vTaiwan: how a civic tech community built an open consultation process for national rules

vTaiwan is an open consultation process, launched in Taiwan in 2014 by volunteers from the civic tech community g0v at a government minister’s request, that brings citizens, officials and stakeholders together online and in person to reach “rough consensus” on national digital regulation.⁠^1,2 Its best known case, on the ride-hailing service UberX, used the opinion-mapping tool Pol.is and preceded regulatory changes that took effect in October 2016.⁠² Its own team and outside observers also document its limits: few topics came from citizens, a ministry could decline to take part, and some outputs, such as a draft bill on online alcohol sales, went nowhere.⁠^2,5

The comparative overview at /research/cases/ covers vTaiwan’s headline numbers and the “tiger without teeth” criticism. This page goes further into its origins, its mechanics, its specific cases, the government petition platform Join, and what the evidence does and does not show.

What happened

vTaiwan grew out of a protest movement and a civic hacking community, then settled into a government-adjacent process whose activity rose and fell with political attention. The dated record below is verified for this page.

• March 18 to April 10, 2014. Citizens occupied the Legislative Yuan in Taipei to protest the opacity of the Cross-Strait Service Trade Agreement, which aimed to liberalize trade in services between Taiwan and China. This was the Sunflower Movement.⁠² MIT Technology Review reported that g0v “had played a leading role in the Sunflower protests”.⁠⁵
• December 2014. Jaclyn Tsai, then Minister without Portfolio of the Executive Yuan, attended a g0v hackathon and asked contributors to build a platform for “rational discussion” of public matters at national scale, which started vTaiwan.⁠² (A later paper by the Pol.is team describes the request as coming from Taiwan’s Digital Minister;⁠³ the earlier account by vTaiwan contributors names Tsai.⁠²)
• 2015. The government launched Join, a separate participation platform.⁠^2,4
• July 15 to August 15, 2015. The UberX case, vTaiwan’s 12th topic and its first to use Pol.is, ran its opinion survey.⁠² The Computational Democracy Project describes this as “the first test of vTaiwan” with Pol.is, in August 2015.⁠⁶
• May 23, 2016. The administration pledged to ratify the Pol.is consensus and amend the Regulation on Automobile Transportation Management; the amendments took effect on October 25, 2016.⁠²
• 2016. After the Tsai Ing-wen administration took office, it withdrew all bills awaiting legislative approval, and the online alcohol sales bill drafted from vTaiwan suggestions “never saw the light of day,” according to MIT Technology Review.⁠⁵
• 2017. Each ministry became obliged to appoint a Participation Officer.⁠⁴ From January 2017 the central government began discussing how to institutionalize a rulemaking process inspired by vTaiwan.⁠²
• January 31, 2018. Taiwan promulgated the Financial Technology Development and Innovative Experimentation Act, a regulatory sandbox law.⁠⁹ The GovLab case study lists a fintech sandbox consultation among vTaiwan cases.⁠⁴
• August 2018. MIT Technology Review reported that vTaiwan’s website claimed 26 cases with 80% leading to “decisive government action,” and quoted a ruling-party legislator saying it had come “close to death” at times.⁠⁵
• 2019. A digital communications bill that would have given the process a legal framework, itself a product of vTaiwan, did not complete its passage through parliament.⁠⁴
• December 2023. vTaiwan’s information page, last updated then, reports more than 28 cases and says it planned to address AI-related regulation in 2024.⁠¹

How it worked

vTaiwan’s process, as described by its contributors in a 2018 paper, had four stages and three working principles.⁠² That paper was written by people who ran the process, including Audrey Tang, so it is a practitioner account rather than an independent evaluation.

The four stages

• Proposal. Weekly “mini hackathons” welcomed anyone. Contributors submitted an issue to a competent government authority, which could accept or refuse it. “An issue will not move into the vTaiwan process without a government authority being accountable for the issue and a facilitator taking charge of the issue”.⁠²
• Opinion. Online opinion was collected with Discourse, Pol.is, Typeform and Sli.do. On Discourse, competent authorities were required to reply to stakeholders’ comments “within seven days.” Each round of opinion collection lasted at least one month.⁠²
• Reflection. A facilitator hosted a livestreamed in-person meeting with scholars, officials, businesses and participants, relaying online questions into the room.⁠²
• Legislation. The consensus was presented as a guideline, a policy, a statement, or a draft bill sent to the Legislative Yuan.⁠²

Transitions between stages were decided by consensus of the vTaiwan community, with no strict rule.⁠² The GovLab case study reports timelines from about three months to more than a year.⁠⁴

Pol.is mechanics

Pol.is is the tool most associated with vTaiwan. Its developers describe the method in a 2021 article in the journal Recerca:⁠³

• Participants vote agree, disagree or pass on short statements, and may add new statements, which are then shown to others. Moderators may screen new statements first.
• Votes form a matrix of participants by statements. Missing votes are filled with statement averages, and participants who voted on fewer than seven statements are dropped from the analysis.
• Principal components analysis reduces the matrix to an “opinion landscape,” and K-means clustering groups participants into opinion groups. The tool then surfaces statements that divide groups and statements most groups agree on.

MIT Technology Review reported Audrey Tang’s explanation of one design choice: “you cannot reply to comments,” which she said “drastically reduces the motivation for trolls to troll”.⁠⁵ GovLab notes that each participant sees statements in a different order “to avoid bias”.⁠⁴ The developers list limitations: the method needs at least dozens of people and tens of comments, works poorly for narrow yes or no questions and for ranking options, and is sensitive to when people arrive, because early comments collect more votes.⁠³

Join, and how it differs

Join is a government platform launched in 2015, separate from vTaiwan.⁠^2,4 MIT Technology Review described it as “a platform for hosting and debating online petitions”,⁠⁵ and the Computational Democracy Project describes it as enabling “petition-fueled agenda setting”.⁠⁶ Join also uses Pol.is.⁠⁴ The difference matters: vTaiwan began with an issue a ministry had agreed to own, then sought rough consensus on its content, while Join lets the public start petitions and debate government policy at much larger scale.⁠^4,5,6 The petition thresholds and response rules for Join could not be verified from an official source for this page.

Measured results

vTaiwan’s results are reported mostly by its own team and partners, and those figures are self-reported.

The UberX case

The UberX case is the best documented. According to the contributors’ paper, 925 participants took the Pol.is survey, entering 145 comments and 31,115 votes; a later public report had slightly higher numbers because the survey stayed open.⁠² Four initial groups (taxi drivers, UberX drivers, UberX passengers and non-passengers) merged into two, with 45% and 55% of participants.⁠² A statement that the government should “challenge the taxi industry to improve their management & quality control system” drew agreement from 95% of participants.⁠² In the reflection stage, 1,875 people followed on livestream.⁠²

The Pol.is developers’ own analysis of the same conversation reports close to 2,000 participants, of whom 1,238 remained after removing those with fewer than seven votes. The first principal component, separating pro-Uber and anti-Uber participants, explained 22% of variance; the second, about regulation, explained 6.2%.⁠³ Participants were recruited through stakeholder groups and through paid and organic social media posts aimed at geographic and gender diversity.⁠³ The sample was not a probability sample.

The amended regulation, effective October 25, 2016, provided that taxis no longer had to be painted yellow, high-end app-based taxis could operate if they did not undercut existing fares, dispatch apps had to show car and driver identification, estimated fare and rating, and per-ride taxation had to be reported to the Ministry of Finance.⁠² vTaiwan’s information page adds that more than 4,000 people helped crowdsource the agenda.⁠¹ MIT Technology Review attributes the outcome partly to face-to-face talks led by Audrey Tang after the online phase.⁠⁵

Totals

No source verified for this page defines “decisive government action” or lists which cases met it. The 80% figure therefore cannot be checked independently.

Failures and criticisms

vTaiwan’s critics include its own builders, legislators from both major parties, and outside researchers, and their strongest points concern authority, agenda and representativeness.

• No obligation to act. MIT Technology Review reported that “the government isn’t required to heed the outcomes,” and quoted opposition legislator Jason Hsu, who helped create vTaiwan, calling it “a tiger without teeth”⁠⁵
• Political dependence. Ruling-party legislator Karen Yu told the same publication that vTaiwan was “not a huge priority” for the administration and had come “close to death” at times.⁠⁵ The 2016 withdrawal of pending bills shows how a change of government can erase outputs.⁠⁵
• Agenda controlled by officials. The contributors themselves write that “very few cases are proposed by citizens,” that the government-funded process “struggles to retain autonomy in the selection of topics,” and that “if a government authority refuses to discuss a public issue, the sensitive topic will not go through the vTaiwan process”.⁠² GovLab adds that not all ministries were convinced the process worked and that the online format favored technology topics.⁠⁴
• Mixed outcomes on specific cases. On online alcohol sales, the contributors’ paper says the draft bill “was rejected by the Legislative Yuan due to its conflict with existing e-liquor policy”,⁠² while MIT Technology Review attributes its failure to the 2016 withdrawal of pending bills.⁠⁵ The two accounts differ, and both agree the bill did not pass. On fintech, GovLab reports that the reflection stage was skipped “because of pressure for immediate legislation”.⁠⁴
• Consensus versus speed. GovLab notes that a focus on consensus may conflict with the need for urgent response.⁠⁴ Contributors describe a mismatch between the community’s fast pace and government procedures.⁠²
• Openwashing risk. C.L. Kao, quoted by MIT Technology Review, warned that ignored recommendations risk the process being seen as “openwashing”.⁠⁵
• Representativeness. Pol.is participants are self-selected or recruited, and the developers acknowledge sensitivity to arrival order and minimum sizes.⁠³ Results describe who took part, not the population.
• Thin independent evaluation. The main published accounts verified for this page were written by people who built or ran vTaiwan or Pol.is,⁠^2,3 by a partner research group,⁠⁴ or by journalists.⁠⁵ No independent peer-reviewed evaluation of vTaiwan’s outcomes was verified for this page.

What transferred and what did not

vTaiwan’s tool traveled further than its institution. The Computational Democracy Project, which maintains Pol.is, lists uses including Austria’s national climate council in 2022, Bowling Green, Kentucky in 2018, where about 2,000 of 65,000 residents took part, a 2020 Demos study in the United Kingdom with a nationally representative sample of 997, and a 2018 German party consultation with 33,547 participants.⁠⁷ These are the project’s own summaries.

The institutional pieces transferred mostly within Taiwan: Participation Officers in every ministry from 2017 and the Join platform.⁠⁴ The attempt to give vTaiwan-style consultation a statutory footing did not pass in 2019.⁠⁴ g0v’s model of open, volunteer civic hacking, which it describes as “a decentralised civic tech community” with a large hackathon every two months,⁠⁸ is a culture rather than a procedure, and no verified source documents it being replicated as a government process elsewhere. Compare city platforms with formal budget authority, such as /cases/madrid-decide/ and /cases/reykjavik-better-reykjavik/.

Lessons for My Vote Forward

The following points are interpretation and design proposals, not findings about My Vote Forward, which has not been tested.

• Mapping agreement is different from counting votes. Pol.is groups people by how they vote and looks for statements that cross groups.⁠³ That is a useful way to report opinions without claiming a majority decision; see opinion and choice.
• Say who controls the agenda. vTaiwan’s weakest point was that officials chose which issues entered.⁠² A framework should publish who can add a question and who can block one; see question design and governance.
• Publish what was done with each result. The 80% figure cannot be checked because the case list and definition are not public in the sources verified.⁠^1,2 Track each output to an outcome; see participation over time.
• Do not present recruited samples as the public. The UberX conversation recruited through social media.⁠³ Report participants against the population; see measurement and collective priorities.
• Institutional durability beats enthusiasm. A change of government and an unpassed bill,⁠^4,5 mattered more than the software. See proposals.

Unresolved questions

• What did the crowdfunding consultation, often listed among vTaiwan cases, produce? No source verified for this page documented it, so it is not described here.
• What role did the National Development Council play in running Join, and what are Join’s petition thresholds and response deadlines? Official pages could not be retrieved for this page.
• Which cases count toward the “more than 80%” figure, and what counts as “decisive government action”⁠^1,2?
• How did the fintech sandbox consultation shape the January 2018 Act⁠⁹ given that the reflection stage was skipped.⁠⁴?
• Is vTaiwan active after 2024? The latest self-report verified here is from December 2023⁠¹
• Would a vTaiwan-style process with a statutory duty to respond perform differently? The 2019 bill that might have tested this did not pass.⁠⁴