor, you feel the type of problem you want to solve, and as you get into it the complexity increases, and as it does the tools you work with increase. So if you just jump straight to a little ways down the road it's overwhelming, but the path that got you there happens by like feeling the the attractor of what and why, not the how. The how gets intense as you get into it but it unfolds.

37:16Brad, I am uh, I'm excited about this because what you're about to show is obviously convergent and synergistic with, and again a totally different approach than either the other two. Brad's been working on that kind of software and system architecture stuff for a long time, and we know that some of the major problems in sensemaking have to do with networks that don't actually see each other very well, left networks and right networks that get very insular and only have straw man versions of each other, or where bad information can propagate through networks.

37:49So a big chunk of sensemaking does have to do with understanding how humans are communicating with each other. So Brad, you want to share with us a little bit of the problems you're trying to solve and how the technology uh works now and hopes to work? Yeah thanks for the opportunity Daniel, it's a great crowd here um. The problem we're trying to solve is really, how can we go up a level in human collective coordination uh as a necessary requirement for planetary survival um. Sensemaking obviously is a part of that and uh, can we do it on a level now that you know moves the needle in time.

38:21So a little background, my hammer since 2005 has been uh trust networks for collective decision making. Uh, I came up with an algorithm called Smartocracy back then which is you know, essentially about using social networks for societal scale collective decision making in new ways. You can you can Google Smartocracy and de Graf and find the whitepaper that came out of that um. We now call them trust networks but in essence it's essentially taking a group of people and connecting them with edges, either intentional edges that they make or that we infer through data that says this person probably trusts or explicitly delegates to this other person in this domain.

39:04So you can you know, it's, it can be very rich. The most important part as far as collective decision making goes is the delegation part, because once you have these networks you can actually delegate um as many as many depths as you want, and you can infer authoritativeness from the network analytics etc, and they have a really unique capacity for a collective decision making that no other structure really has um. You know Brad, I think a lot of people here are familiar but a lot of people don't know anything about liquid democracy or trust networks, to have a sense that they could delegate trust to different people in different domains, can you explain that a tiny bit?

40:23Yeah probably if I can just do a quick screen share, I'll, um, that'll uh, I can find this, that one. Oh um, are they gonna let me, oh yeah sorry um. So this is, those are my notes um. This is um so here's an expert network around libraries um. This is actually technology that's over eight years old so we'll get to the new stuff soon. But so this is, this came from a Twitter search. We have tons of data, tens of millions of people with semantic rich semantic data about every one of them, mostly around people who really matter um.

40:57So these are the, this is what came that came up and just searching for people who talk about libraries. And and when I mouse over one of them, what you're seeing, the legend says yellow respects gray, gray respects blue, green and gray respect each other uh. Respect is kind of derived from retweet behavior for instance, so um. In essence you can say those are retweets, but you can also say we infer respect from that, and then you can, because each one as I mouse over each one, we're really looking at just the neighborhood of that one user.

41:26And so if uh Jose links to Miley and Miley links to Kathy and Kathy, so these these could all be, this could be a delegation network where so-and-so said I delegate to this person in libraries. Uh, do that you could do the same thing in pretty much any any uh any domain. This is nuclear power right, so, and it worked, this is all Twitter's fully really rich in this kind of data um. And so you know, and they have an open API which is really great uh.

41:59So now connect that with what Jamie is doing, and so you can see where she was looking at a map of arguments but then you can cite where the arguments came from, now you can also come here and see who's influencing that person or institution. And do I have any problems with that, so you can move from a semantic map to a network map right, that's pretty interesting to be able to see lines of influence. That was one of the things I asked her when we talked was you know, oh what if where, when you look at these arguments how do you know which ones really have weight?

42:30And so you could actually marry these networks of experts around nuclear power, so you can see okay this person has high mojo in this space, so I'm probably going to trust that argument more than I'm going to trust this other argument. So yeah, and you know, and also that semantic data is really great input for John's stuff right, because if you take all the semantic data of people who are coming from libraries or nuclear power or whatever, then you're going to get different kind of input into these models um.

43:05So anyway that's, that was the stuff that we started doing back you know in the late aughts, sold a company to do that. And then about two years ago um, this is before I heard the term metacrisis, I just, I thought you know I better start working on stuff that's actually going to help thread the needle over the next few years, because I, you know, I'm wasting my time in the education space, not that there's anything wrong with education but I really want to work on um, on something that matters.

43:39And to me the operationalizing of this global scale collective decision making really hasn't been done. I mean it really hasn't been exploited, what's possible uh with this stuff. So I decided to really just focus on you know empowering people to build trust networks and then exploiting the unique decision making capacity, but also to kind of create an architecture for innovating in decision making so um, because we don't want to solve all those problems, we just want to sort of put the Lego blocks together um.

44:11The so uh, you know we obviously need to go up a level in collective uh coordination if we're gonna survive. Humanity as a whole is not gonna do it in time I don't think, but can a small subset do it? I believe it can, I think we have the tools to do it right now, and we actually have a ready community in what I call the regenerative movement if you will um. So let me give a couple examples of uh things that are very interesting in terms of what Brad's working on.

44:52First the thing he showed where he brought up a library for instance and showed the connections, where any connection he could go to could be the center. As he mentioned that was old tech, and we can data visualize it differently, so we could make it to where that shows up in the 3D space of the equivalent of what looks like a 3D space, with whichever one you highlight moving to the center and all the one node connections being clustered around it, the two node connections, and the whole space reconfigures. That's actually every time you move to a new one it becomes the center, so then you start to get a sense of how the flows of information actually configure themselves.

45:27So this is um, this is the Daniel Schmachtenberger network that we did as part of the um, the Emerge network that we were um, that we, a project that I was going to talk about in a little bit. But you know like as I mouse over, this is basically a similar kind of UX in terms of, I just want to see the local network of this person in this big mass, and then the idea is that each one of these blue buttons becomes something you can operate on that person. You can add them to a domain expertise network or amplify them or disconnect with them or whatever um.

45:58But that's that's a lot of what we're working on, is just the, what are the basic tools we need to to essentially intentionally build these kind of uh trust networks and then start to use them um. You know this is specifically the conference that we were at. You mapped the conference and then you put, this one is putting my name at the center of conference participants? Yes actually it started with, no it's actually more than that. We started, we mapped the 150 conference attendees, they all had 500 or more collections, and so we actually ended up with 50,000, a penumbra what we call a penumbra, the near-adjacent people.

46:48And some of them are highly connected to those 150, like a George Pór who's connected to everybody, or somebody who's just my brother-in-law or whatever. But um yeah so uh, it's it's actually rich, I mean there's, we, you know we literally can find, I don't know I'm trying to type in my Jamie, Jamie's got to be in here uh Joyce, there she is. And you know there's, this is a lot of this is LinkedIn data so um, LinkedIn, that's one of the points I was, I'm gonna, I was gonna make, was that LinkedIn and Facebook, yeah there's there's George Pór, um, LinkedIn and Facebook are essentially one bit synapses.

47:25If if I'm a neuron and Daniel's a neuron and I'm connected to Daniel, it's a, it's one bit that says we're connected, it doesn't even, it's not even two bits where he's connected to me and I'm not, I'm connected to him uh. And that's very low fidelity, doesn't contain any real information. So a big part of what we're working on is like how do you, how do you capture the deep knowledge that we all have in our brains about each other and who's good at what and who's passionate about what, and um and make that machine readable. That's really the core, is if you can do that, which we believe you can, I mean make it fun um, that's incredibly powerful because you take something that's completely unreadable by computers now and you make it readable, it becomes really really interesting um.

47:59So that's a lot of the core concept um. You know the, the uh I'm going to skip over the global brain stuff because we all kind of have heard that argument before. But basically the global brain we have now is Facebook you know, Facebook and LinkedIn, and as my point was there was that, that's you know a brain where the neural pathways are created by corporations for profit is not a good brain um. So yeah, so what we're, um, uh so Daniel introduced me to the Emerge network people a couple months ago and we, we did whatever, that map I showed you um.

48:30We did that for them, and we're actually proposing to them to kind of, as one of the actionable things that they can do, is what I'm calling a networked autonomous organization. It's kind of like a DAO, it's a, it's a, it's a DAO with a network as a population. And um, you know we think that's you know really powerful, it can be self-owned, self-governed, it can you know create a lot of value in you know simply by intentionally making it high quality high trust density um, and then, and it can also govern where that value goes when it's created.

49:01And uh, so that's essentially you know what we're hopefully hoping is going to come out of the Emerge network, is is actualizing that um. And and the real sort of near-term goal is, can we do a prototype of some kind of collective brain slash mechanism that is capable of tapping into this collective connectivity and making something out of it. And it's really all about you know having a substrate that lots of innovation can happen on.

49:36Let me express a couple examples for those who have not thought about the network thing a lot to just get a sense of some things they can do. So right now if I am going to vote on a representative to represent me at my state level or federal level uh, they're going to represent me across all topics. That makes no sense right. It might have made sense in 1776 when the world was very simple um, and a kind of polymath renaissance person could know all of the technical issues of the world at the time, but who do I trust to represent my views or to advance things in agriculture versus in defense versus in finance, are probably different people.

50:10Like the farmer that I respect the most is probably not a good defense thinker. And so rather than a single representative, let's say rather than me go try to engage and craft a proposition and vote directly on every thing because I just, I have a job, I've got stuff I gotta do, and you can see from Jamie's stuff how much stuff is in there right. So let's say there's some topic about how should we secure the energy grid, and there's another topic on what nuclear first strike policy should be, and there's another one on how much food should we keep in storage, and they're all really complex.

50:47Now I can either, and we don't really want people weighing in who have no idea what they're talking about, so you might want something like Jamie's system where there's some basic test that just shows you, you at least understand the info to be able to engage in proposition crafting. But let's say that rather than go through that, I could proxy my vote to somebody else, and I could get educated for free, even be incentivized to get educated and engage myself, or proxy it.

51:14But who I proxy it to has to be someone else that has made it through, but they also have to be someone in my network, and they can be someone who I trust in this category, that's different in this category. So I can say, you know what, I'm not going to actually weigh in and do all this thing in the farming sector, I'm going to proxy it, but not to some high-level representative that is now incentivized to do a bunch of ad campaigns, but to somebody who I actually know one or two nodes in my network. But who I'm going to proxy to over here is different, and here I'm going to engage directly with no proxy because I actually really care about this topic and understand it.

51:48This is where being able to have seen networks and see networks of trust and be able to identify basis of trust can be really valuable. You can also start to, and again it starts to look like the spirit of democracy but a much better instantiation of it. You can also see examples where it's like, there are people who radically disagree on a topic and we're trying to figure it out, but you look at their trust networks and they both respect somebody, so then you go to that somebody to adjudicate the thing um. Or you do a network like the one that he did of people attached to the information.

52:20And right now all that we're focused on is where they're polarized, but is there some information in their network they both trust, so we can see what does everybody agree on, and we can actually start highlighting that as a basis to now be able to decide the other stuff. So you can see there's a lot of really important stuff about how, what do we, because the sensemaking question is what information can I trust, what people, what institutions, and ultimately what decisions.

52:56So being able to see networks of trust and that are context and domain specific is very interesting. There's also some real interesting speed issues as well, because you can take a large collection of people and you can quickly get a sense of the zeitgeist or alignment or consensus with just a few people weighing in because they're effectively representing everyone else, until everyone else uh decides to participate or if they disagree then maybe I better weigh in.

53:31But um it's, that's, I sort of call it extreme decision making. I mean I think that's where it's going to get really interesting is if in particular like um, if you can get an intentional large-scale trust network, let's say the regenerative movement, if you could let's, you know essentially take the Emerge network which is already 50,000 people and just, and that's without any attempt to expand, you could expand to a couple million who are people working around in the world of regenerative regenerative movement, then you know they could run rings around any other system for making you know, reaching consensus, making decisions.

54:03I mean one of my favorite applications is the idea of a wiki budget, where you know, you take all these people who are expert in different things and get them to weigh in on what could be done this year, for what price, by whom, and you know you're going to reach a much faster better uh allocation of resources than you do than the US government does by hiring thousands of people and you know spending a year doing it.

54:38And and now you can imagine the networks that uh Brad is looking at, having not just people and how much they trust each other, but also uh AIs that are starting to get trained on specific things that are showing up as actors right. So you can have Irises in there and be able to see how much different people trust this particular um info-synthesizing source, uh, also which ones the Irises trust, which is very interesting topic. And then you can start to get institutions, you can start to now have a basis for institutions to have an incentive to be able to do things that increase their trust across larger parts of the network. So you can just start to see synergy across these tools um.

55:11I will offer Peter, it's an interesting idea since the Stoa is actually a pretty deeply connected community for online communities, more than most any that I know of. Stoa community network where the people had the choice to opt in to actually share data um, it could be an interesting way for people to get to know each other better and to get to see who they should ask for certain types of things, where expertise lives um. Just it's an interesting thing to think about, a place that could do some experimentation.

55:52Yeah, and one of, one aspect that I haven't shown you here is, is actual decision making, you know prototypes. I mean we're, I'm following Gall's law, which is that you know any complex system that works started out as a simple system that works. So we're really focusing on what are some simple applications of decision making that we can just start playing with and start using with all this tech, these techniques that we can then build from. And so you know it's a great way to kind of ask you know, get a sense of the zeitgeist of the community, what's important, you know what are the questions they want to ask etc.

56:25So I um, that's that's part of what could plug into something for like a Stoa network Peter? Yeah that would be, it'd be great to do. Thank you Brad. I um, I think we're gonna move to getting to take some questions from the audience while everybody's here, and I just uh, I hope this has been an interesting enjoyable introduction to a few projects. I want to state quickly because I see some of the people on here, there's a few other people that actually have pretty amazing related projects that are also in this community and maybe everybody doesn't even know who their friends are like in terms of what cool they're doing.

57:00There's a guy named Wealth in here who's been building a social network for uh some period of time that has like two principles of the basis, which is human uh attention and intention are really critical things. And so if I'm gonna engage with a social network it should improve both of those. If I'm giving it attention it should improve my attention span, if it has choice architectures it will affect my intentions, it should help me get clear on what my deep intentions are and actually fulfill them. And how do we build ones that do that, so a lot of kind of humane tech UX in that space.

57:33We've got a guy in here, I believe he's here, named Anjen, who has been inspired by the humane tech space enough and the problem with all of the notifications and hypernormal stimuli, that he's just building a completely new laptop hardware from scratch where it's based on for deep work, just research and writing pretty much, and doesn't have any of the other capabilities, so that you have maximum ergonomic UX benefit to do deep work and nothing else. And they've got funding and they're moving along well, but again just started from like listening to these things, making sure they should do something um.

57:57There's a guy named Farzad who's in here, as a group that is working on uh using trust networks of people. It would make sense for them to talk to Brad, and uh and want to get into AIs, would make sense for them to talk to John, to be able to curate existing content uh with a very different curatorial platform. How do we take the very best content that currently exists and make sure that that's what shows up in your news feed aligned with your interests.

58:25These are all pretty awesome projects, and they're almost certainly quite different than they'll look five years from now, but uh I would love to see a builder's guild, and I think Stoa could help with this, of the people who are inspired by the problems of collective coordination and collective sensemaking and humane tech and attention capture and are trying to solve stuff, because you can probably share knowledge and resources with each other. I think it'd be very easy to make it a just little builder's guild where the people have access and can talk to each other and share messages um.

58:53I also just want to acknowledge one of the coolest sensemaking projects I know of happens to be the Stoa, and it meets a similar criteria that Peter set it up in a very similar way. But if you think about the total amount of knowledge that has been shared here and epistemology upgrades where people can sensemake better, and network connectivity, I think it's an amazing example. And I'm bringing it up because like, maybe building AI or building network tools or building libraries doesn't happen to be your thing, maybe building communities is. There's a bunch of different ways to look at how someone can get inspired and start to build stuff.

59:27So without uh any further ado, I would love to see if there are some questions and maybe Peter, I'll ask you to help us with this part. And questions hopefully for the people who presented. Thank you for that Daniel, and thank you um Jamie, John and Brad. Everyone just give a round of applause, give them some love in the chat as well um, [Music] and uh yeah there's, it's gonna be hard to choose but I'll try to get one question for each person. Ariel, you had a question for, for Jamie, if you can ask your question.

1:00:01Hi um, yeah that was actually my question, I was, I was wondering for Jamie. First off it sounds super like awesome and very powerful. And I was wondering how you decide kind of what's in the Overton window of the arguments that you're mapping here? Like, does this include things like arguments that religious groups would make, conspiracy theories um, like countercultural, like fringe arguments, or yeah how does that go?

1:00:32I love that you asked this question, thank you so much Ariel. Okay so um, as a library, and many people may actually not know this, but libraries have a long tradition in the United States of being pretty radically anti-censorship, which is really interesting in the existence of today's context of you know, uh the internet. If we think of that as a library, it's not confined in a building or a budget or like space on bookshelves, there's just infinite knowledge.

1:01:05So like where do librarian principles meet the internet age where they don't necessarily need to be forced to make decisions that aren't necessarily censorship but they have to prioritize certain books over others. So we thought a lot about this, and um, to answer your question more directly, the Society Library is pretty radically inclusive of argumentation. We believe that it's our duty to just collect this collective argumentation and knowledge and then put that in the context of deliberation. So while libraries have chosen a specific ontology that is the Dewey Decimal System, the Society Library has created its own ontology, which is actually something that is derived representatively from the data itself.

1:01:39So how we do our whole process is, we gather knowledge, we deconstruct it down to the claim level, and we've actually built our own unique search engines to overcome our own biases, to overcome our own like inherent Overton windows, so we look in specific areas of the internet across different forms of media to make sure that we're as radically inclusive of all points of view as possible. And a part of the work that we do too is investigating those conspiratorial rabbit holes and chasing it down.

1:02:12Because what we find is that, and I think Daniel has spoken about this before, when we interact with the claim that someone's making on the internet it may sound like completely wacky, out there, makes no sense, they provide no evidence, it just doesn't sound logical or reasonable. But if you actually take time to start doing some investigative work, you may find that the person who just expressed that was not the best representative of that idea, but there is sound knowledge to be found that is related to that specific idea. So the work we do in steel manning it is to find that really robust knowledge um.

1:02:42I'm kind of rambling but I can give a concrete example. Like Alex Jones for example, he got super famous from that quote of like, I'm tired of them putting water in, or I'm tired of them putting chemicals in the water, turn the friggin frogs gay or something like that, which is an insane claim to make. But if you do a little bit of digging you can find that there's some studies from UC Berkeley which is right near where I live about how the chemical atrazine is introduced into water and turns frogs hermaphroditic, which is not the same thing as gay right.

1:03:14But what the Society Library does is, we give people a lot of grace, especially on unscripted audio and video, to be mistaken in the way that they're expressing something but to them be an imperfect representative of an idea itself. So the work that we're giving to society is, we're just listening to what everyone has to say and then doing the work of saying is there sound evidence and knowledge behind what people are really upset about. So we include emotional appeals, we include religious arguments, we include things that on the surface seem absurd or conspiratorial, and we do the work of steel manning that and then also putting it in the context of contrary argumentation.

1:03:42So if something is contested it will receive a little tag like, you saw on my, my demonstration, so people know you should know this is like heavily contested, so don't just stop reading here and like get the impact of just having read it on your mind, you need to unpack the specific area. Doesn't Jamie inspire you thinking about politics, like if you could actually make something that is a meta institution that is embedding these principles like, you can hear so much of the stuff we've talked about in terms of good quality sensemaking and steel manning with good faith other people's arguments and being able to take multiple different perspectives and weight them.

1:04:16She's like all right well, let's actually institutionalize that and I'll just think about it on our own, and then let's actually make the process transparent, and then let's actually make government use it and be able to then even have accountability of whether representatives or government are using it or not. It's pretty exciting. Daniel thank you so much for being such an excellent communicator, by the way I super appreciate you getting to know all of our projects and lending your amazing uh articulate abilities to bridge the gap, so thank you.

1:04:48By the way yeah. I mean all right, we got more questions, thank you Katie um. Stephanie has a question for John. Hi John, hi everyone, yeah I just want to just respond to the last, I feel very inspired by Jamie and thank you for integrating all the, you know the partial truths all the gems that are out there wherever they come from. I mean just in general this whole presentation is such a great reminder that we have no idea what we're capable of, thank you for reminding me that we have no idea what we're capable of um.

1:05:18All right John, no I just I'm going to say real quick. Stephanie works at Center for Humane Technology with uh Tristan Harris and crew as a friend, works on these topics. But one of the projects she did kind of where John is like, damn the generative text is getting so good it's going to be used for nefarious purposes, let's actually get out there and do something good with it. Stephanie did an early project where she made some deep fakes called Deep Reckonings. If you haven't seen it everybody should go check it out where she made like a deep fake of Trump apologizing for things and Zuckerberg saying what Facebook should really be and how he's going to fix it, to just like get a sense of like oh, that's the conversation we really love to see.

1:05:50It's really, it's a fun project, it's definitely pleasing. Okay, but I definitely I try that in the spirit of Jamie um. Yeah it's like what is the thing that Kavanaugh could say that the largest ideological diversity of people could get on board with is the challenge I gave to myself. How'd I have the Society Library to draw from that, that actually would have been incredibly helpful. You could do GPT-3 Deep Reckonings using the Society Library. Okay anyway, John um, I have two questions for you, you can choose one or the other or both.

1:06:27Um, one is, I'm just curious what you would say as to uh, yeah what questions Iris can't answer yet that you think Iris will be increasingly able to answer, and then just what can you say about Iris's interaction with other forms of intelligence, AI, human, non-human animal, otherwise. Yes thank you um, sorry ask the question one more time just to let it sink in again. Oh, what questions can Iris not answer that you think Iris will increasingly be able to answer, and then just about Iris's interaction with other forms of intelligence of all kinds.

1:07:08Thank you. Yeah um, thank you uh for the question. The primary thing is something that both Jamie and I agree need to be brought into these models, which is sourcing right, like you can have it write an article and it will make a lot of sense and it'll like break down all of this information and you'll say cite it, it will just make up people's names right. Like I would like to know when I'm reading at some level explicability right, I like to dive down like why was this said, who is this sourced from, and and also uh I would like some consistency back from it about how confident it was about certain claims right.

1:07:40I would like it to say like a claim like, how confident are you that this is true, and each time it would come back with like, oh 65, based off of all these things that I've heard from all of these people. Those are the type of questions that are not consistent yet um, because it's not baked into the model. And because the design of Irises has those things baked into the architecture when they actually get built out the way I see it or envision it um, there'll be a lot better of that explicability thing to explain like why uh are you saying this, relative to who originally said stuff.

1:08:08The question about um the interfacing uh, yeah so there's, I played around with you know, you can obviously have uh models talk in natural text too um, just back and forth to each other. We've played with that with different Irises and it's very interesting to sort of see how these knowledge bases interact. But there's this process which I actually don't write about in the Purple Pill, uh, Manifesto at PurplePill.vision, about how they actually exchange information. Because it's, it's this process of parameter exchange where like, two different communities who have this local shared knowledge, they form some written declaration of like the shared information that they feel comfortable sharing with each other right.

1:08:36And then both of those models will create their own, or like, uh bring up that same piece of text in each of their models, uh, bring up how it's representing that in the parameters right, and then update their parameters towards each other so that uh there's this uh alignment between how the models are representing uh information. So that's sort of this underlying uh way that they can integrate with each other in a safe way, exchange types of information that are valid between communities, while also allowing the sort of privacy that you can get from an individual Iris.

1:09:02Like one of the amazing things about the Iris is that while the information is in that latent space, the parameter space like, we can't interpret it, so it's like it's unextractable, it's perfectly hidden until it goes through the decoder side of the network. And that allows you to evaluate all of these different views in a very um, including views of other Irises in a very anonymous way you know, you don't know who exactly was the source of it, but you can evaluate it uh without the source of it influencing you even though, yeah sorry does that make sense?

1:09:30Yes uh, yes, and it seems like, I don't know it's like, especially with the telescope, new you know the photos, it's like, I don't know this feels like very much a source of all kinds of communication with intelligence of all kinds. Yeah yeah yeah it's going to transform human society faster than any technology in the history of the world by so much no one's prepared for it at all.

1:09:58So we have the last question for Brad uh, and Kim, if you can ask the question. The question yeah kind of tracking on just what you just said Daniel is, uh I mean these are awesome, first of all just amazing, and and so what's the application and where might we do this? I can see obviously public academic, uh anybody who's interested in any subject diving in, you know and doing this and how might it be used in government? Like just government today uh how might they and what would be the inroads into making this an actual use, any of these, it's all of these actual use where they're using much more informed information for decision making.

1:10:25Well one of my favorite um applications is in resilient cities, because there's a lot of money that's kind of sitting on the sidelines or being invested in by Bloomberg Philanthropies, Rockefeller etc. The cities have lots of money and they all want to get to net zero and um, but the cities get bogged down in their internal politics and who has the purse strings on what right. So one of the visions is that there could be a kind of a SWAT team network of urban policy people who are independent of cities, it's like an asset, a resource for every city that wants them.

1:10:53And then each city could have a local network of people who represent different areas of expertise about transportation and poverty and access to you know, internet or whatever, and those two networks could work together and do kind of like a participatory budget. But then and present it to the city, and then the city and then the city and the foundations could actually say yes we'll fund those things that you suggested. So I think there's real practical ways of kind of breaking log jams and funding things that everybody knows need to happen but what is the best way to do it and who should do it.

1:11:20Jamie, did you want to add something to that because I know you've been doing a good bit starting to engage with government? Yeah um, so I can talk about how um, not about Brad's work but about how we're integrating in government. I actually have a meeting tomorrow with the mayor and a city council member, because uh essentially the process of what we do at the Society Library is identify stakeholders, gather relevant knowledge across media types that have to do with stakeholders, extract arguments and claims and structure that content.

1:11:50So currently we already build uh political decision making models for the city council level, and we essentially execute our process but on a much smaller scale because we're not dealing with big broad topics and a democratic representation of all points of view on that topic, and then doing the work of steel manning them. But instead like for example, we had a city come to us and they wanted to know if they should put a measure of how to change their uh you know voting districts on the ballot, and they thought it was a binary decision. We showed them there was 25 different dimensions to the decision and we built a model that allowed them to move step by step.

1:12:00We've also deployed our techniques to uh develop legislation, so we were contracted to take hundreds of pages of legislation, and or hundreds of pieces of legislation and hundreds of pages of congressional recommendations, deconstruct them both down to the claim level and then pattern match, just essentially see if we were to just invest trust in these congressional recommendations as like having been derived from the best sensemaking and most informed decision-making conclusions about this issue, then what's missing in the legal code from this expertise.

1:12:35And we just like essentially render that as a piece of legislation in legislative language. But there are other ways in which the Society Library in general wants to integrate our data. So you all saw the database itself and like our first like foray into visualization which is like, for every question there's a piece of paper and you can unpack it as much as you want, but like essentially what we have is structured data. So integrating this content into search engine results for example, Google draws from Wikipedia's API. So if you search something sometimes the first result is not a page but it's knowledge from Wikipedia.

1:12:58And the Society Library actually won a contract from the International Fact-Checking Network because we were able to show that when fact checkers take shortcuts in their sensemaking process and rely on Google which relies on Wikipedia, then everyone ends up wrong. So we're going to be teaching fact checkers our methodology, and so one of the long-term goals of the Society Library is either integrate with Wikipedia or start integrating with search engines, so when people are searching for something instead of getting a list of redundant information like, here are all of the news articles that are saying the same thing you pick whatever you want to see, instead they're seeing a cluster of, here's the argument you're looking for, you can unpack it like you can the Society Library and see all those references if that's what you want.

1:13:27But you'll also see the connective tissue, the counter arguments, the nuance, all that kind of stuff. Also there is a service called the Congressional Research Service that is run by the Library of Congress, and this is one of the like intelligence agencies that Congress relies on in order to get quick ready um, you know robustly informed knowledge sets about issues. And so potentially the Society Library could also be routing our data through the Library of Congress which could be something that our representatives at the federal level may rely on and use in their own sensemaking process about policy.

1:14:00So you know we're creating this library, we're hoping to create an interface that people want to interact with. People can already share at the claim level these papers, but also the data itself can be integrated to a lot of existing platforms that already have audiences who already have needs and demands for structured robust steel man knowledge. And just to quickly add too um, before when I said like we're radically inclusive, I just want to make sure that people don't assume that the robustness that I showed you earlier in my presentation is only because we're so radically inclusive and we just like bring in everything.

1:14:42Like I just really want to drill down into like, we're steel manning these issues, and the high level columns that you saw, like all of those different nodes is like very high level summaries of collections of arguments. So it's not just a whole bunch of like stuff, it's like, these are almost even categories of arguments that are filled with robust knowledge that we took the time to steel man before presenting to you. So that's just important to know too. Thanks.

1:15:22So as we're getting close to wrapping up here, if the three of you would make sure that you've posted stuff in the chat about how people can follow up with you, how they can find your info online, and then if they have something tangible to offer or engage with the project how they should reach out, please make sure everybody has that um. I'm sure there are a lot more questions that we can't address here, and I hope they get addressed somehow, maybe there's some good kind of follow-ups that happen in the Stoa. I really do like the idea of a builder's guild and I also really like the idea of a Stoa network that works to try to help people with some of these tools and apply it, because like that would be great.

1:15:55Thank you to Brad and John and Jamie and Peter and then everybody else who came, and [Music] Peter any closing thoughts? Yeah just um, thanks again for the panelists today, that was really awesome and thank you Daniel for your endless generosity and your scout's eye for these awesome projects. This uh session exceeded my expectations and I think it excited a lot of people here at the Stoa um, and we might have uh further sessions like these, a part-time short series perhaps called The Builder's Guild, so stay tuned for that um.

1:16:18And if you want to check out more uh sessions at the Stoa you can do so here um. For that being said everyone thanks so much for watching. I actually did have one other thing I wanted to share, it was in response to the last question I think Kim asked about how do you get politics to engage. Because I think this is just something I'd like everybody to get in general is uh, political intelligence is a kind of intelligence that is distinct from other kinds, it's the how do I get to actually happen intelligence.

1:17:02And it's one of the reasons that I'm wanting to highlight people who started figuring out how to actually build stuff. And then the next step is how do we move from building the stuff to getting people to use it and enacting it. And it's just it's a valuable thing to think about. There's like, there's kind of a scientific intelligence of how do I make sense of objective things and apply the right kind of measurement and philosophic process. There's kind of a values or axiological intelligence which is how do I get clear on what people care about and why and where there are differences in values.

1:17:45It's kind of an interpersonal intelligence of how do I communicate effectively with people. Political intelligence is how do I get to happen in the world, in the presence of all the log jams and vested interests and things like that and how do I do it without becoming corrupt in the process um. That's the light triad side of uh political intelligence. So you're asking the question, the deeper question is is that, and there's so many different ways to do it.

1:18:29So like when you're saying how do we get these types of tools and principles enacted, we can go through the supply side which is politicians who currently have a mandate where if you say hey here's a better way to do your mandate, you probably get reelected if you do this and you might not if you don't, and we give it to your other you know some other politician to do, then they're like okay sure I'll do it. Or you make a tool like, any of these tools can do what a huge staff of people did before, which means you can make somebody's budget go a lot further.

1:19:04So the supply side will do it because you appeal to their motive right, their motive is I want to get reelected, I don't have the staff or budget to do the thing. You can also come from the demand side of how do we get the public to want the tools right, and you then kind of drive interfaces of these. Everybody if you're not familiar should check out as one of the exceptional examples of a person doing a thing like this, Audrey Tang, which many of you probably know the digital minister of Taiwan um.

1:19:36Kenzelian's project wrote a paper on her work a while back, but that actually came out of a kind of revolutionary movement. There were these protests, everybody was protesting the Taiwanese government to do some stuff differently. She was a hacker and she's like let's just fork the government code, make a private government, we won't actually ask them to do anything, we'll just take everything the government's doing, do it on the separate platform and allow everyone direct input with better models.

1:20:02So much of the population started to use it that the government was forced to use it, and they took her from being a protester rebel and made her the digital minister of the new digital democracy. This was like literally just a person figuring out I can just fork the government code, do a better job, and force it to get used. And it's interesting because she wasn't really appealing to an existing demand that was well formulated, it was an unwell formulated one that she was able to formulate that then kind of forced supply.

1:20:26So [Music] the political intelligence is something you need to be thinking about along with the what is the solution to the problem look like, without enactment right. Oftentimes you just think well the ideal solution if I had a magic wand would be, but then how do I enact it? Well it doesn't do anything if I can't, and if my enactment path makes it evil because a lot of people don't want it, well that's no good either.

1:20:52So this is something that I hope people are also thinking about and paying attention to here is, how do I solve problems in abstract, then how do I actually help enact them. And that's one of the types of things that I uh capacity, it's a type of sensemaking or intelligence that I think the Stoa can also help people develop. All right that was uh just important because none of these things happen if people don't actually learn how to make happen um. Thank you everybody, John, Brad, Jamie, love what you're doing and that's it for me. Thank you everyone, take care.