The Participatory Turn: Side Road or Highway?

This essay is part of a collection of reflections from participants in the Participatory AI Research & Practice Symposium (PAIRS) that preceded the Paris AI Action Summit. Read more from the series here.

“The default path of AI is not democracy.” This was the stark message delivered to hundreds of assembled guests on Monday, February 10, 2025, at a reception hosted by Make.org after the first day of the Paris AI Action Summit. The message that dominant AI technologies neither promote nor presently accept democratic governance is a powerful challenge for those of us who gathered in Paris to explore participatory approaches to AI.

At the Participatory AI Research and Practice Symposium (PAIRS) just a few days before, we heard from over 60 researchers and practitioners with powerful case studies of diverse publics involved in shaping emerging technology: from engineers co-designing locally appropriate AI tools with grassroots communities in Namibia to large scale efforts to engage citizens in setting AI strategies across Africa, or even sourcing collective input into the decision-making process at a tech firm at a global scale. But the ‘participatory turn’ that these initiatives represent risks being more of a side road than the highway unless we can build stronger evidence, shared advocacy, and communities of practice across those working to embed the voice of the public as the driving force in technological development.

Margins or mainstream?

The broad meaning of terms such as ‘democratic AI’ or ‘participatory AI’ has been widely examined. For some, democratizing AI can be about increasing access to tools, or building more diverse and ‘representative’ training data. For us, participation is an active process: one that involves individuals and communities having an informed, meaningful, and powerful say over technology. It is also essential: we cannot have another vast reconfiguration of our public services, workplaces, supply chains, and social lives without a legitimate mandate and democratic oversight and control.

It is rare to hear loud arguments against democratization of AI, except perhaps from certain quarters when democratization is wrongly conflated with open source. However, the activities necessary for making democratic power real are often sidelined with the claim that they would be too slow, complex, or costly to deliver, or because of biases that give undue weight to technical expertise in deciding how AI should shape society. Yet if even just a tiny fraction of AI development budgets were spent on engagement, public input could be delivered at speed and at scale. The evidence is clear that when invited, members of the public can make nuanced, informed, and effective decisions about AI.

In 2023, alongside the Bletchley Park AI Safety Summit, Connected by Data organized the People’s Panel on AI as a citizens-jury style review, working with 11 randomly selected members of the British public to learn about AI governance and propose actions for industry, academia, and government. The process demonstrated public input at the pace of AI governance, and how the public could and should be present in future summits. In the run-up to this Paris AI Action Summit, thousands of citizens input into an online consultation, developing robust proposals for the summit agenda. Over the week, a range of initiatives were launched that seek to demonstrate how public input could guide the future of artificial intelligence at scale, including through distributed dialogue and global collective intelligence platforms. Yet, these activities remained firmly on the margin. Moving efforts like these, and many other local and global participatory projects, to the mainstream is the challenge ahead and was a core motivator for convening PAIRS.

The value of critical friends

PAIRS was structured around three themes: development, governance, and resistance. We anticipated 30 or 40 submissions to our open call, but we were overwhelmed with more than 120 from across the globe, and over 800 people registering for our in-person and online gatherings, including from industry, international institutions, local governments, civil society and grassroots activism.

As in the open data movement, where individuals working with different goals and institutional logics found common ground in calling for a particular way of working, we see a growing collective movement for participatory AI. However, unlike the early open data movement, where differences of agenda, positionality, and power were often ignored, discussions at PAIRS facilitated the sometimes uncomfortable conversations between those working from the inside, and the outside, of sites of AI production and deployment.

Whether it was questioning if, and how, Meta’s governance team can translate experiments with large-scale engagement into meaningful change; reflections from Partnership on AI on the challenge of finding champions for participatory initiatives in firms; or discussion of how far independently organized local dialogues can deliver long-term and at-scale change, the symposium demonstrated the potential for our practice and our movement to be strengthened through dialogue and debate.

Building the field

It is clear, however, that developing and sustaining this dialogue will take effort. In a week surrounded by glitzy tech company receptions, and high-level panels, PAIRS was organized by a volunteer team with a budget of under $5,000. The challenge ahead is to build on the foundations laid: continuing to develop a community of practice by sharing learning across contexts; supporting advocacy by developing a robust evidence base on the feasibility and impact of public engagement around AI; and building the movement by supporting solidarity and collective action.

Being part of public engagement is both a grounding and energizing experience. Working with communities to explore visions of a better future can be as exhilarating as working on the cutting edge of machine learning. But where systemic impacts are often the unplanned side effect of AI development, securing systemic impact from public involvement takes hard yards. To advance that work and shift the default path of AI, it takes community. Building and sustaining that community is central to the work ahead.