Knowledge Representation

Abstract

Knowledge representation has been called the most central problem in artificial intelligence, and every system developed in AI has to deal with it at some level or another. Cercone and McCalla (1987) state that “Knowledge Representation is basically the glue that binds much of AI together”; knowledge seems to be a prerequisite for any kind of intelligent activity. All areas of research in AI from game playing to expert systems, from computer vision to natural language processing, require vast amounts of knowledge about the domain within which the system will be operating. Chess programs need to know not only which moves are legal for each piece but also all kinds of heuristics or rules of thumb for deciding the best strategy, and for knowing when the game is, for all intents and purposes, lost or won. Expert systems, of course, are the very embodiment of an expert’s knowledge and experience translated into a series of condition-action rules (when this condition occurs, then take that action). In computer vision it was found that there is too wide a gap between raw image data and any kind of intelligent use of what is ‘seen’. In order to make sense of these images various kinds of knowledge are necessary; for example, knowledge about brightness and brightness change, knowledge about the relation of these changes to texture, edges and surfaces, etc. Natural language processing also requires vast amounts of knowledge; knowledge about the syntax of language, the meaning of words, knowledge about what is assumed as well-known in a conversation and what is implied by a particular choice of words. As we saw in Chapter 2, both literal and metaphoric utterances go beyond the level of the words in the sentences and involve entire semantic domains. Beardsley talks about the connotations of words, proponents of the anomaly view discuss the semantic categories involved in metaphor, and Black claims that words entail entire systems of commonplaces. All of these are a form of knowledge that is necessary for any kind of language comprehension.