Dear reader,
Today’s conversation is a much-needed whitepill for these blackpilled times.
Audrey Tang is someone I’ve admired since my first trip to Taiwan in 2018. She’s known as the young technical prodigy who led Taiwan’s digital democracy efforts and helped the country expertly navigate Covid-19. Yet our conversations have led me primarily to appreciate her deep social and emotional intelligence, and her relentless but grounded optimism—a quality that has undoubtedly enabled her experiments to succeed (this conversation is full of inspiring case studies). Among other things, we discuss:
Audrey’s unusual childhood
Substack, Bluesky, and Partiful as pro-social media
What DOGE gets wrong about hacking the bureaucracy
How LLMs can scale citizen participation
Going from a consumer to a producer society
Why more politicians should be live-streamers
Watch/listen by clicking above, the Substack app; or add it to Spotify, Apple Podcasts, or Pocket Casts. A full transcript and list of links is below.
Episode transcript
This transcript has been lightly edited for length & clarity.
Technology, democracy, and Taiwan
Jasmine Sun (00:00)
Today on the podcast, I'm excited to have Audrey Tang. She's the current Cyber Ambassador of Taiwan, a principal author of the excellent book Plurality, and a frequent collaborator on technology projects across international lines. She's inspired me for many years to think bigger, more positively, and more expansively about technology.
She's also like no other politician I think you will ever encounter. For instance, Audrey is a middle school dropout and an autodidact. She told me she's been reading LessWrong since when Eliezer was still writing The Sequences. Welcome, Audrey!
Audrey Tang (00:45)
Hello, good local time everyone. Very happy to be on the show.
Jasmine Sun (00:49)
Audrey, you began your career not as a politician, but as a programmer. You contributed to a variety of free software projects in Perl and Haskell. You've started your own startups. You worked on Siri at Apple.
I think it was when you got involved with Taiwan's pro-democracy Sunflower movement in 2014 that you entered activism, and then government. Which came first, your interest in politics or your interest in technology?
Audrey Tang (01:18)
In a sense, I was into activism since I was a very young child. When I was born, Taiwan was still under martial law, and it was not legal to form new parties. My parents, as journalists, participated in the earliest grassroots communities. My mom co-founded the Homemakers Union, which grew to be still Taiwan's largest consumer co-op, and my dad headed the first community college. Activism is in the roots of our family.
After dropping out of school at 14, I immediately started this dot-com entrepreneurship track, but concurrently participated in the activism in Taiwan around experimental education—the right of children to choose their own curriculum.
I wouldn't say that in Taiwan there are two different generations of people: one older, upholding political institutions, and the other younger and more technology-oriented. In Taiwan, it's literally the same people. The same generation that brought us personal computers also brought the lifting of the martial law, and the same generation that brought the World Wide Web are the same generation that voted in the president for the first time in 1996.
Jasmine Sun (02:45)
That's really interesting to see the pro-democracy and the pro-internet generations come up together. Could you say more about how those were intertwined?
Audrey Tang (02:56)
In the very beginning, when I studied dial-up BBS and later the internet, a lot of organizers already used these technologies to communicate. When my dad participated as a spokesperson for one of the candidates during the first presidential race in '96, already we were using the browser and what later would be called Web 2.0 as ways of getting citizen input.
To us, democracy has always been a social technology. It is not some old institution that cannot be changed, that is read-only, but rather it is something that we own and we write together. It's almost like semiconductor technology that you can upgrade every year to different layers, different designs. For example, Taiwan introduced the ideas of participatory budgeting, e-petitions, and sandboxes quite early on. Of course, they didn't originate in Taiwan; they originated in Brazil or in Spain and other places, but we brought them over almost as quickly as they appeared.
Jasmine Sun (04:16)
Someone once described this phenomenon of the internet and democracy being hand in hand as the “Palo Alto Consensus,” associated with the Bill Clinton era and the Atari Democrats. Then people say the Palo Alto Consensus collapsed: we got Facebook and the whole world was connected, but this ended up being bad because people got angry and read too much misinformation. But you still seem like an optimist about some version of the Palo Alto Consensus.
Anti-social and pro-social media
Audrey Tang (04:54)
It’s not the same point in time when people got connected together and when polarization drove us apart. There's a very specific point in time in 2015 when social network companies switched from a common experience into an individualized experience optimized for addiction. That's when we got autoplay, when Instagram switched to an algorithmic feed, when the following feed made itself unimportant, and of course later on with TikTok and such. So there existed a time—Bluesky and the Fediverse are trying to go back to that time—when connecting everybody together did not automatically lead to polarization.
In Taiwan, our domestic equivalent of Reddit is PTT, and it's maintained by the National Taiwan University's Student Club. There are no shareholders; there's no advertiser pressure. So there was never an algorithmic feed push. Because it's open source, it's probably difficult to force something like that onto the users. And because we enjoy a more robust digital civic architecture, we've never seen the full-fledged polarization that was brought in other corners of the world, because we have a counterbalancing force.
Jasmine Sun (06:34)
Why hasn’t there been a company looking at PTT and thinking, “That's an opportunity to build something similar to that, but way more addictive?” Why has PTT been so resilient?
Audrey Tang (06:48)
First of all, people understand that they have a choice to migrate elsewhere. And you can only get away with highly addictive, extractive algorithms if you know that your users are trapped.
A year and a half ago, there was a study that shows that you’d have to pay US undergrads almost $60 a month for them to stop using TikTok. They perceive they lose that much utility from FOMO. But if there was a magic button they can press to migrate everybody away, they're willing to pay you almost $30 a month to do that. Everybody's losing utility, but the first person moving off loses the most. That only happens when you don't have the freedom to migrate between platforms.
Because of our commitment in Taiwan to build such infrastructure as something that is more portable—like how you own your telecom numbers, even if you switch telecom providers—that keeps the pressure towards the top. Platforms compete on quality rather than how much you can capture users into an addictive regime.
Jasmine Sun (08:11)
I assume that this mission is related to the work you’re currently doing with Project Liberty on the People's Bid to acquire TikTok. I'm curious if you could say more about that.
Audrey Tang (08:25)
Sure. It's not just Frank McCourt but also Kevin O'Leary. Now one of the co-founders of Reddit [Alexis Ohanian] has also joined the People's Bid.
The idea is to make sure that when TikTok runs on a US-based team as required by the law, anyone can have a choice of how to present those front-end experiences—meaning that if you post on TikTok, you will be able to view it on Bluesky, on Truth Social, and any other places through federation. People using Bluesky or Mastodon know what I'm talking about. It’s similar to how if you publish your podcast on one index, you do not have control over what players people use to consume your podcast. The platform becomes more like a recommendation engine built by the people, with the people, instead of a single algorithm controlling the people in a top-down or opaque way.
Jasmine Sun (09:34)
We've talked in the past about how you imagine various pro-social ways to build algorithms or technologies. I'm curious to hear more about these tools. How can algorithms strengthen civic infrastructure rather than weakening it?
Audrey Tang (09:53)
For people who have experienced Community Notes, you know that this is a way for people to add useful context to posts. It used to be on Twitter (now X), but it's spreading to YouTube and soon the Meta platforms as well. Each person can upvote or downvote the notes other people provide. But only the the content billed as useful by both the left-wing and the right-wing will win and flow to the top. This bridging algorithm came from a long line of applied research that was first used in Taiwan in 2015.
When we deliberated the Uber case in Taiwan, instead of asking people abstract questions—like choosing between an extractive gig economy and a sharing economy—we just asked people how they feel. Uber drivers, taxi drivers, and passengers all logged on to the open-source deliberative democracy platform vTaiwan. There, people can see each other's thoughts about the Uber situation. For example, a participant might write, "I feel that insurance is important" or "Surge pricing is fine, but undercutting existing meters is not." Other people will upvote or downvote to show whether they resonate with your feelings. Then the Polis algorithm groups participants using k-means clustering, where each cluster represents a faction or tribe. It shows what drives people apart, but also the bridging statements that bring people closer, and participants compete on those. There’s no reply button, so there aren’t trolls. It’s a race to the top, not to the bottom of the brain stem.
People are quite surprised, after three weeks of conversation, to discover that most people agree on most measures with most neighbors, most of the time. Finally, we took those bridging items and made them law. That's how we resolved the Uber situation in Taiwan.
The main takeaway is that polarization is not a function of specific people. It is a function of the space. If you make the algorithm prioritize dunking and viral broadcasts, people will perceive each other as more polarized than they actually are. But if you build systems that prioritize bridge-making, then people will perceive what we call an “uncommon ground”: a rare common ground that can bring people together despite initial difference.
Jasmine Sun (12:48)
I definitely buy that there are often fundamental agreements not being brought to the surface, and technologies like Polis are good at finding those and showing what we can make traction on.
But to some extent, people like dunking. If you run A/B tests, often people like the polarizing content more. They will view more content. They will click “Like” more. I saw in Substack’s data that frequently the top notes on our feed were these very partisan "screw Donald Trump" or "screw the Democrats"-type notes. I would love to believe that people wanted nuance, but oftentimes they chose, even independent of algorithms that push them in that direction, more polarizing content.
So is it always the algorithm's fault, or are people sometimes just choosing to antagonize other people?
Audrey Tang (13:53)
That depends on how you do the measurement.
If you measure instinctive responses, then of course there are all sorts of dark patterns that elicit addiction. On the other hand, you can measure satisfaction, or how much time you want to spend as opposed to how much you actually spend on it. Or as I mentioned, “How many dollars do I have to pay you to keep you on the platform?” And “how much you are willing to pay me if you're migrating everybody to a different platform?”
If you measure it this way, then you find that people have meta-preferences. People prefer to smoke, but prefer not to prefer to smoke. Upon further reflection, people can see that they would prefer more common experiences with their neighbors. But if you just measure the instinctive response, then of course you're getting system 1 fast-thinking signals.
Jasmine Sun (15:00)
That's a good point. I often feel like the challenge of building pro-human, pro-social technology products is that people's short-term and long-term wants are very different. It doesn't even have to be digital technology. Think of junk food, right? If you put a stalk of celery and a bag of chips in front of me right now, I will probably choose the bag of chips. But if I'm thinking in a more long-term way, I don't want to choose that every single time.
The current technology ecosystem prioritizes meeting people's short-term wants at faster and faster speeds—you can get your food delivery or chips faster, you can see more pieces of polarizing content in a shorter period of time. I wonder whether we need a more fundamental rewiring in order to get platforms that are optimizing for different metrics.
Audrey Tang (16:02)
One part of it is just how isolated people are in the first place.
There's a famous experiment from a few decades ago called the Rat Park, where rats self-administer drugs and get addicted to them. But interestingly, they only do so because the lab environment is a confined environment. If they're put into a park with social activities—like in Taiwan, where humans have food together, enjoy each other's company, do civic activities together, even take out trash together and so on. Once people are in that mood, they're much less likely to be addicted to those short-term hits.
So we shouldn’t just configure the tech ecosystem, but also more group-based activities so that people can see see information contextually. Information is not just me seeing this individually, or precision targeting from an advertiser, but rather an enjoyable experience the other communities that I'm in are also enjoying, and that brings people together. This is a fundamental insight.
A recent paper I co-wrote, “Prosocial Media,” basically says that shared experiences are actually addictive in a different way, and perhaps in a more healthy way.
Jasmine Sun (17:28)
Do you use the app Partiful? The events app.
Audrey Tang (17:32)
Partiful? No, not yet.
I checked out your links to Esmeralda, the city-building project, and Socratica, for discovery of common interests. These are the kind of pro-social ideas I have in mind as well.
Jasmine Sun (17:49)
Partiful is this popular app that people use to host events now. It's like if you took Eventbrite, but made it more fun and easy. I realized that this is one of the most pro-social apps I've ever used because it just makes organizing events extremely easy. My friends are probably hosting three times as many hangouts as they used to. It's just party-hosting infrastructure. It makes it easy to invite everyone from your previous event to new events. Instead of email, it sends texts. It counts how many parties you host. And these are all real-life events, it gets people together.
Audrey Tang (18:36)
That's very good. I'm installing it now, and I'll let you know when I host my first party using it.
Jasmine Sun (18:41)
Yeah, it’s great. And that's a useful insight that when people have a real-life social infrastructure that they're embedded in, they’re less likely to get addicted to internet things.
But I'm also wondering: are there incentives for for-profit companies to optimize on these longer-term metrics? It would be great if companies were running surveys on how good you feel after you've used their app for a day, but I'm not sure that they are because their business metrics point them in a short-term direction.
Audrey Tang (19:15)
If it's subscription-based like Substack, then naturally it's more aligned than precision advertisement models, simply because to retain your users for a long time, you probably need to make them successful in some way that is everlasting.
In addition to Substack, I think LinkedIn fits into that. The main funding model for LinkedIn is not targeted advertisement or addiction-building, but the career value you derive. Even though it has a news feed, it penalizes overly addictive short-term attention-seeking, like “Comment here to win a lottery.”
DOGE vs. Taiwan’s digital ministry
Jasmine Sun (20:09)
I want to turn the conversation towards your work with the Taiwan Ministry of Digital Affairs. Remind me how long you were there?
Audrey Tang (20:23)
I joined the cabinet when I was 35, in 2016. I was the digital minister without portfolio, meaning there was no digital ministry at the time. It was an ad hoc team composed of people from all the different ministries. A little bit like DOGE, I guess. We called ourselves PDIS, the Public Digital Innovation Space. In 2022, I became an overall minister. Altogether, I served seven and a half years.
Jasmine Sun (20:58)
One reason I was keen to have you on the podcast at this time is because, like you mentioned, there are some analogues between the stated intentions of DOGE and what you were doing in the Taiwanese government, right? You were trying to make government more effective and efficient, incorporating technology where it makes sense.
But these days, I’d say that most Americans—especially progressives—are very worried about the idea of taking a hacker's approach to politics. In the last few months, DOGE has shut down USAID, is firing tons of people indiscriminately, is freezing funding for university research, and generally causing chaos.
As someone who has experience being an inside hacker trying to make government more efficient, what are the DOGE folks getting wrong? What could they be doing instead?
Audrey Tang (21:54)
When I joined, the US already had established the USDS, the US Digital Service—which is, by the way, the full name of DOGE today. They're officially the US DOGE Service. And the US also had 18F.
In a sense, the idea was the same: to use technologies in the service of the transparency and accountability of government services. When we first came in, we also looked at all the procurement contracts, all the government spending; we made bulletins, comparisons, so people can comment on it, and so on. In those respects, it's not very different.
One crucial difference is that we announced our every move. When I first came in, I made sure that all regulatory changes were open for public commentary for 60 days instead of 7 and 14 days as was previously the norm in Taiwan. For example, when we responded to people’s petition to reduce the tax filing experience from three hours to three minutes, we did so not unannounced but rather in the open. People always have 60 days to digest the repercussions of our moves and offer their commentary before we finally implement it two months later.
That’s a big difference, and won us much more trust from the career public service. They saw us as servants serving the public service, rather than acting against the public service.
Jasmine Sun (23:45)
So you actually lengthened the period of feedback and comment? It used to be much shorter, and you made it longer to make things more efficient in the long run?
Audrey Tang (23:55)
We saw the conflicts and the various comments not as fires to be put out, but rather as energy sources we can channel into co-creation and change. Literally everyone who complained in 2017 about the tax filing system being “explosively hostile” to Mac users was invited to the Ministry of Finance to draw the tax filing experience for the next year together. People already had an idea of what they wanted to be better. It’s just that people have very different ideas. Offering for people to meet in a collaborative space and design the experience together pre-bunks the polarization or the backlash, and lets us design in a way that takes care of different needs.
Jasmine Sun (24:56)
So you’re using that 60-day period itself as a way to collect different perspectives and synthesize them into a plan.
Audrey Tang (25:08)
Exactly. In our national participation platform, join.gov.tw, which has been running for almost a decade now, just recently there were ePetitions that counted hundreds of thousands of signatures. Under each petition, you can see two columns. One is the supporting arguments and one is the contra arguments. There are upvotes, there are downvotes, but there's no reply button. So people do not troll each other, but you can just see the arguments floating to the top and use these as the foundation for bridge-building.
Jasmine Sun (25:55)
That feels like a fundamental difference. Because Elon Musk's DOGE is very much saying, “We know what's best, so we'll start slashing and burning programs that we think are inefficient and bad.” Whereas you're starting from the assumption that people inside the administration already know that there are things that could be improved, and you just want to help leverage that expertise and make it actionable.
Audrey Tang (26:19)
And also to give them air cover. The reformers within the system usually face an uphill battle of coordination problems. They know how problems should be solved, but unless all the different levels of bosses line up, as well as people in other ministries, there is no way to unilaterally force change. This is quite a wicked problem. Most issues within bureaucracy are of this shape.
During the public consultation process, I suspected that many people contributing as citizens are actually section chiefs of the relevant departments. They just participate meronymically. And because we don't have a real name policy, I sincerely don't know who they are, only that they offer very good context on how to navigate this. So the 60 days serves as an ideation period for all the different people working on wicked problems to voice their difficulties. When we use bridging systems to find consensus, we're also addressing career public servants’ woes and ideas, not just the people in the streets.
Jasmine Sun (27:41)
Right now in politics, there’s a big conversation on participation versus speed. People assume that if we ask more people to participate in a process, it’ll become less efficient. DOGE believes this, of course. But on the left, the Abundance liberals advocating for YIMBY policies—their solution is to close down channels of public participation that turn into vetocracies, because community input meetings often get weaponized by wealthy landlords with a lot of time. I agree that’s bad, but I’m also struggling with the idea of shutting down opportunities for voice.
How do you create mechanisms for participation that help things move faster?
Audrey Tang (28:31)
I’ll share an example: Last March, we organized an online citizens’ assembly to solve the issue of fraudulent advertisements online. At the time, if you opened Facebook or YouTube, you often saw an ad posted by Jensen Huang of NVIDIA trying to sell you some crypto. And if you click it, Jensen actually talks to you. Except it's a deepfake.
That’s a big problem—not just because of the scam damage, but also because it crowds out the legitimate advertisements of small and medium enterprises. We knew that people want these ads contained, but they don't want the state to get more censorial powers. Taiwan is the most free in all of Asia in terms of Internet freedom, and people would like to keep it this way. Previous attempts to assign more power to the administration have failed.
Here’s what we did:
We sent 200,000 text messages to random numbers around Taiwan asking: “What do you feel about online fake ads, and what do you think we should do together?” People gave us their ideas, and thousands also signed up to participate in synchronous online assemblies.
We worked with Stanford to use stratified random sampling to select 450 people who represent a statistical microcosm of the Taiwanese population in terms of occupation, gender, age, place they live, and so on.
The selected participants met in 45 rooms of 10 people each. Each room got an AI facilitator. When people are too quiet, the room encourages them to speak up. If people interrupt each other, the room makes sure it's just for five seconds. In real time, we can see one room saying, “We should assume it's a scam unless Jensen Huang actually signs off on that ad.” Another room might say, “If somebody gets scammed out of $5 million, we shouldn't fine Facebook. We should make them liable for the same amount.” And a third may say, "We should slow down internet connection to TikTok until all their business goes to Google."
We then read the rooms’ proposals back at a plenary so that people, after another round of deliberation, can vote on how much they're comfortable with these kind of measures.
At the end of the plenary conversation, we learned that regardless of their party affiliation, more than 85% of the Taiwanese population considers a KYC (know your customer) for advertisers proportionate.
In April, we checked with all the stakeholders—YouTube, Google, AdSense, OpenAI, Microsoft, and Meta—and they said, “Of course KYC is physically possible. It's just going to cost a lot.” But that wasn't the question we were asking.
By May, we wrote out the legislation to the Parliament and it was swiftly approved. This, the Digital Signature Act, was one of the most fast-tracked bills.
This year, if you open Facebook or YouTube, you won't see any fraudulent celebrity advertisements anymore.
This kind of story shows that it’s actually faster if you ask everybody—as long as you can prioritize “conversation networks,” where the ideas that resonate within one group get amplified to other groups; rather than a single voice polarizing the conversation, going viral, and blocking people from entering into conversation. The latter broadcast networks work for indignation, starting social movements, injustice and so on. But we think conversation networks can weave a social fabric together in a very quick way, resulting in decisive action.
Jasmine Sun (33:12)
How did you build the political will inside government to run processes like this? Because I can imagine people reacting with, "That sounds great, but it's time-consuming and requires all these resources."
Audrey Tang (33:29)
The infrastructure view is very important. If we do this on a per-project basis without a shared platform, then every time you want to kickstart something, you have to prove that this is urgent enough to warrant this process, and also that it is not already within the mandate of some department. Not many topics fit these two criteria.
But once we established some success that satisfied both the urgency and the inter-agency criteria 10 years ago, we institutionalized that as a regulation. Now the National Development Council provides this platform. Any city, any municipality, any ministry, and anyone in Taiwan can now start such a conversation if they get 5,000 digital signatures online.
When the bootstrapping threshold is low, people can initiate this kind of process easily by piggybacking on existing infrastructure. But if they have to get the software set up, the facilitator set up, and do everything by themselves, it’ll be almost impossible except for the most urgent cases.
In California, we’ve spent the past couple years working with Governor Newsom’s teams at the GovOps and the ODI on Engaged California to provide such a shared infrastructure. The actual launch topic was on something that is urgent and inter-agency, which is how to recover from the LA fires. Now there are already more than 6,000 people going through this process. If people have a good experience, then it’ll be easier to start conversations around less urgent issues, because they can refer back to the positive experiences that happened before.
Jasmine Sun (35:33)
So you’re building reusable infrastructure. Just like posting a tweet is easy because there's already an entire infrastructure for posting tweets—you’re making the kind of political behavior you’re excited about much easier and simpler.
Audrey Tang (35:48)
Yes, exactly. And it’s also norm-shaping. When people have a good idea and want to get people to support their idea, they can rely on this platform to find unlikely allies across generational, urban, rural, or even partisan lines.
Jasmine Sun (36:11)
I also like that you take a bit longer to collect perspectives so that getting legislation through becomes more efficient. It reminds me of being a product manager. You always have the meetings before the meeting, right? You go around and talk to individual stakeholders and hear everybody out and do a brainstorming session—so that by the time you officially introduce the plan, everyone's bought into it or understands why it's happening, and you don't have a fiery crazy debate.
Audrey Tang (36:43)
This is very common sense in product development. It's just applied to civic and political processes.
Facilitating empathy & agency
Jasmine Sun (36:51)
How do you think that you became such a good facilitator? Because when I talk to you, it sounds like you have a very keen sense of the social dynamics and how to bridge very different groups.
Audrey Tang (37:07)
When I was young, I practiced the Daoist practices of Qi Gong, like inner breathing exercises and many others.
It was a survival skill. I was born with a heart defect. I wasn't sure I would survive until the day of surgery, which I got when I was 12. When I was 4, the doctors told me and my family that I only had a 50% chance to survive to that day. Not only did I go to sleep feeling like I was flipping a coin—if it doesn't come up correctly, I might not wake up—but I also literally cannot get upset. If I have my heart rate goes higher than a certain threshold of beats per minute, I faint and may wake up in the hospital.
That is one large part of this work: to always switch to System 2, the slow mode of thinking, whenever a conflict arises, and taking the time to understand more fully where other people are coming from. If I feel I cannot comprehend what someone is doing, I don't suffer from the upsetness of a racing heartbeat and fight or flight or flee. Rather, I always want to engage them with a sense of curiosity.
In my mind, there are no people who are like me and people who are different from me. There are only people I can share some experience with and people I have not yet shared some experience with.
Jasmine Sun (39:01)
And you're saying that that comes from Daoist practices as well?
Audrey Tang (39:06)
If you think of the Daoism symbol, the Tai Chi, it's two very polarized sides getting into a circle, complementing each other and also finding each other embedded recursively within. That symbolism to me means that instead of saying, "These people are my people, those people are not," I see the shared common ground within any confrontation. I want to basically spend many hours or even days to find out what actually is that perspective—to maybe not see exactly eye to eye, but at least see in a kind of stereoscopic vision.
Jasmine Sun (39:59)
That makes a lot of sense. You also mentioned that your parents are journalists. Personally, I don't have your spiritual tradition. I didn't grow up with that heart condition. But one thing that I've noticed since becoming a full-time writer is how that has helped me become less angry at things that I find wrong in the world. I switch from "I'm mad at this thing" mode to journalism mode—and just start asking questions and collecting information. I may find someone’s beliefs crazy and weird, but will sit down for 45 minutes with 10 people who think this way and ask them questions. And as a journalist, you can't be in convincing mode. You're doing your job poorly if you're doing that. So I actually have to sit down and really hear people out and take notes the entire time. In almost every case, I come out the other side understanding people better.
This is one of the things that makes me happiest about doing writing—I think it makes me a more empathetic person, and I hope it's good for my approach to politics and social change as well.
Audrey Tang (41:18)
I can see you smiling, so it's a real pursuit of happiness. I do think that is where interpersonal joy comes from: from breaking through a perceived gap or alien relationship, then after working through it, seeing the world differently. And it's both sides—not just the facilitator or the journalist side, but the people who were facilitated or who have seen the journalistic outputs of their side and other people's side. Their life also changes.
Jasmine Sun (41:59)
I know you spent some time when you were younger doing education activism. With these principles in mind, if these are the kinds of citizens that we want to develop, how do you think of reforming education to make that more possible?
Audrey Tang (42:16)
What we want is not just literacy, which is passively consuming information, but rather competence, which is the ability to produce information in social context.
In Taiwan in 2019, we switched our curriculum to do just that. What used to be “media literacy” and “data literacy” are now “media competence” and “data competence” classes. Instead of teaching critical thinking about, for example, air pollution, water pollution, and noise level; the students actually measure these together, and use distributed ledger technology to contribute their findings to our common understanding of the environment. Or they fact-check the three presidential candidates in real time as they're having a debate, contributing to the common understanding of what is going on in the campaigns. Or they submit e-petitions. In one petition, students changed the rules so that they don't have to go to school so early. They go to school one hour later because they proved together that one more hour of sleep is actually better than one more hour of study when it comes to their grades.
The goal is to get learners to switch from a consuming mode to a co-production mode. It’s like the Pygmalion effect. If you treat 15-year-olds as adults and imbue in them the ability to set the national agenda for conversations, even retaining some of them as cabinet-level senior advisors—what we call “reverse mentors”—then they’ll feel that they're already an adult and have a lot of good things to contribute. In 2022, the International Civics and Citizenship Study, ICCS, ranked the Taiwanese 15-year-olds as top of the world when it comes to their civic competence and their ability to participate in shaping national conversations on people and the planet. I think that has a lot to do with this new curriculum's emphasis on active competence, not just literacy.
Jasmine Sun (44:46)
Even considering something like TikTok that we were discussing as addictive—I actually think that if more people were making TikToks, I would feel less bad. It’s actually very creative and oftentimes collaborative to think of a funny sketch, come up with an idea to share with friends, or to edit a video along with music.
What I hope for is: how do we get more people producing? Whether that’s something as big as a startup or as small as a sketch comedy video or making a petition. One of my earliest formative experiences was writing a petition at my middle school that accidentally kind of worked. That was a crazy experience of learning very early that you can have a dent on the world and not just be a passive person who the world washes over.
That's also one of the things that I find very inspiring about Taiwan and my experiences being there. It seems like there's a very rich culture of co-producing, whether it's youth activism or opening small businesses.
Audrey Tang (46:07)
Or even just throwing a party using an app!
One of my earliest experiences voting was when I turned 20. I voted for the first time in the neighborhood leader election in Muzha, south of Taipei City. The candidate that I supported won by one vote. It was almost down to a coin flip. I felt that all the time I spent reading up on their campaign promises and traveling from Taipei back to Muzha was very much worthwhile.
I do think these earlier wins are very important, which is why we should not limit democracy to once every four or two years. We need to make sure that there are many ways for young people to express their ideas. Not just opposing, but proposing, and having some proposals work, and work in a well-understood fashion so that people can look up to the young people who shape national conversations.
Jasmine Sun (47:14)
This is the sort of everyday democracy stuff I'm thinking a lot about. There's a cultural practice of contributing to the world around you versus, every four years showing up at the ballot box, coloring in some circles, then forgetting about it for another four years.
Audrey Tang (47:39)
We're seeing more places take up the Engaged California method of digital democracy.
I think language models have a lot to contribute. Previously, once we got tens of thousands of contributions, it takes a lot of time to read through all of them. But now, using language models, we can do what's called “sense-making”: to show very clearly, like a data journalist, what are the overarching themes—always grounded to the individual statements as footnotes—and what are the outstanding divisions. This kind of continuous group selfie was not possible before the advent of language models.
This deployment makes me really optimistic about everyday democracy because previously it required a lot of staff investment. But now even the smallest neighborhoods, intentional communities and so on can run sense-making automatically. It's now built into Polis, the open source tool that we used in 2015.
Jasmine Sun (48:57)
Or like when people are responding to the text polls with a sentence, and suddenly you have 100,000 comments in natural language. Now you can actually figure out what people are thinking.
Audrey Tang (49:10)
Yes, and when we took action, it also allowed us to close the loop. We could tell these exact people, “Because you said this, now we're doing this action together.” The feeling of satisfaction is much stronger because we can individually go back to people and encourage constructive behavior. People may have also submitted a lot of very destructive comments, but as long as you reward only the part that is constructive and led to decisive action, people really do change.
This is what I call troll hugging. This is one of my hobbies.
Jasmine Sun (49:46)
That reminds me being a product manager too. Substack writers were sometimes very angry at us and would send me an email saying like, "10 reasons I think your product is buggy and not good enough and it sucks."
A month later, maybe we’ve only done two of the things and there are still eight that we haven't done. But if I manage to remember and email them again—if I say, "Because you said this, we now fixed these two issues. We’ll still work on the others, but wanted to let you know."—people are generally really pleasantly surprised.
I think a lot of people don't have an expectation that a company or a government will ever speak back to them. People don't expect to be heard by the institution. But when there are these small moments of responsiveness, it really changes the emotional dynamic.
Audrey Tang (50:36)
This is working not just in California. As we speak in Kentucky, in Bowling Green, there is a "What could Bowling Green be?" Polis conversation going on.
As more people become aware of these tools for broad listening, I think broadcasting will not be the only norm when it comes to politicians’ interactions with the population. Instead of just one person speaking to millions and ignoring the millions’ voices, it's equally easy to listen to millions of people and have millions of people form conversation networks. It was not a norm before, but we're seeing more and more successful experiments now.
Jasmine Sun (51:20)
Now I'm imagining like a YouTube or a Twitch Live where a politician is screen-sharing, and they have a Polis-type tool that surfaces ideas and questions from the community. Then the politician is live reacting on stream and answering people's questions.
Audrey Tang (51:34)
That's literally what was done in Tokyo. There was a mayoral candidate, Takahiro Anno. He’s 33 years old and read the book Plurality that I co-wrote, then decided to run for governor of Tokyo one month before voting day. Practically nobody had heard of him. He synthesized his avatar as a VTuber and did exactly that sort of live-streaming that you talk about, and he used the broad listening tools, Polis and Talk To The City. Anyone can hashtag #TokyoAI to form the platform together, call into a dedicated line, and talk to his voice clone about his platform. The platform crowdsourced this way was independently ranked as the best score, even better than Koike-san, the existing mayor who of course won re-election. Koike-san appointed Anno-san as the advisor to GovTech Tokyo, so now you can also hashtag #TalkToKoikei and participate in this real-time sense-making.
Jasmine Sun (52:43)
The example that I have in mind is how AOC uses her Instagram Stories. For example, she'll run the Instagram survey tool. Right after the election in November, she said something like, "If you voted for me and Donald Trump, why did you do that?" She then posted on her story a giant collage of what Instagram users said—without judging, just displaying what the responses were. And then she collected them and a little Instagram live speaking to those people to say, "I hear you, I understand, and this is is what I think."
I thought that was such a good use of technology. A lot of our representatives are not quite as savvy, but I would like to see more of that kind of behavior. It's the evolution of the town hall, because now town halls just get captured by old people.
Audrey Tang (53:41)
Emerging conversation network technologies such as Cortico from MIT or Remesh allow people with no particular training to interact with a large crowd as if it's just a few avatars. Or you can listen to a medley of fragments of people's voices to get a holistic understanding, then respond normally as you would to a few people. I think these will massively lower the threshold for politicians to meaningfully engage because they won’t need to operate 10 different windows and read 500 different comments—but rather just respond as if it's a common voice, thanks to language models.
Jasmine Sun (54:37)
You're very good at this yourself, engaging with people on different social media apps and doing podcasts like this one.
I read the transcript from your appearance on the Laura Loomer show. For people who don't know, she’s a popular MAGA personality and a big supporter of Donald Trump. I imagine that you and Laura have very different politics, and very few Democratic politicians in the US would ever consider doing so. But your approach was very clever. You started by saying, "I want to talk about freedom of speech and resisting authoritarian censorship." I'm just curious to hear more. How did you decide to go on her show? How did it go?
Audrey Tang (55:25)
I decided to be on her show through a common friend, Glen Weyl, who helped Laura get a Bluesky account when she was having difficulties with X.com. She was swiftly banned on Bluesky without even posting a thing, and we helped to restore that account.
The 60,000 livestream viewers on X and Rumble—I got a Truth Social account just for this—were actually extremely supportive of our mission of countering authoritarianism in a way that does not become more authoritarian ourselves. That is an ethos that we in Taiwan take to heart because we cannot afford to go back to martial law. If we go back to martial law, the authoritarians have already won. So we have to counter those threats without resorting to authoritarian means. And that means doubling down on freedom of speech and figuring out how to make more speech heal communities. In a sense, that is also what they want. Paradoxically, I found a lot of validation from the real-time comment box. I insisted on reading them out, and Laura projected it. All in all, it was a very good experience.
Jasmine Sun (57:03)
This is something I have noticed about the way that you approach politics. You always find the uncommon ground. In this case, it's freedom of speech, resisting authoritarianism.
Similarly, you’ll find creative framings and be flexible with language. People react very emotionally to language—if you use the right word versus the wrong word, the underlying action can be the same but people will have very different emotional reactions. Meanwhile, you're very comfortable switching between saying “digital public goods” and “security infrastructure” and “freedom of speech technologies” depending on your audience, but really you're referring to the same thing. Even in the way you say, "My pronouns are ‘whatever,’ call me whatever you want.”
Audrey Tang (57:51)
Yeah, literally whatever.
I truly believe that people want the same thing: life, liberty, and the pursuit of happiness. Who would be against that? Just as in the civil assembly against online fraud, there’s no pro-fraud faction in the political system. There's no constituency that says we want more fraud. In California, there is no constituency that said we want more damage from wildfire. That's literally unthinkable.
What we need to do as politicians and as diplomats is to cherish our uncommon grounds and make them more common. That will very effectively move past perceived polarization, because the actual polarization isn't actually that much.
Jasmine Sun (58:49)
We often have the opposite scenario in the US where you might change the name of a building to be less offensive, but still have exclusionary housing rules. Okay, now the racist guy's name is off the building, but people of different races cannot live together.
Audrey Tang (59:10)
That is the difference between communities taking actions together, rather than just a symbolic representation of the action. It's very easy in the digital age to confuse the two. Only communities banded together can reliably move past just a symbolic phase and into actual actions.
Jasmine Sun (59:35)
Overall, it sounds like you've had a lot of successes navigating these wicked, thorny problems. What you have found most challenging in your political roles? What is unexpectedly difficult?
Audrey Tang (59:54)
One of the most challenging things we overcame together was COVID-19. Nobody expected it. And in early 2020, we faced a twin threat of not just a pandemic, but also an infodemic: a lot of very strongly charged polarized opinions that are not quite right or wrong because the science was not determined yet.
For example, mask use. In Taiwan, we had one faction saying that because of SARS, only N95, the highest grade, is useful. But another faction said that any mask hurts people because ventilation is important and N95 hurts the most. Without some way to find uncommon grounds, people will drift apart even without algorithmic polarization.
But as I mentioned, in every ministry there's a team of participation officers. And it just so happens that our health ministry's participation officer lives with this very cute Shiba Inu—the doge dog, actually. We pushed out this meme of the Shiba Inu dog, putting her paw to her mouth, saying, "Wear a mask to remind each other to keep your unwashed hand from your face." This is truly an uncommon ground, framing a mask as a hand washing indicator. Nobody can say, "I'm against hand washing," there's no such thing. And masks do protect the mouth from your own hands.
Because it's such a good uncommon ground, we actually measured tap water usage and it increased a lot once this meme pushed out. People who have laughed about this meme in a kind of humor-over-rumor way could not get as agitated again by the polarized narrative around mask use. We did that around masks and around vaccines so that we depolarized the society even at a very height of polarization.
Jasmine Sun (1:01:55)
What I’m hearing is that the government needs to get better at making memes.
Audrey Tang (1:02:00)
Exactly. Memetic engineering, essentially.
Jasmine Sun (1:02:04)
How have you evolved your views on social change over the last 10 years?
Audrey Tang (1:02:15)
As an outsider, I thought that the private sector and the social sector are innovative, and the bureaucracy is just maintaining the status quo. But that was not the case. Public servants are actually the most innovative, especially around wicked problems.
All they need is air cover—for example, the presidential hackathon. For that, each team needs at least one civil society, one private sector, and one public sector member. This creates plausible deniability: if an idea doesn't work, the public servant can say, "We're just attending an event; it's the civil society's idea." But if the idea does work—every year there are five teams that receive the trophy—whatever small-scale experiment you just did will get promoted to national infrastructure level in the next fiscal year, and your entire career changes. We provided air cover for public servants to make mistakes and share the credit when they made something work.
That is much more useful as an innovation system compared to letting one or two people airdrop from the private sector or the social sector into the career bureaucracy. Some of that intermixing helps, but I think systemic change needs to be done from within.
Jasmine Sun (1:03:48)
That's interesting because I would have assumed similar. There's a bad stereotype of bureaucrats as just not having any new ideas when you're saying that they have the ideas, they just need air cover. And those are the opportunities that you can create.
One thing that strikes me throughout this conversation is that you are relentlessly optimistic. You don't get angry. There's always a solution or a way out of the most difficult problems. But I feel like you have to humanize yourself a bit for us mortals who get annoyed at things. When was the last time you got annoyed at anything, even just a tiny bit?
Audrey Tang (1:04:30)
Well, I tried this microphone, the HyperX podcast mic. Then the podcast host tells me that I have a metallic, cyborg-like voice. I thought, “Okay, maybe I'll have to replace this microphone after all.” I felt slightly annoyed for half a second. Maybe you didn't see that.
Jasmine Sun (1:04:33)
No I didn't, you were perfect at covering it up.
Audrey Tang (1:04:55)
Well, at 720p, this isn't very good at transmitting microexpressions.
Jasmine Sun (1:05:01)
If that's the angriest that you ever get, we should all learn something.
The last question that I ask all my guests is, what's a research rabbit hole that you’re going down these days? Especially anything more speculative or that you're still learning about.
Audrey Tang (1:05:15)
Recently, I've been surveying various different ways to piece together conversation network technologies. Think about not just one Polis conversation or one Engaged California conversation, but if each participant can take home a snapshot of what was just agreed and disagreed, and start their own neighborhood conversations. I’m thinking of something like RSS that can bridge those different tools—that would be very cool. Some people, like the
project, are funding interoperability work to get all those different projects to import and export in an open format. But how do we weave those formats together in a reliable way that does not require a top-down taxonomy? Again, maybe this is something language models can help with.That is one of my really boring-looking research interests, to look at the existing communication, conversation network, and facilitation practices; and figure out assistive intelligence: AIs that can serve as eyeglasses to facilitators so that facilitators do not need to change their practices, but can automatically facilitate deeper conversations with more people across more scale. That's the subject of a recent paper on ArXiv, "Conversation Networks," that I co-wrote with Larry Lessig and Deb Roy. I encourage you to check it out and let me know any ideas that you may have.
Jasmine Sun (1:07:16)
Awesome, I'll read that. Where can people follow your work?
Audrey Tang (1:07:19)
I'm on Bluesky at audreyt.org; on the Fediverse, on g0v.social, my handle is @au; and I'm still on X.com. Let me post this tweet. "Democracy isn't something you have, it's something you practice daily. Jasmine beautifully captures Taiwan's civic heartbeat and why active participation matters now more than ever." That's a truly beautiful civic muscle post that you wrote. And feel free to follow me there as @audreyt. I'm also on Truth Social.
Jasmine Sun (1:08:02)
A plurality of networks, very impressive.
Thank you so much! This was so fun, and I appreciate you taking time out of your very busy schedule to come on this podcast.
Audrey Tang (1:08:24)
Thank you. Live long and prosper.
Links & books
“garbage in, garbage out” (Jasmine Sun, 2025)
Plurality: The Future of Collaborative Technology and Democracy (E. Glen Weyl, Audrey Tang, and community, 2022)
“The Rise and Fall of the Palo Alto Consensus” (Kevin Munger, 2019)
“When Product Markets Become Collective Traps” (Leonardo Bursztyn et al., 2023)
“Rat Park” experiments (Bruce K. Alexander, 1970s)
“Prosocial Media” (E. Glen Weyl et al., 2025)
“Interview with Laura Loomer” (Audrey Tang & Laura Loomer, 2025)
“Conversation Networks” (Deb Roy et al., 2025)
Thanks for listening or watching!
—Jasmine
Thank you to the Community Privacy Residency for sponsoring this trip to Taiwan and making this work possible!
Share this post