We are delighted that Ambassador Audrey Tang can join our meeting today. Audrey is a world‑renowned figure — as you know, TIME named her one of the 100 most influential people in AI in 2023.

Audrey served in Taiwan’s Cabinet from 2016 to 2024, including as the first Minister of Digital Affairs (2022–2024), and now serves as Taiwan’s cyber ambassador.

A leading voice in digital innovation, open government and civic tech, Audrey has prioritized transparency, inclusion and digital innovation in governance. Today, she continues global engagement on digital democracy, digital diplomacy, AI ethics and more.

My pronouns are whatever — really, whatever works.

We all see that technology, especially AI, is a double‑edged sword: it makes life easier and drives innovation and jobs, but it can also fuel chaos, mis/disinformation, unfair elections, and jeopardize democracy. How do we strike the right balance and build mechanisms to get the best from AI and technology? We’re honored to have Audrey here. I’ll open the floor for conversation. Thank you.

I’ve been looking forward to hearing from you. You’re very concerned about polarization and its effect on democracy—much of it plays out on social media. My understanding is that Taiwan regulates platforms but doesn’t do censorship. You’ve said you regulate them but don’t take content down. How can the negative effects of social media be mitigated?

Definitely. Social media was kind of neutral until, say, 10 years ago. Before that, it was blogging. We followed the people that write. My parents are both journalists, so I follow both of them to pay my filial piety. As long as we followed the same authors, we lived in a shared reality.

About a decade ago came the first large‑scale anti‑social use of AI: the recommendation engine. Major platforms shifted from a following feed to a “For You” feed that maximizes engagement through enragement. Doomscrolling results; it hacks the human brain’s reward system.

The fix is to build bridging systems — pro‑social media that elevate uncommon ground — nuanced, rarely discovered common ground rather than the loudest extremes. The bridging approach we prototyped years ago (e.g., Polis‑style deliberation) inspired what powers X’s Community Notes today and to “Notes” pilots on YouTube; Meta has also tested community note‑style labels attached to trending posts. These surface statements that draw up‑votes from otherwise opposed groups; the original poster can’t remove them. That is bridging.

It is not yet mainstream (main feeds remain largely anti‑social), so we are pushing to make bridging the default. We are working with Bluesky, the Fediverse and others. A key blocker is switching cost: if you move alone from TikTok to another network, you lose your social graph. Studies show an average U.S. undergrad might demand money to switch alone, but would pay to move together with their circle — first movers suffer most, so almost no one moves.

That is why we support social portability in law. Utah’s “Digital Choice Act” (HB 418)—set to take effect July 1, 2026, — requires large platforms to provide data portability and interoperability so users can take their social graph and keep interacting across services. It is the social‑network analogue of number portability in telecom: competition shifts to quality of service, not lock‑in. Taiwan is exploring similar ideas in our Data Innovation and AI Basic frameworks, and the EU’s Digital Markets Act also pushes interoperability/portability. We do notcensor content; we change incentives so the race is to the top, not to the brainstem.

That sounds brilliant, but will pro‑social media really go mainstream? Extremes make money for platforms.

Law helps, and here is a fast case. In early 2024, YouTube and Facebook feeds in Taiwan were awash with deepfake ads — for example, using Jensen Huang to pitch crypto giveaways or stock tips. Scammers outbid SMEs, so platforms profit while public trust suffers.

We texted 100,000 randomly selected residents asking a simple question: How should we protect information integrity together? Thousands volunteered; 450 (random, stratified) joined an AI‑facilitated citizens’ assembly (45 rooms of ~10). Proposals included:

Label every ad “probably a scam” by default unless digitally signed by the endorser.

Make platforms liable for harm from unsigned ads.

If a platform lacks legal representation in Taiwan, require it or apply traffic‑shaping until it complies.

Note: none target content; they target integrity and accountability. Over 85 percent across demographics agreed. The next month we met platforms; by Mayan electronic‑signature measure advanced and by July we passed an Anti‑Fraud Act with enforcement rules for deepfake/scam ads and penalties (including 24‑hour takedown on notice and fines up to NT$25 million for serious violations). Result: those deepfake ads markedly diminished in Taiwan. Bottom‑up deliberation produced rapid, cross‑party regulation.

Why not impose top‑down measures against certain misinformation (e.g., false information about polling places)?

Research shows that in any network with a repost button, polarization emerges structurally. Taking content down often backfires, validating conspiracies and burning bridging potential.

We favor pre‑bunking over debunking — design humor and shared references that defuse polarizing frames before they dominate. Early 2020, we anticipated mask polarization and reframed masks as “protect your face from your own unwashed hands” — with a cute Shiba Inu mascot. As science evolved, people stayed open because the frame was apolitical and shared — humor over rumor.

Recent examples of PRC use of social media to polarize Taiwan?

There is a repeating meta‑narrative: “Democracy leads to chaos and never delivers.” Content is reframed to fit it. During Hong Kong’s anti‑extradition protests, a PRC public‑security account posted a Reuters photo with a misleading caption implying protesters were paid to assault police. That’s a cheap fake — real photo, manipulative framing.

Today, we see malicious AI swarms: synthetic broadcasters read LLM‑generated scripts, pushing 50+ variants of the same negative spin, A/B testing what “travels.” They stitch together facts into speculation and innuendo (e.g., family ties) to maximize negative affect while avoiding falsifiable claims. Hard to “take down” because it’s not strictly false — it is engineered negativity.

It also seems big platforms externalize harm. Thoughts on corporate responsibility?

Sometimes itis less than zero — they profit from harm (e.g., scam ads). Our response is to ensure bridging context appears quickly.

A critique of Community Notes is that it is too slow because it relies on human juries. So Taiwan’s g0v community built locally tuned open models that draft first‑pass context within seconds. See Cofacts for a live dashboard of viral memes with immediate, editable context.

Groups can invite chatbots from multiple trusted orgs — e.g., Trend Micro and Gogolook (Whoscall) — to auto‑suggest context in group chats. Apps like Message Checker can locally compare notifications against a crowd‑sourced fact database. This is epistemic security — bottom‑up, privacy‑respecting, and fast.

On “semiconductor diplomacy”—how does it strengthen Taiwan’s position?

Chips from Taiwan manufacturers (e.g., TSMC) power sensitive computations — medical, military, scientific. That conveys trust in Taiwan technology. As cyber ambassador, and previously as digital minister, we worked with the supply chain on cutting‑edge cybersecurity/resilience standards such as SEMI E187. Our assume-breach approach avoids trusting any single vendor or adjacent stack from the same vendor; we audit open, interoperable protocols to limit lateral movement.

We also build resilience (e.g., fallback to microwave and LEO/MEO satellites if subsea cables are cut). All three hyperscalers engage with Taiwan: AWS launched a Taipei region in June 2025; Microsoft announced plans for a local Azure region; Google Cloud already operates the Asia‑East1 (Taiwan) region. As chip production globalizes, we export trusted tech and supply‑chain resilience with it.

How is Taiwan approaching international AI collaboration?

The discourse has polarized between a vertical race (Manhattan‑Project‑style superintelligence) and a horizontal diffusion (open models steered by local cultures). At the Bletchley Summit, these camps clashed.

With Mozilla and others, we launched a middle path in Paris: ROOST — Robust Open Online Safety Tools — to build decentralized defenses with open models that reinforce security. Example: fighting synthetic CSAM. Attackers are decentralized; centralized defense breaks down. So mid‑sized platforms (e.g., Bluesky, Roblox, Discord) collaborate on federated threat detection and incident sharing using privacy‑respecting fingerprints. Attacks strengthen the system — like an immune response.

You grew up around journalism. What’s its role now?

Journalism is not only a profession; it’s a civic muscle anyone can practice — establish shared facts, present the best arguments from multiple sides. In 2020, Taiwan shifted from media literacy (receiving) to media competency (co-production).

Primary students build low‑cost airboxes to measure PM2.5; middle-schoolersfact‑check presidential debates and see their names in national news. Teens can start e‑petitions on join.gov.tw — at 5,000 signatures a ministry must respond, and some become reverse mentors to Cabinet ministers before 18.

Google’s AI Overviews (Gemini) summarize content and may reduce publishers’ page views. Thoughts?

Ad‑funded models have perverse incentives. Subscriptions are incentive‑aligned. Compare engagement‑driven social networks with subscription‑based ones (e.g., LinkedIn): You see far less polarizing content because users can simply unsubscribe, and scammers do not offer long‑lasting jobs.

We are exploring subscription pools (think Apple Arcade / Spotify for journalism): people pay a monthly fee that is redistributed by time spent, remix value, or other civic‑care metrics, with privacy safeguards. The Ministry of Digital Affairs, or moda, is discussing funding pools with major platforms — an alternative to “bargaining codes.”

Copyright feels shaky in the AI era. How do creatives retain control?

I write like Wikipedia: publish first, then invite critique and remix. Our book Plurality was open from outline to release under CC0. Over 60 contributors and 100+ editors participated. It spawned adaptations — a UK young‑adult novel (Any Human Power), a short film (Good Enough Ancestor), a student‑friendly Japanese adaptation, etc. Paradoxically, by relinquishing individual control early, we achieved better community‑level control and a more inclusive narrative — even blending Taiwan’s ROC/independence threads into a shared history.

Like geology: Taiwan’s Yushan rises ~0.5 cm yearly; we turn tectonic conflict into co‑creation energy.

Small islands lack local capacity for “bridging.” What can we do now, given TikTok’s influence?

Beijing bans TikTok domestically for being polarizing while exporting a version that amplifies extremes — like a reverse opium war. The remedy is structural interoperability: adopt radical portability (like Utah’s law). Require big tech to forward the social graph/events to family‑friendly or pro‑social local platforms for your citizens — or apply measured traffic controls until they comply. The tech is open‑source and not rocket science; small nations can adopt these laws and join a ruleset for fair competition.

Many citizens are passive. How do we scale competency?

Literacy alone cannot withstand superhumanly persuasive AI swarms. We must equip people with defensive AI they can steer.

Also, do not assume universal passivity: on join.gov.tw, the most active groups are ~17‑year‑olds and ~70‑year‑olds — they have time and think long‑term.

We are reinventing polling to be generative. In Bowling Green, Kentucky (with Google), an open‑ended process — What Could BG Be? — let people up‑vote/down‑vote ideas (no retweet button). Over 80 percent of ideas converged on uncommon ground that made most people happier and few unhappy. This is expanding via the Napolitan Institute toward 400+ U.S. Congressional districts — turning NIMBY/YIMBY into “MIMBY” (“maybe in my backyard, if…”).

Taiwan’s digital presence is strong despite geopolitics. Any cyber perspective on recognition?

In cyberspace, .tw is fully recognized and resolved here, not in Beijing. Online there is no geographic proximity, only proximity of values (a lesson from Estonia’s former president Toomas Hendrik Ilves). Countries that do not seek to dominate the Internet prefer co‑regulation with people — the Taiwan model of digital democracy. We have practiced radical openness for a decade.

Attackers seem to have the advantage. How do we protect citizens’ data (e.g., after major breaches like Indonesia’s)?

In the short term, offense dominates: AI finds vulnerabilities faster than defenders can re‑architect. So we assume breach and design zero trust: detect quickly, isolate laterally and deploy autonomous guardians that respond at machine speed. We share threat indicators with democratic partners.

On data protection, collect less. If a forum needs proof you’re over 16, do not ask for birthdate or name. We are rolling out selective disclosure with Taiwan’s Digital Identity Wallet, so people can prove attributes (age, residency) without leaking extra data — built on W3C Verifiable Credentials/SD‑JWT; cross‑border interoperability is planned. Privacy and convenience are not a trade‑off.

Will digital residency blur borders?

I am a Palau digital resident and a Lithuanian e‑resident. You don’t get voting or property rights, but you gain access to digital public infrastructure (identity, payments, markets). Digital resources are anti‑rivalrous — like language, they become more useful as more people use them.

Taiwan’s Gold Card invites contributors to the global commons (open source, Wikipedia, blogging, etc.). After five years of contribution in Taiwan, naturalization is possible without renouncing your original passport. It is a new post-Westphalian model.

Is Taiwan’s digital presence threatened by PRC lobbying in standards bodies?

Not in my experience. At the UN Internet Governance Forum (2017, Geneva), I spoke on the record in the U.N. building via a telepresence robot — the chair said, “We’re discussing policies, not politics.” Since then I have joined many U.N.‑affiliated meetings remotely.

Back to pro‑social media: how are such platforms actually developed and funded?

Treat them like public parks — public squares without ad dependence. Taiwan’s PTT (our Reddit) has run as a National Taiwan University student club for decades: open source, stakeholder‑governed, no ads.

The g0v ecosystem (including Cofacts) is incubated by the National Academy, which is politically neutral. When we created the moda, we joined the Transportation Committee in parliament — because we see ourselves as infrastructure builders enabling civil‑society media to flourish.

How capable is the PRC in the digital realm?

Internally, very. They spend more on domestic “harmony” than on the military; polarization and bottom‑up mobilization are stifled. They export the Great Firewall model to Belt‑and‑Road partners. We don’t compete on censorship; we deliver — better pandemic response, better anti‑deepfake systems, better growth. Their blind spot is that censorship also buries true public‑interest journalism (e.g., Dr. Li Wenliang).

With the shifting world order, can Taiwan’s chip leadership and digital innovation re‑set how problems are solved?

Yes. Taiwan hit peak polarization around 2014–2018, and then built participation infrastructure that helped us become one of the least socially polarized societies among peer democracies by some measures. These are battle‑tested democratic innovations — now ready to share as pressure‑made diamonds.

Your view on American democracy?

Research suggests the U.S. has hit peak polarization — you cannot get more partisan‑polarized. A Facebook‑funded study once claimed more Facebook exposure no longer increases polarization; another showed it is because polarization was already maxed out.

Now, there is nowhere to go but up. I’m working across the aisle — Engaged California with Gov. Gavin Newsom (broad listening for wildfire recovery and government efficiency) about pro‑social media. Both left‑ and right‑leaning institutes are reinventing polling to depolarize it.

Could things still get worse—e.g., dismantling courts or the public service?

That moves beyond democracy into neo‑reactionary ideas (e.g., Curtis Yarvin’s CEO‑state/monarchy). In 2016, as Taiwan’s “OG DOGE” open‑government team, we pushed government efficiency — but by trusting bureaucrats: pre‑announce moves, invite 60‑day public comment, let career public servants co‑create solutions. The result: high approval for former President Tsai Ing-wen and a mobilized civil service during the pandemic.

The difference from “DOGE‑style” corporate reorgs is that we distribute the steering wheel to the frontline, rather than centralize power at the top. With AI taking routine tasks, this distributed model matters even more.

Have you faced personal threats?

Yes. The day I joined the Cabinet in 2016, Gmail warned me I was under a state‑sponsored cyberattack. Taiwan’s government faces about 2 million intrusion attempts daily, so we adopted zero trust and share indicators internationally.

There have been physical threats too. After I said at the Summit for Democracy that “if authoritarian regimes want an invitation, they should democratize,” I received ghost money by mail from Hong Kong (a veiled threat). Soon after, someone with a shotgun gained entry to the moda and blasted the front door before surrendering. He claimed he had bought the gun online; attribution remains unclear, though the timing suggested a warning.

Could you describe Taiwan’s Presidential Hackathon?

It is our pipeline from local experiments to national infrastructure. Each year ~200 projects from towns and agencies propose solutions (tele‑health, air/noise monitoring, zero‑knowledge credentials, zero‑trust networking, etc.). The public uses quadratic voting: everyone gets 99 credits; 1 vote costs 1, 2 costs 4, 3 costs 9, 4 costs 16, etc.—so you seek synergies across projects.

We select the top 20 by votes and analyze their synergy graph to standardize interoperability. Five winners receive a trophy from Taiwan’s president (a micro‑projector that — meta! — projects the President awarding it). Each winning team must include at least one public servant, one civil‑society member and one private‑sector partner. The trophy carries a promise: what works locally will be scaled nationally the following year. Tokyo has adapted the model for a Governor’s Cup Open Data Hackathon.

Leaders often distrust bottom‑up processes as slow or messy. How do you overcome that?

Make deliberation as fast, fair and fun as a poll. We used to need dozens of facilitators and days of synthesis; now language models produce “group selfies” of consensus in seconds.

When we explained to Gov. Newsom that our method mathematically guarantees a coherent package that leaves most people happier and almost none very unhappy, he said he is ready to precommit to implementing feasible outcomes, sight unseen. Pre‑commitment plus the urgency of polarization is the crucible for democratic renewal. Start on a small, urgent issue; if it works, institutionalize it.

Politicians fear pre‑commitment—it may reduce maneuvering room.

Urgency helps. A decade ago former President Ma Ying-jeou’s approval was 9 percent; any process that delivered more trust was worth trying. If you are already at 70 percent , you maintain systems but hesitate to renovate. I am discussing these methods this month at the Victoria Forum with Canadian practitioners; Canada is less polarized, but the need for innovation remains.

How would this work somewhere like India, the world’s largest democracy?

We are optimistic. The same broad‑listening infrastructure scales. The key is to start with an urgent, shared topic and pre‑commit to feasible outcomes.

Thank you, Ambassador Tang, and thanks to everyone for the thoughtful questions.

As a practitioner of Taoist thoughts, I would like to leave you with one core message: To give no trust is to get no trust. Democracies need to move from vertical control to horizontal co‑creation, strengthening civic care.

Thank you all for your time. Live long and prosper!