A multi-agent reinforcement learning framework for exploring dominant strategies in iterated and evolutionary games.

Area of Science:

• Game Theory
• Artificial Intelligence
• Computational Social Science

Background:

• Classic strategies like tit-for-tat are limited by current analytical tools.
• Exploring complex decision-making requires advanced computational methods beyond human intuition.

Purpose of the Study:

• To discover novel dominant strategies in iterated games.
• To investigate the potential of multi-agent reinforcement learning (MARL) in strategy discovery.
• To assess the performance and impact of newly discovered strategies on cooperation and social welfare.

Main Methods:

• Utilized multi-agent reinforcement learning (MARL) to explore extensive strategy spaces.
• Introduced a novel strategy, memory-two bilateral reciprocity, into simulated game environments.
• Validated findings through pairwise interactions, population dynamics simulations, and mathematical analysis.

Main Results:

• The memory-two bilateral reciprocity strategy consistently outperformed existing strategies in pairwise interactions.
• This strategy demonstrated dominance in evolving populations, promoting higher cooperation and social welfare.
• Effectiveness was observed across various game types and population structures (homogeneous and heterogeneous).

Conclusions:

• MARL is a powerful tool for uncovering complex strategies in game theory.
• The memory-two bilateral reciprocity strategy offers significant improvements in cooperation and social welfare.
• This research expands the understanding of dominant strategies in iterated and evolutionary games.