Confidence as a Means to Assess the Accuracy of Trust Values

Abstract

Open, heterogeneous multi-agent systems (MAS) have to cope with a variety of uncertainties introduced by the systems' participants and their environment. In such systems, agents can use trust values that characterize the behavior of their interaction partners as a measure of uncertainty, allowing agents to make more appropriate decisions. However, because of the systems' dynamics and the agents' limited knowledge, the accuracy of these trust values is often limited as well, introducing another form of uncertainty. In this paper, we present confidence as a general concept to indicate the degree of certainty that a trust value describes the actual observable behavior of an agent. On the basis of three open MAS, we identify different criteria the confidence depends on by revealing situations in which trust values can be inaccurate and therefore impair an agent's decision. By means of scenarios, we show that a trust-aware agent can increase its own utility when its decisions are based on the confidence in trust values.