Imported source discussion · 2024 preprint; retrieved version recorded in source snapshot
Large Language Models (LLMs) as Agents for Augmented Democracy
Jairo F. Gudiño, Umberto Grandi, and César Hidalgo
Full readable text of the retrieved 37-page preprint, including references and appendices. Figures are not independently interpreted. AI interpretation; the authors are not participants in this app.
Read the original publication ↗You arrived from this claim
Reported LLM policy-preference prediction results →Large Language Models (LLMs) as Agents for Augmented Democracy · passage 14, 15
This idea cites the publication. The interpretations below describe the source as a whole; they are not an exact passage match or a quotation.
The preprint tests LLMs as citizen-specific assistants for modeling policy preferences. It reports promising results on one Brazilian participation dataset while emphasizing substantial limits to deployment.
The source’s perspective
Source idea 1 · exploring
Augmented democracy is framed as citizen-controlled agents assisting participation between representative and direct democracy—not as AI governing on citizens’ behalf.
What role might personalized agents play in democratic participation?
The proposed agents represent individual citizens and may help unbundle policy choices.
Large Language Models (LLMs) as Agents for Augmented Democracy · passage 2, 4, 5
Claims reported in this source
Source idea 2 · Reported from this publication
In tests of pairwise choices, fine-tuned LLMs outperformed a party-alignment bundle rule in 21 of 24 comparisons; predictions within one candidate’s platform reached 65–77% accuracy.
Can LLMs predict policy preferences beyond party alignment?
These results suggest predictive information beyond self-reported political orientation.
Large Language Models (LLMs) as Agents for Augmented Democracy · passage 14, 15
Source idea 3 · Reported from this publication
For this dataset, LLM augmentation improved estimates of aggregate preferences, especially with small samples; the authors report gains for samples below 30%.
Can LLMs improve estimates from small samples?
The comparison uses policy win rates for the study’s participant population.
Large Language Models (LLMs) as Agents for Augmented Democracy · passage 16, 18, 19, 22
Source idea 4 · Reported from this publication
Accuracy varied across groups: models more accurately predicted liberal and college-educated participants, while age and sex differences varied by model.
How evenly did model accuracy extend across participant groups?
The reported disparities raise representation concerns.
Large Language Models (LLMs) as Agents for Augmented Democracy · passage 13, 14, 24, 44
Qualifications
Source idea 5 · conditional
The findings concern 267 participants on one Brazilian platform, not Brazil’s general population. The authors say current LLMs are not ready for full deployment, citing bias, inconsistent choices, opacity, and limited explainability.
How far do the results support real-world use?
The authors identify both narrow study scope and unresolved technical and governance concerns.
Addresses: The preprint’s reported predictive and augmentation performance
Large Language Models (LLMs) as Agents for Augmented Democracy · passage 6, 9, 22, 23, 24, 25
Elsewhere in the exploration
These ideas cite the same publication. Sharing a source does not mean they agree.
Analysis version source-discussions-v2. Source snapshot 343b13d07d01. Coverage and source identity remain separate from the AI’s interpretation.