Value-Based Rationales Improve Social Experience: A Multiagent Simulation Study
Sz-Ting Tzeng, Nirav Ajmeri, Munindar P. Singh
TL;DR
This work introduces Exanna, a framework where agents incorporate explicit value hierarchies into both decision making and the generation of rationales. By linking values to payoffs and leveraging value-aware rule learning via XCS, Exanna produces rationales that align with stakeholders’ priorities while selectively sharing information to preserve privacy. In a multiagent pandemic simulation, value-aligned rationales improve conflict resolution, social experience, privacy, and flexibility, and encourage normative emergence beyond baseline sharing strategies. The findings support value-driven rationales as a path toward trustworthy, privacy-conscious social AI with practical implications for norm-sensitive coordination. The approach offers a principled balance between transparency, privacy, and social welfare in autonomous agent systems.
Abstract
We propose Exanna, a framework to realize agents that incorporate values in decision making. An Exannaagent considers the values of itself and others when providing rationales for its actions and evaluating the rationales provided by others. Via multiagent simulation, we demonstrate that considering values in decision making and producing rationales, especially for norm-deviating actions, leads to (1) higher conflict resolution, (2) better social experience, (3) higher privacy, and (4) higher flexibility.