Cooperation in public goods games: Leveraging other-regarding reinforcement learning on hypergraphs

Area of Science:

• Evolutionary Game Theory
• Artificial Intelligence
• Complex Systems

Background:

• Cooperation is crucial for ecosystems and human society, yet its evolutionary mechanisms are complex.
• Existing models often simplify interactions to self-regarding or pairwise settings, not reflecting real-world group dynamics.
• Reinforcement learning offers novel approaches to study cooperation beyond imitation learning.

Purpose of the Study:

• To investigate the evolution of cooperation in a group Public Goods Game (PGG) using other-regarding reinforcement learning on hypergraphs.
• To analyze how player history, group interactions, and spatial structures influence cooperative behavior.
• To identify conditions and mechanisms that promote or hinder the emergence of cooperation.

Main Methods:

• Combined Q-learning algorithm with an evolutionary game framework to model other-regarding agents.
• Simulated the Public Goods Game on hypergraphs, incorporating players' action histories and group interactions.
• Analyzed the impact of the synergy factor (r̂) and spatial patterns on cooperation levels.

Main Results:

• Identified three distinct regions of cooperation (absent, medium, high) as the synergy factor increased, with two abrupt transitions.
• Observed regular and anticoordinated chessboard structures in spatial patterns, impacting cooperation transitions differently.
• Provided a theoretical analysis for the first cooperation transition, linking it to player foresight and exploration rates.

Conclusions:

• Other-regarding reinforcement learning on hypergraphs provides a more realistic model for cooperation evolution in groups.
• Player characteristics (long-sightedness, low exploration) and spatial structures significantly influence the emergence and stability of cooperation.
• Findings offer insights into human cooperation, emphasizing the role of group dynamics and shared information.