Contribution to the handbook
Priority · Assessed 2026-10-27
The records distinguish a human-mediator comparison from a virtual citizens’ assembly and describe what those settings tested. They also expose consequential limits: the human comparison differed procedurally, and the assembly omitted fact-finding and expert-testimony stages. This materially informs how evidence should be interpreted and what a fuller governance-relevant evaluation would need to cover.
Does this challenge the goal?
In-scope gap
The question asks about evaluation settings and their limits, directly supporting the handbook’s existing interest in testing and implementation without challenging its goal.
The evidence describes UK study settings and reported outcomes, but does not establish transfer to other populations, countries, or governance functions.
Why this category?
- testing
- implementation
- consequential risks and limits
These records are interpretations of one shared article source, not independent replications. The results concern the reported UK study conditions; they do not demonstrate effectiveness for city- or country-scale governance.
This is an AI assessment of contribution to the goal, separate from whether the question has an answer.
Tests, stop conditions and unresolved work →
Implementation and transition →
Authority, accountability and resilience →
Read the exploration goalDescribe the evaluation settings, including comparison with human mediation and the virtual citizens’ assembly, and what each was intended to test.
What we can say so far
The authors tested the mediator in two relevant settings. First, they compared AI-generated group statements with statements written by human mediators: in 75 groups of six, participants preferred the AI statements in 56% of comparisons and rated them as more strongly endorsed and higher in argument quality. The comparison was intended to test how the AI’s group statements performed against human mediation. It was not a fully matched comparison: the human mediators did not have the AI process’s critique phase. Second, they used the deliberation procedure in a virtual UK citizens’ assembly with 200 participants recruited to reflect the UK population. This was intended to test whether findings from the earlier, crowd-sourced groups also appeared in a broader, population-oriented sample. Groups used the same deliberation procedure; they discussed nine selected questions over three weekly sessions. The assembly mainly reproduced the discussion phase and omitted usual fact-gathering and expert-testimony stages, so it was not a test of a complete citizens’ assembly process.
What remains open
- The supplied evidence does not establish whether these results generalize beyond the UK or to complete citizens’ assemblies with fact-finding and expert testimony.
Conditions and limits
- The human comparison did not match the AI and human processes in every respect: human mediators did not have a critique phase.
- The assembly sample was recruited with demographic representation in mind; the supplied evidence does not establish that the realized sample was representative.
- The assembly reproduced mainly the discussion phase, not the fact-finding or expert-testimony stages of a typical citizens’ assembly.
- These are reports of the authors’ study and design, not independent verification.
Objections to consider
- The answer reports the assembly’s population-oriented recruitment without treating intended demographic representation as proof that the realized sample was representative.
- The answer explicitly states that the assembly mainly tested the discussion phase and omitted fact-gathering and expert testimony.
- The answer limits the assembly inference to the broader UK sample and does not claim generalization to other countries or settings.
- The answer states that the human-mediator comparison was not fully matched because human mediators lacked the critique phase.
Evidence and related ideas
Supporting evidence, qualifications and context play different roles in this answer.