Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generative Social Choice: The Next Generation

Niclas Boehmer, Sara Fish, Ariel D. Procaccia

TL;DR

This work advances generative social choice by introducing cost- and budget-aware slates and accommodating approximate query responses. It formalizes a budgeted slate selection problem with a rich axiomatic basis (cBJR), and provides both theoretical guarantees and impossibility results under query errors. The PROSE system demonstrates practical applicability by integrating GPT-4o for discriminative and generative queries on unstructured data, achieving improved proportional representation and user utilities over baselines. Together, the results push scalable, fair, and representative democratic processes that can adapt to real-world constraints and imperfect AI components. The framework lays groundwork for future extension to participatory budgeting and ordinal preferences while emphasizing responsible deployment considerations.

Abstract

A key task in certain democratic processes is to produce a concise slate of statements that proportionally represents the full spectrum of user opinions. This task is similar to committee elections, but unlike traditional settings, the candidate set comprises all possible statements of varying lengths, and so it can only be accessed through specific queries. Combining social choice and large language models, prior work has approached this challenge through a framework of generative social choice. We extend the framework in two fundamental ways, providing theoretical guarantees even in the face of approximately optimal queries and a budget limit on the overall length of the slate. Using GPT-4o to implement queries, we showcase our approach on datasets related to city improvement measures and drug reviews, demonstrating its effectiveness in generating representative slates from unstructured user opinions.

Generative Social Choice: The Next Generation

TL;DR

This work advances generative social choice by introducing cost- and budget-aware slates and accommodating approximate query responses. It formalizes a budgeted slate selection problem with a rich axiomatic basis (cBJR), and provides both theoretical guarantees and impossibility results under query errors. The PROSE system demonstrates practical applicability by integrating GPT-4o for discriminative and generative queries on unstructured data, achieving improved proportional representation and user utilities over baselines. Together, the results push scalable, fair, and representative democratic processes that can adapt to real-world constraints and imperfect AI components. The framework lays groundwork for future extension to participatory budgeting and ordinal preferences while emphasizing responsible deployment considerations.

Abstract

A key task in certain democratic processes is to produce a concise slate of statements that proportionally represents the full spectrum of user opinions. This task is similar to committee elections, but unlike traditional settings, the candidate set comprises all possible statements of varying lengths, and so it can only be accessed through specific queries. Combining social choice and large language models, prior work has approached this challenge through a framework of generative social choice. We extend the framework in two fundamental ways, providing theoretical guarantees even in the face of approximately optimal queries and a budget limit on the overall length of the slate. Using GPT-4o to implement queries, we showcase our approach on datasets related to city improvement measures and drug reviews, demonstrating its effectiveness in generating representative slates from unstructured user opinions.

Paper Structure

This paper contains 86 sections, 12 theorems, 7 equations, 3 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Our Contributions
  3. Related Work
  4. Model
  5. General Problem Statement
  6. Queries
  7. Axiom
  8. Algorithm and Theoretical Guarantees
  9. Differences from FGPP+24
  10. Warm-Up
  11. Approximate Queries: Guarantees
  12. Approximate Queries: Impossibility Results
  13. Validation in Synthetic Environment
  14. Environment
  15. Results
  16. ...and 71 more sections

Key Result

Theorem 3.1

For exact discriminative queries and $(\gamma,0,1)$-accurate generative queries, $\text{DemocraticProcess}_{C,f}$ satisfies $(0,1/\gamma)$-cBJR if $\{\lfloor j\cdot \frac{B}{n}\rfloor \mid j\in [n]\}\subseteq C$ and $\ell\in f(\ell)$ for all $\ell\in[r]$.

Figures (3)

  • Figure 1: Proportionality of different versions of our democratic process for varying error settings: For a given $c$-value ($x$-axis), the plot shows the maximum $d$ value for which $(c,d)$-cBJR is violated ($y$-axis) averaged over $100$ instances. The shaded gray region represents the area ruled out by the guarantees provided by \ref{['th:fullGuarantees']} for Complex-DemocraticProcess.
  • Figure 2: Comparison of proportionality violations introduced when only modifying one error parameter and keeping the others accurate.
  • Figure 3: Analogous to \ref{['fig:approx-BJR']} but with worst-case error model.

Theorems & Definitions (23)

  • Definition 2.1: $(b,d)$-costBJR
  • Theorem 3.1
  • proof
  • Theorem 3.2
  • Theorem 3.3
  • Theorem 3.4
  • Proposition 3.4
  • Definition A.1: PPS21
  • Lemma A.2
  • proof
  • ...and 13 more