Algorithms for Sustainable Group Decision Making

Group decision making (GDM) is a central activity of human interaction, and also an increasingly important issue in computer science with applications such as multi-agent systems, preference aggregation in artificial intelligence, e-democracy platforms, and group recommender systems. This research project is situated in the field of computational social choice (COMSOC), which is concerned with the study of GDM from a computational point of view. The main focus of this project lies on sustainable, long-term, algorithm-supported GDM within small or medium-sized groups that decide about a varying range of topics. Instead of considering decisions as singular events, we want to take the history of past group decisions into account. This novel viewpoint allows us to tackle a fundamental issue of GDM: how to avoid situations where frustrated participants drop out of the decision making process. For example, if a minority has been repeatedly overruled in the past, it might withdraw from the decision process. To overcome this problem, we will introduce and analyze continuative voting rules: these are voting rules that take the history of previous decisions into account and guarantee fairness across this history. Furthermore, if a GDM algorithm does not provide incentives to compromise, the decision quality might deteriorate and, consequently, users may refuse to use a corresponding system. Hence, we will study measures that foster compromise in GDM and will augment continuative voting rules with corresponding methods. Finally, it may be that GDM algorithms require excessively detailed preference information. In such a case, the effort required to use such a system may outweigh its benefits (such as guarantees on fairness). We will analyze the information requirements of continuative voting rules and reduce these by utilizing information learned from previous decisions. The overall goal of this project is to find GDM algorithms that foster participation and make the GDM process long-term sustainable. To achieve this goal, we will use a large array of methods established in computational social choice. The outcome of this project contributes to a future in which GDM apps and e-democracy systems have "COMSOC inside", and hence make use of the theoretical and experimental expertise established in this field. The perspectives of the proposed researchlong-term GDM, compromise, and reduced information requirementsare essential ingredients for sustaining GDM processes.

Group decision making (GDM) has expanded far beyond political elections into numerous technological applications: group recommender systems, multi-agent systems, e-democracy platforms, and preference aggregation in AI systems. In these contexts, algorithms gather preferences from agents (human or machine) and aggregate them into collective decisions. Our research in computational social choice investigated how algorithms can support fair group decision processes that encourage long-term participation. We examined three key scenarios: **Multi-winner voting** is about selecting a fixed number of items based on voters' preferences. This forms the foundation for many applied scenarios, ranging from dynamic Q&A platforms to validating blockchain transactions. Among our contributions in this area, we developed a classification framework for multi-winner voting rules, helping practitioners select appropriate methods for specific applications, and identified which methods are best suited for shortlisting alternatives or candidates. **Participatory budgeting** enhances democratic engagement in many cities worldwide (Vienna, Paris, Amsterdam, ...) by allowing citizens to vote on projects funded from a limited budget. Our research explored, for example, how different definitions of "satisfaction" fundamentally alter what constitutes fairness in these processes. We also investigated how to incorporate individual donations for projects, expanding the flexibility of participatory budgeting systems. **Perpetual voting** is a novel approach to GDM developed within this project. Its key feature is considering the history of past decisions rather than treating each decision as isolated. Imagine a team of five colleagues deciding on a weekly meeting time - perpetual voting would ensure that if a person is repeatedly overruled, their preferences receive higher priority in future decisions. This approach addresses a fundamental issue: preventing participant dropout when minorities are repeatedly ignored. By accounting for previous outcomes, perpetual voting rules provide mathematical guarantees for fair representation over time, ensuring that all participants maintain meaningful influence and encouraging continued engagement. This kind of fairness is particularly important in small groups, in which diversity is encouraged and where participation of all individuals is valued highly. To summarize, our project yielded four main outcomes: (1) a deeper understanding of fairness across these domains, (2) an analysis of necessary trade-offs between competing fairness properties, establishing theoretical limits of fairness, (3) practical new algorithms with provable fairness guarantees, and (4) open-source software libraries supporting future research and applications.