Decision-Making in Repeated Games: Insights from Active Inference.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Artificial Intelligence

Background:

• Repeated games involve complex decision-making with social uncertainty, mirroring real-world interactions.
• Traditional models like reinforcement learning struggle to unify cognitive processes.
• Active inference provides a novel, integrated approach grounded in free energy minimization.

Purpose of the Study:

• To systematically review the active inference framework's potential for explaining cognitive mechanisms in repeated game decision-making.
• To highlight active inference's capacity to unify perception, learning, planning, and action.
• To explore its application in modeling social uncertainty and mentalizing.

Main Methods:

• Systematic review of the active inference framework.
• Exploration of its theoretical underpinnings in variational free energy minimization.
• Analysis of its application to partially observable Markov decision processes.

Main Results:

• Active inference unifies cognitive components (perception, learning, planning, action) within a single generative model.
• Belief updating and exploration-exploitation are managed through free energy minimization.
• The framework inherently addresses social uncertainty and supports modeling of mentalizing processes.

Conclusions:

• Active inference provides a unified account of social decision-making in repeated games.
• Its hierarchical structure and grounding in free energy minimization offer advantages over traditional models.
• Further validation via simulations and behavioral fitting is recommended.