Communication patterns affect the collective performance of social agents

More often than not, we work in group settings where the communication structure within and between groups governs the flow of information among individuals. This structure can be designed to optimize group performance, enabling individuals to solve tasks in the shortest time or achieve the highest reward. In this paper, we explore the effects of communication patterns on the collective performance of a group of interacting agents. The agents are tasked with performing an action, where the reward depends on their skill in executing that action. At any given time, an agent switching actions has two choices: to learn from the best-performing connected agent (with probability q), or to randomly explore the action space (with probability \(1-q\)). Our findings indicate that decentralized networks enhance collective performance by increasing both the overall group reward and the maximum reward achieved by an individual. Conversely, in more centralized and hierarchical networks, we observe that better connected agents, as reflected by their betweenness centrality, exhibit better performance.