KL‐ONE is a system for representing knowledge in Artificial Intelligence programs. It has been developed and refined over a long period and has been used in both basic research and implemented knowledge‐based systems in a number of places in the AI community. Here we present the kernel ideas of KL‐ONE, emphasizing its ability to form complex structured descriptions. In addition to detailing all of KL‐ONE's description‐forming structures, we discuss a bit of the philosophy underlying the system, highlight notions of taxonomy and classification that are central to it, and include an extended example of the use of KL‐ONE and its classifier in a recognition task.

No abstract

A widely recognized goal of artificial intelligence (AI) is the creation of artifacts that can emulate humans in their ability to reason symbolically, as exemplified in typical AI domains such as planning, natural language understanding, diagnosis, and tutoring. Currently most of this work is predicated on a belief that intelligent systems can be constructed from explicit, declarative knowledge bases, which in turn are operated on by general, formal reasoning mechanisms. This fundamental hypothesis of AI means that knowledge representation and reasoning—the study of formal ways of extracting information from symbolically represented knowledge—is of central importance to the field. In this article, we review some of the basics of this important research area and briefly survey the kinds of techniques typically used for representation in AI programs. We also consider some important current research directions.

A fundamental computational limit on automated reasoning and its effect on knowledge representation is examined. Basically, the problem is that it can be more difficult to reason correctly with one representational language than with another and, moreover, that this difficulty increases dramatically as the expressive power of the language increases. This leads to a tradeoff between the expressiveness of a representational language and its computational tractability. Here we show that this tradeoff can be seen to underlie the differences among a number of existing represcntational formalisms. in addition to motivating many of the current research issues in knowledge representation.Cet article 6tudie une limitation computationnclle fondamentale du raisonnement automatique et examine ses effets sur la repdsentation de connaissances. A la base le probkme tient en ce qu'il peut Etre plus difficile de raisonner avec un langage de repdsentation qu'avec un autre et que cette difficult6 augmente consid6rablement i! mesure que croft le pouvoir expressif du langage. Ccci donne lieu un compromis entre le pouvoir expressif d'un langage de repdsentation et sa tractibilitt computationnelle. Nous montrons que ce compromis peut ttre vu comme I'une des causes fondamentales de la diffkrence qui existe entre nombre de formalismes dc repdsentation existants et peut motiver plusieurs recherches courantes en repdsentation de connaissances.Mots clis : repdsentation de connaissances. complexit6 du raisonnement. logique du premier ordre. schtmas, r5seaux strnantiques, bases de donntes. Comput. Intell. 3, 78-93 (1987)

Abstractclassic is a data model that encourages the description o f o b j e cts not only in terms of their relations to other known objects, but in terms of a level of intensional structurea sw ell. The classic language of structured descriptions permits i partial descriptions of individuals, under an`open world' assumption, ii answers to queries either as extensional lists of valueso ra sd e s criptions that necessarily hold of all possible answers, and iiia n easily extensible s c hema, which can be accessed uniformly with the data. One of the strengths of the approach is that the same language plays multiple roles in the processes of de ning and populating the DB, as well as querying and answering.classic for which w e h a ve a prototype main-memory implementation can actively discover new information about objects from several sources: it can recognize new classes under which an object falls based on a description of the object, it can propagate some deductive consequences of DB updates, it has simple procedural recognizers, and it supports a limited form of forward-chaining rules to derive new conclusions about known objects.The kind of language of descriptions and queries presented here provides a new arena for the search for languages that are more expressive than conventional DBMS languages, but for which query processing is still tractable. This spaceo f languages di ers from the subsets of predicate calculus hitherto exploredb y deductive databases. MotivationA database is normally used to maintain a model of some aspect of reality. T raditional data models, such as the relational one,h a vea c hieved great e ciency in data storage and retrieval by restricting m o d e ling power; in particular, the database is assumed to be a complete and accurate model of the world, where all the individual objects are restricted With Department of Computer Science, Rutgers University, New Brunswick, NJ 08903. 1 to be primitivev alues liken umbers and strings, and all their inter-relationships are known and expressly stated. While undeniably of extensive v alue, this makes traditional data models unsuitable for a number of situations, for example, when complex objects are the naturalw ay of describing the domain; when information about the domaini sincomplete or becomesa vailable incrementally; when the database should be taking a more active role in deducing relationships rather than being just a passive repository of data. These situations include those in which new artifacts are being designed e.g., CAD CAM, con guration, or an understanding of some existing situation is being builtu po ver time e.g., diagnostic situations.The eldo flogic or deductive databases 14 has emerged as one response to some of these weaknesses: incomplete information can be expressed naturally in logical languages using disjunction and existential quanti ers, and the database can infer new relationships through deductive rules. The chief drawback of this approach is computationalintractability: a generalv ersion of this problem is equivale...