Algorithms for Managing Deliberation

Abstract

Citizens’ assemblies, in which ordinary people are randomly selected to participate in the policymaking process, have become increasingly widespread in recent decades. Chosen via a method known as sortition, these assemblies have not only had a profound impact on worldwide legislation but also sparked a flurry of recent research into how they can best be organized, from recruiting and selecting participants to managing the deliberation itself. In this report, we focus on the assembly partitioning problem, in which we wish to partition participants among a set of moderated groups over multiple sessions of deliberation such that 1) each group is representative of the assembly, and 2) participants get to interact with as many new people as possible over the course of deliberation. We first provide an overview of the baseline algorithm that is being used to generate partitions. We then define a model for the problem and propose a greedy, linear programming-based approach to tackling it. Afterwards, we conduct a series of experiments to compare our algorithm with the baseline, demonstrating that our algorithm generates significantly higher-quality solutions on both synthetic and real-world assembly data. We also derive a theoretical bound on the maximum number of unique interactions that can be achieved between back-to-back partitions. Finally, we discuss the advantages of our algorithm over the previous approach, advocating that organizers adopt our algorithm for future assemblies.