This is part of a series of stories on digital participation platforms produced with support from the Friedrich Naumann Foundation’s Global Innovation Hub, Taipei. To read others in the series, see these case studies from Germany and Northern Ireland.

Years ago, vTaiwan became a great example of digital democracy; it highlighted the benefits of a decentralized digital consultation platform designed to bridge the gap between citizens, government, and stakeholders in policymaking. Today, vTaiwan is still dedicated to experimenting with new methods of public deliberation.

However, unlike in its earlier years, vTaiwan is now fully volunteer-driven, without direct government support, and serves as a civic laboratory for participatory democracy. Previously, the Taiwanese government supported vTaiwan as a public-private collaborative project. Despite the reduced governmental support, the community remains motivated to continue developing and using digital tools to enable citizens to provide input on policy issues.

Since 2024, the community has organized a series of “Social Issue Meetups,” combining hybrid events with deliberative discussion and digital participation tools. “Our goal is to provide digital toolboxes and serve as a laboratory for deliberative discussion in Taiwan. We are a volunteer-driven community of five to seven active contributors who design, facilitate, and analyze events. We use Mentimeter to measure active participation. We are currently averaging around 50 participants per event,” as Peter Jia Wei Cui, program coordinator.

One of vTaiwan’s recent initiatives focuses on the recently passed Fraud Crime Hazard Prevention Act. In July 2024, the Act came into law, while the act is meant to strengthen the fraud prevention system and effectively combat fraud crimes, vTaiwan recognized some issues and opportunities. So they created a platform for informed debate on the Act, aiming to balance crime prevention with human rights concerns. In December 2024, they hosted an event to allow their community to discuss the implications of the law.

Cui stressed, “We focused on the conflict between allowing the government to have more power over crime prevention, and the potential human rights infringement. Due to the mass financial loss from the online fraud cases, as well as the public discourse prioritizing crime prevention, the consideration of human rights infringements was usually marginalized during the legislative process. To us, we saw deliberative discussion as a bottom-up effort that could supplement the policy discussion and conversation and form a more multi-stakeholder perspective.”

Leveraging digital tools and AI

During the Fraud Crime Hazard Prevention Act event, the importance of using digital participation tools in deliberative discussions was clear, as they found that the tools strengthened the discussion process. When vTaiwan noticed that most of the participants were typically aged between 20 and 40, with higher education and digital literacy levels, they ensured that the deliberations were hybrid, ensuring that participants from other cities and backgrounds could contribute online. The deliberation discussions were conducted primarily in Mandarin, with limited English interpretation provided by volunteers.

vTaiwan has always demonstrated how civic communities can innovate in democratic participation by combining open-source platforms, commercial tools, and AI. Currently, the community integrates tools such as

Pol.is for large-scale opinion mapping,
Mentimeter for interactive deliberation, and
Generative AI tools such as ChatGPT and Claude for analysis and summarization in their deliberation and participation initiatives.

“Our current approach focuses on how AI technologies work together across the democratic process. We currently deploy different tools at two main stages: first, mapping citizen opinions through Pol.is; then, facilitating deeper discussion through large language model-monitored deliberation. Each stage addresses specific barriers to meaningful participation, whilst we envision expanding this toolkit to include better information-gathering tools,” explained Josh, a vTaiwan contributor.

While vTaiwan is experimenting with AI, during its deliberations, the end-users do not directly engage with the AI tools. However, the vTaiwan team members have used AI tools to circulate opinions, summarize them, and help produce the script for the final reports. As they continue to use digital tools and introduce AI, one aspect that is becoming increasingly important for their work is informed consent and privacy. Cui said, “When we collected the feedback for AI from the participants, we did not collect any personal data from the participants. We also asked the participants if they allowed their opinions to be used to generate new opinions by AI before the event, offering them a chance to opt out. We suggested that if the data to be fed into AI contains personal data, clear notice, informed consent, and the right to opt out are necessary.”

One vTaiwan participant expressed that AI helped her articulate her opinion on social issues, as AI contributed to” smoothing the oral expression and reorganizing her originally fragmented thoughts.” vTaiwan thinks that AI might be able to increase the inclusiveness of their deliberation processes, allowing more people to participate in the discussion with a lower psychological threshold.

However, vTaiwan warns that this comes with some risks.

“One key risk is that AI may distort the meaning of participants’ contributions, especially if there is no effective review or double-checking after AI processing. At the same time, we recognize AI’s potential to enable broader participation, particularly for individuals with language difficulties or disabilities. To mitigate risks, we test the tools before deployment to identify potential issues, and we always include human review as an irreplaceable safeguard,” shared Peter.

Explaining their process, Josh, said, “After generative AI became viral, we also wanted to see if we could leverage AI on some time-consuming tasks and save the contributors' effort on more important tasks, like generating more insights from the deliberative discussion, developing more tools, and last but not least, engaging with people, the core of every deliberation process. From vTaiwan’s perspective, AI works most effectively when different tools handle distinct democratic challenges. Rather than seeking one universal solution, we match specific technologies to particular participatory problems based on what we’ve already implemented and what we envision for the future. “

How it works:

The community combines multiple stages of digital participation in their deliberation discussions, and the use of AI tools saves the volunteers’ time on analysis and reporting while enhancing the inclusiveness of discussions. Before the deliberative discussion, they use polis.tw to collect public opinion (with a seed opinion created by our community members to spur discussion). “We collect both qualitative and quantitative data. Qualitative data include participants’ feedback on the issues we discuss, and the reasons they agree or disagree, while the quantitative data include participants' demographic information, their vote to agree or disagree,” Cui.

Then they use the report generated by Pol.is to see the visualization of opinions, including different cohorts divided by their response, and generative AI to summarize the stance and opinion that different cohorts support, and find out the most divergent opinions, and take those into the second round of deliberative discussion. They use interactive slides on Mentimeter to record the input from the audience, including the reason participants agree or disagree with the divergent opinion, and also to visualize the result in real time. The team described the process as follows:

Mapping Understanding: Polis currently handles broad participation through intelligent consensus-finding that actively encourages perspective-taking. vTaiwan’s existing system does far more than simply map opinions. Polis deliberately nudges participants towards greater understanding of alternative viewpoints by strategically presenting statements from people who hold different views. This creates a dynamic learning environment where people often discover common ground they didn’t know existed, or develop a more nuanced understanding of why others hold different positions. The approach effectively processes large-scale input: by 2020, vTaiwan’s mailing list included 200,000 individuals. Lessons that we learn in Taiwan when using Polis to build consensus are, we need to not only map where people currently stand, but also help them explore different perspectives through respectful engagement.
- For the vTaiwan team, human validation is paramount as it ensures democratic accountability, particularly around AI-generated summaries. “The most critical oversight occurs when AI systems summarise both the Polis cluster analysis and subsequent face-to-face conversations. Participants vote on these summaries to verify they accurately capture what different opinion camps actually believe and what was genuinely discussed, ensuring AI cannot misrepresent human voices.”
- Large Language Models (LLMs) are used to analyze these Polis clusters to understand the reasoning behind different opinion camps. Once Polis has mapped public opinion, we deploy LLMs to examine what distinguishes different groups of participants. The LLMs process the specific statements that each cluster agreed or disagreed with, identifying the underlying values, concerns, or assumptions that drive different positions. This analysis reveals not just that people hold different views, but why they hold them, whether disagreements stem from different priorities, different factual understandings, or different experiences.
Deliberating on Difference and Discovering Consensus: During face-to-face deliberation, LLMs have been deployed to monitor ongoing discussion to track how the understanding of each of the participants evolves. Equipped with insights about different opinion camps from the Pol.is analysis, LLM was used to identify when participants begin to understand different viewpoints or when a new consensus emerges around unexpected solutions. This creates a feedback loop from opinion mapping through analysis to deep deliberation.
Learning through Conversations: We are exploring ways to facilitate and improve the process of gathering information for participants before they begin the discussion, and the goal is to ensure the information is comprehensive and easy to understand. AI agents become our tool for experimentation since it can autonomously research policy questions after a prompt is given and generate interactive LLM-based content that allows citizens to explore issues through conversation-style interfaces. These envisioned tools would facilitate the preparation phase, ensuring participants arrive at both Polis voting and face-to-face discussions with a robust understanding of discussion topics and broader context.

Benefits

According to Cui, hybrid participation has been beneficial for citizens in many ways:

First, for the participants, online participation enables them to join discussions without being limited by the availability of seats and geographical location. “To ensure inclusiveness, we use a hybrid strategy and select proper tools that allow participants online and offline to join in the discussion together; we create a hybrid version of discussion, asking participants if they agree, neutral, or disagree with the opinion, making them vote and see the result in real time, then asking them to explain the reason behind their vote, collecting opinion through the transcript and uploading opinion; using polis.tw and menti.com ensures the whole process is interactive both for the offline and online participants, and reduces the need to have extra work on synchronizing the online and offline opinion.”
Polis.tw or Mentimeter creates a real-time visualization of collected opinions, which helps the organizers have a better understanding of the public opinions and serves an important role in organizing the discussion, while the real-time visualization also prompts participants to explain more about their stance as they are more aware of their stance and others.
The online tools and platforms we use automatically generate the data, category, and transcript we need for the insight and discovery, which saves a lot of manpower and time, enabling us to engage with more productive and substantially important tasks like reflecting on the whole process and improving it without compromising the quality of the final report. You can click here to see our report generated by Polis visualization report function.
Deeper Dialogue: The combination of Polis, AI summarization, and real-time deliberation tools helped participants better understand opposing views and articulate their positions. Some participants found AI particularly useful in helping them clarify and express their opinions.
Efficiency Gains: Online tools provided real-time visualization, automated data collection, and categorization, reducing manual work and allowing volunteers to focus on facilitation and reflection.

Impact of the initiative

Policy Engagement: vTaiwan presented its methodology and insights from its December 2024 AI governance event at a March 2025 consultation hosted by the National Human Rights Commission. “Our engagement with the National Human Rights Commission focused on the issue of artificial intelligence. Some of the feedback from our process has already been reflected in the NHRC’s recommendations for the draft AI Basic Act in Taiwan, which is currently under legislative review,” shared Cui.
Regarding the Fraud Crime Hazard Prevention Act, some members of the ruling Democratic Progressive Party (DPP) have expressed interest in both the deliberation process and its outcomes. In addition, we submitted a proposal through the government’s Join Platform, which successfully passed the required threshold. A follow-up meeting will be held in October 2025 to gather further opinions and refine the policy design. Cui mentioned that the ruling party is interested in the result of our discussion on the Fraud Crime Hazard Prevention Act. “These are all marks that the button-up effort can still have influence, although not so directly, on the government's policy.”
More details are available here.

Navigating challenges

When we use AI for various purposes, some language, transparency, and trust issues appear together. The team found that AI tools performed less reliably in Traditional Chinese, reflecting their lack of high-quality training data. Cui pointed out, “For Taiwanese users, the main challenge remains the lack of a high-quality traditional Chinese training dataset as AI usually compromises the accuracy of the output of the mainstream AI models. In the specific context of policy deliberation, which requires a deeper understanding of local background and nuanced consideration of expression, the disadvantage will be amplified.”

Secondly, participants have questioned AI outputs, highlighting the need for open-source data, transparent prompts, and human validation. Additionally, the team has found that AI adoption can create both opportunities and risks, especially where accuracy and context are concerned. “In our last event, we tried to use AI to polish the uploaded opinions by participants, ensuring that there was no offensive or emotional wording that could arouse others' emotions and lead to the distortion of the opinion gathering. However, we found that such a way of standardization of opinion can sometimes create inaccuracy of opinion, leading to more misunderstanding and confusion for other participants. For example, in an opinion containing "driver" (in Mandarin 車手, which is a term that has multiple different meanings, including a scooter or car driver cashing out from the ATM or victims for money laundering or just normal drivers), AI will change the word to ‘Basic labor of the fraud gangsters’, which is confusing for the locals.” noted Cui.

Another key challenge facing vTaiwan is measuring the effect of its efforts, for example, measuring how its bottom-up efforts are truly supplementing multi-stakeholder policy discussions. “At present, we do not have concrete quantitative data on how many policies have been directly shaped by our efforts. Instead, we rely on qualitative indicators—for example, whether policymakers reference our discussions in their statements, whether government agencies invite us to further consultations, or whether our proposals gain traction on official participation platforms such as Join. These signs suggest that our work is supplementing existing multi-stakeholder discussions. Over time, we hope to develop clearer metrics to better evaluate the extent of our influence.”

Cui concluded, “We think the biggest challenge is that AI adoption can still be unstable from time to time; therefore, human modification and monitoring are still necessary.”

Lessons from vTaiwan

Start small: AI should be introduced gradually into deliberation processes, focused on specific, well-defined tasks.
Human oversight is essential: AI outputs require careful review to ensure accuracy, avoid bias, and maintain trust.
Citizens come first: Technology should serve participants, not overshadow them. Successful digital participation depends more on building trust, accessibility, and responsiveness than on selecting “the right” tool. “Digital participation is not only about adopting fancy tools or adding buzzwords into the introduction. The necessity of digital tools is always the priority, and it should be linked to an urgent or important issue to be solved, including adding more diversity, reducing time-consuming jobs, or scaling up the deliberation. Moreover, when adopting digital participation, it is the citizens, not the issue or technology used, that matter.”
Hybrid models work: Combining online and offline participation broadens access and enables richer deliberation.

Reflections and advice for others

For governments or organizations considering AI-enabled digital participation:

Begin with pilot programs and iterative testing before scaling.
Match tools to specific participatory needs rather than adopting one-size-fits-all solutions. According to Cui, “Challenges usually lie in how to select the tools, as there are many different digital tools developed for different purposes and catering to distinct processes of digital participation. Sometimes, a little difference in the discussion process can complicate the whole project. Therefore, if a government agency wants to adopt digital participation tools, they might need to do the test or pilot rounds to see if the tools can really fit the need. Also, adopting multiple tools and platforms during different stages of deliberation can be an alternative. The vTaiwan community has run several digital participation events in Taiwan, proving that a combination of different free or open source tools and platforms can still cultivate high-quality discussion and form consensus.”
Invest as much in outreach and communication with citizens as in tool development: How to make the digital participation tools/platform more familiar and accessible to the targeted citizens is very critical. After all, if no user trusts and uses it, a perfect tool can still become helpless. Instead of investing a lot of time and resources in tool development or selection, we suggest that more resources should be allocated to the promotion and conversation with the targeted citizens.
Ensure transparency, informed consent, and participant control over how their inputs are used: Finally, public trust and political effectiveness are also important. If you want to make people know the importance of digital participation, you need to show that digital participation can really respond to citizens’ needs and answer their questions. Therefore, besides the tools used in the digital participation, a special team to engage with the opinions collected from the process, tracing how the opinions are received and really alter the policy, is as important as the tools themselves.
When adopting AI-based tools or platforms into the deliberation process, focusing on one single part of the process as the start, never try to use AI to solve every problem all at once. Understand the tasks, target audience, and the achievement you want before implementing any measures. A high-quality output usually requires a long time from discussing, testing, implementing, and reflecting on the whole process. One simple metric to see if a task can be allocated to AI is whether the result will be different if the task is done by different people.

Future plans:

Now that vTaiwan is an independent community, they are working on using strategies to bring the outcomes of deliberations to policymakers’ attention. “Our strategies focus on building cooperation and connections with government agencies and NGOs. For example, we are collaborating with TWNIC on deliberations related to Internet governance. Additionally, I personally serve as a mentor for the Department of Education’s Bureau of Youth Development in its annual Let’s Talk Project, which supports teams across Taiwan in hosting deliberative discussions aimed at cultivating future public service professionals and facilitators.” Cui.

Adapting to the loss of government support: AI has helped increase the productivity of our team, particularly for routine tasks and managing large volumes of data. Without government support, limited human and financial resources have been a challenge. AI enables us to focus more on stakeholder engagement and participant interaction, rather than being consumed by administrative chores.

They hope to continue vTaiwan going, and influence the government to adopt a similar strategy on policy design and open government, creating more space for deliberative discussion and covering broader stakeholders around the country. vTaiwan is testing how technology can deepen dialogue, identify consensus, and reduce barriers to participation. While challenges remain in ensuring accuracy, inclusiveness, and trust, their hybrid model of deliberation highlights how technology can deepen public dialogue and bring new perspectives into policymaking. By centering people first and tools second, vTaiwan is building a participatory process that not only gathers opinions but fosters understanding across differences, an approach that could inspire civic innovators and governments worldwide.

Ready to explore further?

Visit the People Powered Digital Participation Platform Guide and ratings.
Sign up for coaching by an experienced mentor.
Subscribe to the People Powered newsletter if you have not already, to learn about upcoming workshops, courses and other learning opportunities.

Featured