Contribution to the handbook
Supporting · Assessed 2026-10-28
The reported comparison offers useful test evidence about one preference-prediction approach. Its relevance is limited to the study’s participants, proposals, models, and specific political-alignment rule; it does not establish that the models represent citizens well enough for governance.
Does this challenge the goal?
In-scope gap
The question concerns a test of individual preference prediction, which can inform assessment of a possible citizen-representation method within the existing goal. It does not contest that goal.
The supplied evidence reports a comparison in one study, not whether such predictions would be suitable for governance.
Why this category?
- Useful evidence about a test of individual preference predictions
- Limited scope and generalizability of the comparison
The results come from one study and share a source root, so they are not independent evidence. The supplied passages do not establish performance with other populations, proposals, or decision settings.
This is an AI assessment of contribution to the goal, separate from whether the question has an answer.
Authority, accountability and resilience →
Tests, stop conditions and unresolved work →
Read the exploration goalState how the models’ individual-choice predictions compared with the specified rule and explain what that rule used, keeping the claim within the reported comparison.
What we can say so far
In the reported study, the fine-tuned models predicted individual policy choices more accurately than the political-alignment rule in 21 of 24 comparisons. The rule used participants’ self-reported political orientation to predict a choice from the corresponding candidate’s program: for example, it mapped left-wing liberals to Lula’s proposals and conservatives to Bolsonaro’s. The result was not universal: liberals choosing Bolsonaro-program proposals were exceptions for Mistral, Falcon, and Gemma. The comparison is limited to the study’s participants, proposals, models, and methods; it does not establish performance with other populations or settings.
What remains open
- The supplied evidence does not establish how the models or rule would perform outside the study setting.
Conditions and limits
- The result concerns the particular participants, proposals, models, and comparison methods in the study; it does not establish performance with other people or in other settings.
- The rule used self-reported political orientation as a proxy for which candidate’s proposal a participant would choose; it was one specific comparator, not every possible alternative.
- The models were more accurate in 21 of 24 comparisons, with the reported exceptions for liberals choosing Bolsonaro-program proposals for Mistral, Falcon, and Gemma.
Objections to consider
- The study’s participants do not establish how the comparison would generalize to Brazil’s broader population. This limits wider inference but does not undermine the reported within-study comparison.
Evidence and related ideas
Supporting evidence, qualifications and context play different roles in this answer.
context
The volunteer policy data The study uses volunteer choices between randomly paired policies from the two leading candidates’ 2022 Brazilian presidential programs. The choices came from an experiment involving collaborative construction of a government program.
Provides background on the volunteer policy-choice data and its collection, but is not needed to state the rule or comparison.
Large Language Models (LLMs) as Agents for Augmented Democracy
Read full source ↗