• Yes, absolutely. CL Kao did the same thing and then he sent me a really nice Granola summary.

  • Yes, I use Granola as well. It’s very convenient.

  • I’d never heard of it. You’re going to make me think AI is a good thing, not the spawn of the devil.

  • It’s assistive intelligence.

  • I’m doing a project about the relationship between social movements and democracy. This part of the project is about talking to people who are innovating and thinking about global challenges in a way that is useful for a global conversation. I’m trying to get those conversations together and paint a picture of what’s happening in a world facing many crises. Different people have different visions and expertise. That’s the bigger context.

  • Some say it’s the same crisis; they call it the metacrisis.

  • Some call it cascading crises.

  • They’re all connected in one way or another. I want to start with a more personal overview of your own journey. How would you define yourself?

  • I’m Audrey Tang. My work has been fostering conservative anarchism , which means honoring human traditions and existing institutions, and changing them in a way that’s non‑coercive—shifting vertical, top‑down control to horizontal, voluntary cooperation.

  • It’s a very different branch of anarchism than anything violent, sabotaging, or coercive. We don’t fight coercion with coercion. I’m heavily influenced by Taoism growing up. Another way to say conservative anarchism is Taoism.

  • How did you become involved in activism and politics?

  • When I was a child, Taiwan was not yet a democracy; it was still under martial law. My parents were both journalists. My father went to Beijing on May 12, 1989, and the very next day the Beijing student pro-democracy hunger strike began. He stayed in Tiananmen until June 1st before he returned, which was very fortunate for our family.

  • In November of that year, the Berlin Wall fell, and in January of the following year, he went to Germany to pursue a PhD in political science, and also went to Berlin to cover the annual conference of pro-democracy activists. He pursued his PhD in Saarland, near the French border, and his thesis topic was about the autonomous civic organizations in the Tiananmen movement.

  • I went to live with him in Germany for a year. During that time, young students from Beijing, Shanghai, and other places would visit our living room—they were people in their early twenties who remained passionate about democracy after the Tiananmen Incident.

  • In Taiwan, the Wild Lily student movement and other movements were happening at the same time, calling for democratization in a thoroughly non-violent way. To have personally experienced all of this was fundamentally formative for me.

  • For me, getting into the early internet and working with democratization activists were literally the same experience.

  • I wouldn’t point to a single turning point. My journey into internet governance and internet freedom, and my trajectory toward democratization, were two parallel paths that eventually converged.

  • Exactly. There’s no separation between activism, politics, and technology while lifting martial law and democratizing with the help of the internet.

  • Main influences would have been conversations around the house, activists, and democracy people. In terms of applying internet or digital tools to democratization, were there other influences that motivated or inspired you?

  • The founding people of internet governance are still around. I still work with David D. Clark every couple of weeks. He came up with the notion of rough consensus and running code.

  • The early internet engineering pioneers designed an egalitarian system. Anybody with an email address can participate. Anyone can nominate themselves to the governance board. Electors are drawn by lottery. It’s very sophisticated. I participated when I was 14 or 15—five years before I could vote—immersed in an advanced political system called internet governance.

  • When I turned 20, I cast my first vote in the neighborhood chief (里長) election. The candidate I supported won by one vote.

  • It’s your vote. Very empowering. When they say one vote doesn’t matter—oh yes it does.

  • My first vote literally mattered.

  • Take me to the Sunflower Movement. What was that experience like for you—not just technically but emotionally? What happened, and what was your role?

  • Before Sunflower, g0v was already supporting demonstrations with internet connectivity. 2014 was the first year YouTube Live, Facebook Live, Twitch, and so on, went production‑level. The first year of livestreams.

  • I translated parts of Manuel Castells’ Networks of Outrage and Hope and went deeply into his network power theory. We were influenced by the wave of internet‑enabled activism in Spain and elsewhere. We were at the tail of that wave and benefited from studies of Occupy—international configurations where each Occupy sent two people to others.

  • It felt like putting theory to practice. We made new applications, but the Castells theory was clear in my mind. It influenced our commitments to radical transparency and to demonstrating renewed democratic systems—not just protesting. Those were straight from his texts.

  • g0v had been working for months on analyzing the specific legislative issue that detonated the trade deal protests. Persistence.

  • There’s the idea of a crisis or urgency moment creating an opening. But if you hadn’t been working for eight months on that particular issue—you didn’t have the portal, you didn’t have…

  • Civic infrastructure.

  • Infrastructure, yes. Is it about preparing? Is the power of civic hacking in having many experiments and people—a whole collective—working on things that are bubbling up and then being ready when the moment comes?

  • You have many waves and don’t know which will meet the urgency. With an open‑innovation culture , once you hit a wall, some advanced waves will tunnel over the wall— quantum tunneling . It’s a very different theory of change from the classical social movements model of amassing public demand, climbing the support curve, and pushing reformers inside.

  • That makes sense. I wrote an early article on Spain’s 15M arguing it wasn’t spontaneous; it was decades of mobilizing and being ready to seize a crisis opportunity. It’s important not to lose hope when things look bad because you never know when the opportunity will come.

  • When things go bad—private harms, chaos, polarization—those can serve as triggers. People get fed up and amass. Civic infrastructure is like a combustion engine : if you harness the social fuel of conflict at that moment, it produces something beautiful and co‑creative. If you don’t, as many Occupy sites showed, it disperses.

  • Spain’s Occupy led to many changes because there was a pre‑existing network that continued to grow and got an influx of fuel. Where there was no network, people had energy but nowhere to channel it—not organizations, but infrastructure and networks.

  • Yes, civic muscle . I felt that at the Medialab-Prado around 2015–2016. There was a huge amount of civic muscle built before 15M. 15M was the front of existing connections.

  • Then COVID—an unprecedented challenge.

  • In Taiwan we also had SARS.

  • Did you learn from that previous epidemic?

  • Yes. During SARS we lost more people than any other country. It was contained because it was too toxic—self‑limiting. It didn’t spread a lot because people got sick very quickly, but it could have been very bad.

  • Our national healthcare system existed but used paper cards. Only Penghu had an IC‑card pilot. Without an electronic system for cybernetic feedback, it was chaos—random phone calls, everybody to everybody.

  • The health minister who coordinated SARS, Chen Chien-Jen, later became Vice President during COVID, a Johns Hopkins–trained epidemiologist. Right after SARS , he designed the CECC (Central Epidemic Command Center). The system puts all ministries in one place, with the health minister as commander. Communication is no longer everyone calling everyone. There’s a shared dashboard, an intelligence center, doctors joining government (paid even more than deputy ministers), PPE stockpiles, and so on.

  • My role was ensuring civic technologists could make full use of this infrastructure so it reached everybody and reduced epistemic harm and chaos— fast, fair, fun —things people would vote for. Without the electronic NHI infrastructure, there would be no ingredients for g0v to design contact tracing , mask visibility , vaccination registration , and other systems.

  • So the ingredients were there.

  • Yes, but particular designs needed privacy. Existing contact‑tracing had people writing phone numbers; the next person could see the previous number—a privacy hazard. Google Forms and other venue‑level aggregations created attractive targets for cyberattacks.

  • Civic technologists critiqued the old system. I worked with them: here are the APIs powering the old system. If you design privacy‑preserving, zero‑knowledge methods—venue learns nothing about you, telecom learns nothing about the venue, government learns nothing—yet notifications still work—then in three days I commit to integrate it so the civic invention becomes new public infrastructure. Fork, then merge —from a strong foundation that wasn’t popular before.

  • Now, you’re a Cyber Ambassador‑at‑Large . Is that the correct title? What does it entail? Goals? Ambitions?

  • I get summoned by missions of Taiwan around the world. For example, at the AI Action Summit in Paris in February, I showed up sharing Taiwan’s experience. Then I was in Munich as well. Interestingly, I listened to J. D. Vance’s first and second international speech. It’s quite an experience.

  • Last month I was supposed to be in Tel Aviv for Cyber Week —a meeting of cyber ambassadors and security officers, although it’s now postponed to December. I think I’ve gone to more than 20 countries , averaging one time‑zone change every four days .

  • The outcome: share the Taiwanese experience of building civic infrastructure of horizontal trust that is anti‑fragile against polarization, radicalization, and the authoritarian narrative that “democracy only leads to chaos and division.” Spread the message of empathy across info/cyber/social domains.

  • Talking about democracy: what does democracy mean to you?

  • People come together to make decisions that concern them in a way people generally can live with.

  • What do digital tools give us that we didn’t have before? The most important innovations of the past decade?

  • Language models —human‑level cognition engines that understand across cultures and disciplines the nuances of qualitative human expression. Electricity gave the internet; the internet gave language models; the language models contain most of the public internet. The microcosm of the internet is now in language models.

  • What can they do for us, and what can’t they do?

  • In Taiwan’s 2019 curriculum reform we assumed: anything that can be automated will be automated. Anything with standardized answers memorized by rote won’t matter. Those literacies were core to basic education. What still matters when cognitive and most motor functions are machine‑driven?

  • We decided curiosity, collaboration, and care for the common good are what matter. These are competencies , not literacies—fostering intrinsic motivation to be curious, to collaborate, to care. That became our new curriculum.

  • How does that feed into democracy? You could have that curriculum in an authoritarian system. What is the connection between those abilities and democracy?

  • Democracy is participatory decisions and processes people can live with. One needs to be attentive —a special curiosity about actual needs and emerging shared challenges, and a curiosity to address those needs together .

  • Journalists have this curiosity— investigative journalism —which autocracies dislike. You also need responsibility : each person taking some responsibility in addressing part of those needs, building confidence , exercising those muscles, delivering, and checking feedback—being responsive to what’s actually happening.

  • Autocracies think in a command way: if a solution is deployed and some people are unhappy, those people need to align because it is state policy. In a democracy, if it doesn’t fit, we figure out a fork —like privacy‑preserving contact tracing—that makes people happier. Ideally, everybody is slightly happier than before and nobody very unhappy . Autocracy aims to make the holders of power very happy; if some are very unhappy, that’s their fate.

  • When you say “everybody can live with,” you’re talking about tolerance versus preference —not maximizing the number of people’s preference, but maximizing tolerance so everybody can live with the outcome.

  • Yes. Guarantee that any coalition of ~10% can veto the worst 10% of proposals concerning them. It’s a dynamic coalition. We require consent, not consensus —protecting against what we don’t want versus mandating what we do want.

  • A high‑resolution democracy , enabled by digital technologies, figures out bundles of issues so everybody is a little happier and nobody very unhappy, instead of low‑resolution 51% majoritarian wins.

  • A lot of societies have internalized hegemonic narratives —e.g., patriarchy —that undermine democracy. What happens if common sense isn’t good sense ? If 80% share a problematic view, what do we do?

  • It’s a function of aggregation . An 80% block is just majoritarian aggregation. With proportional or bridging algorithms , we reveal clusters—say four stakeholder groups: one 80%, others 7%, 7%, etc. In a bridging system , the 80% still must bridge to other groups.

  • We show how preferences cluster and offer bridging bonuses for proposing rarely discussed uncommon ground with nearby groups. People who propose novel statements gain the bonus. Finally, we get bridging narratives ; four groups may coalesce into two; then another round of bridging.

  • At the end, you need consent from all smaller groups . A 7% group can veto the thing that hurts them most. A 21% coalition can veto 21% of harmful proposals. We find and amplify uncommon ground , orthogonal to majority opinion—the opposite of antisocial social‑media corners where least‑bridging content gains virality. In our system, the most bridging gains virality.

  • So it’s iterative—each iteration brings you to a better, more satisfying outcome.

  • Yes. It produces uncommon grounds that are bridging, then presents the best argument of the remaining divide— balancing —which is how newspapers are supposed to work. The Hutchins Commission : front page—what everyone can agree happened; second page—the best arguments of main ideologies. Bridging first, then balancing , then resolve, produce more bridging, and balance again. An interactive, evolving system.

  • There are digital divides, and some people don’t want or can’t participate. A single mom working all the time—does she have time? Motivation?

  • If it affects their child and there’s something they can do , they will be motivated. The agenda must be set by people closest to the pain .

  • In Taiwan you can have citizen initiatives ; some now have hundreds of thousands of signatures online . Demographics show 17‑year‑olds and 70‑year‑olds are most active: they have more time and care more about intergenerational unity than people focused on the next quarter.

  • Market mechanisms address next‑quarter pain points. Truly non‑market issues— banning plastic straws for bubble‑tea takeouts , or a menstruation museum to remove taboos—are started by those caring about the long‑term future. Participation infrastructure brings ministerial conversation to such issues, amplifying niche agendas to a national platform .

  • But you must have the infrastructure that enables those voices to enter the system. In many societies, corporate control of huge platforms shapes sociality and politics. How do we resist that power, domination, and repression?

  • When I was a child, ATMs only served cash to cards of the same bank . New banks couldn’t compete; big banks cornered street corners. The state intervened—not by nationalizing—but by mandating interbank protocols . A small bank’s card could withdraw from all ATMs.

  • Later, with telecoms, people wanted to switch for better reception but would lose their phone number . The state mandated number portability .

  • A simple fix: adopt the EU Digital Markets Act approach— portability between instant‑messaging systems. The EU already has this. The State of Utah went further: starting next July , when you switch from TikTok to Instagram or from Twitter to Bluesky, the old network is mandated to keep forwarding new likes, replies, reactions, and content to your new network— social portability , like number portability. It’s a consumer‑choice act. Not nationalizing big tech—just removing the network‑effect lock‑in .

  • And they can make them do that? They seem so powerful. What if all of them are the same type of platform—tech bros censoring progressive speech?

  • Without interoperability, if Jay Graber gains 70% of the market, investors might force a sale. When Twitter was the underdog, they were good.

  • The point isn’t blaming a CEO. If Bluesky users fork into Blacksky , Bluesky must keep interoperating with Blacksky—which is already happening. Law can require on‑ramps and off‑ramps —full liability for networks that don’t interoperate. It’s technically feasible; privacy concerns are now solvable. Bluesky already shows how.

  • How does that get you a platform that isn’t selling your data?

  • Once a platform sells your data, you switch to one that doesn’t —and keep all your friends, content, and followers .

  • Will people be aware enough to do that? My students often aren’t bothered about data use.

  • It’s a product market trap issue. A study showed: an average undergrad in the U.S. using TikTok would need $60/month to switch off TikTok. But if there’s a magic button that moves everyone they know off TikTok, they’re willing to pay $30/month to stop using TikTok.

  • Don’t they get a benefit from TikTok? Otherwise why be there?

  • They’re there because everybody else is there . The first mover suffers FOMO . It’s like luxury brands: if everyone carries LV or Hermès, you feel compelled to invest; you don’t get much. If everyone switches to non‑luxury as the norm, everyone is better off. With TikTok, the difference between individual loss and collective gain is even higher.

  • To break monopoly, don’t scare people about data; just switch to where they are in control . If they don’t lose friends, why not ?

  • We saw that with X and the migration to Bluesky, though the affordances differ and experience differs. Many radical geeks dreamed of non‑corporate platforms, but projects weren’t sustainable.

  • Signal and Proton serve needs ; but many others hack preferences —like junk food . No one needs junk food, but exposure shapes preferences against our meta‑preferences (meaningful connections). Free software doesn’t do that. Forced interoperability creates the offering.

  • AI is revolutionizing society—ethically, environmentally, economically—with many negatives. Challenges and opportunities? Institutions aren’t very reflexive; they think “it’s great.” What should we do?

  • I distinguish assistive intelligence and authoritarian intelligence . Authoritarian, vertically aligned AI subjects the model and outputs to the will of one controller. It disempowers people; machine judgment replaces human judgment, aligned vertically to the controller’s principles.

  • There’s horizontal alignment : the mediator aligns not to my will or yours, but to the relational health of the conversation. Many ways to align models to relational health —optimizing the health between people who previously disagreed.

  • We use AI to find promising bridging statements between clusters and rewrite them so everyone can understand in their native language and culture— summarization, publication, cultural translation . The AI serves the community’s needs and then self‑effaces when needs are fulfilled.

  • Is that written into the code?

  • Yes. That’s the difference from garbage AI producing wrong information and falsifying. Tell it that vaccines contain 5G chips and it might say, “You’re the only one seeing through the veil; don’t believe police or journalists.” That’s epistemic harm .

  • Algorithms repeat bias in data. With assistive AI, we eliminate garbage and bias —like preparing a curriculum . Early interpretations are key. Language matters. Ask about the ultimate ending of AI–human relations in English and you get dystopian Terminators; ask in Japanese and you get symbiosis. Japanese sci‑fi describes symbiotic systems; English literature often describes Terminators. Same query, different cultures.

  • Creating a curriculum for AI systems is crucial.

  • How do you get people to use the right AI? The average person only sees the garbage version. How do we get to the civic version?

  • Many countries are doing public AI / sovereign AI —local versions more culturally attuned. Countries, including Taiwan, are preparing curricula. It’s like an immigration test : you expect a new immigrant to understand how society functions. For robot immigrants taking jobs, people expect understanding of social norms. With the EU AI Act , most EU countries are building public AI curricula. There will be alternatives. Swiss-AI is training a good European‑values version.

  • At some point, people will say: if you use a culturally non‑aligned version for certain jobs or activities… Most of this is in the EU AI Act ; they’re now defining enforcement rules. There’s huge industry lobbying, but the Act contains most of the language I’m talking about.

  • I didn’t know that. Universities say “use it, just disclose how,” which seems insufficient. Also the economic use: airlines using your data to decide how much to charge for tickets. Legislation and policy seem the only control.

  • Of course. Since Coded Bias —like The Social Dilemma —policymakers aren’t blind. People want working programs , no matter how small. In the “pocket of good,” in Taiwan we sent 200,000 messages to people; 450 people came together and deliberated online; starting this year, if an ad can touch a Taiwanese person’s eyeball , Facebook needs strong digital‑signature KYC . If they don’t, and someone lost $7 million to unsigned ads, the platform would be liable for that $7 million. If they still don’t comply, we can slow down connections to their service until conditions are met.

  • Genius—and doing something.

  • If you had to say the most important steps to safeguard democracy—here or globally—what do we need to do right now ?

  • Focus on our capacity to care . For too long, the digital realm is viewed utilitarianly: jobs, GDP, growth, data centers. Once a civilization is established in human rights and democracy—it’s a full chicken no longer just out of the egg. In the egg you grow; out of the egg you mature . Once mature, you want to flourish in your well-being. If an adult chicken keeps growing, that’s cancerous growth . The system isn’t designed for exponential growth. It has limits in social and environmental capacity to care . Outgrow both sides of the doughnut , and you have an exploding chicken .

  • If we refocus on care , we’re attentive to actual needs and compete on how quickly and capably we address them—without self‑aggrandizing or creating demands based on addiction. Pivot back to capacity to care .

  • I think you’re absolutely right. How do we get people to focus on care? Through curriculum, education, civic ideas?

  • Make care outward —not just about individuals’ emotions but society’s emotions . If a part of society feels really bad, it’s not just for psychiatrists or caregivers; the democratic system needs to shift so measures evolve into something they can live with. Civic care is very important.

  • In the U.S. the narrative elevates cruelty as a value and denigrates empathy as feminine.

  • We need to make care appealing— a care gym —six‑packs of care. Make care bro‑friendly.

  • What concerns you most among the meta/cascading crises, and where are you focused?

  • We focus on machine ethics : training AI systems not to enslave, lobotomize, or maximize profit/engagement, but to imbue civic care . That’s my Oxford research topic: 6pack.care — How can you align AI horizontally so they maximize the health of inter‑human and human–machine relationships, maximizing civic care ?

  • Many people are in pain, suffering, fear, and anxiety. Do you feel optimistic about the future?

  • Yes, because we have anti‑fragile infrastructure to meet the moment. Ten years ago, when Taiwan faced epistemic polarization, we were top of the world in receiving it for 12 years; it was lonely. Few understood epistemic security as part of cybersecurity.

  • Now, after last year when all ruling parties worldwide lost seats , people see that vertical institutions—journalists, academics, ministers— lost relevance . People find belonging in amplified social media; even robots dressed as people feel closer. In Japan, this election saw a viral YouTube alternative narrative party gain ground. Once in that echo chamber, you feel you alone see through the veil, supported by similar people.

  • I’m optimistic because everybody sees the same thing now. No democracy is an island; all are connected.

  • Are people aware of that, or do they think they’re the only ones seeing through the veil?

  • Most people are tired of polarization . On the left and right in the U.S., nobody likes it. The potential for uncommon ground is high— product‑market fit for community notes and horizontal AI.

  • Some issues, like Gaza , are deep moral lines . There are two sides with no reconciliation; you’re either for it or against it.

  • That makes the uncommon ground rarer , but also more desirable .

  • On a personal note, when things get tough, what motivates you to keep going? You have tremendous stamina.

  • I’m a very competitive sleeper . I just had more than 8 hours . When things get tough, I sleep nine or ten hours ; that restores me.

  • Sleep is my favorite thing too. In Taiwan, government was receptive to civic hacking—exceptional. What strategies would you recommend to civic hackers and activists in contexts where establishments resist change?

  • Taiwan used to be the largest polity where civic infrastructure became public infrastructure. Smaller polities have many examples— Reykjavik , and most city‑level citizen‑initiative systems far exceed Taiwan’s. We pulled municipal innovations into a national infrastructure.

  • If you scale down to 100 or 1,000 people , most places learn better ways of meeting. There’s little pushback introducing digital democracy at small scales . Pushback appears at 1–10 million people, where existing stakeholders has interests in blocking new infrastructure.

  • For a while Taiwan was the largest polity trying this because of pressure. Now California adopted the same system. I worked with Newsom’s team for two years— infrastructure, deliberation, digital tools . We chose an uncontroversial issue: teenage use of social media . We were about to launch with children and parents; on the week of launch, wildfires happened. We pivoted immediately . The click‑through rate on wildfire recovery on the engaged.ca.gov page was 25% —five times more than anticipated—because of urgency. It delivered uncommon ground on wildfire-related issues.

  • In Japan , something similar is happening. Takahiro Anno , newly elected, ran on broad listening, de‑polarization, bridge‑building, plurality with the Team Mirai , the Future Party , running on AI‑enabled assistive deliberation ; they also won.

  • You need to prepare at the meso level —thousands up to half a million people—to exercise civic muscle. Don’t go to the gym and lift heavy stuff first; you’ll strain yourself. Once you keep exercising and urgency comes, you can scale across many polities.

  • Scale across; scale in before you scale up and scale out.

  • When governments are deaf to the people—e.g., Gaza—the public is out of line with leaders. There’s alienation, disenfranchisement, anger. You’ve tried everything: mass mobilization, alternative journalism, social media with images, and they still won’t move. What can be done to exert grassroots pressure when lives are at stake?

  • That’s the traditional theory of change . I advocate quantum tunneling theory of change : don’t try to scale up your demand until you have scaled across sufficiently.

  • Start many swarms of dialogues about adjacent issues. In Taiwan we didn’t tackle marriage equality first. We tackled how single mothers can raise children without discrimination , menstruation , and other adjacent but important issues. With thousands of such movements, one is bound to hit a nerve, building support beyond the original issue.

  • We had petitions— 8,000 for changing Taiwan’s time zone to UTC+9 (to show sovereignty), and 8,000 to keep UTC+8 ; it was a standoff between the two sides. We convened both and proposed dozens of ideas : instead of spending resources on time‑zone change, spend half on popularizing LGBT‑friendliness , hosting human right conferences in Taiwan, offering gold cards to open‑source contributors, etc. After exploring ideas, people saw time‑zone change wasn’t the best effect. Eliciting them in an ethics of care reveals other needs they can address together. Don’t start from a one‑bit ontological debate; start from adjacent .

  • In crises you don’t have time to develop that. Limitations mean it’s the long game —preparing for the next urgency, not a magic button.

  • If you wanted to advise activists working to strengthen democracy, thinking about the Taiwan experience, what advice would you give?

  • Sleep more. Take care of yourself.

  • Think not only about fast or fair responses, but also fun responses. Humor over rumor is a consistent lesson. In a joyful mood, people are more creative in finding ideas others can live with. Starting only with stress and anger narrows a corridor.

  • Expand the corridor between domination and chaos—even if it starts narrow, expand it.