Time-critical decision making with communicating influence diagrams

In this dissertation, we design a system for time-critical decision making using communication. In addition, we present the design and implementation of a multiagent environment (MAE), in which this time-critical decision making system works. The complexity of today's workplace and the amount of information dramatically increasing everyday broaden the role of intelligent agents to alleviate human burdens while working with or without human intervention. Even though many single agents have been successfully created and used in order to improve quality of service and to reduce costs, such stand-alone agents have limited ability, analogously to single human experts, because of limited knowledge capacity and physical limitations. In order to overcome such limitations, in this dissertation, a probabilistic multiagent environment is designed and implemented, in which each cooperating agent is represented by a probabilistic network. The construction of this multiagent environment includes the following: (1) agent design using Bayesian networks and influence diagrams (decision networks), (2) communication between agents, (3) treatment of missing information, (4) reliability checking algorithms, (5) value of information computation in MAE, and (6) time-critical decision making algorithms for a MAE. Finally, these methods and algorithms were applied to controlling the speed of the fleet in a hypothetical operation based on “Eastern Exit,” the noncombatant evacuation operation (NEO) from Mogadishu, Somalia.