Framework for Agency

The primary contribution of this research is to define and evaluate a general multi-agent distributed architecture, in which software agents utilize fuzzy knowledge structures to support agent socialization, collaboration, and cooperation for the purpose of solving a complex decision-making task. To date there have not been any documented coherent multi-agent conceptual frameworks applied to the design of multi-agent environments for handling complex decision-making tasks. The development that has occurred has generally been directed in an ad hoc way, where the needs of the specific applications are given first consideration and the agent framework has been mostly neglected. Each agent environment has been developed to specifically match the application for which it has been designed [Liu, 1998]. This ad hoc approach has not resulted in a common framework that is general enough to be utilized by different applications. To handle the reuse of the agent framework, there must be an agreement on what constitutes the properties of an agent, which is basically undetermined in the AI community. Our research has defined desirable properties for the classification of agency that will allow reusability of the proposed framework. A relational distributed agent solution is developed for the correspondence of knowledge and interaction that occurred between the participants of a multi?agent environment. Versatile relational methods provide means for knowledge acquisition, representation as well as the flexible substratum for developing platforms for various relational computations. Therefore the collaboration/communication necessary for interaction between the participating agents within several different environments is based on: 1) fuzzy relational BK?Products, 2) relational BK Capability Criterion of Generalized Morphisms, and 3) matrix manipulations. The BK?Products are used to compare and analyze the fuzzy relational structures used to represent new interpretations of the original data within the agent's knowledge domain. There are three environments used to demonstrate the reusability of the proposed framework: 1) Crisp Relational Base Model 2) Partially Fuzzy/Crisp Model and 3) Totally Fuzzy Model.