Using the System-Model-Operator Metaphor for Knowledge Acquisition

The techniques for building expert systems have advanced from tools that provided an “empty knowledge base” with a backward-chaining inference engine, such as Emycin,1 to tools that allow for an explicit representation of the domain-general control knowledge necessary for a specific task, such as diagnosis or design.2 6 Task-specific tools incorporate a way of organizing knowledge and an inference procedure for applying this knowledge. As researchers analyze these tools to generalize and integrate different methodologies, we need to understand the relation between specific problems and general ways in which knowledge can be organized and applied. In essence, how should we formalize the reusable part of the knowledge base so its capabilities and limits can be related to new problems? Task-specific architectures need to address the following distinct issues:7 • What system is being modeled? • For what purpose, or task, is the system being modeled? • What subsystems and subprocesses are represented in the general model (that is, in the knowledge base)? • What subsystems and subprocesses are represented in a situation-specific model (that is, in a problem solution)? • What relational networks (what kinds of hierarchies and transitional graphs) are used to represent processes? • What are the operators, or inference procedures, for constructing situation specific models? • How are the relational networks and inference procedures implemented in a programming language (for example, in frame and rule-based languages)?