Interaction and Expressivity in Collective Decision-Making

Umberto GrandiIJCAI 2022 · DOI

TL;DR

This paper explores how to improve collective decision-making by allowing more interaction between agents and giving them more expressive ways to communicate their preferences. It presents ongoing research in two areas: iterative voting, where decisions are refined over multiple rounds, and delegative voting, which allows for more nuanced preference expression on multiple issues.

What Problem Does This Paper Address?

Traditional social choice theory often assumes independent voters with simple preferences over a limited set of alternatives. This model struggles to capture the complexities of real-world decision-making, where individuals interact, influence each other, and often face decisions involving multiple, interconnected issues. The paper addresses the gap between these simplified models and the richer, more dynamic nature of human and artificial agent decision-making. It aims to bridge this gap by investigating mechanisms that allow for:

  1. Interaction: Agents can learn from each other's choices or public information, leading to more informed or strategic voting.
  2. Expressivity: Agents can express more complex preferences, particularly when dealing with multiple, interdependent issues, moving beyond simple "yes/no" or single-choice votes.

The ultimate goal is to pave the way for "agent-mediated social choice," where AI can facilitate more effective collective decisions.

Methods

The paper focuses on two main research directions:

  • Iterative Voting: This approach involves repeatedly applying a voting rule. In each round, agents might receive information about the current winner, other agents' votes, or candidate scores. They can then react by changing their vote or preference, either due to social influence or strategic considerations. This is framed as both an equilibrium selection mechanism and a novel voting procedure in itself. The research assesses the quality of outcomes using metrics like social welfare and the frequency of Condorcet winners.
  • Delegative Voting on Multiple Issues: This direction enhances expressivity by allowing agents to delegate their votes. A key mechanism explored is "liquid democracy," where voters can vote directly or delegate their vote (and any received delegations) to another voter. The research has focused on solving "delegation cycles" and developing algorithms for handling delegations with constraints between multiple issues, often using priority orderings over these issues.

Datasets

Not discussed in the paper. The paper describes ongoing research and theoretical frameworks, with some mention of experimental platforms and behavioral experiments, but no specific datasets are detailed for evaluation.

Key Insights and Ideas

  • Interaction enhances decision quality: Allowing agents to iteratively refine their votes based on information from others can lead to better collective outcomes, potentially escaping classical social choice impossibilities.
  • Iterative voting as a lightweight deliberation: The iterative process can be seen as a simplified form of deliberation, improving decision quality without the overhead of full-blown discussion.
  • Delegative voting offers richer preference expression: Moving beyond simple choices to mechanisms like liquid democracy allows for more nuanced representation of preferences, especially when dealing with multiple, interconnected issues.
  • Handling complexity in multi-issue voting is crucial: Constraints between issues and the potential for delegation cycles are significant challenges that require sophisticated algorithmic solutions.
  • The rise of "agent-mediated social choice": There's a growing need for AI systems that can facilitate collective decision-making, especially in the context of digital democracy and electronic institutions.

Key Results

The paper highlights the beneficial effects of iterative voting, showing improvements in social welfare and an increased frequency of electing Condorcet winners in certain voting situations. While specific quantitative results are not detailed in this overview, the research has led to the development of platforms like "Itero" for organizing iterated plurality elections online and has produced polynomial algorithms for solving delegation problems under constraints.

Limitations and Future Work

The authors acknowledge that assessing proposed mechanisms when used by artificial agents acting as proxies for human decision-makers is a crucial next step. Initial work has been done on reinforcement learning agents in iterated voting, but this area requires further exploration. The paper also points to the need for novel social choice methods that can handle a large number of alternatives and incomplete preference data, inspired by experiments with collective political program construction. The rise of an interdisciplinary community in digital democracy is seen as a fertile ground for uncovering new problems and research directions.

Summary updated

BibTeX
@inproceedings{grandi2022interaction,
  author = {Umberto Grandi},
  title = {{Interaction and Expressivity in Collective Decision-Making}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2022},
  doi = {10.24963/ijcai.2022/816},
}
Figure