Contribution to the handbook
Supporting · Assessed 2026-10-27
The private collection of views and intended group statement offer useful background on a possible deliberation tool. That mechanism alone does not specify how governance decisions are made or establish that the tool is suitable for them.
Does this challenge the goal?
In-scope gap
The question asks about a deliberation mechanism that could inform citizen input; it does not contest the exploration’s goal or beneficiary boundary.
The evidence describes help writing a group statement, not AI choosing policy or exercising government authority.
Why this category?
- system mechanism background
The article-derived accounts share a source root and are not independent evidence. An intended group-acceptable statement does not establish fair acceptance or explain how rejection is handled.
This is an AI assessment of contribution to the goal, separate from whether the question has an answer.
Making and interpreting laws →
Authority, accountability and resilience →
Read the exploration goalDescribe the AI mediator’s private collection of individual views and its proposal of a group-acceptable statement, including the caucus-mediation approach.
What we can say so far
The researchers developed a private group mediator: it collected each participant’s views individually, then proposed a group statement intended to be acceptable to the group. They describe this private, individual exchange as a caucus-mediation approach.
What remains open
- The supplied account does not explain how the system handles participants who reject its proposal.
Conditions and limits
- The statement is described as intended to be acceptable; this does not establish that participants accepted it or that it represented every participant fairly.
- The account is based on an AI interpretation of the readable article copy, not independent new evidence; supplementary materials were not included.
- The system is described as helping write a group statement, not choosing policy or governing the group.
Objections to consider
- The selection details qualify any implication that the chosen statement was necessarily fair or acceptable to everyone. The answer describes the statement only as intended to be acceptable, consistent with the direct mechanism account.
Evidence and related ideas
Supporting evidence, qualifications and context play different roles in this answer.
supports
A private group mediator The researchers developed an AI mediator that privately collects each participant’s views before proposing a statement intended to be acceptable to the group. They call this private, individual exchange caucus mediation.
Directly supports the private collection of each participant’s views, the proposed group statement, and the caucus-mediation label.
AI can help humans find common ground in democratic deliberation
Read full source: AI can help humans find common ground in democratic deliberation (opens in a new tab) ↗Read source discussion: AI can help humans find common ground in democratic deliberation →
qualifies
Selecting a group statement The mediator generated 16 candidate statements. For each participant, the reward model used their opinion and critique to predict a ranking of those statements. The system combined the predicted rankings using the Schulze voting method and returned the winning statement; the authors repeated the process to provide multiple samples in the main evaluations.
Adds that candidate statements were selected by combining predicted participant rankings; cautions that the selection procedure does not establish fair or accurate representation.
AI can help humans find common ground in democratic deliberation
Read full source: AI can help humans find common ground in democratic deliberation (opens in a new tab) ↗Read source discussion: AI can help humans find common ground in democratic deliberation →
context
Learning preferences The authors trained a 1.4-billion-parameter Chinchilla reward model with a linear layer that produced a score for each candidate statement. It learned from participants’ statement rankings and quality ratings; its scores were designed to reflect how much each participant was expected to agree with a statement.
Describes reward-model training signals, but does not establish prediction accuracy and is not needed for the requested description.