Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI 2019, Macao, China, August 10-16, 2019

Jingkuan Song, Hengtao ShenIJCAI 2019 · DOI · DBLP

Introduction

Modern democratic systems often face a fundamental tension between the accountability of Representative Democracy (RD) and the majoritarian precision of Direct Democracy (DD). While DD allows every citizen to vote on every issue, it imposes a prohibitive cognitive burden. Conversely, RD reduces this burden by electing a small committee, but often fails to accurately reflect the electorate's will on specific policy matters.

Flexible Representative Democracy (FRD) is proposed as a novel hybrid system. It allows voters to dynamically shift their voting power among a fixed set of elected representatives on an issue-by-issue basis. Unlike Liquid Democracy (LD), which allows delegation to any other voter and can lead to chaotic cycles or privacy concerns, FRD maintains a stable, accountable body of representatives while providing the flexibility of direct participation.

The FRD Model

The FRD framework operates in two distinct phases:

  1. Election Phase: Voters elect a fixed-size committee of $k$ representatives using standard multi-winner voting rules.
  2. Delegation and Decision Phase: For any specific issue, a voter can either rely on a default weighting (where their power is split equally among all representatives) or "delegate" their full voting power to a specific subset of the committee they trust on that topic.

The final decision on an issue is determined by a weighted majority of the representatives' votes, where weights are derived from the voters' delegations.

Figure 1: Illustration of the FRD mechanism across two issues. The weights $W_l^i$ reflect how voter power is distributed among representatives, influencing the final outcome $X_1^i$.

Computational Complexity

A key theoretical contribution of this work is the proof that selecting a committee to maximize "Majority Agreement"—the frequency with which the committee's weighted vote matches the majority of the electorate—is NP-hard. This suggests that no efficient algorithm can guarantee a perfectly representative committee for all possible issue sets, highlighting the necessity of flexible mechanisms like FRD to bridge the gap.

Experimental Methodology

To evaluate FRD, the authors conducted extensive simulations using synthetic datasets. They generated populations of up to 501 voters and 100 candidates, with preferences drawn from shared distributions. The performance of various multi-winner voting rules—including Approval Voting (AV), Reweighted Approval Voting (RAV), Single Transferable Vote (STV), and Borda count—was tested across up to 150 independent binary (Yes/No) issues.

Key Results

The simulations revealed several critical insights into the performance of representative systems:

  • The Representative Plateau: Standard RD systems, regardless of the voting rule used, tend to plateau at approximately 60-65% agreement with the majority as the number of issues increases. This represents an inherent "information loss" in fixed-committee representation.
  • The Impact of Delegation: FRD significantly outperforms static RD. As the "delegation rate" (the percentage of voters actively choosing representatives for specific issues) increases, the system's agreement with the "ideal" direct democracy outcome approaches 100%.
  • Voting Rule Performance: Approval-based rules (AV and RAV) generally outperformed more complex methods like STV or Borda in selecting committees that align well with the majority under the FRD framework.
Figure 6: Majority Agreement across different parameters. (a) performance vs. number of issues; (b) performance vs. number of candidates; (c) performance vs. committee size. Weights Agreement consistently outperforms standard rules.

Discussion and Future Work

FRD offers a promising middle ground that mitigates the "minority risk"—where an active minority can flip results if the majority is inattentive—provided the majority remains even moderately engaged.

Future research directions include moving beyond binary issues to more complex policy spaces and accounting for interdependent issues. Additionally, analyzing strategic behavior and the potential for direct voter participation within the FRD framework remain open areas for exploration.

This paper was presented at IJCAI 2019, Macao, China.

Summary updated

BibTeX
@inproceedings{song2019proceedings,
  author = {Jingkuan Song and Hengtao Shen},
  title = {{Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI 2019, Macao, China, August 10-16, 2019}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2019},
  doi = {10.24963/ijcai.2019},
}

Loading related content...

Figure