A sequence-modeling approach to cooperation in Public Goods Games via Multi-Agent Transformers

As autonomous and intelligent systems become increasingly integrated into various sectors of society, such as transportation, digital assistance, and manufacturing, understanding the principles that govern their coordination and collaboration is becoming progressively more important. The Public Goods Game (PGG) is a game-theoretic model where individuals must balance self-interest with collective benefit, serving as an ideal framework for exploring collaboration dynamics. Reinforcement learning (RL) has proven to be powerful in the field of autonomous agents, while transformer models have gained significant attention in sequence modeling. This paper introduces a novel approach to studying cooperation by applying Multi-Agent Transformers (MAT) to the Public Goods Game. Leveraging the ability of transformers to capture complex sequential dependencies, we demonstrate that MAT can model more sophisticated inter-agent patterns and reveal deeper insights into the nature of cooperation, thus contributing to the development of effective multi-agent systems.