Social Choice and Machine Learning

ESSLLI 2024
  • Lecturer: Eric Pacuit (website)
  • Venue: Leuven, Belgium

    Ruth Benedict

  • Dates: August 5 - August 9, 2024
  • Meeting Times: 11:00am - 12:30pm

  • This course is an introduction to social choice theory, with a special focus on the use of machine learning to study voting methods. The first part of the course will introduce the mathematical analysis of voting methods, including probabilistic voting methods and new voting paradigms such as liquid democracy. This introduction to voting theory will include hands-on experience with the Python package pref_voting, a set of tools in Python designed to facilitate the computational analysis of voting methods. The main objective of the course is to explore the way that machine learning tools and ideas have been used to complement existing social-choice theoretic results. The main topics that will be discussed include:

    • introduction to mathematical analysis of voting methods, voting paradoxes;
    • probabilistic voting methods;
    • quantitative analysis of voting methods (e.g., Condorcet efficiency);
    • learning voting rules (PAC-learning, MLPs, other approaches);
    • using modern deep learning techniques to generate synthetic election data;
    • strategic voting, learning to successfully manipulate voting rules based on limited information about how the other voters will vote using neural networks (multi-layer perceptrons);
    • RLHF (reinforcement learning with human feedback) and social choice;
    • using large-language models to improve group decision-making; and
    • liquid democracy (time permitting).

    While most of the course will focus on applications of machine learning in social choice theory, time permitting, potential application of social choice theory in machine learning will also be discussed.

    This course will be self-contained. No previous experience with voting theory or machine learning will be assumed. All of the topics in voting theory (e.g., voting methods, voting paradoxes, strategic voting) will be introduced.

Day 1    Introduction to voting theory
Day 2    Characterizing voting methods
Day 3    Stable Voting, Preferential Voting Tools, Learning voting rules
Day 4    Learning to manipulate elections
Day 5    Strategic Voting, RLHF and Social Choice