Chapter 3 - Knowledge Representation

Abstract

The development of automated systems that manifest biomedical knowledge requires a mechanism to encode that knowledge for the computer. Although it is most straightforward for software engineers to program knowledge into procedural computer code, often computer systems need to be able to operate off the knowledge in a flexible manner. More important, our knowledge of the world invariably changes over time. As a result, developers may turn to a variety of declarative, symbolic formalisms for representing knowledge, including logic, frame systems, and rules. Developers also may choose to encode an ontology that enumerates the entities in an application area, and the relationships among those entities, and to use procedural code or Web services that operates on the ontology to produce the desired intelligent behavior. Each of these approaches has relative advantages, and highlights the inescapable trade-off between the expressivity of a formalism for representing knowledge and the computational tractability of the resultant system. Knowledge representation is possible only when the developers of the system and the users of the system agree to the meanings of the symbols that constitute the computer’s knowledge base. Because meaning is something that people must ascribe to a knowledge representation—and not an intrinsic property of the symbols—a principal challenge is for the developers and the users of a knowledge-based system to be certain of those meanings, and to ensure that the meanings do not drift over time.