Public deliberation can’t be an afterthought in AI governance
Read key insights from Claudia Chwalisz’s recent address at the FARI AI Conference in Brussels
Why AI needs democratic (not just technocratic) governance
“Scientific knowledge is an enabling power to do either good or bad — but it does not carry instructions on how to use it.”
Claudia opened her remarks at this year’s FARI (AI for the Common Good Institute - Brussels) conference with this quote by physicist Richard Feynman in 1955, reflecting on the atomic bomb.
Seventy years on, the same dilemma sits at the heart of AI debates. Except this time, the technology is advancing faster, reaching further, and shaping more areas of human life than any innovation before it. And today, the people building these systems are often the same ones attempting to write the rules.
Claudia’s message was clear: AI governance cannot be left to technocrats alone. Democracy isn’t a nice-to-have. It’s fundamental.
Three reasons this is a democratic question
1. Complexity with consequences
AI may be technical, but its applications cut to the core of public values: rights, justice, dignity, and the kind of society we choose to build. Whether to deploy facial recognition in schools, use predictive policing, or automate life-shaping decisions are not engineering questions - they’re democratic ones.
2. Unprecedented speed
AI has diffused faster than any technology in history. ChatGPT reached 100 million users in two months. Governance cannot lag years behind deployment; by the time debate catches up, norms are already shaped by default.
3. Knowledge asymmetry
Expertise is concentrated among a handful of private firms. When those developing AI also hold the technical knowledge required for oversight, we face a democratic deficit. The public and governments lack the tools to participate meaningfully - while commercial actors write the terms of their own regulation.
Three ideas for a more democratic path forward
1. Public deliberation in AI labs - promising, but constrained
Labs such as OpenAI, Anthropic, and Meta have experimented with public input. These efforts matter: they broaden participation and set new precedents.
But Claudia highlighted three limitations:
Agenda-setting remains with the labs. Key questions are often off the table.
Power stays advisory. Public input does not bind decision making.
Process gaps remain. Digital-only formats, narrow samples, filtering out people without AI knowledge, and unclear impact fall short of established democratic standards.
The central question she posed is:
Can deliberative processes commissioned by the very organisations they’re meant to govern ever constitute genuine democratic oversight, or are they sophisticated forms of legitimacy-seeking?
These experiments should be seen as important building blocks, she argued, but not the final architecture of democratic governance.
2. AI localism over globalism
While global bodies produce high-level, non-binding principles, the most practical governance is emerging locally.
Research Claudia co-authored with Stefaan Verhulst shows why local and regional governments often succeed where global efforts stall:
Context-specific expertise
Trust rooted in proximity
Democratic legitimacy through citizens’ assemblies
Ability to test, iterate, and adapt quickly
More than 150 examples of “AI localism” are already documented worldwide - from municipal algorithmic impact assessments to participatory processes guiding the use of AI in local services.
This shouldn’t replace international coordination, Claudia noted, but enrich it. The most grounded, legitimate governance may come from the bottom up.
3. A new social contract for data
Finally, Claudia addressed a growing contradiction: the open data movement - once celebrated for democratising knowledge - has been repurposed. Public repositories, Wikipedia, academic archives, and government portals are now scraped at scale to train large AI models controlled by a few corporations.
As Stefaan Verhulst has compellingly argued, this has triggered a “data winter”: institutions retreating behind paywalls to avoid exploitation.
Drawing on Verhulst’s work, Claudia advocated a shift toward a reinvented data commons, built on:
Governed access rather than default openness
Benefit-sharing for contributors
Participatory oversight reflecting community interests
Technical accountability like provenance tracking
This is not an abandonment of openness, she stressed, but an evolution toward a civic, equitable data ecosystem.
Preserving democratic agency as we innovate
Claudia closed where she began - with Feynman’s reminder that we must “leave the people of the future a free hand.”
AI governance isn’t about slowing down innovation. It’s about making sure that as we innovate, we preserve democratic agency – the ability of communities to shape the technologies that shape their lives.
The experts can tell us what’s possible.
But we need public deliberation to tell us what’s desirable.
These are questions that we’re thinking about and interested in exploring further at DemocracyNext. Please get in touch with Claudia if you’d like to partner or support our work on AI governance.
📡 On the radar
Writing in The New York Times, Eric Schmidt and Andrew Sorota warn that deploying AI in government and politics isn’t inherently progressive - unchecked, it could accelerate authoritarian-style control rather than strengthen democracy.
The New York Times also covered the citizens’ assembly in Ostbelgien, which has quietly become a testing ground for rethinking democracy - its “citizen-by-lottery” council turns everyday residents into real contributors to regional policy.
Liverpool’s Children’s Parliament (Australia) marks 10 years of giving 9–12-year-olds a real seat at the table - shaping policy on parks, safety, schools and more.
The Wire write about Zohran Mamdani’s approach and how he reframes politics as an act of listening first, treating conversations in living rooms and community centres as the real engine of representation.
The Urban Activist interview Simona Levi, who suggests that Europe’s push for digital sovereignty should put citizens back in the driver’s seat when it comes to control over data, infrastructure and democratic innovation.
Lakshmi Venugopal and Shrishtee Bajpai highlight that our democratic obligations shouldn’t end at people - we need to expand representation to include non-human species, treating them as part of our shared democratic community.
🍂 Events
23 & 24 January 2026, Berlin, Germany
Claudia will be hosting a workshop with Kyle Redman from the AI & Democracy Foundation on deliberative technology during the Political Tech Summit.
6-7 May 2026, Massachusetts, USA
Claudia will deliver a keynote address at the Third Annual Symposium on Constructive Communication at the MIT Media Lab. This year’s theme is Trust & Human Connection in the Age of AI. Audrey Tang, a member of our International Advisory Council, will also be speaking as a keynote.
7-10 May 2026, Athens, Greece
Claudia will be speaking at the World Beautiful Business Forum, the most human gathering for the more-than-human world, next year. As the House of Beautiful Business turns ten in 2026, the event will explore how to shape a humanist future.
Maybe use full for this work: how the City of Amsterdam used a citizen & narrative based approach to compose its vision on AI https://openresearch.amsterdam/en/page/108899/using-socratic-design-to-build-amsterdam-s-vision-on-ai-with-citizens
A incluse form of sensemaking which i think is a strong deliberative democracy approach