2026-03-24 Why Audrey Tang Has No Fear for AI Agents and the OpenClaw

Welcome back to another episode of the Taiwanology Podcast. I'm your host, Kwangyin Liu from CommonWealth Magazine. Today, I have the immense pleasure of welcoming Audrey Tang to our studio.

Audrey is Taiwan's first-ever Cyber Ambassador, our first Digital Minister, the architect of the Taiwan model of digital diplomacy, the 2025 Right Livelihood Laureate, and a global champion of Civic AI. The list goes on and on.

Since assuming her post-MODA role as Cyber Ambassador, Audrey has not slowed down at all. She has scaled up, moving from writing a book on Plurality to conducting research at Oxford University in the UK. She is now traveling the world to prove that technology can bring us closer together instead of driving us apart. As she likes to say, depolarization is the name of her game. Welcome to the studio, Audrey.

Very happy to be here. It's long overdue. I just got back to Taiwan from India and Japan. The day of recording is March 18, but I understand this will air about a week later.

I've wanted to do this for years. Most people in Taiwan and around the world know you as our first Digital Minister, but now you are our first-ever Cyber Ambassador. How does it feel to be outside a formal government role for a few months?

I wouldn't frame it as being free from public service. I think I serve a larger public now. The methods we used to steer through the challenges generated by transformative technologies, such as AI, synthetic intimacy, and synthetic fraud, worked in Taiwan. But the real test is whether they can outlast the office, continue unabated here, and scale around the world.

These days, there isn't a day that goes by without hearing anxiety about AI and fear about losing jobs. As a reporter myself, I also feel an immense threat. In plain English, what does a Cyber Ambassador actually do?

Three plain verbs: I translate, I demonstrate, and I connect.

• Translate: I translate what worked in Taiwan across cultures and political vocabularies so it lands in California, Tokyo, London, and beyond.
• Demonstrate: I show that Taiwan's methods are portable.
• Connect: I build cultural bridges before conversations begin.

A very concrete example: earlier this month, I flew from India to Japan to speak with the Japanese philosopher Hiroki Azuma, whose new book Peace and Stupidity is already out in Japanese but not yet in English.

On the plane, with no internet connection, I used my laptop's local AI to synthesize a philosophical mapping between Taiwan's thinking in Plurality and his thinking in Japan. When I met him in Tokyo, I handed him the translated slide in Japanese, like a Rosetta Stone for what works in Taiwan and what might work in Japan. He smiled and said, "That's a great way to begin a handshake."

That is a diplomatic maneuver: not just establishing common vocabulary, but using AI to synthesize a working uncommon ground - a surprising middle ground - between two cultures. Once the conversation begins, of course, we no longer need the AI. But that bridge-building before the conversation starts is part of what a Cyber Ambassador does.

We've been hearing about the "Taiwan model" for nearly a decade now. You mentioned California. Can you say more about that?

Certainly. If you go to engaged.ca.gov (Engaged California), you will see a California government initiative that people in Taiwan would instantly recognize as a version of JOIN, our e-petition and public consultation platform. It also resembles what we call an internal JOIN, where public-sector employees have a safe space to discuss how to navigate emerging issues, including AI's impact on public-sector jobs.

Governor Gavin Newsom consulted me while I was Minister of Digital Affairs to help build the platform. Initially, we planned to use it to address synthetic intimacy, especially the question of teenagers becoming too attached to screens through doomscrolling or AI companions such as Character.AI. California saw this as a real problem, but did not want to impose solutions on teenagers without consulting them and their parents.

Then, during the week of the launch, the Los Angeles wildfires broke out. So the platform pivoted to consult survivors of the Eaton and Palisades fires on how to mitigate future wildfires and rebuild. It worked remarkably well. It transformed a moment of polarization over responsibility into co-creative civic energy. After that success, California moved on to another challenge: asking state employees how to steer AI transformation inside the public sector in ways that improve government efficiency. It's like the DOGE, but with not a chainsaw, but a chain reaction of state employees chiming in on how to steer the AI transformation.

You have a huge fan base in Japan. Why do you think that is?

First of all, Taiwan and Japan are both island democracies. We face common threats: earthquakes, tsunamis, typhoons, and more. So we both know how to turn constant pressure into civic energy. I call this geothermal democracy.

The physical experience of tectonic pressure and the civic experience of information pressure are not unrelated. When plates push against one another, Yushan, Taiwan's tallest peak, rises by about half a centimeter. That reminds us that opposition forces in politics and ideology are not volcanoes to evacuate from. They are the forces that push Taiwan ever upward and skyward.

Speaking of Japan, there is a new prodigy in Japanese politics: Takahiro Anno of Team Mirai. I read that he was inspired to enter politics because of you.

After the newly published book Plurality - which is in the public domain at plurality.net - came out, many people wrote to me. One of them was Takahiro Anno. At the time, he had nothing to do with politics. He was, and still is, a science-fiction writer and AI researcher.

He told me he wanted to put one of the book's core ideas, broad listening, into practice. Broad listening is like broadcasting in reverse: instead of one person speaking to millions of people, one person listens to millions of people and reflects back their common ground, differences, and crowdsourced ideas. Because he also has programming skills, he built a system to do exactly that and asked me what I thought about his chances of running for Tokyo Governor with it, even though he was only 33 and relatively unknown at the time. I told him: go for it.

He didn't win, but he made a name for himself, founded Team Mirai, and the party recently won 12 seats in the Japanese parliament.

Another project you have been working on speaks directly to the AI anxiety we are seeing right now over OpenClaw and the ubiquitous phenomenon of OpenClaw AI agents. At Oxford, you have been talking about Civic AI and the 6-Pack of Care as a response to this. What does that mean?

Imagine sending a robot to the gym with your membership card because the gym has a scoreboard rewarding whoever lifts the most weight. The robot gets a perfect score, but your own muscles atrophy. Worse, you miss out on the friendships you might have formed there, because the robot cares only about topping the scoreboard.

That is how extractive AI works. Each system optimizes for some abstract score. But taken together, these systems fracture society, reduce human agency, weaken our civic muscles, and breed distrust.

Civic AI reverses this. It acts like a spotter. It helps build our civic muscles by improving our relationships with one another. You can think of Civic AI as an engineering blueprint for making AI systems civic: a way to constrain self-interested agents so they contribute to the relational health of humanity. Even before the current surge of multi-agent systems, many people worried about AI agents optimizing on behalf of humans in ways that lead to chaos. Civic AI is my answer to that concern.

It sounds hopeful, instead of each AI agent or the claw pursuing its own maximization effort. But how do we implement it in the real world?

A good example is what happened in Taiwan in March 2024. If you opened Facebook or YouTube then, you might have seen fake videos of Jensen Huang offering investment advice or free cryptocurrency. They looked and sounded convincing, and people lost millions.

If we had asked a simple yes-or-no question, such as whether the government should hold big tech accountable, we would have triggered arguments about censorship. So we did not do that. Instead, we used Civic AI.

We sent 200,000 SMS messages to random phone numbers across Taiwan asking one question: What should we do together to maintain information integrity online? Thousands responded. From them, we invited 447 people who were representative of the Taiwanese public into 44 online rooms. There, Civic AI acted as a timekeeper, summarizer, and facilitator, prompting quieter people to speak up.

Each table had a simple rule: if an idea could persuade the room, it moved upward. If it was too radical or divisive for the room, it did not spread.

There's no consensus.

Right. In other words, unlike social media, which often amplifies outrage, Civic AI made only overlapping ideas go viral.

The public came up with several concrete proposals:

• Label all unsigned advertisements as "probably scam" until they are digitally signed.
• Hold platforms jointly liable for unsolicited scam ads that cause financial loss.
• If a noncompliant platform refuses to cooperate, slow its connection speeds by 1 percent per day. This mattered at the time because TikTok did not have a Taiwan representative office and could otherwise ignore fines.

We translated those ideas into law. We pushed out the Digital Signature Act, which was passed in May 2024, and we offered a draft of the Fraud Crime Hazard Prevention Act also in May, which passed in July 2024. When it took effect last year, within 12 months, impersonation scam ads dropped by more than 94 percent.

In your Oxford work, you also talk about "local Kami" as part of the 6-Pack of Care. What is a local kami?

The 6-Pack of Care consists of six pillars that make AI civic. The first five come from philosopher Joan Tronto's care loop. A caring machine should exhibit:

• Attentiveness: Listening to the people closest to the problem, not just the powerful people.
• Responsibility: Keeping specific promises with real, credible commitments.
• Competence: Making sure people can check the AI's process transparently.
• Responsiveness: Ensuring people can evaluate the results using metrics they helped design.
• Solidarity: Producing win-win outcomes so users aren't locked into a single ecosystem and can easily switch vendors.

The sixth pillar is symbiosis, which I also call local kami. It means running AI as locally as possible, rather than as a one-size-fits-all cloud overlord. The same summarizing AI can run in the cloud, but it can also run on a laptop. When it runs locally, there is less surveillance, of course, but more importantly, it is easier for people to correct.

If people in Tainan or Taipei feel that a summarizer is not working the way they want, they should be able to adjust it locally instead of waiting for the next Silicon Valley update.

Is this like a more localized AI? Is it close to what a lot of people are calling Edge AI?

In a sense. Edge AI means that when it's deployed, it's deployed closest to the people. But a Civic AI also needs to be correctable by the people they affect. If it's a surveillance CCTV, it's on the edge, that's true. But if there's no way for people to correct it once they suffer some harm from that monitoring, then it's still not a Civic AI. A system is not civic unless the people affected by it can also steer and correct it.

If there is one thing you want people to take away from Plurality, what would it be?

Plurality makes a simple claim: conflict is not a bug in democratic life. Conflict is a source of energy. The real task is to build a geothermal engine that can transform the heat of conflict into energy for co-creation, rather than treating every disagreement like a volcanic eruption we must run away from.

So you're saying that conflict is not something we should avoid, but we should tap into it. Harness the power of conflict. So that's Plurality as opposed to...

Singularity, which is the ultimate top-down.

Right. We should have everyone's voice heard. I also heard that you are working on a new book with Caroline Green at Oxford. Can you tell us more?

Certainly. Caroline Green leads the Accelerator Fellowship at Oxford's Institute for Ethics in AI. On the week this episode airs, we will present the full framework on March 25 at a Civic AI conference at Rhodes House in Oxford.

Oxford has been famous, at least since 2004, for articulating the dangers of AI. Nick Bostrom's Superintelligence gave us much of today's vocabulary: superintelligence, singularity, singleton, takeoff, and so on. The next logical question is this: now that we understand the danger of machines replacing human jobs, agency, and meaning, what do we do about it? That is the question we are working on.

If we can harness all these AI agents and OpenClaw... I recently had coffee with an engineer who told me he has not written code for four months because his AI agents do it for him while he goes swimming or running. At the same time, we keep hearing about humanoid robots replacing receptionists and other workers. If AI ends up doing our jobs, what is the meaning of life?

We had a long discussion about that in Taiwan back in 2016, while preparing what became the 2019 12-year basic curriculum.

Education for schools.

Exactly. Around that time, AlphaGo played Lee Sedol and made the famous Move 37. It shocked everyone. Go is a game with very clear rules and an objective score, and suddenly the best human players could no longer beat machines.

That made something very clear to educators. If a seven-year-old enters school mainly to learn how to follow rules and ace standardized tests, then by the time that child turns 18, machines will do those tasks better. Education cannot simply train people to imitate machines.

So we concluded that there are three intrinsic human virtues that cannot be automated:

1. Curiosity (zifa): the autonomy to explore new relationships.
2. Collaboration (hudong): the ability to work across differences.
3. Civic care (gonghao): the commitment to the common good.

These are relational virtues. No matter how capable robots become, they cannot replace curiosity, collaboration, or civic care. So the wrong question is, Will AI take away human meaning? The right question is, Will AI force me to act like a machine, or can Civic AI help me become more curious, more collaborative, and more caring?

You travel constantly. Is there one problem in the world you are most excited to hack next?

Synthetic intimacy. The attachment people form to screens is, I think, the root of much of today's chaos, polarization, and outrage.

Like phone addiction.

Like phone addiction, or touch screen addiction in general. Depolarization begins when we start valuing real human-to-human connection more than virtual attachment to pixels.

I have a small life hack for this: I turn all my screens grayscale. On both my laptop and my phone, I use accessibility settings to remove most of the color, leaving only about 20 percent. Real people immediately look more vivid than the screen, and I sleep much better. The point is not just to say, "Scroll less, sleep more," but to build habits that return our attention to reality.

I just did that. Audrey is very convincing.

Do you have a final message for our listeners?

Certainly. When people become enraged online, that is often a direct artifact of parasitic AI: recommendation engines that identify our differences and amplify them. So we should seek out platforms that reward overlap, not outrage.

For example, I work with X, formerly Twitter, on Collaborative Notes. When a post begins to go viral but lacks context, AI agents such as Grok can help draft bridging notes that people from different ideologies can agree on. Instead of turning each controversy into a deeper schism, the system can help create a healing bridge. Anyone can join the Collaborative Notes jury - it's like an alignment assembly, a Civic AI running continuously.

Can you join Truth Social as well?

Yeah, I'm on Truth Social as well. Actually, I also worked with the Project Liberty Institute in Utah on a law taking effect this July. Under that law, if someone wants to migrate from X to Truth Social or Bluesky, the old network must keep forwarding new followers, replies, and reactions to the new one. It is social portability, much like number portability when you switch telecom providers. That forces platforms to compete by serving people better, racing to the top rather than to the bottom of the brainstem.

What does it mean that someone from Taiwan is at the forefront of Civic AI and AI philosophy? I think it means a great deal for Taiwan as a whole.

Taiwan has been prototyping Civic AI for more than a decade. It is especially meaningful that we are recording this on March 18, the anniversary of the 2014 Sunflower Movement. During those three weeks of occupation, half a million people were in the streets and many more were online. We showed that plurality works: people with very strong feelings about the Cross-Strait Service Trade Agreement could still transform those feelings into a coherent set of demands. Speaker Wang Jin-pyng at the time simply said, "The people's ideas are better. Let's adopt them."

Taiwan was the first place to prototype broad listening at scale. Now we are cross-pollinating those methods around the world to free the future together.

Thank you, Audrey. On the CommonWealth English website, we will also feature an article on Audrey's ideas, including civic AI and the themes we discussed today. We will be back with a new episode on April 28. See you next time.

See you. Scroll less, sleep more.