No worries — we have this place until five.

But everyone is on LinkedIn.

Taiwan is rare: we do not ban TikTok outright. We ban it in the public sector (harmful cyberproducts) and in basic education environments. But if you are 18 and using your own device and network, you are an adult. Interestingly, TikTok installs in Taiwan have been gradually declining, and we ...

Europe’s <em> Digital Markets Act </em> already mandates interoperability for messaging; they are consulting on extending it to social networks. In Taiwan our AI Basic Act just passed first reading with similar provisions, and I am proposing a Data Innovation Act patterned on Utah.

The State of Utah is paving new ground as a family‑loving state: it passed a <em> Digital Choices Act </em> . Starting next July, if you are a Utah resident and you move off TikTok to, say, Bluesky, TikTok will be legally required by Utah to keep forwarding new likes, ...

It only becomes a problem when an open society allows a monopoly that becomes impossible to exit — then the monopolist sets the rules. TikTok is a prime example — the only large‑scale social network in Taiwan that is also a cyberarm product. Research showed that if you ask an ...

It is not a trade‑off if the system is <em> anti‑fragile </em> : the more “virus” you get, the stronger your immune system — people exercise <em> civic care </em> and grow a <em> six‑pack of care </em> .

We did not require much from platforms beyond no paid amplification of political/social content by non‑Taiwanese — treat it like lobbying. Taiwan was the first jurisdiction, I believe, where Facebook’s Civic Integrity team implemented strong KYC in election seasons against foreign‑sponsored content.

In 2021, we published a scoreboard of preferences for four vaccine brands by age and district. Some elderly people did not like AstraZeneca (news of blood clots in Europe) and preferred Moderna. Some preferred BNT (Pfizer‑BioNTech, secured via TSMC); younger people preferred Medigen (homebrew), others distrusted it. The scoreboard turned ...

Using <em> uncommon ground </em> , we found both sides were okay with the message: wear a mask to remind yourself to wash your hands — because a mask keeps your hands away from your mouth. Nobody disputed that. So we pushed a cute Shiba Inu meme: <em> “Keep your ...

Early in the pandemic, we saw polarization: some believed only N95 masks work; others said it is about ventilation and masks <em> harm </em> . Science was not conclusive, and the virus kept mutating. Taking content down can backfire when the science shifts.

Finding de‑polarizing content and uncommon ground is step one. Step two is making it more viral than polarization — <em> more popular than the populists </em> . Humor helps.

Social media’s <em> engagement‑through‑enragement </em> machine prioritizes those voices.

In Taiwan the people understand I am at a Lagrange point between social movements and government — working with government, with the people, not for them. I resist capture by either orbit. My focus is on civic‑care capacity. I do not have a platform other than the platform. I do ...

I still am an anarchist. <em> Anarchism </em> is counter‑coercion, not anti‑governance. If you have a good <em> multi‑stakeholder </em> institution — the way the Internet is governed — Kropotkin would be happy. I am against top‑down, coercive takedowns, shutdowns, lockdowns.

When I joined the Cabinet in 2016 we were like the OG “DOGE” — radical transparency, publishing scoreboards, changing procurement rules to cut unnecessary spending. But we worked with the bureaucracy, not against it. We announced every move and waited 60 days before regulatory changes; most ideas were crowdsourced from ...

We also close the loop quickly. Citizen initiatives are guaranteed a response within 60 days; during the pandemic we sometimes responded in 24 hours. The faster you close the loop, the more trust you build — a flywheel. Once your petition is realized in 60 days, you are more likely ...

Not at all. It is no coincidence we have among the lowest affective polarization scores in OECD‑equivalent societies — across urban–rural, ethnicity, gender. We literally leave no one behind. In rural communities, outlying islands, high mountains — broadband is flat‑rate, all‑you‑can‑eat for about 14 euros per month. For my 92‑year‑old ...

We send text invitations to about 200,000 random numbers. People volunteer; then we use stratified random sampling — exactly how a rigorous pollster would.

By the end of the day — although at that time no party had more than 50 percent of seats — we were able to show a comprehensive slate to the Legislature. They passed it without a cooldown period: the Digital Signature Act in May (after a March deliberation) and ...

So, even mid‑conversation, we prepare a slate that is mutually comprehensible and reinforcing — think <em> Overton window </em> . We read it back to participants and experts: e.g., “Most of you want all political ads to carry a digital signature and advertiser KYC; most of you also want platform ...

Those 45 rooms produce transcripts concurrently; we run <em> sense‑making </em> in the background with language models. By the time they are done with a topic, we can calculate the proportional veto core or surface the uncommon ground — packages that make <em> everybody </em> slightly happy and <em> nobody ...

The software we use is from James Fishkin’s lab at Stanford’s Deliberative Democracy Lab. It is a <em> deliberative polling </em> system. Functionally, it is like Zoom with breakouts of 10 people each (say, 45 rooms at once), but the software <em> facilitates </em> — no human moderator. It reminds ...

Civil society now expects this kind of <em> sense‑making </em> . The U.S. is adopting it too: Bowling Green, Kentucky’s <em> What Could Bowling Green Be? </em> initiative uses the same code (Google Jigsaw sense‑making + Polis) for bridging conversations. Work is underway to roll out similar <em> bridge‑making </em> ...

Nowadays, with language models and AI, this is no longer a problem. They are now more accurate than humans — not just at <em> transcription </em> , but also <em> summarization </em> and <em> translation </em> . In <em> digital deliberations </em> , people expect real‑time captions and transcripts — ...

Yes — quite a lot. Initially it was just me and my office, because it was costly to transcribe in real time. Someone would record video; a stenographer — a court reporter — would sit next to us, typing everything. We adopted that in Cabinet meetings and other important meetings, ...

The idea of <em> humor over rumor </em> eventually dominated public conversation in Taiwan, so polarization attacks became <em> anti‑fragile </em> : we grew more resilient the more we received those external attacks.

If there is only one thing I would point to, it is the 2019 education curriculum reform, where we shifted from <em> media literacy </em> (consuming information) to <em> media competency </em> (co‑producing information). Primary schoolers learned to measure <em> air quality </em> and create shared knowledge about pollution sources. ...

This co‑creation — this “ <em> uncommon ground*” that discovers common ground across ideological camps — needs to be *more viral </em> than polarization. We have to be more popular than the populists. <em> Radical transparency </em> helped investigative journalists, civic teachers, and others build <em> civic care </em> into ...

Even this — <em> radical transparency </em> — was very important early on. When I joined the Cabinet it was right after the <em> Sunflower Movement </em> , in which President Ma’s approval rate was at 9 percent, mostly because people had lost trust in <em> vertical </em> institutions. For ...

Shall we record this?

Late last month, a large number of Taiwanese became the latest victims of AI-driven scams. According to media reports, this started when groups tagged thousands of Facebook pages in their posts. For over 3,000 of these page owners, the moment they tried to remove the tags, Facebook's algorithm automatically flagged ...

Thank you—very happy to be here. Live long and prosper.

Globally, more communities are pivoting from antisocial to pro-social algorithms, building bridges and rejecting extreme polarization. People are ready to move on.

Taiwan went from high polarization and low trust 10 years ago to one of the least polarized societies across many dimensions. We treat external pressures like vibe attacks as challenges that strengthen our democratic resilience.

At the Paris AI Summit, Mozilla, Project Liberty, the Eric Schmidt Foundation, and others launched <em> Roost </em> (Robust Open Online Safety Tools). It’s open-source foundational work enabling smaller operators to detect and monitor online harms, including CSAM and deep fake intimate images—threats amplified by modern AI capabilities. As harmful ...

The AIAGE connects election officials, policymakers, and AI experts worldwide to share best practices—like meonymity and provenance—without resorting to surveillance states or falling into AI swarm chaos.

Interference is shifting from disinformation (fake content) to malicious AI swarms (fake people). These accounts post genuine-looking content and engage organically, conducting “vibe attacks” to subtly shift public perception.

Yes. In Taiwan, starting this year, advertisers on Facebook and YouTube must pass KYC (Know Your Customer) checks. Ads without meonymous digital signatures disclosing origin are fined under the crowdsourced Anti-Fraud Act, which involved 450 randomly selected citizens.

Pure anonymity invites malicious AI swarms; pure real-name policies deter vulnerable voices. The middle ground is <em> meonymity </em> —partial identity disclosure. For example, someone can prove they’re a California resident without revealing more. Platforms can set topic-based thresholds, ensuring real deliberation without storing personal data.

In antisocial media, algorithms prioritize addictive engagement, often amplifying extreme voices. Pro-social media reverses this: we give virality to bridging ideas, not extremes. Community Notes and polls function this way—rewarding uncommon ground and dampening hostile takes.

A real-world example is Bowling Green, Kentucky, where around 10% of the population engaged in a pro-social media experiment. Rather than extreme voices dominating—as in traditional town halls or antisocial corners of the internet—bridging ideas went viral, producing nuanced statements. Reports show over 70% of statements were highly agreed upon, ...

Extending that to social media, there are initiatives like Community Notes on X.com, YouTube, Facebook, and others, where people can add missing context to viral posts. Only notes that receive bipartisan upvotes get shown—encouraging bridging rather than polarizing content.

The idea of pro-social media is inspired by platforms like Wikipedia. On Wikipedia, you see the main article (the common ground) and the talk page (where ongoing debates are documented fairly). This bridging and balancing of content helps depolarize conversation.

For example, <em> Roost </em> is one of the projects I help advise. It makes it easy for anyone hosting any content site to scan for child sexual abuse materials, non-consensual intimate images, and other harmful content in a way that preserves privacy and avoids reliance on a central provider. ...

It includes not just Taiwanese Mandarin or Taiwanese Holo, but also Taiwanese indigenous languages. Through this participatory input, Mozilla aims to prove that the internet becomes safer as it becomes more open. Just as in Taiwan we say we’re making the state transparent to the people—not the people transparent to ...

Mozilla has been around for more than two decades now. For many, it’s the steward of the Firefox browser. For people working in AI, it hosts <em> Common Voice </em> , a dataset that includes more than 200 languages. Anyone speaking any language can contribute their voice to help AI ...

Very happy to be here.

Thank you. Let’s free the future together.