Reinforcement learning accounts for moody conditional cooperation behavior: experimental results.

Area of Science:

• Behavioral Economics
• Computational Social Science
• Game Theory

Background:

• Human participants in social dilemma games frequently exhibit conditional cooperation (CC) and moody conditional cooperation (MCC) behaviors.
• Previous computational research indicated that reinforcement learning (RL) could explain these observed CC and MCC patterns.

Purpose of the Study:

• To investigate the presence of MCC across different game types (Prisoner's Dilemma, Public Goods Game).
• To determine if reinforcement learning principles can account for observed MCC behavior in human participants.

Main Methods:

• Conducted repeated multiplayer Prisoner's Dilemma and Public Goods games with human participants.
• Analyzed participant behavior to identify patterns consistent with MCC.
• Compared the explanatory power of RL models against MCC models for the observed data.

Main Results:

• Observed MCC behavior in both game types, though with variations from previous findings.
• Found that a participant's own prior cooperation, rather than others' actions, significantly influenced their subsequent cooperation levels.
• Reinforcement learning models demonstrated comparable accuracy to MCC models in describing the experimental outcomes.

Conclusions:

• Reinforcement learning appears to be a significant proximate mechanism underlying moody conditional cooperation.
• The study highlights the role of individual past actions in shaping cooperative behavior within social dilemmas.