Learning Reliability Models of Other Agents in a Multiagent System

This paper describes how agents in a multiagent system can use local domain information to learn reliability models of other agents. Using these reliability models agents can evaluate the quality and validity of messages communicated to them. The dynamically learned, explicit knowledge of the quality of transmitted information is significant in an intelligent system where the autonomous agents have only partial and incomplete view of solutions and the state of problem solving activities, and where any cooperation or coordination behavior relies on communication. We describe how agents learn reliability models, and how these models can be used in multiagent problem solving coordination and coherence.