Cheonan, South Chungcheong Province. On a Friday night in October 2017, a man named Jeong-ung sat at his kitchen table reading a hundred-page policy briefing packet.
He was a corn farmer. He knew nothing about nuclear energy — could not tell fission from fusion, had no prior opinion on spent fuel disposal. Yet spread across his table were documents arguing for and against the resumption of construction on the Shin-Kori Nuclear Reactors No. 5 and 6 (신고리 5·6호기). The economic analysis submitted by the pro-construction side and the safety assessment submitted by the anti-construction side sat side by side, equal in thickness — the Public Deliberation Committee (공론화위원회) had guaranteed both sides identical page counts.
When someone asked him why he was reading all of it, Jeong-ung answered:
"It's a sense of responsibility — I felt I had to make my voice count, so I wouldn't fall behind the others."
There were 471 of them. A citizens' jury randomly drawn from across the nation (based on in-person participants). The sample was designed as a microcosm of Korean society, balanced by gender, age, region, and prior stance on nuclear power. They were given three months. Read, listen, question, deliberate, decide. Caught between the president's campaign pledge to phase out nuclear power and a reactor already more than half built, they were asked to render judgment on behalf of fifty million citizens.
The Deliberation Committee surveyed participants' views at four intervals. In the first survey (immediately after orientation), 36.6 percent supported resuming construction. As participants studied the briefing materials, attended expert lectures, and discussed in small groups, opinions shifted gradually through the second and third surveys. In the fourth and final survey, conducted after three months of deliberation and a comprehensive plenary debate, support for resuming construction had risen to 59.5 percent — a swing of 23.5 percentage points.1 The implication was clear: when people are equipped with information, opinions change. And opinions that people arrive at on their own terms are opinions they can live with.
Kim Ji-hyung (김지형), a former Supreme Court justice who chaired the Deliberation Committee, described the process with a single analogy:
"Think of it like a doctor presenting several treatment options for cancer. The doctor explains the choices, but the patient is the one who decides."
Experts provide information; citizens decide. The doctor-and-patient model. This was slow justice in its most elegant form — a corn farmer gradually becoming literate in nuclear policy. It was not fast. It took three months. But the decision those three months produced withstood every subsequent political attack. Because 471 citizens had made it with adequate information in hand.
1. A Corn Farmer Studies Nuclear Power
As one participant put it, contrasting public hearings with public deliberation: "A public hearing is a lavish feast of one-sided arguments where like-minded people gather to talk past each other. Deliberation, on the other hand, is a process of finding solutions that anyone can accept as reasonable."
The Shin-Kori 5 & 6 deliberation was South Korea's first full-scale experiment in deliberative democracy. In July 2017, the Moon Jae-in administration faced a political deadlock between its campaign promise to phase out nuclear power and a reactor already under construction. An unprecedented decision was made: ask the citizens.
Five hundred were invited; 471 accepted. At a two-day plenary forum, pro-nuclear and anti-nuclear sides each presented for twenty minutes. Small-group discussions followed. Participants wrote questions on whiteboards, and expert panels answered them. One participant, a woman in her sixties, left this testimony:
"At first I thought we should stop construction, no question. But after hearing the explanations, I understood that more than half was already built and that halting it would cost even more. I simply hadn't known."
It was the moment information asymmetry dissolved. The shift from 36.6 percent support in the first survey to 59.5 percent in the fourth was not the product of propaganda. It followed a process in which both sides received equal briefing materials, heard expert presentations, discussed in small groups, and submitted questions on whiteboards. In Chapter 1, we saw Rome's smallholding farmers, unable to make their voices heard in the Senate — a structure in which the reality of harm never reached the institutions. The Shin-Kori deliberation inverted that structure. It returned information to the citizens. When citizens received information, ideology retreated and judgment advanced.
Yet the experiment drew criticism as well. Could 471 people truly represent fifty million? Had the president used citizens to break a political impasse he could not resolve through conventional channels? How does deliberative democracy, which bypasses the legislature, coexist with the principles of representative democracy? [See: Hélène Landemore, Open Democracy, Princeton University Press, 2020 — on the tension between deliberative and representative democracy]
These are important questions. But one fact precedes every critique: for those three months, a farmer who harvested corn stayed up late reading policy documents. No one told him to. He did it out of a sense of responsibility.
2. Audrey Tang and the Three Principles of Digital Democracy
Six thousand five hundred kilometers away, in Taipei, one person was working on a different version of the same question: can consensus be engineered?
Audrey Tang (唐鳳). Born in Taipei in 1981. Her father was a journalist and columnist at the China Times. She taught herself programming online at fourteen, dropped out of school at fifteen, co-founded a startup in Silicon Valley. At thirty-three she changed her legal gender to female. At thirty-five she became a Minister without Portfolio in the Tsai Ing-wen administration — the youngest cabinet member in Taiwanese history and the world's first openly non-binary government minister. In 2022, she was appointed the inaugural Minister of the newly established Ministry of Digital Affairs (數位發展部), a post she held until 2024.
Tang's philosophy of digital governance can be distilled to three phrases: radical transparency, civic participation, and rough consensus.
"That's radical transparency, civic participation, and rough consensus. And surprise, it's working and it's transforming our society."3
"It's working and it's transforming our society." This was not bravado. There was evidence.
Every one of Tang's official meetings was recorded and made public within forty-eight hours. She refused to meet with any stakeholder who requested confidentiality — a rule she never broke after becoming minister. The accumulated public transcripts run to thousands of pages. Any citizen can see who the minister met and what was discussed.
"Power comes from secrecy. Eliminate the secrecy and you eliminate the power."
This was what Tang meant by radical transparency. It stood at the opposite end of the spectrum from the world of Brooksley Born that we saw in Chapter 3 — a $27 trillion derivatives market labeled a "dark market," left in a regulatory blind spot. The structure in which those who monopolize information hold power — information asymmetry — Tang declared she would dismantle it with technology.
While most politicians focused on defending democracy from the worst of AI, Tang was working in the opposite direction — using AI to strengthen democracy. In 2023, Tang partnered with the Collective Intelligence Project to launch "Alignment Assemblies." The structure allowed ordinary citizens to voice opinions on the risks and direction of AI development, and those opinions were fed into the strategies of AI companies. It was a reversal: not AI judging citizens, but citizens judging AI.2
And then there was Tang's most famous declaration:
"When we see the Internet of Things, let's make it an Internet of Thee. When we see virtual reality, let's make it a shared reality. When we see machine learning, let's make it collaborative learning. When we see user experience, let's make it about human experience. And whenever a singularity is near, let us always remember the plurality is here."3
It was a new framing for the age of AI. In Chapter 6, when Sam Altman warned a Senate hearing that "if this technology goes wrong, it can go quite wrong," his proposed solution was a government licensing regime — control from the top down. Tang's solution was the reverse: participation from the bottom up. And the technology to make that participation possible.
"Democracy is a technology. It gets better when more people try to improve it."
"The superintelligence we are looking for is already here. It is us."4
3. vTaiwan and Uber — A Machine for Building Consensus
In March 2014, Taiwanese students occupied the Legislative Yuan (the national parliament). The Sunflower Movement. Fury had erupted over a cross-strait trade-in-services agreement with China being pushed through without procedural legitimacy. One coordinator recalled the moment:
"People started to feel it — that if we really come together, we have the power to change the government."
In the space the Sunflower Movement left behind, vTaiwan was born — a digital deliberation platform created by civic-tech activists. At its heart was a piece of software called Pol.is.
The innovation of Pol.is was simple. It removed the reply button.
"We discovered that when you take away the reply button, the trolls disappear. Trolls thrive on reply buttons — because they can attack people, not statements."
Traditional online debate follows a predictable spiral. A posts; B rebuts; C attacks B. Division is confirmed and amplified. Pol.is worked differently. A participant wrote a short statement, and others could only agree, disagree, or pass. No replies. An AI clustered participants by opinion similarity and visualized the resulting terrain.
The results were striking. Instead of highlighting the most divisive statements, the system surfaced statements on which nearly every cluster agreed. Hidden spaces of consensus emerged.
"With Pol.is, you can clearly see that most people agree with most of their neighbors on most things."
In 2015, the tool was put to a real test. Taiwan's Uber controversy had exploded. The taxi industry called Uber illegal; riders wanted the convenient service. Under traditional democracy, the dispute could have dragged through years of legislative hearings, lobbying, and protests with no resolution — just as the EU AI Act in Chapter 4 took three years to reach a thirty-six-hour marathon negotiation.
vTaiwan opened a different path. Pol.is collected opinions at scale. Nine hundred and twenty-five people voted on statements in the Pol.is phase of the UberX discussion (online participants), and over four thousand joined a separate agenda-crowdsourcing stage (online participants). The UberX discussion alone produced 31,115 votes and 145 submitted statements.5
When Pol.is visualized the opinion landscape, participants split into two large clusters — taxi-industry supporters and Uber supporters. But certain statements drew agreement from both: "The government should establish a fair regulatory framework." "Private vehicles should be registered." A binary vote would have produced "legalize Uber vs. ban Uber." Pol.is instead identified seven points of consensus — above all, that both sides wanted a clear regulatory framework.
In May 2016, the Taiwanese government adopted this consensus as official policy. The mandatory yellow color requirement for taxis was abolished, and app-based premium car services were licensed. A decision that could have taken years was reached in months.6
Here the paradox emerges. The cost of consensus. The structure of the Roman Senate we saw in Chapter 1 — where a single tribune's veto could block an entire reform — was a paradox in which the democratic mechanism of checks and balances stifled innovation. The cost of reaching consensus was so high that no one could move. vTaiwan inverted that paradox. By revealing the landscape of agreement before the battlefield of disagreement, it lowered the cost of consensus while raising its quality.
The experiment attracted global attention. When OpenAI launched a grants program seeking ways to "democratize AI," vTaiwan — under the name "Recursive Public," in partnership with Chatham House — was selected as one of ten teams from roughly one thousand applicants across 113 countries. One vTaiwan operator described the new possibility:
"After generative AI went viral, we wanted to see if we could use AI for the time-consuming parts of the work — so that contributors could focus their energy on what matters most: engaging with people, the core of every deliberative process."7
Technology not replacing deliberation, but becoming a tool that lowers its cost. It was the concrete realization of what Tang called "collaborative learning."
4. Estonia and Iceland — Success and Failure
Two contrasting cases mark the landscape of digital democracy.
Estonia. A Baltic nation of 1.3 million. When it gained independence from the Soviet Union in 1991, its administrative systems were virtually nonexistent. The new government led by Prime Minister Mart Laar made a radical choice: instead of repairing the twentieth century, build a digital state from scratch. Three decades later, Estonia had become the world's first fully digital nation. In the 2023 parliamentary elections (Riigikogu), approximately 51 percent of all votes were cast online (officially 50.97%) — the first national election anywhere in which internet voting exceeded half of all ballots cast (by online voter count).8 An AI-driven employment matching algorithm raised job retention rates from 58 to 72 percent, and the annual time savings amounted to an estimated 2 percent of GDP.9 There is an irony: the Soviet legacy is what made Estonia's digital transformation possible. The more there is to discard, the easier it is to start fresh.
On the other side stands Iceland. After the 2008 financial crisis, Iceland decided to crowdsource a new constitution. Nine hundred and fifty randomly selected citizens identified core values at a national forum. Twenty-five constitutional council members then drafted the document in public, using social media. More than 3,600 citizen comments and 360 amendment proposals were submitted. In a 2012 national referendum, 67 percent of voters approved the new draft constitution (turnout 49 percent; approval rate 67 percent — the two figures are separate metrics).10
And then the parliament refused to adopt it.
A document shaped over months of collaborative drafting and 3,600 reviewed comments was never brought to a vote before the parliamentary session expired — blocked on procedural grounds of insufficient quorum. One council member put it bluntly: we built the house, and the parliament padlocked the door.
A constitution approved by 67 percent of citizens had been stopped at the wall of existing institutions. The lesson was clear: the energy of citizen participation alone is not enough. Without institutional linkage to existing structures of governance, every deliberation remains merely advisory.11
vTaiwan did not escape this limitation either.
"vTaiwan was never codified into law. The government has no obligation to use it."
After 2018, vTaiwan was no longer used for major policy decisions. Platform operators called it "difficult to use," and public interest waned. Critics charged that "legislators don't take it seriously."12
Yet the verdict of "failure" is premature. On specific issues — consumer internet regulation, economic conflicts like Uber — vTaiwan succeeded as a proof of concept. The lesson is not "digital deliberation doesn't work" but rather "digital deliberation is sustainable only with institutional embedding."
This is precisely why the Shin-Kori deliberation succeeded. The Deliberation Committee's process was conducted under a prior commitment from the president and the administration to accept its conclusions. The institutional compact that Iceland lacked was present in the two-day forum in Cheonan.
5. When Humor Beats Rumor
Audrey Tang had one more weapon.
In 2020, as the COVID-19 pandemic spread, misinformation about masks was flooding the public. Instead of issuing conventional government rebuttals, Taiwan deployed a Shiba Inu mascot — an adorable dog wearing a mask. The caption was simple: "Masks protect your face from unwashed hands."
The image spread faster than the original misinformation.
"We favor pre-bunking over debunking — humor over rumor."
During the COVID-19 pandemic (2020–2021), the Taiwanese government's average response time to fake news was sixteen minutes.13 The maximum was two hours. After the 2018 elections, every government ministry had established a dedicated unit tasked with rebutting misinformation within two hours, and the pandemic pushed that infrastructure into overdrive.14
Sixteen minutes. Compare this number with the other timelines in this book. Sixty-four years in Chapter 2 — the time it took for British factory legislation to address child labor. Three years in Chapter 4 for the EU AI Act to be born. Years upon years in Chapter 6 during which the United States operated without any federal AI regulation. If the thesis of this book is that institutions are slower than technology, Taiwan offered a counterexample. At least when it came to combating misinformation, the institution outpaced the technology.
Here, two of the five structural patterns running through the entire book intersect.
First, the paradox of the cost of consensus. Deliberation is slow. Four hundred and seventy-one people had to spend three months; vTaiwan's Uber negotiation took several months as well. Yet that slowness actually reduced the cost of consensus. The outcome of the Shin-Kori deliberation was never politically challenged — because it had been forged through sufficient deliberation. Had the decision been made quickly through a binary vote, the losing side would have resisted. The longer it took to reach agreement, the lower the cost of conflict afterward. Slowness, in the end, was speed.
Second, the resolution of information asymmetry. Rome's smallholding farmers in Chapter 1 had no information and no channel through which to be heard. Jeong-ung received a hundred-page briefing packet. A woman in her sixties from South Gyeongsang Province realized "what I hadn't known." Pol.is revealed what people actually agreed on — the terrain of consensus, rather than the polarization amplified by media. Returning information to citizens: that was the essence of digital deliberation.
But digital deliberation faces threats. Structural threats.
First, bots and opinion manipulation. In the 2017 U.S. FCC net neutrality proceeding, a single industry-funded campaign submitted more than 8.5 million comments using stolen identities. In total, roughly 80 percent of the approximately 22 million comments — more than 18 million — were found to use fabricated identities.15 Genuine citizen opinions were buried under automated noise. In the 2016 U.S. presidential election, the ratio of pro-Trump bots to pro-Clinton bots was approximately five to one, with many attributed to Russian operations.16 Digital platforms are vulnerable to organized manipulation that physical town halls are not. Identification, evidence, attribution, enforcement — none of these four challenges has been solved.
Second, the digital divide. Low-income populations, the elderly, rural residents, and people with disabilities — those most exposed to algorithmic harm — are also the first to be excluded from digital participation. This is not a temporary gap that will close on its own; it is a structural barrier bound to socioeconomic stratification.17 When educated, urban, digitally fluent participants are overrepresented, the result is policy consensus in which the interests of the most affected groups are systematically undervalued.
Third, the paradox of scalability. Taiwan's success was possible in part because it was Taiwan — internet penetration of roughly 93 percent, a small nation of 23 million, recent democratic transition, government commitment to digital infrastructure. The conditions that make deliberation deep (small groups, sustained engagement, professional facilitation) and the conditions that make it democratic (large numbers, representative participation, low barriers to entry) exist in structural tension.18
Digital deliberation is not impossible. But what is needed is not design for participation alone — it is design against attack: adversarial design. Identity verification, bot detection, deliberative rather than polling formats, human facilitation. And unless equity is built in intentionally — digital-plus-offline hybrids, asynchronous-plus-synchronous formats, structures that actively recruit marginalized communities — digital deliberation risks becoming consensus among those who already have the loudest voices.
6. South Korea's Choice
In December 2024, South Korea's National Assembly (국회) passed the Artificial Intelligence Basic Act (인공지능 기본법) by a vote of 260 to 0. An overwhelming consensus. But why had this law not gone through a citizen deliberation process the way Shin-Kori did?
One analyst's observation was cutting: "Both ruling and opposition parties aligned their positions in the direction of AI technology companies' interests. Not a single lawmaker fought to protect citizens' safety and rights from AI risks."
Meanwhile, where were the people who would be directly affected by the law?
Mr. Park, whom we met in Chapter 8, was still grappling with the same problem in 2024. He still could not learn why the AI credit-scoring system had denied his loan. The "black-box rejection" was never placed on the agenda as a central issue during the AI Basic Act deliberations. The voices of 5.5 million self-employed business owners like Mr. Park were absent from the 260-to-0 consensus.
Driver Lee, whom we met in Chapter 9, watched the algorithm cut off his ride dispatches at two in the morning that winter as well. The AI Basic Act contained no provisions addressing platform-labor algorithms. "Eight hundred and eighty thousand platform workers whose dispatches and ratings are determined by algorithms" — this figure never appeared during legislative review. Just as Jeong-ung the corn farmer had sat at his kitchen table reading a hundred pages before the Shin-Kori deliberation, Driver Lee too had a right to participate in that decision. That right was never granted.
South Korea already possesses the infrastructure for digital democracy. Since its introduction in 2020, the National Petition System (국민동의청원) has received 2,106 petitions, of which roughly 0.4 percent resulted in actual legislation.19 A K-voting system operates for non-public-office elections, a participatory budgeting program exists at the national level, and Seoul's M-Voting system is used for citizen participatory budget votes. The infrastructure is there. But the conversion rate to political influence is dismal. What the 0.4 percent figure symbolizes is a sobering reality: digital participation infrastructure does not automatically translate into political power.
Was the Shin-Kori deliberation a one-time event, or did it become institutionalized? The answer is uncomfortable. It was closer to a one-time event. No deliberative experiment of comparable scale and depth has been repeated since Shin-Kori. During the passage of the AI Basic Act, the voices of Mr. Park and Driver Lee were absent. Seven years separated the Shin-Kori deliberation from the AI Basic Act — and in those seven years, Korean society failed to solve the challenge of institutionalizing deliberative democracy.
Democracy is slow. This book has said so from the beginning. One hundred and thirty years in Chapter 1. Sixty-four years in Chapter 2. Three years in Chapter 4. Slowness was the problem.
But Chapter 11 tells a different story. Slowness is not necessarily weakness. Jeong-ung's three months proved it. The consensus forged by 471 people through slow deliberation was more durable than any decision a quick majority vote could have produced. Tang's Pol.is proved it. When you discover the landscape of agreement before the battlefield of disagreement, slowness itself becomes a tool for faster consensus.
The problem is not slowness itself. The problem is who is granted the right to be slow. Slowness granted to senators was regulatory capture by the entrenched. Slowness granted to citizens was deliberation. The same slowness produces entirely different results.
Tang expressed this in her own words: "Social innovation is the co-creation of social value that happens when people with different views choose to work together." Co-creation. Not something imposed from above, but something that gathers from the sides. That is what Jeong-ung demonstrated, moving back and forth between his cornfield and the policy briefing packet. And the fact that Mr. Park and Driver Lee were excluded from the 260-to-0 vote means that the opportunity for co-creation has not yet been adequately designed.
We have now seen two alternatives. Regulating through technology (Chapter 10's adaptive regulation) and citizen participation (Chapter 11's digital deliberation). Both are responses to "fast order." Both are imperfect. Adaptive regulation keeps pace with technology but lacks democratic legitimacy; citizen deliberation has strong legitimacy but is slow and vulnerable to institutional fragility.
Now it is time to ask history. Was fast order ever good order? The fast order of Augustus that we saw in Chapter 1 — two hundred years of Pax Romana — what price did it ultimately exact? If we can buy time for slow justice, how much is that time worth?
In the next chapter, this book moves toward its conclusion. What does a better speed look like?