Over the past few months, InnovateUS and the Reboot Democracy team, in partnership with the GovLab  and the Allen Lab for Democracy Renovation, have convened a public workshop series exploring what it takes to make public participation work in practice. The final workshop explored how AI-enabled civic technologies, when combined with thoughtful process design, can reduce polarization, surface common ground, and make democratic decision-making faster, fairer, and more engaging.

Click here to watch the recording of the conversation between Audrey Tang, Cyber Ambassador and former Minister of Digital Affairs, Taiwan, and Danielle Allen, Director of the Allen Lab for Democracy Renovation at the Harvard Kennedy School.

Below is a transcript of the conversation. Please note that the transcript is AI-generated and may contain errors. 

Danielle Allen: Thank you so much for joining us. You have so much experience using new technologies to improve democratic experiences, participation, and governance in Taiwan. We're gonna spend a lot of time talking about the solutions that you've built and deployed, but before we do that. I actually just wanted to start with a little diagnostic work. 

We all recognize that we're in a moment where democracies today are challenged, they're overwhelmed by complexity, it often seems that they're unable to sense or respond quickly enough to changing realities. Trust in institutions is eroding, there are all kinds of misinformation and disinformation problems. So sometimes I think it's helpful to actually go ahead and name what are some of the sort of failure points of democratic practices and institutions as we currently know them. I think that if we can name those failure points, then it can be easier to understand how to target solutions. And I know that, you know, your thinking about solutions is really, really powerful.

I want to help people really understand the kinds of problems that your solutions are solving. So, when I, when I list, current points of failure for democracy. Things I would put there include the tension between wanting to be efficient and get things done versus having participation and inclusion. People feel like there's a sort of trade-off between those things. I think there also can seem to be a tension between, effortfulness and utility in data creation that is sort of more effortful. Data production can have more signal in it, but then that effortfulness, again, weighs against efficiency. There can be a misalignment between the impacts of the population that is impacted by a given policy, and then who has actual decision rights over that. So, for example, often when we think about water management. You know, you have to sort of a river will often cross jurisdictional boundaries, and so it's not always clear how do you govern water resources when you have all these different jurisdictions, and one jurisdiction can make a decision that affects others, and maybe they didn't have a role in that. I'll just name two more. 

There's also the problem of the sheer volume of content that our democratic systems have built up over time. Minutes and laws and all the proceedings behind decision-making, and all of that sort of text is understood as material that people should factor in when they're making a decision or implementing a decision, yet the volume is sort of overwhelming for ordinary humans. 

And then fifth, lastly, there are the problems of misinformation that pollute both participation moments, they pollute decision making, and they pollute implementation. So, all five of those things. In other words, the sort of efficiency, participation tension, utility, effortfulness, tension on data, misalignment between those affected and those with decision rights, volume of text, and problems of misinformation seem like pathologies that our democratic institutions need to overcome in order to maintain effectiveness and trustworthiness to the population legitimacy going forward. I'm curious what you think about that list of pathologies. Would you put other things on that list of pathologies? What problems were you tackling when you came into your work in Taiwan?

Audrey Tang: Yes, I really love how you frame, like, forced dilemmas, right, as the main, pathologies. Efficiency inclusion, data utility effortlessness, and so on. I would add, perhaps one. I think it is a core pathology for people to treat democracy like a finished building. Rather than a piece of storytelling, a piece of software, if you will. Because to treat it as something fixed, democracy, as passed on by hundreds of years of traditions. That means also that we do not see the bugs, like misinformation, polarization, and so on, as something that we can patch, something we can upgrade. We can make democracy better simply by investing more thoughts into it. Rather, we would then say, oh, actually, this lack of maintenance is a respect to our, like, fore-parents, that these are just the things that we need to cope. And this kind of coping, the inability to see democracy as an evolving technology, I think is a core pathology, and that leads directly to the high latency, because the current democratic institutions are designed in such a way that people still think, like, 60 days is a good iteration to have, while the challenges we're facing are numbered in literally minutes. And so, when the democratic institution, the due process, and so on, understandably take a couple months to get anything done, people feel that the effort they put into it do not get any delivery, any result. And because of that, it leads to cynicism. And in a place of cynicism, then populism, extremism, and so on. So I think if we fix the latency problem. So that people can get democracy-delivering results here and now. I think that democracy can be fast, fair, and fun again.

Danielle Allen: Did you say fast, fair, and fun?

Audrey Tang: Yes.

Danielle Allen: Yeah, great, I love that. Fast, fair, and fun are a terrific sort of set of ideals, for the experience of democracy. That is serious updating, of the core design principles. So now let's talk about what you did in Taiwan, to achieve democracy that was fast, fair, and fun. Can you share for our audience of public officials more about the technologies and processes that you tried in Taiwan? For example, with vTaiwan. Can you walk us through the process of what that is, what it took to build and deploy and successfully leverage that tool to inform policy?

Audrey Tang: Definitely. So, we built vTaiwan, back in 2014. That was a year where the president of Taiwan, Ma Ying-jeou at the time, was enjoying a 9% approval rating. So in a country of 24 million people, anything the president said, 20 million people against him. And of those 20 million people, we're deeply, polarized at the time. One side want to free China, another said, want to be free from China, and they're really at each other's throats. 

And at that point, we invented this system, not because we enjoy, like, experimenting with democracy, but because the legitimacy of the trade deal with Beijing that year was simply below the zero line, so that half a million people occupied our legislature together, three weeks of nonviolent conversation. And we always say we're not protesters, we're demonstrators. Protesters are against something, like, against a trade deal. Demonstrators are demo for something new. So to show that half a million people on the street can nevertheless agree to some core principles after 3 weeks of deliberation. So we used what's called the citizen assembly method. People in smaller groups having conversations about any aspect of the trade deal. For example, I talked about cybersecurity, over-dependence on Huawei and CTE, and so on. But most crucially, we identified the uncommon ground, surprising common ground that people of different ideologies can nevertheless agree on. And every day, we announce, through live streaming what was agreed, surprisingly, the previous day, and what was still remaining to be discussed that day. 

And after 3 weeks, we just converged on a set of very coherent ideas that the Speaker of the Parliament simply said, okay, the people's ideas are better. And so I was 33 at the time, and I was then tapped as a reverse mentor, a young person advising cabinet minister, who's always under 35, the young person, I mean, that's designed, this kind of system to overcome, problems, like Uber, sharing economy versus gig economy, and so on, without the Ministry of Transportation getting occupied again. So we replicated the occupied process that we ran in person, but we brought it online. And then, in place of the human in-person facilitator, we use an AI system at the time called Pol.is. And what it does is, again, for each sentiment your citizen has on the Uber case, for example, you feel that the search pricing is fine, but undercutting existing meters is not. Okay, you see the statement, you can like it, you can unlike it, you can pass, but there's no retweet, there's no reply, so there's no room for trolls to grow. And again, there's a visualization of where you are after a few likes or unlikes. And then you can see, this group and that group never agree on something, and then if you propose something, surprisingly, that both sides agree on, they will put a leaderboard that pull those uncommon grounds viral. 

So basically, instead of the more extreme part going viral, like in the antisocial corner of social media, this is the more overlapping part going viral. And again, after 3 weeks, we agree on a core set of rules of Uber regulation, and then we make that our new law, and we de-escalated that conversation. So after hundreds of such collaborative conversations, by 2020, the approval rating of Dr. Tsai Ing-wen Was more than 71%, and so a very big jump, just by simply trusting the people to co-create policy together.

Danielle Allen: It's a very, very powerful story, Audrey, and many of us have been inspired by it. And I want to just name various features of the design, see if I'm getting right my understanding of it. 

So, for starters, you recognize the value in public sentiment as a source of learning for the political system. Tapping into that public value in a fast, fun way has always been a challenge, right? We vote, but that's every four years or whatever. You can't get that public sentiment or public knowledge immediately into a decision-making process. So, the deliberative structure, citizens' assembly structure, does provide a way of making sure that public sentiment and knowledge can be continuously input into governing processes. However, sort of in-real-life citizens' assemblies are very laborious, time-consuming enterprises to establish. So, in that regard, focusing on the fast part of it, you found a technological substitute for the sort of in-real-life version. It sounds as if then what you also did was, design that technological substitute to tackle some of the other pathologies. So, I didn't name as a pathology trolling, for example, or outrage. I probably should have added that to my pathology list. Because you made key design choices to keep that under control. You let people like or dislike, right? Sort of thumbs up or thumbs down. So people were able to form kind of groups of sort of shared perspective, but you turned off the kind of commenting function, so that what you got was simply affiliations of people. And then, I think the recognition that you had was just that basic information of sort of, like, distributed patterns of affiliation. could make visible common ground that hadn't been seen previously. And so, it was really the discovery of that common ground, it sounds like that was the real goal of the platform. And maybe that's the sort of fair part of it. So if the fast part was using technology to talk instead of in real life – the fair part was surfacing common ground. And then, so what was the fun part?

Audrey Tang: The fun part is that people thought they would never agree on something. At the time, Taiwan was so polarized that if you tell anyone on the blue side that there's something on the green side that you would vehemently agree on, they're like, that's not going to happen. And then, at the end of three weeks, when we list the scoreboard, the top nine, each of them enjoying, like, 80%, more than 90% of agreement, this is very surprising and profound. family fun. People really enjoy discovering such uncommon grounds. They're very noteworthy. And I would also share a link to the chat called We the People 250, where we use a very similar Pol.is-inspired way with the Napolitan Institute and Google Jigsaw in the U.S, with 2,000 people, five from each congressional district. Statistically, a microcosm of the U.S, and surprisingly, no matter whether they identify as conservative or liberal or red or blue, actually 97% of people agree that good values start at home, where parents are first and most important job of teaching children, respect, kindness, how people should be judged, should be not by their appearance and background, and so on. So there's this whole list of more than 90% agreements that people simply have different words for it, but the core sentiment is shared by all American people.

Danielle Allen: Okay, that's fantastic. So it turns out that fairness is fun, actually, that it's the fairness kind of thing that brings us common ground, makes that visible, is fun and rewarding. 

And in addition, as I think many people were putting in the chat, the other aspect of fun that you're really highlighting is a sense of connection. So, in that regard, maybe that's another pathology I might have named. We have the pathology of atomization and alienation. And what your platform was doing was helping everybody see their whole society all at once, but also then feel connected to it, because you can see where you fit inside that landscape. 

It actually reminds me of one of the most important technologies developed in ancient democratic Athens. Ancient Athens invented the theater. the physical space that was in the round, so that when there was a play occurring, citizens would react, but you could also see how everybody else was reacting. And that was a really important part of making the democratic population sort of cohere in ancient Athens. So, you have kind of discovered another version where we can, like, see ourselves in the round. Know who we are as a people, but also, really importantly, see those connections among us. So that's really helpful, I appreciate that, and I love recognizing, you know, what was fast, fair, and fun, about that, work. So, now let's talk about how you, you know, how does one change people's imagination of what's possible? You know, you said there was a demonstration or, you know, occupation of the parliament. Now, most of our public officials here probably don't want that to be the starting point for transformation, they would probably rather do it in a way that doesn't involve as much, kind of, turbulence. So, but tell us a little bit about when, vTaiwan and Pol.is who are coming into existence as possibilities, things that might be used, so before people had started using them. How did you communicate about the process you were proposing, and the technologies you were proposing in order to garner public support and participation?

Audrey Tang: Yeah, definitely, that's a great, great question. So, usually, it takes two things in order for this to really happen. One, it takes air cover. It takes a pre-commitment from politicians. So, for example, for the police conversation around Uber, we publicly say that the Ministry of Transportation, of Economy, of Finance, all have very different takes. And so, we're genuinely curious what people would agree that the three ministers wouldn't agree among themselves. So if you can garner more support than 85% across groups, then all the three ministers are committed to respond. Of course, this is still, like, not a referendum, so if it is something, like, physically impossible, or legally constitutionally impossible, they still get vetoed, maybe. But actually, people come with extremely nuanced ideas, so to justify this pre-commitment, that wasn't difficult at all, so this is the first thing. And the other thing is that we need to show people that this is independently verifiable. So, the whole vTaiwan one uses open source technology. So, Uber, for example, provided their own analysis on the open data, which doesn't contain personally identifiable information. And because people can't independently analyze it, and the clustering algorithm is really easy to replicate and all open source, people independently run this kind of nodes. This is very similar to how community notes works today. They actually, they use exactly the same idea of bridging algorithm on social media, like X. If you see a post and you want to attach a clarification to it, again, you upvote, downvote, and it finds the cluster that don't vote, equally, they always reject each other's votes, and then it finds the few uncommon grounds and attach that as the company notes, but the entire algorithm and data set is open source, so people can independently validate it, so that they see that this is even more transparent than the, for example, a rigorous survey. It is like the poll, except that the poll questions are written by the people themselves. It is a wiki survey, so it's trustworthy because it's trusted citizens, and also it gets a pre-commitment. From the power holders.

Danielle Allen: Okay, that's great. So it's trustworthy because it's open and accountable, and so that's an important design change. Our existing analog governments are often extremely opaque. You often describe yourself as embracing radical transparency. I imagine that when you were doing reverse mentoring, mentoring older cabinet members. And, perhaps you were trying to mentor them into an embrace of radical transparency. How did that go? What worked, what didn't work, as you tried to bring older generations into practices of radical transparency?

Audrey Tang: Yeah, I think it's easier, to show, rather than to tell, so I always model this, on my own conversations. On the website sayit.archive.tw you can find, as a digital minister, and even before that, for the past 10 years, I talked to 8,000 people in over 400,000 utterances in more than 2,000 locations, and all of it is in public domain. So, as a minister, and before that, as a reverse mentor to a minister, I accept any lobbyist visits, any journalistic interview, if and only if that state commit to radical transparency, and we publish always the transcript into the public domain. And people can see immediately, two effects. First, the lobbyist, like David Bluth, who at the time was lobbying for Uber, visited me. And he talked exclusively, pro-social outcomes, like a better, I don't know, climate, better, like, less congestion, and so on. He never talks about something that is good only for Uber and me, and not for everybody else. He talks exclusively how good it is for the next generation, like, for public good, because everybody knows that the next generation is watching, or will be watching, these public domain transcripts. So first, it shifts the incentives to talk about, like, long-term, public benefits. And the second is that it cuts through the polarization. Because too much of the online polarization stems from people, taking, like, out-of-context ideas, another out-of-context ideas, and then just amplify the extremes. By embracing radical transparency, basically it makes it easier for people like investigative journalists and students and so on, to see the whole context than it is to, to make this out of context, and add it to the language models today, which means that anytime you ask any language model to search online, more likely than not, these public domain ideas are already trained in their pre-training, and so they have a very accurate representation of the way I think, and so on. And so all these are real dividends. polarizes people, it enables journalism, it makes it much easier for lobbyists from various different competing stakeholders to actually agree on public goods, and so on. And so after the other ministers see that regulatory transparency has this much good, the only question remaining for them is that how much does it cost? And fortunately, thankfully, with language models, the cost is now almost zero.

Danielle Allen: Okay. So, at the end of the day, that you felt that they were able to make the sort of shift in their own mental models pretty easily, they just needed facilitation in delivering on radical transparency? 

Audrey Tang: Yes, yes, and then I'm not livestreaming anything, really, so the point here is not to scoop the journalists. The journalists still publish first, it's just that as soon as they publish, I publish the entire context.

Danielle Allen: I see, I see, interesting, okay. Thank you, that's super helpful to share that. So let now… I mean, I think, that was beginning to open up the question of what is required institutionally and culturally for the approach with technology that you've modeled to work. I would love to dig into that more. At this point now, you've also helped other jurisdictions adopt similar approaches. For example, in California, you helped set up Engage California. I think it's a very long list of places that you've helped people deploy these kinds of public engagement tools. Let's talk a bit about the best practices that you've learned along the way about the enabling conditions. So, you know, how do you shift norms? We got that a little bit with that example about the reverse mentoring, but let's spend more on that. So, yeah, what are the enabling conditions when you're coming into a new place? What does it take to get this going?

Audrey Tang: Yes, so I talk about the openness, accountability on the infrastructure level, and I talk about the pre-commitment, the air cover on the political level, but having these two around doesn't mean that a process will get started. To get a process started, you also need urgency for clarity. Uber is a great example, because nobody had clarity of what Uber even is back in 2015. Or last year in Taiwan, we encountered a lot of bigfake scam advertisements online. Whenever I scroll Facebook, I see Jensen Huang, you know, the NVIDIA Taiwanese CEO's face, talking to me, saying that he wants to give back by giving me free cryptocurrency or investment advice, and how to overcome that without resorting to state censorship. Also, nobody has a clue, actually. Because generative deepfakes is a very new phenomenon. And so in these situations, the existing silos within the public sector would not even pretend to say that they know all the answers. And therefore, there's a far larger canvas for people to paint together with pre-commitment and also with accountable, open tools. 

And in California, as you mentioned, I've worked with the Newsoms, Gavin and Jennifer, for a couple years, because they want to answer this question, like, how should young people use social media in a way that's better for their mental health? And again, this is a classic case of something about them without them, because certainly they did not vote in the legislators. So they thought, okay, maybe it's a better idea if we just ask the young people and their parents, and offer a pre-commitment, if they can figure something out. And so, we made the system, the platform, Engage California, it all tested, and on the week of launch, we did not launch, because the Los Angeles fires happened. And then we pivoted immediately to talk to survivors of Eaton and Palisades on how to overcome the NIMBYism that has prevented effective mitigation and prevention of wildfire. And it turns out, again, if you poll people individually in traditional polls, people are very extreme. But if you, again, put people in groups of 10, and have them react to each other's feelings, suddenly, instead of YIMBY or NIMBY, everybody becomes MINBY. Okay, maybe in my backyard, if you do this, if I do that. And again, the pro-social algorithm really shines because we say, okay, of this cohort. If you manage to convince each other, then your collective idea, your surprising, uncommon grounds. can cross-pollinate into the final reports. But if you keep, you know, doubling down on extreme ideas and you cannot convince your cohorts, well, then it's just your ideas. And so this, again, this pro-social bridging algorithm resulted in a very comprehensive report on how to overcome future wildfires. That's a great success. And now, we're moving on in California to have a conversation with state employees, so that they can, in a button-up way, to figure out how to introduce AI in the public sector in a way that's fair, fast, fair, and fun for the public servants, without, you know, top-down, chainsaw-style government efficiency stuff.

Danielle Allen: That's great. So, let's just talk a bit more about the pro-social algorithms. I'm not entirely clear whether or not, as you describe Engage California, whether the pro-social algorithms operating there are the same as the ones you described for Pol.is, or if there are additional pro-social algorithms.

Audrey Tang: Well, it's the same in spirit. It's called Ethelo. However, again, it calculates the Ethelo score, which means that in order to win on the leaderboard, you have to let like, everybody slightly unhappy, but nobody very unhappy. So this is a more, like, a proportionate veto core, to use the academic term, a kind of negative way to say the same thing that Pol.is is saying positively, but it's the same spirit.

Danielle Allen: Got it. All right, but in the Engage California case, were people able to comment on each other? So that's different, so they're in small groups, but they are commenting in their small groups?

Audrey Tang: They're commenting, but that's a conversation. At the end of the day, after the commenting, they still vote on each other's options. And then it selects the bundle of options that maximize the Ethelo score, which, again, means that it can unify people who otherwise would never agree.

Danielle Allen: Right. Well, there's another ancient Greek theme there, because of course, Ethelo means I want, or I desire, so I guess that's where the name comes from, that it's sort of surfacing what people's hopes are, aspirations, and then you're looking for those clusters. In order to find that common ground. So that's, that's really good. Now, sometimes when my team has engaged with you in talking about how do people get started with this, you've shared lessons like, start hyper-local. scale up, so starting small, you've said, doesn't require as much air cover or political commitment. What do you mean exactly by starting hyperlocal? Can you provide an example of that?

Audrey Tang: Definitely. So, in Taiwan, we start exercising our civic muscle really young, in what I call civic gym. Which are literally gymnasiums, because high schools. And so, we changed our curriculum in 2019 so that, the young people in Taiwan, as young as 7 years old. start learning how to measure air quality together, water quality together, noise level together, to co-create the civic data pipeline so that they can inform their parents whether they should go hiking because of pollution levels, and so on. And when they're slightly older, like high school level, they're invited to crowdsource policy changes, because on our national participation platform, join.gov.tw we only ask whether you're a resident in Taiwan. We don't really care how old you are. So the most active participants on Join, the e-petition part, is 17-year-old and 70 year old. So they form this beautiful intergenerational unity, because they both have more time on their hands, I guess, and also care more about future. And so, for example, the high schoolers started to petition. Any 5,000 people can force a ministerial conversation. So they say, oh, let's go to school one hour later, because science shows one more hour of sleep gets you better grades than one more hour of study. Which is real science. And then they got what they want, right? 

And there's also a high schooler that petitioned for banning plastic straws from our bubble tea, our national drink. And she became my reverse mentor, even before she turned 18. There's also a young lady who crowdfunded the nation's first menstruation museum and removed the taboo in just 2 years, and so on. And so these young innovators, before they are even of age to propose a referendum. They can try the civic muscle in the civic gym, in their gymnasiums. And with, the new curriculum, we reward anyone who can come up with this kind of, what we call PBL, purpose-based learning, so that instead of learning to out-compete each other individually, the score, the university exam, and so on, are all revamped, so that they, only care about the three core competencies, which are curiosity, collaboration, and civic care, because we reasoned back in 2019, after Gen Alpha go, only these three virtues are going to matter in the future. Any other literacies are going just to be automated by robots.

Danielle Allen: Hmm, wow. Okay. Let me… I want to build on that, and, you know, thank you for that, example of the way in which young people can start, and I guess that's an example of starting small. I'm also wondering about communities that don't have, you know, an existing level of comfort with technology. You sometimes have said, use tools that the community already uses to meet them where they are. Does that mean, you know, literally sort of starting with Facebook in those places where Facebook is the one thing that people are using? Or, you know, what do you mean by that idea of using tools that the community already uses?

Audrey Tang: Well, sometimes it means literally printing out QR codes and going door-to-door, and visiting people, and where they are, like, in Bowling Green, which I just posted the reports, the What Could Bowling Green Be? They engaged, I think, around 10% of everyone in the Bowling Green City, area of Warren County. That's about 8,000 people, and they used a very analog tool of a printed QR code, and going door-to-door, and having, like, living room conversations. And then during those conversations, they ask people, what are your future, like, in 2050 imagination of Bowling Green? And I think it's really good that they're trying Pol.is now because this year, Pol.is does not need a human moderator to figure out what topics people are talking about. 

So if you click on the report of the Bowling Green and scroll down, you can easily see community identity, equity and inclusion, infrastructure, transportation, and if you click into identity, you see community pride and image, and so on, and none of this requires any It is all figured out by a language model, and the entire stack is open source. And so people can easily see an evolving group selfie, which makes it a selfie movie, I guess, and see their places in it without manually interpreting those individual statements. So I think, to your point of hyperlocality, I think the Bowling Green experiment succeeded precisely because it's not just a top-down, like, mayor thing, but rather they work with all the listening partners from all the different in-person organizing ways. For example, they work with the Boys and Girls Club of the Bowling Green area, the storytelling group, the housing group, public health group, talent development group, so on and so forth, the curators, and so on. So each of them have their own hyper-local way of bringing people together, sometimes through food, sometimes through sport, sometimes through spirituality, and sometimes through fashion. And all this then have their own kind of subtopic, but they all share a grand, holistic conversation that can do topicalization and summarization in real time.

Danielle Allen: Thank you very much. I think I could carry on asking you questions all day, but we have a lot of questions in the Q&A as well, so I'm gonna go ahead and pull some of those questions in now. I want to bring in a question from Akinori Oyama, who's in Japan, Akinori says, Audrey, I see your emphasis is shifting a little bit toward having live experience and community conversation that help a lot to lessen problems such as polarization. And there are studies from MIT and others that indicate that in-person information will earn respect and trust. How do you think about the relationship between the, sort of, the tech-based and the in-real-life, elements of engagement and deliberation?

Audrey Tang: That's a great question, as often expected from Aki. So, I think for many people, they need a peak, a sheer peak experience to see that polarization is just an illusion. And the way we do this differs from the places with a relatively low PPM, that was just in Sweden. PPM stands for polarization per minute, in my lexicon today. It also means CO2, which is important. But anyway, so in a place with low PPM, you can afford to have less of a live conversation and have an asynchronous conversation. For example, Finland used Polis for the entire Finland. What do Finland think about each other, about future, now about the future of AI, and so on? But the most polarizing statement in all of Finland's, Pol.is is just that, the older people feel that the younger people are correcting their way of using words. Seems like some words they used to say cannot be said anymore. Okay, if that's the most polarized statement, you have a low PPM environment, so you don't need that much of a live, like, lived experience on a hyper-local community scale. However, in a place with very advanced PPM situation, a lot of geothermal energy, as I like to say, because polarization brings energy, like the US, with a lot of free energy going around for the geothermal engine, then you do need this kind of life experience so that people can check using nonverbal expressions, or maybe literally at the same dinner table, to show that actually polarization is an illusion, because asynchronously, there's just too much Psychological projection, to project, ill intentions on each other's lists.

Danielle Allen: Yeah, that's certainly my experience. I appreciate that clarity about the difference between high polarization context and low polarization context, that seems really critical. I'd like to pull in a question from Cassandra Wilbur, who asks, How did you manage the need for transparency with competing priorities for confidentiality and flooding the zone with too much information that might, in its volume, cause confusions for the citizenry?

Audrey Tang: That's a great question. So, in Taiwan, we practice the art of pre-bunking. So instead of debunking, which is when polarization, disinformation, this kind of foreign interference attacks, which, by the way, Taiwan, according to the VDAM, for the past 12 years running, has been the world's top target for such foreign interference. So, whereas in other countries, you have to pay for penetration testing, in Taiwan you will get that for free. So, we are very familiar with this too much information flooding the zone, causing confusion, and overwhelming our subsidiary. And in that situation, a saturated information environment, debunking doesn't work. 

Actually, debunking sometimes only fuels the conspiracy theory. Because the framing effect already takes shape. So, in Taiwan, we moved to prebunking, which is before any polarization, disinformation attack flooded the zone. We call out that attack and offer something that is even more viral than the attack. For example, we pre-bunked the mask/anti-mask conversation in early 2020. We detected two viral strains of memes online. One that says only N95, the highest grade of mask, is useful, and everything else, including the medical mask the state was handing out, was placebo, and they're not useful, they're polluting. The other side says, oh, it's aerosol, it's ventilation. Any kind of mask hurts you, and N95 absolutely kills you, right? So both are sensational, and the science wasn't that clear, which side is potentially right or wrong. So if we do the debunking, it's not useful at all. So what we did is, almost immediately, we rolled out this meme with the Shiba Inu, our spokes dog, a very cute dog, putting her paw to her mouth, saying, wear a mask to remind each other to keep your dirty, unwashed hand from your own face. So it's literally personal protection equipment against your own hand. And the tap water usage did raise up, and because the dog was so cute, it went far more viral. than the two strains of disinformation, or polarizing information. And so, the point being, if you pre-bunk, then people associate the mask thing with a cute dog, and then they de-escalate the conversation. And if you wear a mask, I don't wear a mask, well, you're just reminding me to wash my hands. What's the big deal?

Danielle Allen: Right, okay, great. Also from Cassandra, where did the idea for reverse mentoring come from, and how do you actually implement that?

Audrey Tang: So, reverse mentorship comes from the occupiers, many of which are just undergrad students. And they, the cabinet members at the time, really feel that the young people have a different mode of trust, whereas at the time, the 50-, 60-year-olds trust institutions by default. In Taiwan, we have very high respect for, like, academic institutions, like our National Academy, journalism, ministers, for sure, and so on. However, they witnessed in 2014 that the 20-somethings, the people who occupied the parliament don't really care. They may have reverence for those invertical institutions, but those vertical institutions don't have relevance to the young people. The young people's mind, first and foremost, are putting the relational health with people who sound like them. who vibe like them, who post on the comments section instead of the headlines of the newspapers, and so on. So that's where their attention is. They shifted to a horizontal peer-to-peer mode of trust instead of a vertical mode of trust. And so the reverse mentorship is meant to get the people who get the most trust on this peer-to-peer horizontal way. And invite them into the cabinet to set the direction for the country. It's like a Pygmalion effect. If you trust a very young person who wants the horizontal trust from their peers to set a direction of the country, then they have to mature up and live up, rise up to the occasion. But if you say, oh, you're too young, you're, you know, not even 18, you should not have a conversation about politics then of course, they go into the extremist, cynical, populist way. So the reverse mentorship is a design back in 2014, to be more popular than populism.

Danielle Allen: Hmm, hmm, that's fascinating. From Charles Pfeffer, I don't know if I've said that correctly, what's the smallest scale example you have for applying pro-social technology approaches?

Audrey Tang: Well, in small scale, that literally means a scale of 1, right? So, so Iuse this kind of method, like, personally. I practice active listening, so that when I have a one-on-one conversation. I don't internally judge the other person. I just listen, and make sure that I can enter their worldview. And as a minister, for example, when we deliberate about Uber, I went to all the Uber blocks in Taipei at the time to really share the ride and have a conversation with the driver. 

When we did Airbnb, I lived in those new Airbnb hotels, questionably legal, and when we did online alcohol, even though I don't drink, I ordered alcohol online. And so I always strive in a personal way to take all the sides, but that doesn't mean that it's a both-sidedism, but it means that I enter the horizon, the perspective, the viewpoint of the other side. And then, I then also entertain the arguments, the lived experience of the opposing side, and then I hold both, kind of, intention, in a geothermal way, like, played against plate that pushes Taiwan up half a centimeter every year, literally. And so then I go to sleep. And then I sleep for 8 hours or more. If there's a huge conflict, I work extra time, I sleep for 10 hours. And then I wake up with something that's more like an uncommon ground, like a bridging statement, a coalise in my mind, that let me cohere the two individuals within me that may not always see eye to eye, but I wake up. and they become my left and right eye, so I can see in stereo. So this is a personal practice that I do, and I encourage people to do this kind of active listening, because it really exercises your inner civic muscle as well.

Danielle Allen: Mmm, that's beautiful, I appreciate that, and so I think then the, you know, questioner was really probably wondering about technical solutions, but your point is that every technical solution is actually just a way of, concretizing a human practice, and so what matters most is the human practice. So that we understand it, you can do it yourself, one person, you can do it with two people. three people, then once you've got the practice going, you will build platforms and tools that help you do that practice, because you appreciate the practice, you understand it, you're going to want to strengthen it. It will actually help your tool building to have just done that practice as a personal practice in the first place. That's really powerful. Thank you, Audrey. We don't have so much time, but I want to get to a couple more questions. And, let's see, from Sonia Putzel, what do you see as the most common motives for people to engage for the first time? And what about keeping people engaged as a habit? Are these two different things?

Audrey Tang: Well, I think, people really love the ability to set the agenda. So, people, if you ask them what should be the, like, economic solution to the Uber problem. Or what should be the actual parameters to counter defects online, and how should watermark work? Of course, these are dull and boring, and nobody really wants to spend their time debating about this. However, if you open it up and say, you know, how do you feel about Uber? How do you feel about deepfake online? People are really experts in their own feelings. And now, with, like, the We The People do 50 experiments, they use, the Habermas Machine, and Anthropic has a competing product, the Anthropic Interviewer, that can then follow up, with, qualitative questions, like the investigative reporter, a good journalist, and ask, like, what is the kind of lift experience that brings you to say this, to say that, and so on. So, people love their ideas being taken into account through their feelings. And people also love to see their feelings resonating with people who otherwise they would never agree with, that they think of, like, the others, the aliens, and so on. So this first icebreaking experience, as soon as possible, let people see the uncommon ground. I think that is the hook. Because once you're hooked to the uncommon ground, the polarization, the hate, the slop are really boring. Right? They're like the trolls, they offer the same kind of junk food, you know, relational slop. They offer some dopamine hits, I'm sure, but it does not actually lead to any insight. And when people see that we're using polis, using this kind of pro-social technology, you can get that dopamine hit, because the uncommon is genuinely new, and in the same fast and fair fashion. Your feelings can also resonate and ripple, and maybe even become agenda-setting, then that empowerment is even stronger for people to keep coming back.

Danielle Allen: Oh, great, okay. Now, I'm gonna ask you now two, sort of, just technical questions, if you don't mind, of sort of implementation. The first, and I'll just ask them both, and then you can respond to them both. The first is, what's the difference between Pol.is and Ethelo, and when would you use one over the other? So, comparison of those two platforms. And then the second one is about how you conclude the experience for people once they've participated on the platforms. Do you use a standard exit survey to evaluate the experience people had, the quality of the engagement, whether the event was worthwhile to people or not?

Audrey Tang: Yes. Yeah, we use standard access surveys, for sure, and we also measure whether people's ideas, even if not taken into account, even if they like, when I say 85% of people agreed with the final package, obviously 15% of people feel a little bit disappointed, but then we ask, like, was it legitimate? Can you live with it? I know you're not happy with it, but can you live with it? And that is also a very important exit survey. And in fact, I would argue it's even more important than the 85% agreement. When we did a deliberation on countering deepfakes, people in 45 different video rooms, 10 people each, each brainstormed a different idea. One say, if Facebook posts an advertisement without digital signature, let's display, probably, skin, like a cigarette label. Another room says, if somebody lost $7 million from this, unsigned, advertisement that Facebook refused to take down, let's make Facebook join liability, pay the same $7 million. And the other one says, if TikTok ignore us by not setting up a local representative legal office, then let's slow down connection to TikTok. 1% every day, they ignore us. These are brilliant ideas. So obviously, there are some other rooms whose ideas did not get included into the package, simply because they're not as strong as these core ideas. But even the other 15% overwhelmingly say they can live with it. They think the core ideas are quite relevant, and also they can see the legitimacy, the process, generating it. So that's that part of the question. Actually, we use all the tools all the time, so this is not like if you use Pol.is for agenda setting, then you cannot use the Stanford Deliberation, or Frankly, or Ethelo, or whatever, for exploring the options. Pol.is is great if you're just discovering stuff, like, really a free-form thing, whether you think about 2050, but then once you need to settle with a final package, a bundle of policy proposals than something that is more nuanced, that offers this kind of pick-and-choose combination of options, like Frankly, or Stanford Delibration, or asynchrony. Ethelo is a very good counterpart to the Polis. I think of it like a diamond, like in a double diamond. This part, the Discover part, is good for Polis, and the definition part It's good for some other tool.

Danielle Allen: Fantastic. Audrey, as always, this has been a wonderful conversation. I learned so much from you every single time we connect. I am incredibly grateful to you for your vision, your creativity, your beautiful human heart, your philanthropy for humankind. You are just a towering inspiration for all of us, so thank you so much for joining us today. Welcome back, Beth, what'd you think?

Beth S. Noveck: I could not have said it better myself. Thank you for your generosity now, not just in this hour, but as Danielle put it, your generosity, now not just in Taiwan, but around the world, inspiring everybody with your words and your deeds. Danielle, do we want, as this is the last of our sessions, I thought we might take one minute, not just to reflect on what we've heard from Audrey, but what you feel you've heard over the last few weeks. As we wrap up our session. And while we do that in the chat, I'll invite the team to put links to where you can find all the sessions and past recordings of everything that we've heard, and we'll be sharing more resources as we go along. But any closing words from you on the semester, if you will?

Danielle Allen: Well, I think it's been a terrific set of conversations, Beth, and I think that we may be on the cusp of a kind of breakthrough moment for democracy. The idea that we can have, I mean, I love, I mean, I think fast, fair, and fun is a good way of capturing the vision of what everybody's aspiration is, everybody who's been on these calls, and we've heard a lot of good ideas about how to get there. I'm thinking about Claudia, most immediately at the moment, who has been doing so much work with Citizens Assemblies also. There are a lot of synergies to the conversation today. I think also it's really important that the conversations have made this all very real, right? These are not abstract ideas from theoreticians. People are doing all of these new things right now in the world, and it is making democracy faster, fairer, and more fun for people.

Beth S. Noveck: So I would agree, and I will say that I was yesterday on Capitol Hill invited, unbelievably, by members of Congress to a hearing on AI and public engagement. So the fact that they wanted to have the conversation, that they've restarted the conversation, that they were painfully aware that I think that 9% approval rating, Audrey, that you quoted for the former president of Taiwan is probably a high number for Congress. Of how painfully low the trust is right now in Congress and in our democratic institutions more broadly, how riven we are with polarization and with political violence, in ways that are truly scary. So but I think these series of sessions have left me extraordinarily hopeful about what we can do. Lots of practical conversations about AI and how that can help us with issues of scale. So all the excuses we've given before about, we can't do this because we'd be too overwhelmed, the issue that we've had with the web, with too many people speaking, making it difficult for institutions to listen, we've heard a lot just very practically about tools you can use to overcome that problem. But I think the thing that we've also heard very clearly is that it's really not about the tools in the end. It is about the trust in people and their ideas, their experience, that active listening that you talked about. The best places have been doing this kind of listening. You heard Taiwan has been doing this… Audrey, I still think of you as 32. I know you're not anymore, but you've been doing this a while. 

And, you know, in Brazil, they have been doing large-scale citizen engagement in lawmaking for more than 10 years in ways that not only involve adults, but also involve kids. So it doesn't take AI, it's about process, and it comes back to what Danielle said in our very first session about the spine of our representative democracy, the three vertebrae of participation that you talked about on day one. That facilitates both information sharing, opinion sharing, that ability for decision makers then to deliberate and listen and take on board that input, and then the ability to connect that again to how we solve problems in the world. And it is connecting those vertebrae that you talked about on day one that I think really set the tone for what we talked about, which is not just how to do deliberation and participation, but how to connect it to decision making. So, with that, we have to thank our audience, and all of their expertise and comments and questions. We have so many people here who have shared their know-how and learning, and more importantly, shared their interest and curiosity in trying new things. So, if we've done anything here to, empower you to do more of what you're doing, which is to go out and listen to people, we're excited about that, and we'll be coming back after the holidays, both with sharing what we've heard here, and talking about, making these resources more available to more people globally. So, thank you for that. Danielle, any last words from you?

Danielle Allen: No, just another round of thanks for everybody. Really appreciate people signing on, and Audrey, you're being here today. Thank you.

Audrey Tang: Beautiful.

Beth S. Noveck: Happy Holidays and Happy New Year to everybody.

Danielle Allen: Bye. Take care.