Skip to content
Citizen Infrastructure

Five dimensions of scaling deliberation

People say AI will let us do deliberation “at scale” — but scale of what? More people? Bigger decisions? More impact? A 2025 paper by Sammy McKinney and Claudia Chwalisz (DemocracyNext) argues the phrase is too vague to be useful, and breaks “scale” into five distinct dimensions so practitioners can say which one a given tool actually serves.

The five dimensions

Section titled “The five dimensions”
  • Scaling out — more citizens in a single process. Directly (deliberative polls now run with thousands, not tens) or indirectly (letting the wider public feed in stories and perspectives). AI here: facilitation across many groups; Polis- or Cortico-style sensemaking of mass input.
  • Scaling up — to higher levels of governance: transnational and global deliberation on issues that cross borders. AI here: translation across languages.
  • Scaling across — more deliberative processes across a system: more issues, and beyond government into workplaces, schools, and cultural institutions. AI here: an “Assembly Assistant” that helps organisers through the repetitive set-up steps.
  • Scaling deep — more impact: institutionalising assemblies and connecting them to the public sphere, so they shift real power instead of being tokenistic. AI here: clearer public communication; more implementable recommendations.
  • Scaling in — higher quality of deliberation within the process: better facilitation, more inclusive learning materials, surfacing perspectives that are missing from the room.

Manon Revel (Meta FAIR / MIT) proposed a sixth — scaling within — for the private, reflective moment where someone reconsiders their own view before deliberating, after Robert Goodin’s “deliberation within.”

The warning: headcount can hollow it out

Section titled “The warning: headcount can hollow it out”

Most of the field’s energy goes to scaling out by headcount — mass online platforms and AI facilitators. McKinney and Chwalisz caution that this can hollow out the very thing that makes deliberation transformative: the slow, relational work of building trust, owning the agenda, and bonding between sessions (the conversations over coffee, the walk between venues). :::caution[Scale is not legitimacy] Numbers are only one source of legitimacy, alongside fairness, being well-informed, and sortition. Equating “more people” with “more legitimate” is an inheritance from electoral thinking, not a law of nature. :::

Two more claims follow:

  • “Cheaper” often hides a cost. AI deliberation is frequently only cheaper because participants aren’t paid and deliberate for far less time.
  • No single dimension is free. Push hard on one and you trade off against another, so a holistic view of scale should guide which tools to build.

Not a technological fix alone

Section titled “Not a technological fix alone”

Their second contribution: scaling deliberation is not susceptible to a technological fix. It needs civic infrastructure — the relational, time-intensive work of connecting assemblies to power and building the networks that sustain them. Their leading example is Arantzazulab, a democracy-innovation lab in Spain’s Basque Country (launched 2020): five years ago the region had no assemblies; today it has them at local and provincial levels, and Arantzazulab is even bringing sortition into the Mondragon cooperative.

A related provocation from discussant Oliver Escobar (Edinburgh): build deliberative principles into the AI itself. We tend to treat AI as an oracle we ask for answers; a deliberative AI would behave more like a facilitator — asking as many questions as it answers, helping make connections, eliciting friction when a shallow consensus is forming. For the deeper tool-by-tool test, see can AI scale deliberation?; for the risk of fake stand-ins, synthetic participation.

Sources

Section titled “Sources”
  • Sammy McKinney & Claudia Chwalisz, “Five Dimensions of Scaling Democratic Deliberation: With and Beyond AI,” DemocracyNext, June 2025: demnext.org
  • Paper-launch webinar (with Oliver Escobar, Kyle Redman, Manon Revel; Office of Eric Schmidt), 2025: youtube.com/watch?v=mOfvIfg2XlE
  • Sammy McKinney, “Integrating artificial intelligence into citizens’ assemblies,” Journal of Deliberative Democracy, 2024.