Exploring Generative AI in vTaiwan
Shuyang Lin, Joshua Yang, Eli, Peter - all from vTaiwan community
15th December, 2023
1
Ideation deck
vTaiwan
2
A public opinion collection process facilitated by an open community
Data Source: info.vtaiwan.tw vTaiwan Slack
Topics on vTaiwan
3
Digital regulatory reforms
2024 生成式 AI 指引 Generative AI Guideline
2024 AI 基本法 AI Basic Law
2023 AI 人權 Democratic input to AI
2020 開放國會 Open parliament
2019 電動滑板車 Electronic personal assistive mobility device (EPAMD)
2017 私密影像 Non-consensual pornography
2017 無人載具 Uncrewed Vehicle
2017 共享經濟 Shared Economy
遠距教育 Remote education
2017 醫師心理師透過網路平台提供服務 E-Clinic
2017 遙控無人機應用與管理機制 Uncrewed Aircraft Vehicle
2017 公司法：董監事選任 Company Law Rewrite
2017 企業資產擔保法
2016 金融監理沙盒 FinTech
2016 網路賣酒配套措施 Online Liquor Sales
2016 公司法：社會企業 Company Law Rewrite
2016 公司英文名稱登記 Company Law Rewrite
2015 Airbnb Airbnb
2015 UberＸ UberX
2015 開放資料例外收費原則 Open data
2015 網路霸凌 Cyberbullying
2015 資安管理 Cyber Security
2015 電傳勞動 Telework
2015 遠距健康照護 Telemedicine
2015 零售 Consumer Protection
2015 網路交易課稅 E Tax
2015 數位生活：遠距教育 Distant Education
2015 群眾募資 Equity-based Crowdfunding
2015 閉鎖性公司 Closely-held corporations
Data Source: vTaiwan 議題盤點 vTaiwan Discourse
Digital tools supported deliberation
4
Community
Process
Start
End
Technology
Content of discourse
Machine- readable output
Human- readable output
Data Source: Bridging Recursive Public
AI tools supported deliberation
5
Community
Process
Start
End
Machine- readable output
Human- readable output
Technology
Community’
Content of discourse
Community
Process
Start
End
Machine- readable output
Human- readable output
Technology
Community’
Content of discourse
AI as stack support
(e.g. AI as transcriber)
AI as bridging support
(e.g. AI as translator)
AI as discourse support
(e.g. AI as facilitator)
How can we help the community steer more effortlessly?
What are the takeaways from the last conversation?
How can we help next discourse with a list of taxonomy?
How can we identify opportunities for consensus building?
How can we identify statements bridging across different opinion groups?
Data Source: Bridging Recursive Public
AI tools supported deliberation
A. Topic preparation
When starting a topic, can we use AI to help us find stakeholders, gather diverse opinions, go through studies
B. Issue Onboarding system
When stakeholder come to vTaiwan, we can use a chatbot to help the stakeholder make his/her thoughts become more tangible, with stakeholders, notes, and potential context.
This topic might contain the following sub issues…
Data Source: MSR x vTaiwan collaboration in discussion
AI tools supported deliberation
C. LLM-supported summarisation of opinion clusters
Within a discussion, this LLM tool would effectively condense opinion clusters, making it simpler for participants from varied backgrounds to quickly comprehend the prevailing viewpoints in a deliberation. This summarisation could aid in grasping the essence of complex discussions and highlight the key themes and commonalities between different opinions. This approach facilitates a deeper understanding and engagement across varied perspectives.
D. AI supported moderator
When the opinion collection deals with some controversial issues, AI moderator can remove the emotional content as well as hate speech, making discussion less distracted.
This group thinks…
This group believes…
Data Source: MSR x vTaiwan collaboration in discussion
AI tools supported deliberation
E. . Consensus Tracing (共識 Tracing) Using LLMs
After a consensus is achieved, the tool could use Large Language Models to analyze the evolution of opinions, revealing the logic behind majority decisions and making the path to consensus clear. This approach could improve understanding and acceptance of the final collective decisions
Community
Process
Start
End
Technology
Content of discourse
Machine- readable output
Human- readable output
Data Source: MSR x vTaiwan collaboration in discussion
AI tools supported deliberation
F. Expanding the scope to articulation
When stating one’s opinion about a topic, it’s very hard to condense one’s opinion to only a few words or in a few sentences. Giving an essay-like format for user’s articulation about a topic gives them more space to articulate and organize their reasoning and explanation with contextual information.
G. Assisting the process of shaping one’s articulation
In a human-led, AI-supported process of shaping articulation, an articulation assistant (AI) could:
submit
Why do you think so?
A counter opinion might be…
Do you mean this..
Data Source: MSR x vTaiwan collaboration in discussion
AI tools supported deliberation
H. Analyzing suggestions and polling potential common grounds
When users finish their articulation pieces, an generative AI tool could analyze their suggestions (while persevering the context connected with the suggestion). In doing so, another generative AI tool could come up with swappable vocabulary (or rephraser) more frequently used by diverse groups and conduct a round of poll and collect the users’ agreements or disagreement stats.
I. Analyzing what makes people change their mind
In iterations of articulation submission, people might change their mind or even stance. An AI tool analyzing how that happens is also a very important element to finding common ground. For example, some topics might be more difficult to find common ground because it's to do with a debate of beliefs while on other occasions we mind finding compromise or giving in taking place more often cause it’s a mild conflict of interest. The experiment could be constructed with a scale of debatable topics, from conflict of interest to conflict of beliefs, and analyze what it takes for people to change their mind, be it rephrasing, understanding, empathy, or other interest given.
Here are 3 statement extracted from your articulation…
Do you approve?
approve
disapprove
Here are a few way to rephrase your statements, do you agree?
rewrite
Data Source: MSR x vTaiwan collaboration in discussion
Thank you. 🙏