How could lower-cost AI facilitation expand human-to-human deliberation while keeping participation available to people facing literacy, access, or technology-wariness barriers?

Contribution to the handbook

Priority · Assessed 2026-10-30

The question advances the handbook by specifying participation routes, comparing AI-assisted, human and mixed facilitation, and testing whether people are heard. Potential savings and exclusion are consequential but not established outcomes [draft-proposal-9-c216d06f370a; s-landemore-inclusive-democracy-1bdcc292ae8336fe; s-landemore-inclusive-democracy-ed0922d39d52bc60].

Does this challenge the goal?

In-scope gap

The question seeks practical ways to make deliberation accessible and assess its costs and reach within the existing goal. Its human-discussion boundary is a choice for this inquiry, not a challenge to the exploration definition.

The best participation arrangements and their effects remain open.

Making and interpreting laws →

Authority, accountability and resilience →

Implementation and transition →

Tests, stop conditions and unresolved work →

Compare total facilitation costs and achievable reach under AI-assisted, human, and mixed arrangements, accounting for accessibility support; distinguish projected savings from tested results.

What we can say so far

A qualitative comparison is possible, but not a comparison of total costs or achievable inclusive reach. Human-facilitated discussions incur facilitator time and can be slow; established small-group formats can nevertheless convene dozens or thousands. AI assistance might reduce some facilitation or synthesis work, potentially making larger processes affordable, but the proposed savings are not costed or tested at scale. A mixed arrangement could use AI to assist human-led discussions while retaining human inclusion, translation and communication support; those services would still need to be included in its total cost. Remote phone or video meetings can reduce travel barriers, but digital participation can exclude people with low literacy or technology wariness. Supporting different communication abilities and active inclusion may improve availability, though the supplied sources do not cost or test particular combinations. Larger or faster discussions can also make it harder to hear and understand different perspectives, so nominal headcount is not the same as inclusive reach. One study reported AI-generated group statements in seconds versus several minutes for human mediators. This is a task-time result, not a total-cost or population-scale comparison. Its proposal to support groups of hundreds was theoretical and dependent on models handling longer texts.

What remains open

• Comparable total costs for human, AI-assisted and mixed arrangements, including accessibility support.
• Measured inclusive reach and outcomes for each arrangement at comparable scale.
• Evidence on which accessibility combinations work for particular groups and what they cost.
• Evidence that proposed savings and inclusive facilitation persist at large scale.

Specify nonexclusive routes for people to contribute and be heard without reading well, having direct technology access, or being comfortable with the tool, and explain how their contributions enter the same human discussion.

What we can say so far

A possible combination is to offer in-person small-group conversation alongside phone participation for people who have phone access. A human facilitator can support turn-taking, listening and translation, and take notes so speaking does not depend on reading or using AI. People wary of the tool could contribute to the human discussion without operating it themselves. The chapter describes sharing and synthesizing notes across small groups; it also describes carrying permitted highlights from recorded conversations into further discussions. To make that wider discussion accessible, a facilitator could read shared points aloud and invite participants to correct them. That oral return-and-correction step is a proposed extension, not a procedure established by the supplied excerpts. The excerpts do not specify an end-to-end process that ensures each route is accessible and contributions are faithfully carried back into the same discussion.

What remains open

• A source-specified, accessible procedure for returning shared contributions orally and enabling participants across routes to check and correct them.
• How organizers obtain permission, check shared highlights, and preserve viewpoints when combining contributions.
• How to select and resource the combination of participation practices for particular groups and settings.

Test who joins, contributes, is heard, and remains involved across affected groups, rather than treating enrollment, statement production, or faster discussion as sufficient evidence of inclusion.

What we can say so far

The supplied evidence does not establish whether AI facilitation improves inclusion across affected groups. In one study, participants preferred AI-written statements in 56% of comparisons with human-written statements, but the processes differed and that result does not show who joined, contributed, was heard, or stayed involved. A test should compare these outcomes across affected groups, including people facing literacy, access, or technology-wariness barriers, and against an appropriate non-AI deliberation process. It should examine whether participants’ views are reflected in discussion and outputs, including views that remain in disagreement. Faster statement production or enrollment alone would not demonstrate inclusion.

What remains open

• Comparative evidence on who joins, contributes, is heard, and remains involved, disaggregated across affected groups.
• Evidence on whether particular practices reduce barriers for people with limited literacy, access, or trust in technology.
• Evidence on whether minority or divergent views remain visible and are heard as AI-assisted deliberation scales.

Maintain AI’s facilitation role as distinct from simulated people speaking in participants’ place; state how the process identifies whose contributions are human.

What we can say so far

The process should expand discussion among people, with AI facilitating rather than simulated people speaking in participants’ place. Landemore’s account makes that distinction. The supplied material does not specify how the process would identify which contributions are human; that identification procedure remains an open design question. An alternative proposal envisions AI representatives and faster model-based exchanges, but that goes beyond this human-discussion boundary.

What remains open

• A concrete process for identifying which contributions are human is not supplied.