Artificial Intelligence. Representing Knowledge in AIMDS

AutoreN.S. Sridharan
CaricaAssociate Professor in the Department of Computer Science at Rutgers University, New Brunswick (New Jersey, USA)
Pagine201-221

    [Editor's Note] This article reproduces the paper presented by the Author at the International Conference oe ´Logica Informatica, Dirittoª (Florence, April 1981) organized by the Istituto per la documentazione giuridica. The final draft of this paper, having arrived too late to be included in the Proceedings of the Conference at present being published by North Holland (Amsterdam), is printed here because of the importance to our readers of the themes discussed.


Page 201

@1. Logic, Artificial Intelligence and Psychology

The purpose of this paper is to provide a tutorial on artificial intelligence: especially on methods of developing a knowledge base. However I have a second aim that colors the manner In which I conduct this tutorial, namely, to introduce to you a high level computer language, AIMDS [2] [3], that I have been developing over the past five years. As I introduce this language I shall emphasize the facilities for the representation of knowledge.

AIMDS has been used In the development of TAXMAN Il system [4]. This paper also serves as a companion paper to [4], using the examples from TAXMAN II, but attempting to give the reader a general idea of the techniques of knowledge representation.

The language that I'm going to talk about is based on logic and set theory, but It is a programming language. It will be somewhat different from computer programming languages that you might be acquainted with. It arises out of an Interesting reconciliation between the power of logic and the limits of a computer!

AIMDS is guided in Its development by the Influence of a psychologist, Professor Charles Schmidt, with whom I am collaborating on developing a psychological theory of recognizing intentions and understanding of certainPage 202 actions. In developing such a theory, modelled as a computer program, we have attempted to deal specifically with limitations of memory and limitations of procesing power that humans have. But people do exhibit extraordinary intelligence. They can organize their information, focus their oration on what is relevant and structure their processes to overcome - voir own limitations to meet the demands of a task. This raises the prime challenge for the field of Artificial Intelligence - how do we make the rori:putef do the same.

Thus logic attempts to set the ideals in rational thinking. Models of thinking developed in Artificial Intelligence take into account the need to conserve computing resources. Cognitive models in Artificial Intelligence go further in attempting to display processing characteristics which are similar to those of humans.

@2. Statement of Purpose

The structure of my talk would be to give you a quick overview of artificial intelligence; and come to focus on the idea of knowledge as being the essence of how our intelligence operates, and then to go directly into the language AIMD8. i shall draw examples and illustrations for this presentation trom the TAXMAN application, so that you will see some representation of, legal concepts and facts. I am keenly aware of the diversity of this audience, therefore. I am trying to keep my presentation at a very simple level, and I hope to communicate as much as I can. My purpose will be served if I arouse enough interest in you that you will follow up the newly developing field of Artificial Intelligence.

In the ideas underlying the language there are some solutions, at least some suggestions about solutions, to some of the problems that are being discussed at this conference, For example, we heard discussions about the distinction drawn between ´descriptionsª and ´prescriptionsª, between ´propositionsª and ´practitionsª. In AIMDS we also make the distinction between simulating the effects of an action and the description of an action (whether past or contemplated, suggested, accepted or rejected). In further discussions it would be fruitful to examine the nature of the distinctions being drawn and to relate them to the analyses oi the problems presented at this conference. Similarly, in AIMDS there is an object level with descriptions of some domain of discourse (for TAXMAN dus is the world of corporate structure and transactions); and there is a meis-level that describes the conceptual organization oi the object level. It's quite impossible to go into details or these issues which the conference participants have brought to the forefront this morning.

Page 203

@3. Brief Survey of Concepts in Artificial Intelligence

There are 4 basic terms that I would like to introduce to you at a survey of artificial intelligence. The first idea is that of a symbol. The beginnings of artificial intelligence are rooted in the realization that the computer is primarily a symbol processor not just a number processor. Symbolic information can be such things as names, text material, rules, definitions and so on. A symbol standing alone in isolation is not of much value. So we bring in the next idea which is structure - symbol structure. Symbols are combined to form symbol structures with specified connections among its parts1. Aggregation is the most familiar method of building symbol structures. Take a common example - in an address label, the name, the street' address, the city, the region and the country all appear together. That is aggregation of different pieces of different symbols to make a symbol structure. Association is another form of bringing the symbol structure together which is a very useful concept in organizing a computer memory. The use of a pointer or an address is a way of achieving this in a computer. At the end of your paper you include citations; that is association. If we take a set of papers from these conference proceedings and follow the citations we see the pattern of association. These are the two main ideas in how we construct symbol structures with symbols.

Symbol structures can be transformed computationally into other symbol structures. Now a set of symbol structures and their transformations constitutes a space. One important space for artificial intelligence is the search space. A problem is formulated to a computer when you define the search space, and you define the criteria by which a solution must be judged. Very early work in artificial intelligence counts success in various forms of game playing and proving theorems in the prepositional calculus. Search spaces for game playing, for. example Chess, can be large. That is an understatement; they can be enormously large. Years ago, it was fashionable for various popular writers to describe how the fastest of the conceivable computers, if they could operate a thousand times faster than they can, would take longer than the putative age and lifetime of the universe to search such spaces exhaustively, So, exhaustive search through a search space is clearly infeasible as the core idea of artificial intelligence. Thus enters the idea of doing heuristic search. Heuristics are used in two ways. Firstly, at the outset, generate a subspace for search, that is generate only a small fraction of the search space, not all of it. This subspace is a plausible subspace; solutions are more likely found there. Secondly, use the idea or pruning the search space. After you consider a candidate you can eliminate a fragmentPage 204 of search space and therefore can confine the search to even a smaller fraction. Programs such as the Geometry Theorem Prover and DENDRAL are excellent examples of heuristic search that exploit both Ideas2. Ineffective search of a space is a mark of Ignorance, I adopt a pragmatic view of knowledge. Knowledge is characterized as that which leads to effective reasoning and effective action. Can knowledge be exhibited only as effective behavior, Recent work in Artificial Intelligence attempts to ;be systematic In describing such knowledge in the form of symbol structures. Knowledge Representation consists In the systematic attempts to provide symbol structures and transformations for encoding knowledge, and forms a primary focus of current work In Artificial Intelligence. Knowledge acquisition and validation forms another focus.

Table 3-1: What is Knowledge?

Knowledge is that which leads to effective reasoning and effect i ve action.

The basic method of reasoning involves SEARCH; knowledge leads to abbreviated search.

All structures we deal with are, at present, discrete and symbolic.

@4. Knowledge, its Types and Levels

What kinds of knowledge are there We formulate four kinds of knowledge: Knowledge of Individuals, knowledge of classes, ´how toª knowledge and meta-knowledge,

Knowledge about individuals: We have to represent knowledge about Individuals. We have to talk about their attributes and properties, and we have to describe relationships among Individuals. Knowledge about a given collection of Individuals and their relationships constitues knowledge about a state of affairs (minimally).

Knowledge about classes: Individuals belong to classes. We have to' represent knowledge about classes. Such knowledge. Includes properties of classes, relation to other classes, relationships between classes and Individuals,

Knowledge of procedures; Recognition of membership of an Individual In a class, comparison of Individuals for similarity and differences, existence ofPage 205 an individual with specified combination of attributes, modeling of state changes are some examples of procedures that utilize knowledge. Knowledge about these procedures include descriptions of what the procedures accomplish and how the procedures operate.

Meta-knowledge: The above may be considered knowledge at the object level. At the tneta-level we talk about the scope and limits of what is known about instances...

Per continuare a leggere

RICHIEDI UNA PROVA

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT