Date of Award


Level of Access Assigned by Author

Campus-Only Dissertation

Degree Name

Doctor of Philosophy (PhD)


Computer Science


Roy M. Turner

Second Committee Member

Clare Bates Congdon

Third Committee Member

Philip M. Dickens


Agents in open multiagent systems (OMAS) are likely to have a social trust responsibility that is not found when multiagent societies are primarily closed. Open systems are those that allow agents to come and go as they wish, while closed systems rely on a stable society of agents (usually under the control of a single administrator). The open systems agent must make decisions involving its level of trust of other agents in the system against a constantly shifting arrangement of participants, behaviors and situation.

Based on the agent's goals, another agent's behavior may be viewed as undesirable – either through malicious intent of its designer or through some mismatch between the system and the self-interest of the undesirable agent. We refer to these undesirable agents as miscreants.

In the dynamic environment of an open system with self-interested agents, it may be difficult or impossible to find a single strategy for determining whom to trust that is effective under all circumstances. In such a dynamic environment the agent may find the need to switch trust strategies to continue to be successful.

Context-aware agents gain an advantage over agents that rely only on current information by recognizing the situation and bringing to bear strategies that have proven effective in such a situation in the past.

This dissertation describes how such an agent can improve how often it makes correct trust choices in an OMAS beset by miscreant agents by using observations of the system, inference about the behaviors of other agents, and a base of knowledge about the known contexts. It describes a series of experiments performed to quantify how much improvement was realized by a context-aware agent in making correct decisions about trusting another agent. This dissertation describes how effective this approach was in one domain and describes how such an approach might be expanded to more complex systems.