Contribution to the handbook
Supporting · Assessed 2026-10-28
The rule’s mapping and the separate treatment of same-program proposal pairs help readers interpret one empirical comparison. This is useful evaluation background, but it does not itself establish an AI governance mechanism or validate the models for public decision-making.
Does this challenge the goal?
In-scope gap
The question clarifies a benchmark and its test procedure, relevant to evaluating preference-prediction claims within the defined exploration.
The evidence describes a study-specific comparison, not a general measure of representativeness or governance performance.
Why this category?
- Supports interpretation of an empirical test and comparison alternative.
- The reported result is limited to the study’s participants, programs, and methods.
The supplied result concerns this particular benchmark; outperforming it does not establish adequate representation or suitability for real governance.
This is an AI assessment of contribution to the goal, separate from whether the question has an answer.
- How did the authors define and test the party-based comparison rule?
- The benchmark models party platforms
- Individual choice predictions
- Why the experiment used pairs
- How the proposals were collected
- Defining the party-based rule
- Testing individual predictions
- Anonymous pairwise comparisons
- Where the data came from
- The volunteer policy data
- Models beat a simple party rule
- Models outperformed a simple rule
Tests, stop conditions and unresolved work →
Read the exploration goalNew evidence is being assessed. Any answer below is provisional; parts whose evidence changed or was withdrawn are withheld.
Specify how the rule mapped participants’ reported political views to predicted proposal choices.
What we can say so far
The authors’ party-based benchmark mapped a participant’s self-reported political view to a predicted proposal choice: a self-reported left-wing liberal was predicted to choose a proposal from Lula’s program, while a self-reported conservative was predicted to choose one from Bolsonaro’s. In the described comparison, the authors compared choices from the corresponding candidate’s program with choices from any other program; same-candidate-program pairs were excluded from that comparison and examined separately.
What remains open
No material gap is recorded within this part’s current scope.
Conditions and limits
- The mapping and comparison are limited to the candidates, programs, and choices used in the study.
- The source describes the benchmark and its comparison; it does not establish that the rule captures all voter reasoning.
- The evidence is an interpretation of the retrieved preprint, not independent new evidence.
Evidence and related ideas
Supporting evidence, qualifications and context play different roles in this answer.
supports
Defining the party-based rule The benchmark predicted that a self-reported left-wing liberal would choose a proposal in Lula’s program, and that a self-reported conservative would choose one in Bolsonaro’s. For liberals and conservatives separately, the authors compared choices of proposals from Lula’s or Bolsonaro’s program with choices of proposals from any other program. They excluded pairs from the same candidate’s program from this comparison and examined those pairs separately.
Directly states the mapping from self-reported political views to predicted proposal choices and describes the comparison.
Large Language Models (LLMs) as Agents for Augmented Democracy
Read full source: Large Language Models (LLMs) as Agents for Augmented Democracy (opens in a new tab) ↗Read source discussion: Large Language Models (LLMs) as Agents for Augmented Democracy →
context
The benchmark models party platforms The party-based benchmark represents the idea that voters in a representative democracy choose among bundles of policies offered by parties or politicians.
Describe the proposal-choice comparisons used to test the rule, including the treatment of same-program pairs.
What we can say so far
The authors’ benchmark mapped self-reported political orientation to a candidate’s program: it predicted that left-wing liberals would choose a proposal from Lula’s program and conservatives one from Bolsonaro’s. They tested the rule separately for liberals and conservatives by comparing choices of proposals from the corresponding candidate’s program with choices from any other program. Pairs drawn from the same candidate’s program were excluded from that comparison and examined separately. The authors also report that models had significant predictive power for those same-program choices, with 65%–77% accuracy.
What remains open
No material gap is recorded within this part’s current scope.
Conditions and limits
- This describes the authors’ comparison in the study; it does not establish that the rule captures all voter reasoning.
- The reported comparisons are limited to the study’s participants, proposals, candidates, and methods.
- The supplied material is an interpretation of the retrieved preprint; figures were not independently interpreted.
Evidence and related ideas
Supporting evidence, qualifications and context play different roles in this answer.
supports
Defining the party-based rule The benchmark predicted that a self-reported left-wing liberal would choose a proposal in Lula’s program, and that a self-reported conservative would choose one in Bolsonaro’s. For liberals and conservatives separately, the authors compared choices of proposals from Lula’s or Bolsonaro’s program with choices of proposals from any other program. They excluded pairs from the same candidate’s program from this comparison and examined those pairs separately.
Directly describes the rule’s mapping and the comparisons for liberals and conservatives, including the separate treatment of same-program pairs.
Large Language Models (LLMs) as Agents for Augmented Democracy
Read full source: Large Language Models (LLMs) as Agents for Augmented Democracy (opens in a new tab) ↗Read source discussion: Large Language Models (LLMs) as Agents for Augmented Democracy →
context