Contribution to the handbook
Supporting · Assessed 2026-10-28
The study’s 267 participants were the group whose preferences it aimed to model, and the authors state that the data cannot establish what Brazil’s general population prefers. Its sample-stability checks therefore offer useful context for interpreting preference-model tests, without specifying a governance design.
Does this challenge the goal?
In-scope gap
The question concerns the limits of evidence about citizen preferences, relevant to testing whether a system serves citizens; it does not contest that goal.
The study’s population and within-sample stability checks do not establish representativeness beyond the participants studied.
Why this category?
- consequential risks/limits
- testing
The reported stability comparisons concern the study data and do not establish how representative a future governance system’s data would be.
This is an AI assessment of contribution to the goal, separate from whether the question has an answer.
Tests, stop conditions and unresolved work →
Authority, accountability and resilience →
Read the exploration goalNew evidence is being assessed. Any answer below is provisional; parts whose evidence changed or was withdrawn are withheld.
Identify the participant population studied and distinguish it from any wider population the results cannot establish.
What we can say so far
The study’s results describe the preferences of its 267 participants—the group whose preferences the authors aimed to model. The authors state that the data do not support estimates of what Brazil’s general population prefers. The results therefore should not be treated as establishing the preferences of people outside the study.
What remains open
- The supplied evidence does not establish whether the 267 participants represent any wider population.
Conditions and limits
- This is the study authors’ stated scope, based on the supplied text from the preprint; figures were not independently interpreted.
- The study’s sample-stability checks and comparisons against the full participant sample do not establish representativeness of people outside the study.
Objections to consider
- The full-sample comparison is limited to the study data; the answer does not use it to generalize beyond the participants.
- The augmentation results do not establish preferences outside the study, consistent with the answer’s stated population limit.
- The stability results are not evidence of wider representativeness, and the answer makes no such claim.
- The sample-stability procedure does not establish broad representativeness; the answer distinguishes the study participants from the wider population.
Evidence and related ideas
Supporting evidence, qualifications and context play different roles in this answer.
supports
The study’s scope The study asks whether models can learn preferences within a specific group of 267 participants, not estimate what Brazil’s population prefers.
Directly identifies the 267 participants as the target group and states that the data cannot support estimates for Brazil’s general population.
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 →
qualifies
The data used in the comparison The authors organize preference data in a matrix: each row represents a participant, and each column records a preference between a pair of policy proposals. They sample and augment this data, then compare how well those samples reproduce the full sample's aggregate preferences.
Clarifies that the comparison to the full sample is confined to study data and cannot establish a match to people not studied.
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 →
qualifies
Predictions improved some sample matches In the Type I B test, the authors fine-tuned six models on the same preference data: GPT-3.5 Turbo, LLaMA-2 7B, Falcon 7B, Mistral 7B, LLaMA-3 8B and Gemma 7B. They tested initial random samples of 5%, 25% and 50%, adding predictions for 20% of the people who remained outside each sample. For the 5% sample, R² rose from about 30% to about 75%; for the 25% sample, the gain was about 7 to 10 percentage points. The supplied text gives no separate numerical result for the 50% sample.
Limits reported augmentation results to tests on the study data; they do not establish preferences of people outside the study.