AI Impact Assembly
The AI Impact Assembly is Fung's proposal for empowered participatory governance at the municipal level. It adapts the citizens' assembly format (Ireland, France, Canada) to the specific challenges of ongoing AI governance. The Assembly is standing rather than ad hoc, reflecting the continuous nature of AI governance; members serve staggered terms (approximately eighteen months, one-third rotating every six months) to balance institutional knowledge with fresh perspectives. It has access to a dedicated technical translation team whose function is not to simplify but to connect AI's technical features to their community implications. And it possesses formal investigative powers — authority to request information from companies operating within the municipality, hear testimony from affected workers and communities, and commission independent research. These powers give the Assembly genuine governance weight within the municipal system.
In the AI Story
The design responds to three specific challenges AI governance poses to deliberative institutions. The speed constraint: AI capabilities advance on timescales of months, requiring standing bodies rather than ad hoc assemblies convened in response to specific decisions. The complexity constraint: AI governance questions involve technical content that requires translation rather than simplification. The scope constraint: AI deployment affects populations in ways that span jurisdictional boundaries and require connections between local and higher-level governance.
The Assembly's standing structure addresses the speed constraint. A body that maintains continuous engagement with the governance landscape can provide deliberated input at the pace AI decisions demand, without sacrificing deliberative quality. The staggered-term design ensures both institutional continuity (one-third of members remaining from previous terms) and regular infusion of new perspectives (one-third rotating in each term).
The technical translation team addresses the complexity constraint. Rather than requiring participants to become AI specialists, the team connects technical content to community implications — the same principle that made Porto Alegre's participatory budgeting accessible to residents who were not public finance experts. The team's role is translation, not advocacy: connecting capability to consequence in terms accessible to non-specialists without distorting substantive accuracy.
The investigative powers address the consequence condition. An assembly with no authority to demand information, hear testimony, or commission research cannot produce the substantive deliberation that distinguishes genuine governance from theater. The powers are not advisory — they are institutional capacities that give the Assembly genuine weight within the municipal system. The design assumes that the municipality, not national or international authority, is a viable starting point for participatory AI governance: municipalities can create such bodies without national legislation, generating the demonstration effects that drive broader diffusion.
Origin
The Assembly design synthesizes elements from multiple precedents: the Irish citizens' assemblies (standing format and sortition methodology), the French Citizens' Convention on Climate Change (policy focus and technical translation), the Paris climate assembly (ongoing engagement), and Porto Alegre's participatory budgeting (municipal scope and binding authority).
Fung's specific adaptation for AI draws on the Ash Center's workshop findings (December 2024) and the WeBuildAI project (2019), which demonstrated that participatory frameworks for algorithmic governance could produce outcomes reflecting considerations that expert-only governance had systematically neglected.
Key Ideas
Standing rather than ad hoc. Continuous engagement addresses the speed constraint that would otherwise disqualify deliberative governance from AI-scale decisions.
Staggered terms balance continuity and renewal. Institutional knowledge accumulates while fresh perspectives prevent ossification.
Translation, not simplification. The dedicated technical team connects technical content to lived consequence without distorting substantive accuracy.
Formal investigative powers. Genuine governance authority — to demand information, hear testimony, commission research — converts the Assembly from advisory body to institutional participant.
Appears in the Orange Pill Cycle
Further reading
- Hélène Landemore, Open Democracy (Princeton University Press, 2020)
- OECD, Innovative Citizen Participation and New Democratic Institutions (2020)
- Min Kyung Lee et al., "WeBuildAI: Participatory Framework for Algorithmic Governance" (Proceedings of the ACM on Human-Computer Interaction, 2019)