I’m really happy to meet you. I will not waste any time and start by asking: what is your vision as a cyber ambassador?

Okay. Before I answer anything, you probably already know this, but I will make a local transcript. Once you publish, I also publish the full transcript in the public domain on the SayIt website, if that’s okay with you.

It’s okay. Are you going to publish the full transcript?

Yes. But, of course, if there are parts that you want to modify or change, we can negotiate. You can edit. Usually, I put it on a shared document, so you also review it as you write. But I will not scoop you. That is to say, you always publish first, and then I publish.

Great. I’m grateful for that, and for the ability to edit it, because my English is not perfect, or I might make a joke that I will regret.

Of course. This is not a livestream. This is not to put anyone on the spot. This is just to make a record so that future journalists can also reuse this conversation. We published more than 2,000 conversations with more than 8,000 people this way.

Great. Just so you know, I will not publish the whole transcript. I will edit it. I have a word limit, so I have to pick and choose. Okay. So let’s start with your vision as a cyber ambassador.

Yes. My vision is centered on Plurality. It’s about fostering collaborative diversity enabled by technology. As Taiwan’s cyber ambassador, my mission is to facilitate global connections to share Taiwan’s model of digital democracy and resilience, and also forge partnerships based on shared values.

As you know, “cyber” in Greek means kybernetes —to steer. Together. So we need to steer technology so that it advances freedom instead of removing freedom. This involves initiatives in the Krach Institute for Tech Diplomacy, where I chair the trusted tech standard, defining what it means for technology and the supply chains to be trustworthy—respecting human rights, privacy, and democratic governance. All in all, it’s about moving beyond defensive measures to advancing technology, steering technology to align with democratic values.

That sounds interesting, but for someone who is not proficient in anything that has to do with digital democracy, I would ask you to elaborate. For instance, let’s start with the question: why is a cyber ambassador, why is your vision and your mission necessary at this certain moment in time?

I think even people who are not online all the time understand that the challenge we face from the Internet has no borders. For example, the spread of polarization, the way that people see their neighbors not as a single community, but many different sides that dunk on each other. This is a direct consequence of the antisocial corner of social media that amplifies polarization.

Of course, the centralization of data by a few tech giants is very far from Europe. That is also something that people understand: if a Silicon Valley company changes its mind, it affects the livelihood of people here in Europe.

And also about our culture, our knowledge. If our culture is far from Silicon Valley’s, then the AI models trained in Silicon Valley by non-European entities will give very different answers from our cultural norms when you ask them questions. When everybody asks a few large AI models the same questions, it has a colonizing effect on European cultures—not to mention Beijing models that do the same.

So a cyber ambassador helps bridge these divides. We, as a diplomatic community, facilitate dialogue between government, civil society, and the private sector across cultures, across jurisdictions, because we need to collectively ensure that the digital infrastructure we rely on can support the local communities and cultures, not take them away and undermine these cultural values.

You mentioned the diplomatic community. Would you tell me a bit about this community of which you are a member?

Definitely. For example, earlier this year, in Paris, we were part of the Paris AI Action Summit. Many people in the diplomatic community joined together and launched ROOST, the Robust and Open Online Safety Tools project.

What it tries to do is that instead of a single, centralized entity somewhere in Silicon Valley making everyone subscribe to the same standards—what is acceptable speech, what is not acceptable speech. Until now, only wealthy tech giants could afford to build this kind of machine to keep the platforms safe. But the problem is that what counts as safe is then also determined by those wealthy tech giants.

So ROOST links together the stakeholders so that anybody can access safety as a public good, like a vaccine. Like AstraZeneca, right? No patents. No restriction on who can manufacture it. It is available to everyone. So people can build the immune system for any online community. For example, something that scans live data to block spam attacks, scams, and things like that, based on transparent rules set by the local community.

Of course, we also care about child sexual abuse materials (CSAM) and many other local specific laws and community guidelines. But again, this is a community that gives everyone the tool to make sure the online experience is safe, but we do not centralize power and dictate what is safe or unsafe for each community.

How can I use ROOST if I’m an Internet user? I understand that ROOST has especially to do with preventing sexual violence against children online?

It is one use; there are many other uses. But yes, CSAM prevention is one use.

How can we use it? For instance, how can a family use ROOST? Can a school? Can a country? How does it work?

For example, if you are a school, and you have your local forum where the parents, teachers, and children can have online conversations. On the forum, you have a code of conduct: what is permissible speech and what is not. ROOST provides the software that you can ask your forum moderator to install. Currently, the tool called Osprey is built by Discord, which is a very popular chat room many young people use. Bluesky is also using it—again, a quite popular, although not super popular, social media.

The tool is available so that if you run your own server in the community forum, instead of having Discord or Bluesky manage your forum, your local forum moderator can also use Osprey. The great thing about ROOST is that as long as the parents, children, and everyone agree on the code of conduct—that is just a text file, like a Google Doc—you can give Osprey this Google Doc. And then the brain, the safeguard model, will understand this code of conduct. Each post is then compared against this standard. If it does not match, it can provide a citation of exactly which words in the standard of conduct this post is violating, showing the whole thinking mechanism.

To run this, you do not need to upload everything people posted to some cloud somewhere. This belongs entirely in the community. If it made a wrong decision because the wording was ambiguous or something, you just change the text file, and it changes its judgment. You do not need to pay a subscription. You do not need to wait for somebody else to change the rules for you.

This is very interesting and exciting. If I understand correctly, you don’t believe that large language models are the best way to make good use of artificial intelligence, right?

I believe in community models. A language model running in a far-off data center where there is no visibility of why it is changing, how it has changed, whether it causes new harms or makes new opportunities—there is no way to be attentive to the actual people’s needs.

But you can take that model, and it fits on a USB stick. It’s actually very small in file size. You can run it on your school’s computer. You can even run it on your laptop. If it misbehaves, you just change a text file, and you change how it behaves. This is not saying that I am against cars. I’m for cars with a steering wheel. I’m not for cars that only accelerate and stop and have no steering wheel.

The analogy is very helpful. Let’s go back in time. You helped shape g0v (gov-zero) in Taiwan, and then you created the vTaiwan platform that used Polis. At the time, were you already aware of polarization? I have read that your tools helped a lot. Were there voices on the extremes of the spectrum that were hijacking the movement?

Of course. In 2014, Taiwan’s people were already some of the most connected. I think there were already more Facebook accounts than the population. It was already hybrid connected. The government found that the communication capability was under strain, and public trust was very low, around 9%. The movement was a reaction to the loss of legitimacy of the government.

The government was looking at the population and found it had lost trust because of very high PPM (polarization per minute). Every minute, there was a lot of polarization making communication very difficult. The air was very thick. You could not see each other well.

The movement was not just a protest; it was a demonstration of how live streaming, collaborative documentation, and structured deliberation could cut through the high-PPM environment and facilitate consensus among half a million people on the street and many more online.

The government, especially the senior leadership, recognized this gap and invited us, the civic technologists, not to take over the government, but to work with the government. I always say I don’t work for the government; I work with the government, bringing the spirit of the movement—working with the people, not just for the people—into the halls of power. I accepted this on the condition of radical transparency; that is to say, everything I do, the process needs to be open to the public.

How did you promote radical transparency as a minister?

As I mentioned, I have the SayIt website that I set up even before I entered the parliament. For the past ten years, if you check out this website, you can see 8,000 speakers, more than 400,000 speeches, in more than 2,000 meetings. This helps because once people understand the why and how of policymaking, it makes it impossible for them to get polarized by conspiracy theories filling in the void. Polarization only grows in the void of participation.

Also, social media platforms in 2015 started switching to so-called “engagement” algorithms. Prior to that, if we subscribed to the same journalist, we saw the same content. But since 2015, they have a “For You” feed that figures out our differences and does engagement through enragement, amplifying the outrage and the division. The vTaiwan process and radical transparency were explicitly designed to counteract this tendency by holding a space for prosocial media, not antisocial media.

The dialogue, like Polis, is designed not to broadcast. In broadcast media, the extreme voices hold the megaphone. But in broad listening in Polis, it focuses the conversation on shared feelings. If your feelings resonate across the political aisle, across ideology, then that reaches people. But if your ideas are very extreme, it actually does not reach people. It’s the inverse of the antisocial corner of social media. It makes the uncommon ground—the surprising common ground—viral, not the extreme. The tools to bridge divides are the way for us to talk with the people directly and be more popular than polarization or populism.

Would you explain to me how you did that with social media? Facebook is an antisocial media. It promotes polarization. X, even more than that. So how did you utilize those social media? Or are you talking about different platforms altogether?

Ten years ago, we built alternatives. That’s our demonstration. For example, Polis. It works very simply.

Let’s see that I understand. Polis was a social media platform?

It is, but it is prosocial.

Okay, so it’s a social media platform where citizens can upload their opinion, their thoughts, or a question?

Yeah. Do you want me to give a very quick demo? Do you see this screen?

Yes.

This is a recent case of using Polis in the state of Kentucky, in a town called Bowling Green. What it does is that many people in Bowling Green, which is a small town, go online on Polis and share their ideas.

This looks like a short tweet. You can see it says, “Bowling Green needs more indoor attractions for families, such as education and museums that are free or low cost to the general public.” You can like, you can unlike, you can pass, but there is no retweet button. Without the retweet button, you cannot attack one another. You cannot make it viral. All you can do is to resonate or not resonate. You see this one, public transit going to county parks—

Okay.

—by far, 90% of people like this idea compared to people who don’t. Within 33 days, about 10% of Bowling Green people went on Polis and interacted more than 1,000,000 times. A language model summarized their common ground so that everybody can see the highlights of what was agreed. They’re automatically clustered into topics of infrastructure, community identity, and so on, without human intervention. This is entirely automatic.

People can see, for example, community pride and image. Nobody wanted to say, “Let’s all speak Russian only.” But people agreed that we need to have development across the railroad tracks. No billboards except along the interstate. Again, the language model summarized within this subtopic the common need to be seen as unique to Nashville or burying utility lines, and so on.

This is called uncommon ground because people coming in expect they’ll be very polarized around private education, around marijuana, around many polarizing topics, but actually, they managed to agree with their neighbors on more than 80% of the ideas. After that, it really depolarized and deescalated. The trick here is that we give the “bridging bonus”: the more distance some ideas can bridge people with differences, the more virality it receives. Completely different from Facebook or X.com. Am I making sense?

Yes, absolutely. But I’m playing devil’s advocate a little here. Won’t the same people that go on the platform to talk and debate about local issues also go on Facebook and go at each other’s throats at the same time?

I would say that this is like a gym. A civic gym, where people exercise their civic muscles. Of course, there are also people who go to the Colosseum for gladiator fighting. But it doesn’t mean that gyms are not useful. It means that gyms are essential. People understand there are places for entertaining dunking, but once we want to do something as a polity—for example, in Bowling Green—we have the committed listeners. That is to say, not just the local political leaders, but also the cultural leaders, the public health group, talent development, and so on.

In traditional social media like Facebook or X, the political class will dominate people’s oxygen. They would not have the ability to talk about Boys & Girls Clubs, storytelling, and so forth. It would just be crowded out. But on Polis, these ideas surface so that they can respond to the part that produces the rough consensus that can be acted upon.

I think both have their merit. Ten years ago, when we came up with this, we published the algorithm, the way the ranking works. Today already, for Twitter (X.com), YouTube, Facebook, and Threads, you see Community Notes, which is a way for the community to make footnotes. It uses the same bridging system design that we published. So we’re already influencing mainstream social media.

Okay. And this work during the Sunflower Movement and later, it actually drove polarization down and trust went up, right?

Yes. By 2020, the approval rating was more than 70%. In a short span of six years, we made the approval rating grow from 9% to 70%. In 2022, the ICCS (International Civic and Citizenship Education Study) ranked Taiwan 14-year-olds as top of the world when it comes to civic knowledge—the civic muscle they have. Before they turn 18, they already feel that they steer the country’s direction. This is why Taiwanese young people do not fall into cynicism, apathy, polarization, or populism.

I’m very interested in the success of Taiwanese society. How do you do that at school? Don’t kids go on Instagram and TikTok?

First of all, Taiwan is one of the rare countries where young people’s use of TikTok did not grow recently. They, by and large, prefer to go to the Fediverse, the interoperable part of the web, so they are not controlled by the TikTok algorithm. For example, Threads.net, the Meta implementation of the Fediverse. There are more active Taiwan Threads users compared to any other country, which is significant because we’re just 23.5 million people.

It shows that we’re the least socially polarized because we face the challenges together. We invest in the civic infrastructure of dialogue so that the young people, once they go to school, are not just individually learning media literacy or critical thinking, receiving information. No. They exercise media competency. They fact-check together. They start petitions together. They measure air quality and water quality together. This is really the antibody to polarization: the ability for collective action and mutual civic care.

I want to ask again because I didn’t understand correctly. You said that young kids in Taiwan are not on TikTok as much as their peers in the rest of the world. They prefer other platforms. Threads and what else?

They prefer interoperable platforms. Interoperability means like a podcast. If we record a podcast and we publish on Spotify, and somebody wants to listen, they don’t need to go to Spotify. They can go to Apple Podcasts or any other podcast app. In fact, many podcasts end with, “Subscribe to our podcast wherever you get your podcasts,” which means it’s interoperable.

People can post on Threads, but they post to the Fediverse, so it’s federated. For example, g0v (gov-zero) has their own forum, g0v Social. They can subscribe to my Threads posts on g0v Social. There are many, many nodes, but they all interoperate with one another.

This media literacy and fact-checking education that you have in schools, did it start at the time that you were digital minister? Was it part of your work?

Before I joined the cabinet as digital minister, I served as a reverse mentor, a young adviser to cabinet ministers. We have a system where each cabinet minister—usually older than 35, with the exception of our sports minister who is 30—must work with people younger than 35 as reverse mentors to learn from the young people. When I was a reverse mentor, I joined the curriculum committee that determined Taiwan’s basic education curriculum.

My contribution to the curriculum was the shift from literacy to competency. The argument at the time was that because of AlphaGo, we already saw that anything routine, anything automatable, will be taken over by robots. If we train a young person when they are seven in those individual skills, they become very competitive. But by the time they are 18, robots take that over. They will hate the education system.

So we asked ourselves, what are the competencies that are human and relational and will never be taken over by robots, and we should pivot education to that. We determined curiosity, collaboration, and civic care. These are the three competencies that the robots cannot take over because they are relational. We changed our education to that. That was my contribution even before I became the minister when I turned 35 in 2016.

Has this made a change regarding the Chinese disinformation campaign? I know that Taiwan is targeted probably more than any other country by these campaigns from China. Is that correct?

That is correct. Also, we’re the least socially polarized—meaning ethnic, religious, urban/rural polarization. So all those attacks actually make us even more resilient, anti-fragile.

Does this have to do with the kind of education that children, and also older people, receive?

Definitely. We involve everyone. This is called “digital resilience for all,” in the strategy of rapid response and prebunking. Not debunking only. Debunking is when an attack comes, and then we say, “Oh, that’s not the case.” Prebunking is anticipating what kind of attack will come and using “humor over rumor”—funny, truthful clarifications so that they become even more viral than the attack. It travels faster than the rumor itself.

This inoculates the public, which is why every age bracket needs to work together on this core democratic process to build trust. For example, all age brackets participate in the vote-counting process. Anyone can use their phone to document the entire paper-ballot-only counting process. Even if we face the same attacks as other democracies about vote rigging and deepfakes, in Taiwan, they’re uniquely quickly dispelled by citizens. We prebunk by asking citizens to have their own recordings. When the conspiracy theories and the vote-rigging accusations come, they just release their recording, and people understand it’s actually entirely transparent.

So vote counting is something that any Taiwanese citizen has access to?

Yes. And it is paper ballot only.

And how is it possible that everyone has access to this process?

You just observe the counting. It’s not only open. We pull one paper out, we show it to all the different angles, and you see people with their cameras or phones from very different directions taking in the entire counting process.

Is this the same level of transparency in all government work? Does it happen in every ministry, for instance?

Of course. We make sure that anything related enjoys this high participatory transparency. We also publish open data, including real-time open data of the government’s infrastructural systems. For example, during the pandemic, every thirty seconds we published a real-time inventory of mask availability in all the pharmacies.

When people see, for example, there’s an unequal distribution—because we designed the system such that each person is the same distance, on average, from the next available mask. But then, opposition party legislators said, “I work with the OpenStreetMap community, the open-source community, and analyzed it. One kilometer in the city is not the same as one kilometer in the rural area. By metro, one kilometer is very easy. But in the rural area, you have to wait for a bus, maybe one hour.” The same distance equality is not true equity.

When they made this interpellation to the Minister of Health and Welfare in the parliament, Minister Chen simply said, “Well, MP, you are a data science expert, and you have the same real-time data as we do. So teach us what to do.” And then she, MP Kao, analyzed it and suggested a better distribution method based on preordering, and we implemented that the next day. The point here is that without radical transparency—imagine if we published this number only every month—the opposition party is simply critiquing. But because we publish every thirty seconds, they become co-creators. It turns the negative energy of a volcanic explosion into a geothermal force for good.

There are other things that you have done which are quite impressive. I was reading about the digital signature and how you were able to fight the scam epidemic on Facebook. Would you elaborate on how it happened and how you succeeded in implementing it and convincing them?

Definitely. The scam issue is proliferating as we speak. Last year, in March, if you scrolled Facebook or YouTube in Taiwan, you would see Jensen Huang, the Nvidia CEO—somebody looking like him—wanting to give you investment advice or some cryptocurrency. If you clicked, that fake Jensen Huang talked to you very convincingly, probably running on an Nvidia GPU. What this means is that individual critical thinking is not enough anymore because the synthetic media is too convincing.

But if you ask people individually, they also say the government should not censor speech. Taiwan has the most internet freedom in Asia. So we cannot just censor people. We sent a text message to 200,000 random numbers around Taiwan from the single trusted number, 111. It comes from the government; people know it comes from the government. We asked people, “What should we do together?” They gave us ideas, and thousands of people volunteered to join an online citizen assembly. We chose 447 people statistically representative of Taiwan demographics—a mini-public. In 45 rooms of 10, AI systems facilitated the conversation.

Again, the extreme does not propagate. You have to convince the other nine people in the room of your idea in order for the room to pollinate that idea out to the plenary.

For example, one room said, “Let’s display all advertisements on social media as ‘probably scam’ until somebody digitally signs it, saying why they signed this message, and then you take that label out.” A good idea.

Another said, “For unsigned messages, if I had Facebook pushed to my eyes, and I lost 7 million dollars to an investment scam, Facebook should be liable for that 7 million dollars.” A very good idea.

Another said, “TikTok did not have a Taiwan office at the time. What if they ignore our liability? Well, let’s not censor them, but let’s slow down connection to their video. One percent every day they ignore us.” Again, throttling. A very good idea. It’s about reach; it’s not about speech.

All these ideas were woven together. People looked at it, and experts commented on the feasibility. At the end of that day, people voted. More than 85% of people said this package from all these rooms was a very good idea. The other 15% said that they were at least happy that the process was legitimate. They found it acceptable; they could live with it.

By last April, we checked with all the big tech about feasibility. Last May, the Digital Signature Act amendment passed. Last July, the Anti-Fraud Act passed. It’s been eleven months since the law took effect. No deepfake scams anymore — scam ads went down more than 90%, and the problem is pretty much solved. The point here is that what’s proportionate depends on how you ask people. If you only ask them as individuals, they go to the extreme. If you put their conversation in rooms of 10, they’re very proportionate and have very good ideas.

How did you get Facebook to do this? Taiwan is probably the only country where Facebook actually does not have the scams anymore.

Yes. Other countries, like Australia, implemented some of our measures after we passed the law. But they apply not to all advertisements, but only on, for example, investments or financial-related ads, and so forth. Only in Taiwan, it’s all-spectrum—all advertisements.

Why did Facebook not push back? First of all, the legislators know that this is the common knowledge of the people. More than 85% of people think this is a very good idea; the other 15% can live with it. Everybody knows that. So they cannot pretend it is just some extreme policy. This is already known as consensus by the entire country.

This is a cultural effect. Once everybody knows that everybody else also knows that this is the way to go, any legislator that votes against it will be seen as corrupt or pro-fraud. Nobody wants to be like that. Even though at the time, we had three parties in the parliament, none of which had more than 50%. In fact, the Digital Signature Act amendment was fast-tracked without the usual one-month cooldown period because nobody wanted to be seen as obstructing the progress of fighting fraud.

I was reading that very soon, I think by the end of the year, most content on the Internet will be created by artificial intelligence.

You’re talking about slop. Yes.

Yes. So what can we do? How will we be able to tell between truth and fiction, truth and false?

As I mentioned, if a synthetic video looks just like Jensen Huang, nobody can tell the difference. The way you tell is by the digital signature of Jensen Huang. We must focus on provenance on the actor level, not the content level. We also understand that digital signatures can sometimes dox people. You are a journalist, and you work with your sources. Your source might be a whistleblower. They are understandably afraid that if they provide a full name in their digital signature, although they can authenticate to you, it also means they can get doxxed and persecuted if they are in an unfavorable situation in terms of power asymmetry.

In Taiwan, we developed what is called meronymity, or partial naming, so that you can make an attestation credential that says, “I prove that I’m over 18 years old,” or “I prove that I’m a Berliner,” or that I graduated from this school, and so on. But they do not reveal anything else. It’s called zero-knowledge so that you can present this credential, and anyone can verify this credential, but in doing so, they don’t learn anything more about you.

This is important because, for human-to-human communication, we want to know that this person is not a bot. But we don’t always want to doxx ourselves just to prove this. I think this is the happy middle ground between full anonymity and slop, and full real name and doxing.

Now that you mentioned verification, I’m wondering—very soon OpenAI will make erotica available to adults. They might mean hardcore pornography, for all I know. They will ask for age verification, but what would stop a 14-year-old from saying, “Yes, I am over 18”? What kind of age verification do you use in Taiwan so that a 14-year-old cannot access a pornographic site?

First of all, in Taiwan, we want to make sure that you can have this kind of credential without “pinging home,” meaning that the government should not learn that you are now presenting your credential. If you are buying alcohol in a grocery store in the paper world, you present your physical ID, which is not optimal because there’s too much information on it, like your name. But at least the vendor does not report back to the police.

Right.

They just check whether you are an adult. But in many countries, if you do the age signaling online, the issuing ministry receives something that shows that somebody like you is trying to prove your age to an alcohol vendor. This is bad because nobody wants to be tracked all the time. There are some jurisdictions that make the citizen transparent to the state, but in Taiwan, we want to make the state transparent to the citizen, not the other way around. When we make the verifiable credential, we take extra care so that when presenting it, it doesn’t even need an internet connection. It’s not pinging back.

To answer your question, if I am 18 years old, and I have in my credential wallet a diploma or a driver’s license, I can just choose one part of it—the fact that I’m over 18—and present that. If you are 14 years old and you don’t have a driver’s license, there’s no way you can forge such a credential.

That’s wonderful. Do you use ChatGPT or Grok in your everyday life?

For the past two years, I trained a local model on my computer. Initially, I worked with Llama, which is the Meta version of an open model. But recently, thanks to ROOST, OpenAI published the open model called the GPT-OSS safeguard. That’s the safeguard model that I was referring to. It runs entirely on my computer. I run it on this laptop. I can give it a policy document, and it can compare any text against that policy and tell me why. In doing so, it runs in airplane mode. It does not need an internet connection. It does not ping home. Sam Altman does not know that I’m using this model.

I’m a proponent of this kind of local AI, edge AI. Just like when I was born in ‘81, everybody was connecting to mainframe computers using a terminal. The mainframe computer operator knew every single keystroke that you were typing into the computer. But then, Taiwan helped the personal computing revolution. You could have a personal computer, PC compatible. You could install your own spreadsheet, your desktop publishing. You could do anything, and IBM knew nothing about what you were doing. This is a very similar moment we’re going through.

Two years ago, I had to pay a lot of money for a top-of-the-line laptop to run this local AI. But this year, it’s very simple. Nobody needs to pay extra. Any new laptop can run it.

Could I also do it? I’m not proficient with coding.

Of course, you can do it. You just download software. If you are on a Mac, you download LM Studio. Once you download it, it will ask you what model you want to download. You say, the GPT-OSS, the open version of GPT, and that’s it. You can start having a conversation with something like GPT, but with no subscription fee. It doesn’t need a network connection, and Sam Altman doesn’t know what you’re typing.

Okay. That’s good. OpenAI. somehow, does not advertise this.

I wonder why.

I have very few minutes now, so I want to ask you a few questions. Before the personal questions: have you found a big community in Europe and in the United States? Are local politicians or activists interested in your vision?

Definitely. I don’t think anyone wants to be put into this hamster wheel, like a human-in-the-loop of AI, exercising a lot, feeling very hooked and addicted, but not going anywhere—not being able to steer the direction of technology. This is profound powerlessness. Nobody likes that feeling. Everybody I met is actually opposed to the overconcentration of power.

The only challenge for them is that they thought you need to make a trade-off: you can have a steerable local version, but it’s inferior compared to the centralized version. Like a local IBM PC at the time, maybe you remember, could only run one application at a time. You had to change the floppy disk.

But our work in ROOST, in Mozilla Foundation, and in Project Liberty Institute, is to show people that actually, because you can tune them, these models work better—better quality than the large, centralized models. This is not well known. This is very surprising to many people. Frankly speaking, this is only true this year. It was not true last year. So this is newsworthy as well.

Okay. So audiences in universities and people in the academic community are very receptive?

Yes. Very receptive to the vision of personal supercomputing.

Okay. Now let me come to the personal questions. You were the first nonbinary minister in the world, right?

In the world. Yes.

In recent years, you have watched how culture wars have broken out in Europe and the United States even more. What do you think? Are you surprised at this as someone coming from Taiwan? Did people talk to you about it?

First of all, I think as humans, our experiences inherently are plural. Instead of saying I identify as this or that, I say that I experienced puberty when I was 14, I experienced puberty again when I was 24. We share maybe some part of our experience, and my pronouns are whatever. It’s very simple.

The weaponization of this issue is often a symptom of the polarization we discussed, because they are fueled by the algorithms that thrive on outrage and division. When we lack the space for genuine listening and dialogue, then you see a very high polarization per minute, high PPM, around these issues. I remember Finland ran a national Polis conversation, and the most divisive statement was the feeling that nothing can be said anymore without offending other people.

Sí.

This is a generational shift regarding language. The solution is not more conflict; it is more bridging. It’s more active listening, recognizing the shared humanity and shared experience across all differences. This social translation, this bridging, is what we need to focus on.

Growing up, you were a gifted child. You were obviously aware of how different you were in that respect. You were four when you started reading literature, I think?

Yes, especially Taoist literature, because my uniqueness was also my heart condition. Any strong emotion—happiness, anger—could cause my heart rate to spike, and then I would faint. I needed to practice qigong energetics, meditation. It was not just philosophical, like reading the scripture. They were survival skills necessary to keep my emotions and heart rate stable until I had surgery at the age of 12.

Is it okay now that you experience strong emotions? Are you safe?

Yes, of course. I even sought to ride a roller coaster just to know what it feels like. But the point here is that Daoism, for me, is a way to have strong experiences, but not be captured by those experiences, not amplified with outrage. I think this is also what gave me the ability to take all the sides—to share some experience with this side and share some other experience with some other side without getting fixated.

I saw the documentary about “The Good Enough Ancestor.” This was really interesting. I saw that you grew up in a house where there was lots of political debate and many Tiananmen dissidents, political refugees from China, right?

Yes. They were very young, like 20. That was when I was in Germany. I stayed there for a year with my father, who covered the Tiananmen movement up until June 4, 1989. He then went to Germany and was there in Berlin on the anniversary of the fall of the Wall with many exiles from Tiananmen at the time.

Okay, so you lived in Berlin for a year?

In Germany, in Saarland, near the French border.

How difficult was it growing up as a gifted child? Did you have a feeling of what you planned to do very early on in your life?

I had three kindergartens, six elementary schools, and one year of middle school before dropping out. Every year was adapting, building new connections. But I don’t feel that I am rootless; I feel that I’m anchored by my curiosity. My parents are both journalists. They nurtured this curiosity and collaboration using the Socratic method. They didn’t impose answers on me, but asked follow-ups—interviewed me—that helped me explore various different perspectives and turn conflict into co-creation. This really helped me navigate these challenges of constant physical moving by focusing on understanding and bridging.

You were exceptional, so this helped. My last question is, I read that you were in Silicon Valley and that you left and returned to Taiwan. Would you tell me about this phase of your life? When and why you went, and when and why you left?

I was in Silicon Valley working on innovation in the so-called open-source movement. When I was in the Valley, that was the year when the term “open source” became prominent; before, it was “free software” or “software freedom.” I became increasingly drawn to the open-source idea of innovation in a way that is democratically governed. This idea is very attractive.

But I returned to Taiwan because Taiwan was arguably a more fertile ground for this kind of innovation for democracy. The Taiwan grassroots movement that I helped start around the turn of the century became the core of the g0v (gov-zero) movement. We didn’t call ourselves g0v at the time in 2002. We called ourselves the Elixus Network, but it’s more or less the same core people. It’s dedicated to creating tools for civic engagement and government transparency.

The unique opportunity in Taiwan was that Taiwan itself was experimenting with democracy. We only had our first presidential vote in ‘96. Applying the technical skill to upgrade democracy—like semiconductor design every other month—is very natural in Taiwan. Whereas the same vision, carried somewhat in the US with Code for America and many civic tech movements, I think encounters more resistance, institutionally speaking. I went back to Taiwan partly because Taiwan was more experimental on the front of democracy.

Taiwanese politicians are—it sounds simplistic—but they’re obviously more open. I cannot imagine European politicians being very open to being radically transparent about what they do.

We learned from the Nordics. We learned from Estonia. We actually also learned from Ukraine when they were doing Prozorro, Euromaidan, and so on. The Swiss digital direct democracy methods, the Better Reykjavik system of Iceland, the Consul and Decidim platforms from Madrid and Barcelona—I can go on. I believe in European values.

The values, absolutely. The politicians, not so much. Okay, that was it. My time is up. I’m really grateful that you took the time to talk with me. I’m using a transcription model, but I bet that your transcript will be more accurate.

I can give you a shared document for the transcript that I make, and we can collaborate on it. I have another meeting, so I will have to do it after this meeting.

No problem. Just so that I can work on my interview. Thank you so much.

Thank you so much. Live long and prosper. Great questions.

You too. Bye.

Bye-bye.