Contribution to the handbook
Supporting · Assessed 2026-10-28
Prior work offers useful context about claims that models can reproduce some responses or subgroup patterns, but the supplied evidence lacks enough methodological detail to make it a standalone design specification or test result for this governance system.
Does this challenge the goal?
None identified
The question asks for background about earlier simulated-response research. It provides context but does not challenge the goal, beneficiary boundary, or identify a missing design detail, mechanism, or test.
The supplied evidence attributes findings to earlier studies but gives limited methodological detail, and those findings are not results of the paper’s own experiment.
Why this category?
- useful background
- limited direct design contribution
These are authors’ summaries of prior work within one shared source root. Their methods, reach, and transfer to civic governance cannot be assessed from the supplied records.
This is an AI assessment of contribution to the goal, separate from whether the question has an answer.
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 Any answer below is provisional; parts whose evidence changed or was withdrawn are withheld.
Identify the earlier work the authors mention and describe its use of AI-generated survey responses and any findings the authors report, distinguishing it from their own study.
What we can say so far
The authors cite Argyle and colleagues’ work, which used language models to reproduce survey-response patterns across human subgroups. They report that its predictions of women’s preferences were about one percentage point more accurate than its predictions of men’s. They also cite Santurkar and colleagues, whose models reportedly reproduced preferences more accurately for young, low-income, politically moderate people with less than a high-school education; in that study, accuracy for men was slightly higher than for women, by about one percentage point or less. The authors also mention earlier work simulating some opinion-poll responses, moral judgments and election choices using detailed personas. These are findings attributed to prior studies, not results of the authors’ own study of policy preferences among participants in Brazil’s 2022 election.
What remains open
- The supplied evidence does not name the other studies behind the general references to opinion-poll, moral-judgment and election-choice simulations, or provide their methods in detail.
Conditions and limits
- The earlier-work findings are reported through interpretations of the same preprint, not independent corroboration.
- The small, opposite-direction gender differences belong to different earlier studies; neither establishes a general gender pattern.
- The reported subgroup results should not be assumed to hold across other models, questions or settings.
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.
Describes the volunteer policy-choice data used in the authors’ own study.
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 →