Constraint satisfaction as a support for decision making in software agents

The U.S. Navy has been trying for many years to automate its personnel assignment process. Periodic assignment of personnel to new jobs is mandatory according to Navy policy of sea/shore rotation. Though various software systems are used regularly, the process is still mainly done manually and sequentially by Navy personnel, called detailers. An Intelligent Agent, IDA has been designed and implemented, which applies cognitive theories and new Al techniques to produce flexible adaptive human-like software. Inside IDA, the constraint satisfaction module is responsible for satisfying the requirements of Navy policies, command needs and sailor preferences. In order to enhance decision quality various methods are created, investigated, tuned and implemented, which are of primary concern to this dissertation. These methods combine benefits of operational research, cognitive science, neural networks, fuzzy systems, and statistics. Results show that high level human like decisions are possible in a noisy, rapidly changing environment under time pressure within an intelligent agent framework.