Reinforcement learning account of network reciprocity

Area of Science:

• Evolutionary game theory
• Network reciprocity
• Reinforcement learning

Background:

• Evolutionary game theory predicts network structures enhance cooperation in social dilemmas.
• Empirical evidence shows human cooperation on fixed networks is not always enhanced.
• A gap exists in explaining observed network reciprocity patterns in human social dilemmas.

Purpose of the Study:

• To explain the observed network reciprocity and its absence in human social dilemmas.
• To investigate the role of reinforcement learning in network cooperation dynamics.
• To bridge theoretical predictions with experimental observations in social game theory.

Main Methods:

• Utilized the Bush-Mosteller model, a reinforcement learning framework.
• Analyzed social dilemma games with varying benefit-to-cost ratios.
• Examined the impact of node degree within network structures.
• Extended previous numerical simulations of network reciprocity.

Main Results:

• Reinforcement learning (Bush-Mosteller model) approximately explains observed network reciprocity.
• The model accounts for the lack of enhanced cooperation in certain parameter regions.
• Results align with experimental observations of human behavior in social dilemmas.
• Extended previous numerical findings significantly.

Conclusions:

• Reinforcement learning provides a viable mechanism for understanding cooperation dynamics on networks.
• The Bush-Mosteller model successfully reconciles theoretical predictions with empirical data on network reciprocity.
• Network structure and game parameters (benefit-to-cost ratio, node degree) critically influence cooperation outcomes.