On 24th May we organized a workshop in Brussels to explore design options for an inclusive global assembly addressing AI, as part of our design lab building off the Global Citizen Assembly Coalitions challenge paper.
The slides used to introduce the workshop are here and below, including four mini case studies of existing examples of global or AI focused deliberation, looking at the approach that each has taken to participant recruitment, learning, deliberation and ‘docking’ into other processes (to draw on a term from the Global Assembly on Climate’s independent evaluators).
Four example processes
Global Assembly: transnational representative mini-publics engaging in deliberation towards shared position, feeding into multilateral process.
We the internet: distributed dialogues with common and locally adapted elements, generating common evidence and local action, feeding into multistakeholder forum.
Meta community forum: deliberative polling with top-down questions and statistical disaggregation to understand naive and informed public perspectives, feeding into product teams.
Anthropic collective constitution: bottom-up solicitation of statements and perspectives to identify clusters of consensus/conflict feeding directly into AI model training.
By examining these four processes, we were able to see the diversity of ways that global deliberation can be designed: from bringing people into live transnational conversations with interpretation, to hosting parallel local dialogues, or using technology to mediate asynchronous interaction around top-down or bottom-up statements of policy.
Over two short hours we had an incredibly rich discussion, leading to a number of the points below shared in our discussion and closing reflections round*:
- Importance of Narrowing Topics: Choosing a specific and focused topic is crucial for effective discussion and deliberation.
- Global Representation and Recruitment: Ensuring global representation across educational, socioeconomic divides, and gender is necessary for meaningful citizen assemblies. Consider multiple mini assemblies or a federal system to bring diverse voices together.
- Stakeholder Involvement: Include a wide range of stakeholders such as economic actors (OECD, UN nation states, continental unions), tech giants, technicians, and operators. However, avoid giving control or undue influence to any single group.
- Learning and Curriculum Setting: Carefully consider who provides the inputs and oversees the learning process. This could include stakeholders from different knowledge groups, but avoiding bias is essential and for some topics such as LAWS (Lethal Autonomous Weapons Systems) may involve offering some contributors anonymity.
- Output and Impact: Focus on creating concrete outputs and impacts from the deliberation process. This includes committing to varying levels of action and publicly committing to the outcomes.
- Technology and AI Integration: Utilize AI to aggregate ideas and facilitate global discussions. AI can help bring together perspectives from different discussion groups and highlight trending topics.
- Human Interaction: Maintain the element of human interaction in deliberations. Digital assemblies can dilute the human aspect, so consider in-person meetings or hybrid models when possible.
- Legitimate Actors and Power Dynamics: Address the contrast between who has the power in AI conversations and who is more directly affected by the technology.
- Citizen Empowerment: Make citizens feel heard and empowered by including their lived experiences and personal stories. This helps in making the outcomes resonate with the citizens themselves.
- Practical Outcomes: Ensure the outcomes of citizen assemblies are practical and actionable, not just theoretical discussions.
- Capacity Building: Focus on building community capacity to be part of the debate. Long-term impact on participants is significant, helping them become more informed and engaged.
- Operationalizing Impact: Find concrete ways to operationalize the impact of the assemblies, such as influencing global polling and surveys on AI-related issues.
- Continuous Engagement: Discussions should lead to action. Follow up on these conversations with concrete steps and ongoing engagement to maintain momentum and achieve results.
- Legislative Theatre: Use methods like legislative theatre to combine lived experiences with expert perspectives, making the discussions more relatable and impactful.
- Funding and Logistics: Address funding and logistical challenges for bringing diverse voices together, including travel and accommodation for participants.
- Integrating Diverse Voices: Think beyond just data extraction from participants. Empower them to understand and engage with technology, potentially leading to them producing their own scholarly work.
- European AI Office: Consider integrating a permanent citizens assembly as a subgroup within the advisory forum of the European AI Office, leveraging existing structures for broader impact.
- Cultural Sensitivity: Recognise the importance of cultural sensitivity and inclusivity in framing discussions and recruiting participants, avoiding a solely Western-centric perspective.
- Legitimacy and Ethics: It is important to consider who has the legitimacy to participate in and oversee discussions.
Thanks to Anthony Zacharzewski of DemSoc for kindly hosting as at BeCentral and to all the contributors who shared their insights. We will be sharing details of future workshops for the GCA on AI design lab soon.
Workshop participants: Anthony Zacharzewski (DemSoc), Connor Dunlop (Ada Lovelace Institute), Gryffin Wilkens-Plumley (Missions Publique), Laura Sandor (The Data Tank), Tim Davies (Connected by Data) (facilitator), Polen Türkmen (The Data Tank), Raegan MacDonald (Aspiration Tech), Sophie Tomlinson (Datasphere Initiative)
Disclosure: On-device text transcription from a Google Pixel phone, and ChatGPT GPT-4o image recognition of post-it notes was used in preparation of these notes. Notes were cross-checked by Tim Davies on the day of the workshop.