Thanks so much. I will briefly begin by by just giving a bit of background. I have a a kind of a a strange strange path to tech. I graduated from University of Connecticut in 2007 with a double major in political science international relations, and I was really interested at the time in collective intelligence. And I didn't didn't really know what to call it at the time. I there was I had not I hadn't found during my time there, you know, the hook into that community. And even now, I think it's still kind of kind of hard to know where to where to where to where to hook into that and and what brings people together. It's really interdisciplinary, and I really love that about it. And I I'm kind of a generalist. So anyway, I I ended up coming into tech basically to build I really wanted to build Polis, and I I had and I I did really general idea, but I had, at the time, no network and no product experience, and and I didn't program. And so I between 2011 and present, I've spent, you know, a lot of time learning web technology side and machine learning, and and a lot of that is is collaboration with people who aren't in the room. And so I'll I'll try to reference some of that. My my and and some of those some of the work of of the people that I've collaborated with to to bring about what we're gonna talk about today. But my background is is and interest is really from the collective intelligence side and how do we how do we consider that we might use communication systems to help people do smarter things in groups. It's pretty pretty general. And the specific way that that's that's kind of manifested for me is in what's what's effectively in short in a a system for gathering opinions, which asks the people who are being surveyed to create the survey themselves. And that is a a a more democratic, it could be said, because the dimensions of the survey and the dimensions upon which population is considered are created by the population in their own words, and they can they have some more agenda setting power than they would if power is asking what power is interested in in the words that power wants to ask it. And so that that agenda setting that agenda setting shift has been something that I've been really excited about as well as trying to to to bridge the divide with between formal official, you know, quantitative survey methods, which, of course, rely on qualitative statements, but those statements are created by by those in power. And the kind of fuzzy ethnic ethnographic qualitative interviews, which, you know, come from come from the words of of the populations being interrogated. And so I, you know, I I think that, that that's that's particularly fertile right now because we have so much so much more mechanism, to play with and so many more so many more entry points with mobile, to be able to say, well, you're on the bus in the morning. And if the mayor has a question for the whole city or a random sample of the city, it's totally realistic for the mayor to ask the question is, ask the city a question in the morning, one to one to 10,000 person random sample, and to have something coherent and meaningful three hours later. And so if if if that's an assumption, how does that change policy, and how does that change how does that change regulation? And the more and more we can automate that, and I think that, you know, there are really big questions about how we integrate the voice of the public into into the political systems at large. The big theoretical and and mission piece for me that I am that that is part part theory and I I think part pathology, and I'm part I'm happy to to I'm happy to answer any questions about this. For me is that I genuinely think that in the next hundred years, it's possible for us to do away with the political party as a mechanism for reconciling complexity. And so there's a in in systems of democratic governance. So I'll go into this just a little bit because so we can talk about the tool and what it does and why it does it. And that's quite independent about whether I'm right or wrong about the some of the more, theoretical implications that I think it has. So those are, the the on the theoretical side, I I my view, is that what happened in Taiwan specifically, and if that's where most of the coverage has been been focused, is interesting because it took a single issue out of the national conversation and dealt with it with stakeholders and with citizens and, in in a deliberative process and resolved it in its own environment. And the result of that was that it took one issue off of the national conversation. And then if you consider the national conversation as a kind of bundler party platforms as these kind of bundlers and aggregators for thousands of issues, that that that serves as a kind of mechanism for for simplification, but also gross and destructive oversimplification of issues, and the polarization is a result of that. If we have if we bundle up all the issues and we say, you have, you know, you have you can either vote on bundle a or bundle b, then then you have polarization based on the based on on the the affordances. So these are the affordances for conversation. So kind of, like, my metaphor for this generally has been, like, Netflix, you know, comedy and drama, and then you fall into one of those camps, and I'd say, well, I I suppose I'm more of a drama person. Right? In fact, Netflix has 13, 1,300, to case preference categories and 79,000, areas. And so we have of of, genres. 79,000 genres and 1,300 taste clusters based on viewing preferences, and these span nationality, and they span, you know, gender, and they span so I think it's it's a we we already have this incredibly high resolution, resolution similarity and interest group, you know, categories in in the marketing space. This is this is well trod territory in fact. We you know, marketers do not think in terms of of of, like, two categories. It's it that's not refined enough, and neither do political marketers. And so in in a in a sense, our political system is already advanced to the point we have better categories. And so the opportunity to to have a communication system that that automatically in real time clusters people in in higher dimensional higher dimensional spaces gives us an opportunity to bring that to the deliberation space. Use the same recommender technology in a different application space. And so I'll very briefly summarize what that interface is if if you're unfamiliar with polls in general. Very simple. The tool allows people to anybody who's participating to submit up Twitter length old Twitter 140 character length statement. The idea is that these are kind of atomic, and you everyone else participating can agree to disagree or pass. This forms a sparse matrix of of one, negative one, and zero for agree, disagree, or or or passed, and then did not see would be a null case. And so perhaps you could consider it four states. And then we do dimensionality reduction in clustering on that and find patterns of people who voted similarly. And then across conversations, you know, we we find a pattern emerges that almost always there are you know, we we find the groups that we knew existed. You find liberal conservative. You find Uber driver and Uber, you know, and and taxi driver in that in that debate. And then you also find some surprising things that that that groups agree on and that and that and that are consensus. And so scaling the ability of civil servants to talk to and listen to thousands of citizens or tens of thousands of citizens at one time, or a newspaper to engage audience, have served as a an interesting and passion and and kind of passionate pursuit, for me for the last, last decade. It started out, I should note, as a for profit, kind of survey oriented, social socially good oriented startup. And I think we got the technology, and then we open source the technology. I think we got the technology mostly right given our goals, but the organization was wrong. And we have since in 02/2017, shut down the for profit company and spun up a nonprofit. And the nonprofit is called Math and Democracy. And so that is the organization which stewards the which stewards the technology and the mission of bringing data science to deliberative democracy in general. So I'll I'll pause there and and answer some questions I see in chat. And then, I we can take this in any direction. Hopefully, I've opened enough threads there, in the, between the practical and theoretical side, and we can we can just take it wherever.

Great. Sure. So thanks for that. So, yeah, I'll briefly say that the the technology was was picked up by a sedate technologist in Taiwan in twenty in twenty, let's see, 2015, 2014, and and integrated into a process of legislation called vTaiwan. And this is the highest profile. There have been many other uses of the technology by the governments of Singapore and governments national federal government in in in Canada. But the the highest profile usage was in Taiwan. And the it resulted the process that came about was the result of protests and an occupation of parliament called the sunflower movement. And that won a it won concessions, and one of the concessions was binding facilitated rulemaking. And that was a process that was turned over to the civics technology community to opt to run and operate. And the the the result of that was that they they used a the poll is as a system of deliberation to get thousands of people into a common space to discuss to to deliberate and to have a and then and then they did facilitate a rulemaking in in small group. And so I I will I could share a number of articles on that. I'll share at the outset so everyone can skim, and I'll share the media coverage page here. But, basically let me let me find the chat. There is a if you're familiar with Roam Research, there is a a public read only Roam Research instance that is a knowledge base for the community, for people who are onboarding to become facilitators or people who are running these these using the technology, academics, peep civil servants, people in media, and and and people in the in in the NGO sector. And it aggregates some of the the major questions about both both practical and academic about the the technology and its its its implementation. I'm just gonna go I'm gonna go go ahead. Sorry. Go ahead.

Yeah. I guess I'll ask so I think I'm first in queue, so I can I can pose the question? Yeah. So does pull so I'm actually I'm I'm relatively familiar with Hola Hola. I guess, because way back when we tried to act actually, we were actually in contact or maybe Andrew Smith put us in touch Yes. Because we're trying to teach that teach Polis in that class at Harvard. And I I was okay. I guess, like, responding directly to something you said, like, does Poll. I s Poll. I s is it more about button link information for politicians? Because that seems the way that it's sold, right, when you sell it to, like, to to cities to use, as an example that you gave me about the mayor on the bus. Or is it more intended to bundle information for users and that kind of follows along with your hypothesis about replacing political parties?

This is the primary tension in the in the adoption cycle, and it's been adopted by in the case that it's adopted by, Scoop News, an independent newspaper in New Zealand that has an an audience, or it's adopted by a academics in Columbia University in Bowling Green, and run-in Bowling Green, Kentucky with Bowling Green Daily News. In that case, it's the it's an academics and news bringing people together, and then showing people showing the public where they're aligned, where they're not, and giving them that information to then bring the power. In the case of Canada's federal government picks it up and uses it as a better comment box, and then says, hey. Thanks for the thanks for the the input, and we'll we'll take what we want and we'll toss what we don't. That I think that's the primary tension that that neither of those is actually perfect. One lacks a a a connection to power and the other, lacks, you know, the kind of activist transparency that you would want. And the reason that vTaiwan specifically was interesting was that it was a connection to power that was binding facilitation that was that was kind of coming from a broad, segment of population and and kind of demonstrated, oh, wait. You know, if you if you if you did have a more coherent kind of public input process, and you did have a better deliberation process, then you you might have this this might imply something new about the way that we handle regulation. And and and so that what you're what you're getting out there is a is is the kind of a core, a core consideration from our nonprofit side about how we steer the technology towards adoption. I I personally believe that the most mission aligned organization in The United States is, independent news. And I believe that in the other, and we work with NPR stations in San Francisco and in Louisville. And I think it abroad, there are there are organizations that are more able municipalities that already have, for instance, console. I'll type the data. If you if you look for console as a deliberation platform, if the municipality already has that infrastructure civic infrastructure in place to get ideas and generate conversations in the public about potential regulations, then we're just kinda plugging into existing civic, municipal processes. And that that's that's that's controversial. I think it's it's very, very difficult to come in in The United States and, to at the government level, given how much conservatism there is about engaging the public and expect that it would be anything more than a than a better comment box. It's very, very dependent on the process.

You know, that's actually really curious because, just as a follow-up. Sure. The, so we actually, a couple months ago, we had, Miguel, who was the, the lead developer of Console Yeah. Come talk to the seminar. And he was talking about how, you know, like, be because Console's designed specifically for cities and it's, like, something that installed by a municipality.

Right.

And these, like, these people hate giving up control to any kind of technical system. I mean, they'll give, like, a tiny little bit of money, you know, that says, like, okay. Depending on the boat here

Right.

In the console system, like, maybe we'll maybe we'll allocate it. Yeah. And I find it really this this idea that you're pursuing that, okay, you wanna sort of, like, focus on these more civic institutions, deploy Polis as a tool in their toolbox. Like, I guess, like, in your interactions, have you seen the same pattern, I suppose, when your interactions with municipalities that they don't want to give up

control? Absolutely. Absolutely. Yes. Yes. I in a way, I have so many different answers and and so many different case studies running through my head. Let me let me think of let me try to pull out one. I spoke to the federal government a number of times, and one of the times I spoke to someone at a a major I'll give you two examples actually. They're both they're both pretty funny. The first example would be a Consumer Financial Protection Bureau, and they said, you know, oh, we have two mandates, and one is to protect the consumer. Right? $600,000,000 organization. Right? 1 300,000,000 towards protecting the consumer, 300,000,000 towards bringing their voice into, you know, national policy making process. And they're like, you know, really, the the second mandate is funded, but but we don't it's like there's there's, like, a $300,000,000 mandate to bring the voice of the public into the policy making process. And they're like, we don't really know what to do. You know? And it's it's like, even when you have the money, even when you have the will, even when you have the mandate, it's hard. It's it's it's it's hard to to do it meaningfully, and I I think that there are a number of considerations there. Like, for instance, I I am a believer in, in expert knowledge, but I think it's much more distributed than we think. I think, like, for instance, a good example of the of New York City used it with the Department of Homelessness and Housing, and they had janitors, frontline caseworkers for homeless people who are being asked for, like, the first time. This is the kind of thing that, like, these people are, you know, they're they're cleaning up the room after the person threw up after an overdose. Like, they know. They know what's going wrong. Right? Like, they are so tuned into the day to day processes and mechanisms that they're inside of, And we don't kind of ask the leaves. Right? We have fear or or, you know, teachers and students and administrators and parents, like, education process, I think, should involve experts. And and I but I think that we need to broaden that to lived experience, and I I and that I think this if we can do ethnographies much more cheaply than, kinda good enough, then it does that's my view is that it gets us further further in that direction. The second example is, to go back back pop the stack back to the, the the example of of the fed talking to the federal government. Second example in the federal government is, okay. We've read this. We understand what you what you're talking about. We understand what happened abroad. Super interesting. The problem is that this would create the perception in the public that we would, take their voice into into account, and we're really not set up to do that thanks thanks and click. And, you know, that was that was, admirably blunt. I mean, you know, the the and it's it they just they have processes, and their processes don't involve, you know, that they're you know, they get some of them get a million I've talked to people who in the government that they have a million comments come in. Plain text, survey survey, you know, advocacy stuff. They they, you know, they took a template and clicked pressed, you know, go. And and and then they they it sent 400,000 of the same template at the you know, to the to the to the endpoint. And and they there's, like, one PhD on the other side of that who's just really eager and underpaid and, like I mean, I and I I've had conversations with them, and it's like, can you make that more meaningful? And it's like, no. I have a different system to put in place of that, but I can't make that I can't make, you know, an unformed free text, you know, process more meaningful. Right? We layer information. When people type, it goes back out randomly. People agree, disagree, pass. You you get a sparse matrix. You do dimensionality reduction. You find meaning. We make meaning based on interactions that they don't have. They just have a million free text comments on GMO. And so, you know, the the more that it's a trash pile, the more that we can that they can say, well, it's trash pile. So, you know, we have 30 industry lawyers that know exactly what they want, and they're they're in the next room over. And you got, you know, you've got a trash pile of a million comments, and you've got a and then you've got a a a 30 industry lawyers who know all the regulations. They know all the line numbers. They know they're paid to be there in the room. And to me, that's the fight in in in a very practical sense for the civil servants, even if there's someone there who's tasked with reviewing those comments, even if they want to, the the amount of time that it would take for them to make meaning. And can they make a statement of legitimacy? Okay. I read through the million comments and, you know, if it says this, okay. Well, did is that consensus? Did did that represent one person or all all all, you know, all hundreds of thousands of people that participated? The worse it is, the easier it is to discard, and and that's a we've avoided it. We've avoided those we've avoided trying to operate in that setting because it's just not, it's not really feasible in The US to there's too many layered problems there in The US for us as a nonprofit to have it successfully attack it. So in The US, I think it's most realistic to work with independent news and and kinda come from a civic institution side. That is not the case. I just got off with The Netherlands, and they have a grant, and they already have municipalities using console. And they're just like, we have a specific problem, which is that some of these conversations are too big, and there's too many people engaging, and we wanna make it more meaningful. And The Netherlands are like, you know, can we can we just you they have, like, a a a a grant for democracy innovation to help municipalities that are already using console do better. And it's like, so we slot in there in The Netherlands. And in The US, I mean, it's just totally blocked. So I know. I've been over speaking in The Netherlands. I really love love it there.

And It's a great country. I just wanna find out if it's out.

Yeah. We were

talking to, like, people that, like, so many people are talking about this problem. Like, Miguel is literally at the Turing Institute in The UK, like, working on this problem right now Right. Bunch of machine learning people. It's yeah. It's so important. Yeah. I'm sorry. I didn't mean to It's Those are my questions. Great. Thank you.

It's a great question. Hopefully, I'd answer it. Hopefully, I've given you some insight into our thinking and limitations.

Did do you mind do you mind choosing, and I'll I'll I'll go through?

So I would say I would make one correction that they're just that my understanding is that because they've used Google, Google Ads or Facebook Ads to distribute the conversation geographically, they'll put up a link and say, hey, come participate. And, yes, you're right that the Civic Technology community is the closest to that spigot. So they're the ones that are gonna see that that information first and that call to communicate and participate first, and they are very active. And it probably does bias it. But also, they they do blast it out more widely, and they pay to get it in front of people. And I think that that's a meaningful thing to do. If you're if you if you wanna buy a Google ad for, you know and and the category is literally anyone in San Francisco and it's untargeted, that's a super cheap ad. You know? And so if you put a $100 towards that, you can get a lot of people in San Francisco demographically balanced to come say hi. And and, you know, and and vote. And we see people in that kind of a setting vote tens, hundreds of times. We have people vote 600, 800, a thousand agree to three passes per person. Those are the that's the long tail, but it happens. And, I think it's just super interesting to me, you know, that that that people would do that if they feel that there's legitimacy. And I think that's kind of where that that that that really is the calculation. And that that's kinda my answer, Janet, is that if you people are making a very rational calculation that that comment box on that federal government website, that's trash file, and my that is not honoring my voice. Whereas if if I see a a process that's coherent and lets me know how my voice is gonna be integrated, I'm gonna make a different, a cost benefit analysis about how to spend my time. And and seeing someone vote 600 or 800 or a thousand times in a polls conversation connect to a connected to a public deliberation makes me feel that we're we are making some progress on that, on how people are are are making that calculation about how their voice is gonna be integrated. That's my hope anyway.

Do you wanna pick you wanna pick a big one so that everyone could benefit here or just is it one that it's burning?

It's all good.

Got it. So if I can resend the question, the question is, what is the is this a question about the facilitation around Polas in that context? In in the context

That's a great question. So our platform is I would say, we are just the tech. And we have tried to make it general enough so that it would fit in a lot of different circumstances. And so we've been relatively successful, I think, in in in being kind of a chameleon for in terms of I like how it's integrated into prophecies, but some have been more successful than others. The one there was court was based on something called Cornell regulation room. You heard of that?

So I'll say I'll share a link back into the this is what Audrey's facilitation method was. And so if you go to the vTaiwan page as well, I made a graphic and then they kind of, you know, they they kind of expanded the graphic. But on the front page here of the vTaiwan site, you could also see a it says how it works. And this is like a, you know, kind of process dialogue. And and also this process dialogue kind of was the first thing that that, you know, Divya and I Divya and I were talking about because it's like, what how are we formally modeling the processes about when you, how when and how you engage people. And that kind of facilitation stuff, I'm I am so I am so not that guy. And and I, you know, I I can I've kinda I've it's been the reason we've been successful, I think, is in in terms of a technology provider here as an open source project is that we've just really tried to listen to the facilitators. And Audrey is just an incredible, like, world class facilitator that having having her working across from it and then having a few context that we made sure we weren't overfitting really helped us to try to to say, okay. We don't know what you know, we don't know who's who your participants are gonna be. In the case of a German political party, you know, it was like more of a walled garden. We have an email list. And then we just wanna kinda get a general sense about the platform. In the case of alternative that a political party in The Netherlands, they or sorry. In Denmark, they they had a a kind of a walled garden. So only people inside of their log logged in system could participate, and then, you know, they had a a decision making process for internal party decisions. They were kind of running by the whole membership in a participatory format. And so I think that the the, whereas in the case of of New Zealand independent news, when they would, you know, get that feedback, and and kinda put a report together, they would feed the report to the government, and it would be more of more of like an, an opinion poll that was just based on the membership of this independent newspaper. Wasn't meant to be demographically balanced. And then if I I'll share one of one very recent, but also very well put together report. This is much more formal. This is a national sample in The UK, and then a and then a, you know, a literature review of of of kind of situating polis in in the methodology and the overall landscape of of of of survey methodology and and purposeatory processes. And kind of give it go do a deep drill down into the content of the conversation as well. And so that last link from demos and they're a think tank. Carl Miller is in that think tank and and wrote a book called death of the gods, which I'll share as well in like, which is a a a book. It's available and I'll yeah. There's a whole chapter on VTaiwan and Polis. What he's probably the best interpreter in the West of of everything that happened there. He spent a lot of time on the ground in Taiwan as well. But but, you know, short answer is that the processes are very varied. There's a lot of different of different context in which polis has been operated in, And we try to be very agnostic about what happens next because we're trying to learn from the facilitators of, regarding, you know, when do they contact and how do they contact, and how do they let people know who participated that things have looped you know, how your voice was integrated in decision making processes. These are all things that I think we there's there's people that are just so much more expert at that than than than we are. We just we just try to help.

Right. You mean specifically so can you talk a little bit more Okay. So so finding.

Mhmm.

Yeah. Sure.

Barricading parliament worked really well in Taiwan. You barricade yourself inside the par parliament for a couple weeks, and they really, really, they just they perk they perk right up. I mean, very effective. I think you have to tear it from their that's like tear it from cold cold dead hands kinda thing.

Right.

Sure. Right.

In when you say inside the system, which system in inside

of Like,

Great. Okay. So in in in Bowling Green, Kentucky, we had a this is Bowling Green Daily News, Columbia University, American Assembly, and I'm just gonna I'm just going to share that website with you. If you go to Civic Assembly, and then you go to the Bowling Green conversation and you go to the report, this was this is a a program started by Eisenhower. They used to get a a whole bunch of, like, old white men together in a castle up in Upstate New York, and all of Eisenhower's buddies would move you know, commission reports that would bipartisan reports that would move the needle and, you know, and the, American Assembly's offices when I went there the first time were a time capsule of reports that had been commissioned by Eisenhower when he was president of Columbia University with all, you know, they would they would get together, they would think about things, and then they, you know, left and right, and it was a bipartisan conversation. That was the intention anyway. And so that didn't move the dial at all by the nineties, and so they were trying and they were read by the, you know, two thousands. They're they're trying to reconstitute it, and this was one of the ways that so polis is kind of the root of what they they they reconstituted around doing this in The States. One particular moment that I found just just probably the the thing that no. Far and away, the most exciting thing that's happened in The United States was in Bowling Green, it's a purple district. There's there's there's recent immigrants that have been kind of settled there by the, you know, by by The US and refugees. There are, there's a college town in Bowling Green, at a university, and then there's a whole bunch of, like, you know, good old boy Kentucky. And that all kind of exists up up next to it, you know, next to each other. And in that, it's easy when that's happening for politicians to focus on the wedge issues. And one of my kind of these are the political party. My big point that I tried to make over and over again is we know we know what the consensus is. The parties know that's not useful to advertise. You can't gain power telling people what everybody agrees with. And so the only way you gain power is to focus on the wedge issues. You do the market research, you know where the consensus is among everybody, and you toss it. And so that's not we don't we can't have that. Right? We gotta have a system in place which which, in terms of, like, an abstraction layer, which governs, which which integrates that as, like, an exciting bit. It has to be exciting that we're actually, like, unified. It's not exciting to the parties when when people are unified. And so the civic assembly does this conversation in Bowling Green. They run it for weeks. They send letters out to the civic groups, the churches, and the students groups, etcetera, and they get people together. Who's responsible for the cable rates and Internet and in the city. And, I mean, it's very clear, left, right, LGBTQ issues, immigration. I mean, it's just left, right, conservative split. It's very, very clear. But it's also very clear that, like, almost 100% of the con of the people in that room, there's 250 people in a town hall, who who all have the report that we had generated in front of them about what the various opinion groups were and and where the consensus was between them. Everyone was upset about cable rates. And he couldn't do it. And he showed up and he he had and it's on YouTube. This exists. I can share the video. It was just a a city official, and he just spoke to them as adults. There was no bickering in the crowd. Everyone knew they were on the same page in United and put all the other issues aside. Liberal, conservative, young, old, it didn't matter. Everyone had the same opinion about cable rates, and they were an Internet in the city. And, he just sat down and he said, here are my constraints. Here's what I can do. Here's what we can do. Here are the options. Here's why we haven't done these various things. And he just started informing people. And he started informing them of what his constraints were on making the decisions that would be implied by the consensus view. I thought that was really powerful because it meant that it it disarmed. It was it disarmed the kind of it's like, well, you know you know, public public, Internet is is socialist, bullshit. Well, obviously, no one thinks that. So put your talking point away. Right? And, like, talk to us as adults about why about why you've made various decisions to tank, you know, public fiber. This is a a really, I think, powerful mechanism, because it's it's the primary weapon that's used against us. And I think that that's if the information weapon is is is find the wedge, discard consensus, find the wedge, discard consensus. It's like, yeah, we're all together though. So now talk to us. And that that's that's a that I I I I I keep that in my head all the time.

Yes. Thank you. Also, I already said congratulations on Twitter. Right?

Yes. Congrat congratulations again.

Yes. You're welcome. So I'm gonna share an issue, and I would love as a connection. If you're going to invite me to your Slack, if you would, I'll explain what this issue is about, and and I would love it, to have all of your feedback on on this idea. Basically, what this issue is, I'll explain briefly. And, Amy, I I I think it's a really, it's a media it's a media issue for us. Really media. So the if agree to disagree pass, and I'm going if agree to disagree pass and and you could submit, you know, everyone could submit a statement, but what we end up with is, like, sky is blue. Okay. Everyone agrees. You know, dogs are really adorable. Everyone agrees. Right? But those are not significant or important or central statements. And the system doesn't have any ability to, differentiate those. And they're not they're not necessarily spam comments. They may be on topic, but they're just not significant. And they may be consensus. Or more importantly, they may be divisive, but they're not significant. What we're considering, doing, and we're we have a a specification here and some in plot and some some, some math for it as well, is to have a checkbox which says, this statement is significant to me. The reason that we have to get the wording on this right, and it's really important, is that even if they click disagree, if it's like, we really, we should not have public Internet. If the person is really passionate about that and disagrees, we want them still to check the box to tell us that it's a significant issue to them. So if they have to be able to click that whether they agree or disagree. His statements are phrased sometimes in the negative, which means that disagreeing is the etcetera, is the positive or or no matter whether you agree or disagree, etcetera. So we have a we we have a proposal here for this. And if if any of you are familiar with quadratic voting, you know, there's a similar concept here which in the math side where if if a participant continues to click it on every statement, it gets diluted. The effect of it would get diluted. And if they click, if they only click it once but vote on a thousand comments, then it's quite concentrated, right, because they've only signaled that one of these things was important to them or significant. And so looking at this over, how diluted it is and and how it was used by all participants gives us a second matrix computationally. So you have one matrix, which is all people by all comments, and that's, you know, one negative one, zero, etcetera. Big matrix, which is sparse matrix, which which fills out with votes on comments. And then you have another matrix on top of that, which is did the person have a checkbox, which says, like, this was significant. And if you overlay those, you know, you can multiply the matrices, and then that can inform clustering because then you're not clustering just on one negative one, but you're potentially clustering on continuous values. And so we're, this issue, you know, if any of you have a GitHub account and want to and want to, you know, make a comment on this issue, there's math. There's there's interface there. There's math there, and this will affect a ton. We this is very early, and we don't know how we don't know what we haven't done the Jupyter Notebooks. We have not gathered any data with this yet. We haven't tried the phrasing to see if if people agree and disagree, you know, if they only if people are if our phrasing changes and then people use it evenly between agreeing and disagreeing, that's good. If our phrasing chain if we use the phrasing and people only use it when they agree with things, then we know we've got it wrong. We haven't done any of that kind of research. So, you know, there's AB testing there that that we really wanna try. And then it has to also translate because this is used in dozens of, like polls are used in dozens of languages. So we have to get it right a lot, and it's gonna be I think it's gonna be kinda tricky. So this one is a a this is, also important because it it potentially informs the clustering. And and and the the reason this is potentially interesting is that we we want to consider experts. So if there are, if there's a conversation about, that involves an an aspect of, like, you know, aviation and or aviation machinery, and they're doing, they they're doing a you know, government of Canada is doing a random sample of, or let's just say they put it out to interest groups. They put it out to pilots. They put it out to, you know, to to executives. They and but there are a few mechanics that are, like, joining. Right? A few aviation mechanics. And they'll say there's only three of them out of 2,000 participants. And they know something. We wanna be we we have to get to the point where we can pull them out of the crowd. This is a really interesting and hard problem because you want to weight expertise at the scale of we want there's two constraints that are opposed. Right now, we've had a 30,000, 35,000 people in a German political party participate, 2,000 comments, like, a 100 average votes per person. It's a lot of data that we gather. Right? But even in that amount of people, even in 35,000 people, we, you know, we have a a really we have an interest. If there's a sub issue on, you know, drugs, if someone has been, if someone has been a researcher on on, you know, on on on issues related to that their entire career, it's really interesting to be able to highlight it. And, and and and that's something as we get as we get forward, we, we have, I have done some research on this, and I'm some I have some public Twitter threads on GPT three. GPT three can I I gave the the people the bio and the people who the bios of the people who follow the person, and I suggest what the expertise of the person might be? And GPT three immediately nails it. It's it's incredibly good. If you give GPT three an example and you're like, you know, and it's like, is the does the can the you know, what does this person this person might be an expert in, and it's like aviation. Like, what it immediately nailed it. So I think it's it's also promising. And part of the reason we have, like, a this is a the idea of math and democracy is a is like a hundred hundred year organization because we feel that that that the the the advance of data science and machine learning over that period will continue to be interesting to collective behavior or collective intelligence over that time scale. And so we we really think there's a like, GPT three is a great example. We have we have an opportunity to say, okay. Oh, and bye, Daniel. Thank you so much for being here. It was great to great to speak with you. You know, we have an opportunity to say, this this new technology helps us solve one of the core problems. So we keep collecting problems, and then we, you know, potentially over time, a decade in, GPT three comes out and can automatically summarize and give us, you know, tokens, and it does. Like, you you can scrape public Twitter, follower information, show me the people who follow this person, and then show me their potential expertise. And then you could, plot that and cluster and label and tag on an interactive visualization to say, this group over here has expertise generally in this word cloud of expert of of topics. These people have experts here. If everyone connects Twitter, we can do that completely automatic now. That's already that's already today. I haven't built it, but, like, that's already, you know but then it's it's completely possible now. And so, you know, we have a a a we have a lot of considerations of what is possible and what should be done, and that's a good transition to Amy. So I'll pass it back to Amy on the on the what should we do.

If so I'll actually share I'm going to share the GPT three issue, which is it's it's there is a GitHub issue. And if you follow the tweets on that thread, there are five, six discrete potential integrations, like summarizing comment topics and suggesting tags. So, you know, this is a topic about it's very, very incredibly good. It's like zoning, you know, and it it it's like it said nothing about zoning, but it's just like, yeah. The abstract category, this is zoning. And, from very little prompt, it can it can tag comments. So I think there you know, that's one, but there's maybe six five, six, seven. Would love to have your thoughts on that on that issue.

Mhmm.

your thoughts. I'm I'm concerned too.

Sure.

I don't have anything to add. I think you the the question was pretty much there.

Right. One of the things that we are we are trying to drive the adoption of the tool and have been successful so far to drive it not towards existing parties. I I have I will not spend a second of my time helping parties build build their platforms. I I think that, you know, that sequesters the the the public into bad abstractions, and we're particularly interested in in creating new abstractions for for the the collective will. I think that's the that's that's the big opportunity. That said, political parties have used it, but they were but they were, say, like, political parties like Alternative ette were interesting early. Let me I'll find I will find you a just a second. Unfortunately, it's in German, but there's transcript.

Yeah. Because I I was reacting because you had mentioned, like, the earlier town hall where everyone agreed that cable was bad, where it had clearly divided on party lines. I'm curious as you develop and continue to do this, if you'll still see as clear of a division or if people can start to, like, break out of those holes.

Right. Right. So I think that that's we see that the more specific we get. It's suddenly when you get to, like, homelessness issues, suddenly, you just don't see parties anymore. There there are, like, eight, nine groups and they're all mixed up on various, you know, comment commentarily. If you have if you have, like, 2,000 people and, you know, 600 statements, it turns out there aren't just two ways to to feel. There's, you know, there's a whole bunch of sub interests and and we see that. And so that that's kind of the data analysis that we're that we're working hardest on is using I'll share one Mathy. We let's see. Here's a here's a here's a paper. You know, if if you use UMAP on this, the UMAP dimensionality reduction algorithm and and use leaden, you know, you can you're using the same k neighbor k neighbors community detection graph, and you could say, okay, you know, you you know, this is a we're finding really detailed community communities, small communities of people that feel certain ways and and are are you can think of it as combinatorially different. Right? You know, the this this group agrees with this other group on this issue, but doesn't agree with this one. This group over here agrees with these two issues, but not this other one, you know. And and so you you end up finding, like, big tendencies, but then also, you know, really kinda fine grained on certain issues, contrary interest. So, yeah, I think that, you know, Uber Uber is the prime example that gets used for polis in Taiwan. And in that one, yes, you discover, yes, you discover taxi drivers and Uber, and it's like, there's a statement that's like, we should make Uber illegal, and, yes, that one splits the group evenly. But then you have issues that are consensus like, taxi cab shouldn't have to be painted yellow and people agree, you know, and everyone should have insurance and everyone agrees that everybody should have insurance. And and so there's there's consensus there. And then it the interesting bit about it is how much other variance is in there. There are almost always many issues which then split the existing groups and imply, you know, other arrangements. And that's, that's really where kind of it it gets interesting. So and that that's where there's variability.