Title:Now that I Have a Good Theory of Uncertainty, What Else Do I Need?

Abstract:Rather than discussing the isolated merits of a nominative theory of uncertainty, this paper focuses on a class of problems, referred to as Dynamic Classification Problem (DCP), which requires the integration of many theories, including a prescriptive theory of uncertainty. We start by analyzing the Dynamic Classification Problem and by defining its induced requirements on a supporting (plausible) reasoning system. We provide a summary of the underlying theory (based on the semantics of many-valed logics) and illustrate the constraints imposed upon it to ensure the modularity and computational performance required by the applications. We describe the technologies used for knowledge engineering (such as object-based simulator to exercise requirements, and development tools to build the Knowledge Base and functionally validate it). We emphasize the difference between development environment and run-time system, describe the rule cross-compiler, and the real-time inference engine with meta-reasoning capabilities. Finally, we illustrate how our proposed technology satisfies the pop's requirements and analyze some of the lessons reamed from its applications to situation assessment problems for Pilot's Associate and Submarine Commander Associate.