The Roles of Reliability and Reputation in Competitive Multi Agent Systems

In competitive multi-agent systems, the act of requesting recommendations to evaluate the reputation of an agent is a cost. On the contrary, an agent computes the reliability of another agent without paying any price. This implies the necessity for the agent to decide in which measure to use the reputation with respect to the reliability when he must select the most promising interlocutors. In the past, two main schools of thinking emerged, one assuming the necessity of always exploiting both reliability and reputation, the other declaring that the use of reputation requires, in terms of cost, more disadvantages than advantages. In this paper, we try to evaluate the role of using reputation with respect to reliability, depending on some characteristics of the involved scenario, as the size of the agent population, the percentage of unreliable agents and the percentage of agents provided with a reputation model. To this purpose, we introduce a novel framework that contains both a reliability-reputation model and a mechanism to integrate reliability and reputation in a unique synthetic value. We have used this framework to build a competitive agent, which we have run on the well-known ART platform, and we have performed several experiments to study how the characteristics of the scenario are correlated to the exploited reliability and reputation model.