Dialogue games and trust for communicating agents

Multi-agent applications are primarily based on agent interactions, which are constrained by the trust of participating agents. Two important issues in these applications are how agents can communicate in a flexible and efficient way and how an agent can authenticate information conveyed by other agents in the system. In this thesis, we present a new communication framework and trust model addressing these issues by considering three factors. The first factor is about the flexibility, complexity, soundness, and completeness of the communication protocol. The second factor is about the classification of agents from a trust point of view using direct interactions. The third factor is related to the categorization of the agent's chains through which the information is transmitted. Such a categorization is based upon the reliability of the agents in the chain. The model aims to examine all available data in order to determine the trustworthiness of agents as transmitters of information. This approach is the first attempt in multi-agent systems towards classifying agents in order to accomplish trust. We also propose a thorough set of criteria and policies to assign different degrees of trustworthiness to each agent and consequently to the chains in which they appear. Agents are considered autonomous and they interact flexibly using a set of logical rules called dialogue games. Termination, soundness, and completeness results of the communication protocol are proven and its computational complexity is addressed. The proposed approach is also evaluated. Keywords: Trust, Dialogue Games, Multi-Agent Systems, Agent Types, Agent Characteristics, Chain of Agent