Knowledge representation - an AI perspective

Most researchers to date in artificial intelligence has been based on the knowledge representation hypothesis, that is, the assumption that in any artificial intelligence (AI) programme there is a separate module which represents the information that the programme has about the world. As a result, a number of so-called knowlege representation formalisms have been developed for representing this kind of information in a computer. This text discusses the most popular knowledge representation languages - logic, production rules, semantics (networked and frames), and also provides a short introduction to AI systems that combine various knowledge representation languages. The knowlege representation hypothesis has been challenged by the re-emergence of a new style of computing, variously called parallel distributed processing, connectionism, or neural networks. These approaches are discussed in a separate chapter, and the arguments in favour of and against parallel distributed processing are reviewed.