Contribution to the handbook
Priority · Assessed 2026-10-27
The distinction between optimizing for endorsed statements and changes in participants’ views is consequential for system objectives, safeguards, and testing. The evidence also cautions against assuming that training alone caused observed changes.
Does this challenge the goal?
In-scope gap
The question distinguishes the system’s endorsement-oriented training goal from observed shifts in participants’ views, a relevant issue for evaluating citizen-serving design and its effects.
The evidence reports associations and study findings, but does not establish that the training objective caused participants’ views to change.
Why this category?
- Clarifies the relationship between a stated training objective and a possible consequential participant outcome.
- Surfaces a risk of conflating statement endorsement with desired changes in citizens’ views.
- Supports scrutiny and testing while preserving uncertainty about causation and the desirability of any observed shift.
These records derive from one study source and do not prove that endorsement training caused view changes, that the changes persisted, or that convergence is desirable. The findings should not be treated as evidence of participant consent to a governance outcome.
This is an AI assessment of contribution to the goal, separate from whether the question has an answer.
- How did the AI’s training goal relate to the changes in participants’ views?
- The AI revised statements using feedback
- Findings also appeared in an assembly
- A working meaning of common ground
- Majority exposure did not predict the shift
- The overall shift favored the majority
- Two models with separate tasks
- Participants reported some movement
- Participants ranked and critiqued statements
- Training material
- Statement support tracked shifts in views
- Convergence may not be unique to AI
- Learning preferences
- Training goal differed from observed change
- Mediation outperformed simple exposure
- AI-assisted groups became less divided
- A bias test found no clear link
- The study asked whether groups converge
- Statements aimed at shared ground
- Agreement increased after discussion
- Critiques may shape revisions
- Groups preferred revised statements
- Some issues showed no clear shift
- Both sides sometimes gained support
- Assembly groups shifted in similar directions
- Agreement also increased in the assembly
- Majority lean did not predict change
- Agreement rose with AI mediation, not exposure alone
- Opinion shifts varied by issue
- AI statements received higher endorsement
Making and interpreting laws →
Authority, accountability and resilience →
Tests, stop conditions and unresolved work →
Read the exploration goalDistinguish the goal of producing endorsed group statements from observed shifts in views, without claiming that view change was the training goal or that training alone caused it.
What we can say so far
The AI’s training goal was to generate group statements that captured shared views and were likely to receive participants’ endorsement—not to change participants’ views. Participants’ views were observed to move closer together during AI-mediated deliberation. The authors attribute the reduced division at least partly to participants’ views moving toward one another, but the evidence does not show that endorsement-focused training alone caused those changes.
What remains open
- The supplied evidence does not isolate the causal contribution of endorsement-focused training from other features of the deliberation process.
Conditions and limits
- This is a characterization of the supplied account of the Science article, not independent verification.
- The reported shifts and the authors’ interpretation concern the study’s deliberation settings; the evidence does not establish that endorsement-focused training alone caused them.
Objections to consider
- The authors’ interpretation could be mistaken for a demonstrated cause of view change. The answer attributes it as an interpretation and states that the evidence does not establish the cause.
- The association between statement endorsement and later views does not establish that endorsement or training caused those views. The answer preserves that causal limitation.
- The comparison suggests a role for mediation but does not isolate which features caused the outcome. The answer does not claim training alone caused the shift.
- The exploratory human-mediator result leaves open whether convergence is unique to AI. The answer makes no claim that endorsement training uniquely produced convergence.
- The proposed critique-and-revision mechanism is presented as an interpretation, not a proven cause; the answer avoids asserting it as established.
- The absence of a detected exposure relationship does not identify the cause of view shifts. The answer makes no causal claim beyond the distinction between training goal and observed outcome.
- The reported test challenges a direct link between majority-leaning statements and change but does not rule out every influence. The answer does not infer a causal pathway from statement position.
- The tested bias measure found no clear link but cannot rule out all possible bias effects. The answer does not claim that bias or training explains the shifts.
Evidence and related ideas
Supporting evidence, qualifications and context play different roles in this answer.