Tracking reliability and helpfulness in agent interactions

Abstract

A critical aspect of open systems such as the Internet is the interactions amongst the component agents of the system. Often this interaction is organised around social principles, in that one agent may request the help of another, and in turn may make a commitment to assist another when requested. In this paper we investigate two measures of the social responsibility of an agent known as reliability and helpfulness. Intuitively, reliability measures how good an agent is at keeping its commitments, and helpfulness measures how willing an agent is to make a commitment, when requested for help. We discuss these notions in the context of FIPA protocols. It is important to note that these measures are dependent only on the messages exchanged between the agents, and do not make any assumptions about the internal organisation of the agents. This means that these measures are both applicable to any variety of software agent, and externally verifiable, i.e. able to be calculated by anyone with access to the messages exchanged.