Collectives and what makes an assembly?
Five different ingredients pour into my reflections for this set of week notes:
(1) Over the last few weeks I’ve been iteratively working on a set of resources that try to distil some of our learning from our Community Campaigns on Data project. One of the key elements I’ve been coming back to (in particular, thanks to insights from a chat with Dirk Slater on Tuesday) is that our goal is not just supporting effective campaigns, but also rooting them in community and collective action, and the construction of ongoing collective power.
(2) Thanks in both cases to BlueSky posts from James Plunkett, my reading this this week has been enriched by a wonderful collection “What Makes An Assembly?”, looking through a set of theatrical, architectural, political, sociological, and distinctly French lenses at the history and present of political and citizens assemblies, particularly following the revolutionary moments of Tahrir Square, and Occupy; and by picking up Ostrom’s Governing the Commons: The Evolution of Institutions for Collective Action
(3) On Friday I was interviewed about experience of the People’s Panel on AI by Mhairi Aitkin for PVAI (Public Voices on AI) research on different approaches to participatory engagement with artificial intelligence. It was a good opportunity, now almost a year on, to reflect on learning from the process (documented here, and in the independent evaluation), and in particular got me thinking about questions of visibility: making citizens visible within decision shaping and making spaces.
(4) On Saturday I had the opportunity to observe Meta’s latest Community Forum on Generative AI, run as a deliberative poll using the Stanford Online Deliberation platform. I’ll write-up more on the substance of this soon (observers agreed not to publish detail of the deliberation until after the official report is out), but will note it was fascinating to be able to tune into many of the 80+ online rooms of live discussions between small groups of citizens connecting from their kitchens, living rooms, markets and workplaces in five different countries. There was something quite striking about this moment of parallel conversations at scale: emotional to observe.
(5) Today I’ve been reading this paper in Science which purports to show the potential of LLMs as a tool for scaling citizens assemblies, and finding common ground in democratic deliberation. Digging behind the headline and summary claims that the an AI model outperformed ‘human mediators’ (only later on explained to actually be non-professional randomly selected crowd-tasking participants), and that the approach was tested in a ‘virtual citizens assembly’, we find that assembly was in fact 3 x 1 hour online sessions, with no education phase, covering many different topics, and with no direct conversation between participants at all (their interaction, as far as I can tell, mediated solely through writing and voting on text based statements). Not least when, this week, FIDE North America have published their guiding principles for Citizens Assemblies, emphasising the importance of adequate time and diversity of information, the appropriation of the language of citizens assembly for a process of ‘caucus mediation’ in small groups alone, is particularly problematic.
Visibility and collectivity
Drawing on the architectural insights of What Makes An Assembly (the book is full of photographs of different gatherings, and different proposals for how to physically configure an assembly space), I reflect that in neither the ‘caucus mediation’ described in the Science paper, nor in Meta’s Community Forum, are participants deliberating or engaging with decision making as members of a larger collective. They have, ultimately, little visibility on the wider group of participants they are part of.
As an observer on the Meta process, with the privilege of dropping into each of the parallel discussion sessions, I could roughly gain a sense of the totality of participants, but individuals taking part could only see the 5 - 10 people they were put into a discussion room with. Contrast this to the powerful images from the Global Assembly on Climate and Ecological Emergency of a zoom room bringing people together across the globe invited to deliberate as a collective and given visibility across the whole.
With growing interest in technologies to allow participatory processes to scale, we should think critically about whether processes claiming to act as a forum, or assembly, are actually providing a shared meeting space for participants, or simply a mediated aggregation of individual or small group experiences.
Building community & collective action
Going back to the Collective Power Playbook’s play template, and in particular, the set of play finder outcomes it references, provided a bit of a breakthrough as I looked to sequence materials in our draft Community Campaigns on Data resource. Recognising the need to consider:
- Outcomes for the group (trust, solidarity, relationships, team cohesion);
- Outcomes for the project, topic or movement (contribution, co-creation, collaboration); and
- Outcomes that create momentum (lasting connection, social learning)
was really useful to think about both the prototype learning activities we need to document, and the resources we should signpost out to for anyone thinking about how to run a community campaign around data. Just as our reflections of our Campaign Cohort have highlighted the value of bringing people together to learn from each other, and simply feel they are not alone in their campaigning, I find I need to regularly return to the importance of understanding campaigning (and participatory) processes as not just about an end-goal, but also as about a process of building understanding, empathy, solidarity and shared progress.
Other bits
- I’m on the look out for case studies of campaigns that have successfully secured a place on data governance structures for community representatives. While I’ve come across cases of bodies deciding to include affected community representatives, I’ve not yet found a good example of this happening because of an organised community-led campaign for this to be the case. Any leads or ideas - do drop me a line .
- Tickets are now available for Gloucestershire Data Day #2 which will take place with expanded capacity for up to 250 participants at the fantastic Parabola Arts Centre in Cheltenham. We’re on the look out for a few more event sponsors who could help support bursary places to let us remove barriers to participation, artists commissions and webcasting of the event. Details here or again, do get in touch for more info .
- I enjoyed learning about Sense about Science’s Responsible AI Handover Framework on Tuesday: highlighting the importance of the procedures by which AI technologies are handed from researchers, to developers, to deployers. As they describe: “The Responsible Handover framework provides all the important prompts to script a handover conversation at whatever level of detail is necessary, when a tool passes from one person or organisation to another” emphasising transparency not as a matter of documentation, but as a matter of conversation, responsibility and accountability between parties. Although developed in context of biomedical and health AI research, lots relevant here for work around procurement of AI in the public sector here too.
- I had the pleasure of chatting to Gregory Metcalf from Children’s Parliament about their work involving children in dialogue around AI and was excited to hear from Mhairi Aitkin about the Children’s AI Summit planned for Feb 4th 2025 in London, just ahead of the Paris AI Summit.
- I put in an initial response to the AI Summit Expert Consultation from Connected by Data, mainly pointing to the importance of embedding public voice, and to our developing Participatory Governance of AI Symposium plans.