Recursive Publics (C. Kelty, 2008) are "collectives... capable of speaking to existing forms of power through the production of actually existing alternatives", the kind made possible by new digital technologies. Bridging the long-standing gap between decision-makers and recursive publics is a challenge that AI must meet, and that might be met by leveraging AI.

In the build up to the Global AI Summit hosted in London this year, the international vTaiwan community and Chatham House propose a collaborative research and deliberation project to pilot a bridged recursive public, to evaluate the ways in which existing democratic deliberative processes could enhance and be enhanced by AI tools, and to present the findings to leaders at the Summit.

vTaiwan is a world-leading methodology for engaging citizens in policy discussions and decision-making. vTaiwain’s international community has been a driving force behind some of the most successful efforts to enable participation in policymaking. Chatham House is a world-leading policy institute based in London and brings long-standing and evidence-based expertise on democratic processes and rule of law, and brings expertise on the piloting of deliberative democratic tools in government.

### Project Background

Public participation in shaping the governance of technologies that will have a seismic effect on societies around the world is critical. AI tools are both precisely this kind of technology and simultaneously a tool to facilitate public participation. The development of AI tools can revolutionise the methods through which public participation by wide and novel publics is possible, and inform the design of the governance institutions and processes of the future.

This project seeks to bridge these two axioms. It takes an existing community of expertise and methodology and both evaluates its applicability to AI policy discussion and tests where applied AI might further improve its methods. This pilot is designed to have a life beyond the timelines of the initial OpenAI grant.


### Project Summary

We propose a two-track research programme, reflecting these two complementary questions.
* **Track One**: How can the Recursive Public facilitate discussion on the responsible deployment of AI?
* **Track Two**: How can Recursive Publics be enhanced by utilising AI?

**Track One**:

AI is a broad, fast-moving and complicated topic, will impact a wide range of stakeholders, will create responsibilities across industry and governance institutions, and impact underrepresented communities. For these reasons, the topic represents a new challenge to existing successful deliberative democratic methods.

In this proposal, we suggest an adjusted approach to operating vTaiwan as an a) international and b) continuous platform to meet the new demands of an AI deliberation. Historically, deliberations would end with a formalized consensus moment. But to engage on AI, a new approach is required that allows for long-term community engagement without a fixed end date. We propose to test a solution that continuously feeds deliberative outputs back into the deliberative process while allowing for 'snapshots' of public wisdom and opinion at any moment. We call this the Recursive Public process.

#### The Recursive Public Process

The Recursive Public draws on the team's expertise in building vTaiwan. A model is shown here: https://info.vtaiwan.tw/#four.


Our deliberation will begin with one key question, as proposed by the OpenAI team:

`*What principles should guide AI when handling topics that involve both human rights and local cultural or legal differences, like LGBTQ rights and women’s rights? Should AI responses change based on the location or culture in which it is used?*`

The team will then follow the established vTaiwan process - establishing a steering group, inviting stakeholders and experts, producing educational material to ensure participants are familiar with the subject and methods, and begin an international online deliberation using the pol.is platform with unique invites capturing basic demographic information distributed to participants. We expect Pol.is to meet requirements around effective moderation, the ability to surface minority opinions and legibility. Finally, the outputs of the process - the "snapshot": an analysis of consensus positions, minority and majority opinions, emergent issues and bridging statements - will be reported on publicly around the Global AI Summit in London: to the community and to our audience at OpenAI, in government, and in the wider AI development community for whom this question is critical. Finally, this snapshot will be fed back into the Recursive Public process, both opening up new avenues of deliberation and allowing the community to further refine or change their thinking on the position over time.

Community members will be joined by experts from Chatham House's AI Taskforce and wider network, including representatives from twenty countries, from the UN, Partnership on AI and a range of academic institutions, government and industry bodies. Results will be presented as recommendations and a values-based decaration for a range of stakeholders, including those architects developing AI tools.

Reporting will support a robust evaluation process. The process will be qualitatively evaluated against established vTaiwan standards, and quantitatively against metrics including:

- Scalability and Applicability
- Consensus Mapping over Time
- Inclusivity and Diversity

With success, the project will form a lasting foundation for community inputs into high-level AI discussions over time, bridging the gap between communities and decision-makers, and doing so in a way that reflects the international nature of AI deployments. This recursive approach both complements and contrasts the range of deliberative projects in this space, including the Collective Intelligence Project's Alignment Assemblies in Taiwan itself, by testing and evaluating the strength of non-governmental and international deliberations.

**Track Two:**

AI tools have enormous potential to support the design and redesign of governance processes and institutions. Track Two will test this. Chatham House will lead a high-level discussion on the current participatory governance challenges that AI solutions could support or improve. The discussion will take applied learning from track one, evaluate their applicability to established governance bodies and approaches, and present a public-friendly provocation on the roles AI-enabled technologies might play in bridging established governance models and institutions and recursive publics.

The report will draw heavily on our experiences in running an AI-related deliberation in Track One, and will focus on processes that AI technology could improve. These may include:

- Generative AI-enabled portals to participation
- Participant upskilling and information gain
- Scalability
- Capturing the benefits of face-to-face discussion
- Facilitation
- Translation
- Transposition of deliberative processes into new contexts

The discussion will be held in September 2023 in the build up to the Global AI Summit. We welcome participation from OpenAI. The findings will support a provocation paper and public event to draw attention to the roles AI might play in the governance models and institutions of tomorrow.

### Provisional Project Timelines

Kickoff: July 2023

Track One, Snapshot 1: October 6. 2023
Track One, Snapshot 2: November 24, 2023

Track Two, Provocation Paper: October 6. 2023
Track Two, Event: October 20, 2023

### Project Outputs

We propose five project outputs. Outputs are designed to be forward-looking, solutions-focused and of use and value to AI development teams, civil democratic projects, and global decision-makers at the Global Summit on AI in London.

Outputs will be supported by public communications and briefings for key stakeholders.

1. The **proof-of-concept Recursive Public**: A digitally-connected, ongoing deliberation bridging international community members, experts and ultimately decision-makers.
2. A **value-based declaration**, made by participants, for AI Labs, thought leaders, and decision makers in the space, based on a Recursive Public snapshot.
3. A **5,000 word summary report** of the vTaiwan deliberations on routes to democratic participation in AI governance.
4. A **4,000 word provocation paper** building on the team's expertise in democratic deliberation on applying AI to the design of democratic governance processes and institutions.
5. An **online, public-facing livestreamed event** in Q4 2023 presenting the research and findings to a global policy audience. We are open to platforming other grantees at this event.

### Project Contacts

Alex Krasodomski and Yasmin Afina lead the Digital Society Initiative at Chatham House's work on new models of digital governance. Prior to Chatham House, Alex led CASM at Demos in digital policy work and implementation of deliberative technologies in the UK government.

Shu-Yang Lin, co-founded Public Digital Innovation Space (PDIS) in Taiwanese government. She is a long-term contributor to vTaiwan of g0v.

Carl Miller co-founded the Centre for the Analysis of Social Media at Demos and CASM Technology. He's a writer and journalist.

Flynn Devine's background is in participatory governance processes, working at different scales and in different countries. His specialty has historically been citizens’ assemblies, with a recent shift to online deliberation and participation.

---
### Other Information

Proposal Information:

1. How long has your team been working in the democratic / consensus-building space?

vTaiwan is an open consultation process that began in 2014 to deliberate on laws and regulations in Taiwan. Shu, an early contributor, later co-founded PDIS with Audrey Tang. Notably, vTaiwan played a significant role in hosting discussions on regulating Uber, which influenced decision-making and regolatory reform. Carl co-founded CASM in 2009, and with Alex led the design of a deliberative collective intelligence unit within the British government. Chatham House has stood at the forefront of democratic dialogue since 1920.

2. Please tell us in a few sentences about prior projects done by each of the team members that they are most proud of.

Co-founded PDIS in Taiwan government that re-architects channels for civic engagements, including establishing Participation Officers's Network for cross-ministerial policy-making. At the same time, collaborating with grassroots civic tech communities and contributing to vTaiwan on digital regulatory reforms, for example sharing economy, uncrewed vehicles.

Alex's work leading the strategy, design and roll-out of the UK's Collective Intelligence Network (COIN) was a key moment in the UK government's engagement with democratic deliberative tools. Since then, we have used these methods to help steer critical pieces of policy, including the UK's Online Safety Bill and the White House declaration for the Future of the Internet.

Flynn was a founding team member of the Global Assembly, a civil-society led, UN and UK government supported project that delivered the world’s first ever global citizens’ assembly. This was the first ever participatory process of this kind done at the international scale and successfully facilitated the presentation of the ‘The People’s Declaration for the Sustainable Future of Planet Earth’ to COP26. This involved collaborating with over 100 organisations in more than 50 countries.

3. Please list any other commitments (work/school) that each of the team members has until October 20, 2023

The project team will commit between 0.2 and 0.5 FTE each to the project, balancing delivery with ongoing professional commitments. Our delivery plan and timelines reflect this.

4. Are any team members covered by noncompetes or intellectual property agreements that overlap with this project? Will any team members be working as employees or consultants for anyone else?

The team is not covered by noncompetes. Reports and events will be open access, hosted and published by Chatham House under open license. Team members will be employed by their respective institutions.

5. What question(s) are you most interested in piloting? It can be from the list of options below or a different one that you decide to pursue. Questions should be decision relevant for developers, users and affected non-users of AI systems.

What principles should guide AI when handling topics that involve both human rights and local cultural or legal differences, like LGBTQ rights and women’s rights? Should AI responses change based on the location or culture in which it’s used?

6. Why did you pick this question (or questions) to work on?

This question is of significant policymaker interest, which aligns with our ambitions of public conversation around the project. Second, it lends itself to a deliberation following the vTaiwan methodology while testing the capacity of the system to surface consensus focus areas. Finally, it demands an internationally comparative conversation, the effectiveness of which this project is tasked with measuring.

7. Why do you think that this question (or questions) are well suited to broader public input? What do you think labs, developers or others might change as a result of input on these questions?

The answers to this question, either interpreted directly or through the proposed values-statement - can form inputs into some of the most fundamental challenges facing AI developers (and technology companies more broadly) in the rollout of generative AI tools, namely the extent to which different users should experience the tool in different ways. International agreement on technology regulation is sparse and slow-moving: in the short term, it is probable that national approaches will dominate. The answer to this question may go some way to informing developers of the Recursive Public's position on how a diversity of national approaches might impact their technoogy, and may support bridging international publics and moving the dial towards more global regulatory approaches.

8. Process overview: Please provide an overview of how the process that you envision building will work. Please touch on participant selection, topic overview, provision of additional context, content moderation, voting/commenting, aggregation of viewpoints, and provision of feedback to participants.

Main Proposal Goes Here.

9. Participant selection: How do you plan on obtaining a sample of participants for your experiment? How do you think about questions of representativeness and how they might matter for your question and method? Note: OpenAI can advise on methods or resources for obtaining a sample.

For this proof of concept, we propose to build out a sample of international voices through the international vTaiwan community and our own through international networks, including the global AI taskforce at Chatham House which brings together expertise from over 20 countries around the world. There are important trade-offs in selecting participants in a deliberation: a sample of new users frequently lack the skills to participate effectively over the short period for which the grant runs. We are opting instead to work with communities that are literate in deliberative processes, and working to test the transferability of the process among new publics. We are an agile team, and have set out mitigations to run the process with an alternative community should our preferred Taiwan public not meet expectations, and would be happy to work with OpenAI on deploying the method to an alternative sample if required, and have published public-facing policy on methods for representative use of deliberative technologies (https://demos.co.uk/wp-content/uploads/2020/12/Polis-the-Political-Process-NEW.pdf).

Diversity metrics will be reported on. Users will be set a short demographic survey before being connected to the Polis X-IDs. This method builds on work by Josh Smith at Demos (https://github.com/compdemocracy/polis/issues/526).

10. Tooling: Tell us about your plan for the tooling or infrastructure you’ll use for your experiment. Will you use existing tools or build new tools?

As part of the vTaiwan toolkit, we will utilize various existing civic tools to engage people in this decision-making processes.

We are planning to use:

* We will use Polis as our core deliberative platform: Polis is an established digital platform used for hosting online discussions and gathering collective insights asynchronously and internationally. It allows participants to express their opinions, engage in deliberation, and build consensus on various issues. We value the potential of Pol.is to onboard participants in deliberation, as well as it's design in encouraging bridging statements.

* Collaborative documents: collaborative documents, such as Hackmd or Etherpad, allow multiple participants to contribute and make real-time revisions, ensuring inclusivity and transparency. We plan to use collaborative documents when curating topics and co-editing seed statements.

* Streaming devices and platforms: we will integrate streaming services and platforms like YouTube to support educational material and to host online discussions.

* Online Email and Survey Tools

11. Limitations: What do you expect to be the biggest limitations of your approach? (e.g., potential for process gaming, types of questions your process would be unable to help answer)

We recognise three risks or limitations to the proposed approach.

First, that the experimental recursive deliberation may challenge the capacity of particpants to engage over the relatively short time. We mitigate this risk by ensuring the process properly briefs and trains participants in line with the vTaiwan process, and by focusing our efforts on a community with experience in deliberative participation.

Second, that operating deliberations internationally brings challenges in coordination. We have opted for asynchronous tools where possible, and will host online sessions twice to ensure all time zones can access the process.

Third, that our choice to focus on community leadership and the lack of a clear, government-mandated outcome may reduce the incentive for participants to take part in the deliberation. Should this impact our preferred sample recruitment, we mitigate this through flexibility and by opening up the process to a wider, international group of stakeholders in consultation with OpenAI. We will emphasise the potential impact of the work to participants to boost uptake further.

12. Resources: How would you plan to use the grant for your experiment?

In summary, the grant funds will primarily cover the time of the project team.

We anticipate approximately 80% of the funding will cover the costs of the technology and facilitation team, 15% will cover the costs of event hosting and digital publication and 5% will cover the costs of digital infrastructure. We are committed to using Free Software solutions to reduce overheads.

13. In your view, what are the top three benefits that AI technology brings to society?

- AI technology has the capacity to fundamentally reshape citizens' capacity to engage with democratic government as individuals and communities by mediating meaningful engagement in ways that would otherwise overwhelm traditional approaches to engagement.
- The scalability of AI governance solutions offers a tool to combat growing balkanization of technological governance and regulation. Collaboration by states on technical governance remains thin, and AI tools can strengthen both the ties between national publics, and the institutions charged with governing their technological landscape.
- Deployed responsibly and equitably, AI tools can empower good actors in much the same way Free and Open Source Software (FOSS) has done, strengthening the capacity of communities and civil society to deliver positive change for societies through the creation of new tools and services.

14. In your view, what are the biggest drawbacks or risks associated with the widespread use of AI technology?

AI will accelerate technological disruption in a significant number of fields:
- In work, AI is set to cause profound disruption to the labour market. AI will also affect and challenge current practices throughout employment, from hiring to performance assessment, at the expense of many workers and in particular those from a vulnerable background.
- In security and defence, AI will enable new means and methods of warfare and law enforcement that necessitate robust oversight mechanisms to ensure accountability and uphold applicable laws; without which risks of human rights violations and the commission of war crimes will increase dramatically. Risks of disruption will also emerge (e.g., through information hazards, disinformation campaigns, mass surveillance, etc.)
- In social justice, AI inherently carries both novel risks and will exacerbate longer standing, deeply-rooted ones (e.g., harmful biases against vulnerable populations).
- In international affairs, AI is set to play a central role in the great power competition. This carries profound polarisation risks and tensions akin to those during the Cold War and for 'middle powers' standing in-between.

Equitable AI - where the benefits of this new technology can be wielded by as wide a set of actors as possible - is essential.

15. What do you see as the most significant challenges in responsibly implementing AI technology, especially in the context of democratic decision-making systems?

Democratic decision-making systems are ancient machines. Innovation in their design and practice is often slow and the subject of public scrutiny. The major challenges therefore might include:

- A requirement around openness, auditability and legibility in deliberative systems to ensure public participation in the process and trust in the results.
- Governance institutions must ensure integration of AI-enabled systems factors in intersectional considerations; preventing situations where segments of the population are barred from such participatory, democratic processes (e.g., populations with minor technical literacy, those with little to no access to the internet, etc.)
- The resilience of AI systems used in democratic decision-making systems will be tested by non-democratic forces, and their resilience is paramount: this creates a web of dependencies around AI implementations that demand care and resources beyond the tool itself.

### Links

https://openai.com/blog/democratic-inputs-to-ai

https://openai.com/blog/democratic-inputs-to-ai#policy-statements-under-consideration

https://openai.com/blog/democratic-inputs-to-ai#application-review-factors