Bootstrapping Trust Evaluation Using a Trust Certificate Model

Trust plays a vital role in the decision to initiate any interaction. Rational agents may use past experiences and other agents’ opinions to decide to trust, but due to the nature of open multi-agent systems, where agents can dynamically join and leave the system at any time, agents may find themselves dealing with complete strangers whom neither they nor their friends have encountered before. This situation forces the agents to choose partners randomly, which significantly increases the risk of encountering unreliable agents. For instance, service requesters may become reluctant to initiate communication with newly-joined service providers. And when the newcomers are service requesters, who are willing to exploit the environment, service providers may also hesitate to start any connection with them. As a result, newcomers are excluded from the competition and old agents lose the possibility of interacting with better agents. In this thesis, we address that issue by creating a Trust Certificate (TC) model in which each agent is equipped with a certificate that works as a reference by providing information about its holder. The information is obtained and stored by the agent itself and is available to other agents who request it to evaluate the holder’s trustworthiness for a potential interaction. The stored information is about the agent’s role in the society and its performance in past interactions. The TC model allows agents to retrieve reputation information and make initial trust evaluations when evidence is unavailable. It also helps agents to avoid the need to make random partner selection due to the information scarcity. We show how this model enhances the interaction process between agents by evaluating it in the context of a simulated multi-agent system