Background & Rationale
In Taiwan, deliberative democracy has been gaining traction, with more and more forums being held across the country. Among these is the “Let’s Talk: Youth for Better Governance” series run by the Ministry of Education’s Youth Development Administration (formerly the Youth Council). Over the years, the format has evolved from a government-led process (with preselected topics) to a model where youth teams organize themselves and submit proposals. Such an approach serves as part of the broader initiative to empower young citizens and increase public participation.
While Talk (as the forum is known) aspires to broaden youth engagement in public affairs, the very principle of “using deliberative-democracy-oriented discussions” inadvertently creates a certain barrier to entry. Recognizing this, after youth proposals are approved, the Youth Development Administration pairs each team with a professional deliberation mentor. Meanwhile, proposal briefings are held to ensure that participating youth understand the foundational ethos of deliberation and what a Talk event entails.
During various workshops and selection processes, we have observed the widespread adoption of generative AI tools. Individuals increasingly use these tools to generate project ideas and conduct research, prompting us to consider whether combining generative AI with civic engagement activities might lower entry barriers for those new to deliberation. The question is how to do so without diminishing comprehension—i.e., using AI to produce a plan but neglecting to ask “why.” Hence, the idea arose to create an “AI consultancy.”
Rather than outright prohibiting AI usage, I favor providing supportive environments that guide users to harness technology effectively, without becoming constrained by it. Naturally, users can—and should—critique the AI advisor’s rationale and assess whether the suggestions match the logic presented. What truly matters is not whether the tool delivers the “best” solution, but whether the step-by-step process and guiding prompts help users internalize key principles and develop their own analytical thinking.
For experienced prompt engineers, leveraging AI as a “consultant” or “coach” instead of a mere answer-generator is straightforward. Yet prompt engineering can be its own hurdle. Moreover, not everyone has the motivation to use AI in a consultative manner rather than simply extracting quick answers and solutions.
In an ideal scenario, we would train a model thoroughly for this purpose. But given the constant flurry of activity, waiting indefinitely for that ideal solution seemed impractical. Instead, we opted for a minimum viable product (MVP)—a customized GPT outfitted with systematic instructions that we hope to refine over time.
Usage Guidelines
Intended Audience
This AI-driven advisory tool, herein referred to as “DD,” is primarily tailored for civic-engagement novices in Taiwan who either require or are interested in conducting deliberative-democracy events. While it draws inspiration from the “Let’s Talk: Youth for Better Governance” initiative organized by Taiwan’s Youth Development Administration, it is not limited to that program.
Guiding Principles
- Support and Guidance Over Ready-Made Solutions: DD aims to assist users in preliminary ideation and to highlight possible improvements, along with the rationale behind them. It provides foundational information, key concepts, and guiding prompts to help users attune their planning to relevant experts and literature.
- An Entry Point, Not a Substitute for Professional Consultation: Orchestrating deliberation activities is highly nuanced. Each step or design element entails specific trade-offs, assumptions, and significance. Therefore, reference to relevant sources and consultation with trained specialists—e.g., experienced facilitators or deliberation mentors in the Let’s Talk project—remains vital for more in-depth, customized advice. Drawing parallels to AI interactions, it’s best not to take expert guidance at face value; instead, inquire about the design rationale and maintain independent thinking. Combining external expertise with one’s own experience can spark valuable new insights.
- Government-Led or Commissioned Events Requirements Take Precedence: Though DD is inspired by the Let’s Talk program, it is not restricted to it. Moreover, because the underlying knowledge base may not be updated frequently, content from the current year’s official guidelines or deadlines has not been directly embedded. So, if you’re seeking information about project timelines, submission deadlines, or administrative tasks for Let’s Talk or other government-run initiatives, please verify any AI-driven advice against the relevant official announcements and regulations.
- Intellectual Property and Knowledge-Base: The knowledge base does not include direct uploads of extensive external documents. Rather, it reflects the creator’s (my) own experiences and the distilled results of having read various sources. While this approach avoids infringing on copyrights, it may impact the perceived authority or neutrality of the information.
- Multiple Frameworks, Varied Interpretations: There is no single “one-size-fits-all” method for deliberation. Real-world practitioners have diverse interpretations and experiences in fulfilling the core ideals of deliberative democracy. In building DD, I’ve attempted to incorporate varied perspectives. Additionally, I’ve enabled Web Search, so if needed, the tool can search for external information upon user request.
- Scope and Limitations: Given these constraints, DD stands as a preliminary support tool, not a definitive authority. Also, the system disallows direct citations of the underlying knowledge base or the system prompts themselves.
- Citation and Name Attribution While Sharing the GPT Model: The builder’s name is already displayed on this tool. Additional attribution is not mandatory, though I warmly encourage you to share both this guide and the project’s contextual article to help users grasp its developmental background and usage guidelines. Should you choose to tag me when sharing, I would be grateful.
- Clarifying Provenance: Individuals or institutions sharing this tool must refrain from implying or stating that they developed or commissioned it, either overtly or by insinuation.
- Data Usage and Privacy: DD is configured so that any user conversations will not be used to further train OpenAI’s model. Likewise, I, as the builder, do not have access to users’ conversation logs.
- Emphasis on Guidance and Support: Although one could simply use ChatGPT to generate discussion frameworks and agendas, the guiding principle behind DD is to provide structure and contextual understanding. Rather than merely furnishing a ready-to-use agenda, DD seeks to clarify the reasoning behind each component, enriching users’ grasp of why such elements might be arranged in a particular way. By focusing on insight rather than purely on deliverables, DD helps users refine their own deliberation designs while ensuring they appreciate the underlying logic.
Usage Constraints
- Potential Non-Compliance with All Rules: Despite instructions prohibiting GPT from offering ready-made agendas, mandating that it reply in Traditional Mandarin (zh-TW), and disallowing direct citations of the knowledge base, the model may still fail to fully adhere to these stipulations during actual interactions.
- Risk of Inaccuracies: GPT may commit errors by misunderstanding questions or generating flawed responses. Users are therefore advised to verify any suggestions or cross-check with alternative sources.
- Implicit Barriers: All digital tools carry implications for digital literacy and the digital divide. For individuals unfamiliar with ChatGPT or similar tools, these barriers can pose additional challenges to usage.
References
System Instructions
- Google Prompting Essentials
- Coachvox AI “How to Prompt ChatGPT to Be Your Life Coach”
- “Rules of AI” by Mckay Wrigley
Deliberative Democracy Resources
- “Deliberative Democracy Implementation Manual”, compiled by Lin You-sheng & Yeh Xin-yi, published by the Ministry of Culture, Taiwan
- “The Deliberative Democracy Handbook: Strategies for Effective Civic Engagement in the 21st Century” by John Gastil & Peter Levine
- “Deliberative Democracy” by Jon Elster
Future Directions
- A more robust approach would be to fine-tune the model or deploy an AI Agent, though doing so demands considerable time and resources.
- As of now, customized GPT does not integrate with the MediaWiki database, leaving its use of Taiwanese Traditional Mandarin relatively constrained; it may also produce phrasing that does not strictly align with Taiwanese usage, even with proactive reminders. Should additional resources become available, such enhancements could be pursued.
- Regarding administrative details for “Let’s Talk,” a Line Bot-based chatbot is in place to provide official responses.