Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

DY: Fachverband Dynamik und Statistische Physik

DY 5: Focus Session: Physics of Behavior (joint session SOE/DY)

DY 5.9: Vortrag

Montag, 9. März 2026, 12:15–12:30, GÖR/0226

Critical Transitions of Reinforcement Learning Dynamics in Social Dilemmas — •Balakrishna Prabhu B N and Wolfram Barfuss — Center for Development Research(ZEF), University of Bonn, Germany

Understanding how cooperation emerges and persists among self-interested agents remains a crucial question in the human, animal, and machine behavioral sciences. Specifically, the aspect of the timescales required to reach a cooperative outcome has received little attention.

While the field of equilibrium game theory has addressed the possibility of cooperative outcomes, it offers little insight into how agents select and reach these equilibria, or the timescales required to do so. Evolutionary game theory and reinforcement learning have addressed some of these questions, but are yet to examine the temporal aspects of strategy adaptation and the critical transitions that occur with changes to basic payoff structures.

In this work, we develop a framework based on deterministic dynamics of reinforcement learning to study critical transitions between different social dilemma games. We find that boundaries involving the Chicken game exhibit strong criticality, whereas transitions involving the StagHunt game do not. We also explore convergence times and equilibrium selection and their variations across these boundaries for static and dynamic systems.

By uncovering the dynamic behavior between game transitions, our work lays the foundation for an integrated theory of coupled social-ecological tipping elements.

Keywords: Critical Transitions; Social Dilemmas; Tipping Points; Reinforcement Learning; Game Theory