Printer Friendly

The effects of cognitive style on the design of expert systems.

The Effects of Cognitive Style on the Design of Expert Systems

Introduction

Two basic approaches to Artificial intelligence (AI) research have evolved: product directed or technological and theory directed or scientific [22]. According to George, Artificial Intelligence is a branch of cybernetics [9], whereas Bonnet considers AI the discipline that aims at understanding the nature of human intelligence through the construction of programs that imitate intelligent behavior [4]. The difference between these two approaches is important because AI's recent debut in the public arena has centered around an outgrowth of product directed AI, namely expert systems which donot require a complete understanding of the human decision making process.

Whereas traditional systems take relatively small quantities of input and create large quantities of output, an expert system does just the opposite. It takes a large amount of data and produces a small amount of information by integrating and analyzing the data to produce a decision or some type of recommendation for the user [21].

As Cross hs indicated, expert systems by following known facts can guide the user through a series of rule-of-thumb procedures to solve problems [5]. The primary goal of expert systems is to perform tasks that normally require intelligent behavior, for instance, making deductions and solving problems. With the use of computer programs, the knowledge and actions of an expert are captured and converted into a vital resource for problem solving. A true expert system compiles the input as an expert would, draws inferences, and creates a small amount of high quality output [15].

It is possible to highlight a number of criteria by which human intelligence can be judged, for example, the ability to make abstractions or generalizations, to draw analogies between different situations and adapt to new ones, and to direct and correct mistakes to improve future performance. The AI researcher must first select a specific activity that requires human intelligence and insight unique to the chosen topic. Next, the researcher frames some hypotheses concerning the information and reasoning process used in this activity. The process of developing the hypotheses includes drawing as much information from the expert as possible.

Although the experts may consciously try to give all of the steps in logic they consider when making a decision, a researcher may still have to theorize about each specific person's thought process [17]. The hypotheses once developed are incorporated into a computer program where the researcher can observe the behavior of the decision process in the program. A study of the limitations of the program will suggest modifications both to the theoretical base and to the program. A problem with which AI deals is the combinatorial explosion of all of the possible solutions. The basic elements of expert systems are symbols, which correspond to facts which are organized into a vast tree-like relationship diagram that represents all known interrelationships in the given situation. In accord with this symbolic representation, at each step there is a "Yes" or "No" choice, which by some is thought to have a neuronal equivalent.

Expert Systems and Al Programs

According to an article in Infosystems, the expert system goes beyond the conventional information system and tries to duplicate the decision making process of the experts by integrating the experts' heuristic knowledge with informational styles of reasoning [21]. An expert system has three main components: (1) the knowledge base, (2) the inference engine, and (3) the user interface.

The knowledge base contains both declarative and procedural knowledge. Declarative knowledge consists of facts, whereas procedural knowledge is information pertinent to courses of action. The inference engine controls how and when the information in the knowledge base is applied. The user interface is that part of the system that interacts with the user.

For a data or knowledge base, the information is gathered from the expert and stored, to be used as a reference when solving problems. A main determination of the performance level of an expert system is the quality of the knowledge base, which depends on basically three types of knowledge: factual, heuristic, and meta. Meta knowledge, which is "knowing what you know," is essential for an expert system to refine itself and learn from experience.

The inference engine is separate from the knowledge base. This component can be generalized as a reasoning mechanism which interprets the rules in the knowledge base and performs logical inferences. The user interface component provides the expert system with the capability of explaining the reasoning to the user in a language the user understands. The system must be able to "know what it knows" and explain how it applies that knowledge.

The knowledge acquisition subsystem (knowledge base component) acts as a knowledge base editor. The knowledge acquisition subsystem is essential, as it allows the knowledge base to be easily modified for incremental growth of the expert system. A distinctive feature of an expert system is the domain specific knowledge, which is used to build each expert system. Domain specific knowledge may be understood as the particular area chosen for problem solving. The other components can be packaged into expert system frameworks (expert system shells), which have been developed for use on personal computers and which provide the inference engine, the explanation subsystem, and the user interface.

The development of an expert system begins with a prototype system of a knowledge base and a knowledge acquisition subsystem, which is then expanded through several iterations until it is debugged and reasonably complete [15]. Prototype systems can be thought of as the initial "skeletal" model of the unrefined expert system. It is the separation, as clearly as possible, of the knowledge base from the mechanisms that use this knowledge, which is characteristic of the methodology of AI programs [4]. The extraction of this information from the expert is a difficult process for the knowledge engineer, whose primary concern is to gather the knowledge for the expert system.

Knowledge is not intentionally withheld from the researcher, but the expert may simply not be fully aware of how each specific decision is reached [17]. To draw out the information from the expert, the researcher may often use a technique called scenario building. The knowledge engineer describes sample situations to which the expert will react with his insight [18]. What is finally written in a programming language is the set of mechanism that integrate these items [4].

Although expert systems can prove to be a valuable asset to a company, these programs have a long design and implementation time [15]. AI programs attack problems for which no general algorithm is known, choosing a method of attack which seems promising while keeping open the possibility of changing to another alternate program. For example, a program for solving quadratic equations would not be counted as an AI program, because the general solution is know. In contrast, a program for symbolic integration would rank as an artificial intelligence program, for such a program would try one change of variable (i.e., one method) in order to simplify the function to be integrated and then another change of variable would be tried if the first variable did not succeed.

Computation or Cognition?

McCulloch and Pitts discussed "feedback" and "feed-forward" concepts in proposing the belief that there was a sense in which "machines could be made to think," which means following up modeling methods such as neural nets [20]. In practice, the only universal model capable of realistic use is a suitably programmed computer. There is an important distinction between the simulation of human behavior and the synthesis of cognitive activities for such a field as control engineering.

"AI is heir to the contraption line of gears and ratchets, steam engines, hydraulics, with one major difference ... AI is big time ... Why" [12]. The real issue has less to do with advanced technologies (or corporate specialties) than with deep theoretical assumptions. According to Western philosophy, thinking is essentially rational manipulation of mental symbols. AI is new and different, because computers actually do something very like what minds are supposed to do. "Indeed, if that traditional theory is correct, then our imagined computer ought to have a mind of its own, a genuine artificial mind, with the real issue being whether we are computers ourselves ... and whether Nature herself is `written in mathematical characters' (shapes, sizes, and motions)" [24].

This attitude derives from Hobbes, who interpreted ratiocination to be computation. "When a man reasoneth, he does nothing else but conceives a sum total from addition of parcels or conceives a remainder from subtraction of one sum from another ... These operations are not incident to numbers only, but to all manner of things ... for as arithmeticians teach to add and subtract in numbers ... logicians teach the same in consequences of word, adding together two names to make an affirmation and two affirmations to make a syllogism and syllogisms to make a demonstration ..." [14].

Carl Jung devised a classification for cognitive styles, which has been adapted over the years by numerous researchers [16]. The modifications described for managers by Blaylock and Winkofsky appear to be appropriate for individuals interacting with a computer [3].

* Extroverson (E)/Introversion (I) - two opposite

orientations toward life * Sensing (S)/Intuition (N) - two ways of judging * Thinking (T)/Feeling (F) - two opposite attitudes for dealing with the environment.

Different cognitive styles lead to different interests, values, and particularly to different problem solving techniques. For this article, only the second and third styles from Jung's original definition will be considered in relation to the ways in which individuals are expected to create and deal with expert systems.

The sensing/intuition (SN) dimension describes differences in the way decision makers prefer to perceive. Sensation is perception based on input from the five senses. Sensing individuals prefer detailed facts and seldom trust inspirations. Conversely, intuitives prefer to make jumps of thought beyond known information.

The thinking/feeling (TF) dimension deals with judging or reaching a conclusion. Thinking is a logical process aimed at impersonal feelings, and thinkers enjoy analyzing and using logic to make judgments. Feeling involves appreciation and bestowing on things personal subjective value, possibly sacrificing efficiency for harmony. Accordingly, there are four types of cognitive styles that a manager can use in adopting innovation: ST, NT, SF, and NF.

The ST type with a high sensing component will probably not skip any steps in his heuristic descriptions, but because he is concerned with the "knows" or established, he may not take easily to innovations such as expert systems. The ST manager may prefer hard facts devoid of any theoretical content that details the immediate usefulness of a product or idea. Because of his reluctance to accept new ideas, he may prefer current products and problems over new ones [20].

The NT type, on the other hand, is a research oriented individual who because of his intuitive dimension tends to explore new possibilities and ideas. However, he may pay little attention to details, and in determining his own heuristic approach to problem solving, he may skip steps.

The SF type, although concerned with factual information, prefers facts about people as opposed to facts about ideas. This is primarily due to his feelings (F) dimension, which may tend to make him accept or reject ideas based on his personal likes or dislikes. Because the SF has sensing and feeling dimensions and is people oriented, he is concerned about new technology acceptance by his subordinates. The SF may be over concerned with attaining harmony at all costs. Also, because the SF bases his evaluation of a project on his personal value systems, he may be slightly unpredictable in what he does and does not approve. His fear of "rocking the boat" may make him afraid of taking the risks necessary for innovation.

The final type of individual is the NF, who because of his intuitive (N) dimension is concerned with the future and new ideas, but because of his feeling (F) dimension is concerned with the "human factors that influence decisions." However like all intuitives, he may pay little attention to details, and like feelers, he may have a great desire for harmony. The NF manager, while at heart an innovator, may suppress his innovative nature if he feels that a change would upset the harmony of the organization. Although he could be a great motivator, someone else might have to deal with the details of the implementation of the innovation.

In summary, personal systems play an important role in determining whether or not innovation will occur. Certain aspects of a person's cognitive style can determine whether he will stimulate or hinder innovation, and implementation depends on the final analysis of people's acceptance of the innovation.

There is an even deeper issue than the acceptance or rejection of technological innovation such as expert systems. Earlier it was suggested that the inference engine of an expert system might be able to be used with multiple knowledge bases. The cognitive styles described above may affect the knowledge base, how one stores information and describes it heuristically, and also the way in which the inference engine actually runs not in the computer, but in the mind of the informer. For some researchers in AI, principles of organization are more important than speed of calculation. In the field of AI, mathematics enter more at the logical level, and while the best understood logical system is that of deductive logic, this is certainly less important than inductive or inferential logic for most of our intelligent activities [3].

Undoubtedly, it would be highly advantageous for expert systems if the brain could be shown to be basically identical to a digital computer of if the psychology could be satisfactorily based on behavioristic principles. "People don't go through every possible move every time they move (inchess); they intelligently produce far more sophisticated algorithms than we have been able to figure out how to give a computer" [22]. A program can have rules for diagnosis and still not understand in any real way the nature of the domain within which it is dealing.

Therein lies a dilemma for AI workers. Should a program attack a problem the way a human would or use a better way if it "knows" it [4]? Dreyfus put forth the notion that there are two central concepts in expert systems: simulation and synthesis [6]. The term simulation is defined as the construction of models that purport to achieve the same ends as human beings and by the same means. In contrast, synthesis involves the construction of models that purport to achieve the same ends as human beings but do not claim that this is achieved by the same means [6].

"The computer is often valued for its property, unlike the human mind, of forgetting nothing, but it is precisely the ability to forget that gives man his ability to learn, by putting aside unimportant details and replacing individual facts by procedures -- clearly intelligent -- that enable him to recover these facts when he needs them. Thus the problem of giving the computer a learning ability similar to that of the human being is that of simulating in the machine the processes by which man distinguishes between facts that are important and are therefore to be remembered and those that are not important and can therefore be forgotten. The ability to extract the essence of a set of facts rather than to store them all systematically is one of the great human strengths" [4].

Conventional Programming vs.

Expert Systems

Conventional programming techniques have been used to create computer systems capable of processing large volumes of data by complex algorithms or step-by-step procedures that guarantee a conclusion will be produced with the correct data input. Symbolic programs or expert systems are quite different. A user can halt the processing at any time and inquire why an exact line of reasoning is being pursued and how a specific conclusion was reached. In many cases, a system may make recommendation that are neither correct nor incorrect but are considered plausible [11].

Expert systems use an inferential process that allows them to arrive at an answer by invoking rules-of-thumb and symbolic language. In addition, expert systems arrive at different conclusions by applying various rules based on a specific problem. Expert systems also use what is know as backward chaining. Unlike conventional programs which are data driven or forward chaining, the inference engine starts at the goal and works "backward" through an intermediate solution in an effort to arrive at an answer. If the possible results are known and if they are reasonably small in number, backward chaining is very efficient [25]. Also, unlike conventional computer systems, expert systems can recommend a solution even though they have incomplete and inaccurate data.

A key benefit of an expert system is the increased power of the computer system to interpret data, make decision, capture critical knowledge, and make it easily accessible. Another advantage to an expert system is that it is easily modified. Although quite expensive, expert systems have the capability to make routine decisions which result in cost and labor savings, increased productivity, and improved customer service. Expert systems are also being used to train new personnel. Possibly the most important advantage of an expert system is that managers and other key decision makers have the capability to solve a class of problems that could not be solved previously [10].

In addition to the cost factors, expert systems cannot be relied upon in the same manner as the conventional computer program, since they have not been tested for every possible contingency. An additional concern is the need for security about information about specific corporate operations, which would lead the designing systems analyst not to fully understand and meet the needs of the user. Additionally, management resistance may be a formidable issue [2].

Expert Systems in Today's Market

Many corporations are looking at expert systems to contribute towards a common goal: to increase efficiency and profits by taking on the role of a consultant, a checklist, a training aid, a communication medium, or as a combination of these. Probably the most important aspect of an expert system is that vital information can still be retained within the organization, even after the people chosen as experts physically leave. In a sense, the corporation has developed a knowledge "bank," where the expert's knowledge can remain without having to maintain the expert's physical proximity.

This new technology has been successfully implemented in many companies and government sectors, with the most recent market for expert systems being the financial service area where they are expected to play a major role. Although a relatively small number of expert systems are presently in use, many are in development, and analysts are enthusiastic with the market's demand [13].

According to Dun's, accounting firms were the first financial companies to utilize this new technology, not only as users but as consultants and vendors of systems [8]. The banking, insurance, and stock market industries are also recognizing the potential benefits of expert systems. From the results of a survey within the financial services industry, 43 percent of the firms studied have applications that now use or are planning to use expert systems. Of the companies studied, 25 percent already have an expert system in place [7]. In another study of the top 100 insurance companies in 1986, 30 percent of these companies had expert systems in use or under development. In 1987, the percentage of the top 100 firms had increased from 30 percent to over 60 percent [23].

Also, expert systems can be beneficial in the stock market by collecting large quantities of information and contradictory theories on how to use and examine this information and offer unique conclusions.

Conclusion

The use of artificial intelligence, specifically expert systems, will continue to have a tremendous impact in the future in various industries. Many corporations perceive expert systems as having a strategic benefit, which will enhance the development of additional quality expert systems used in information systems. The nature of the computer of the future and the impact it will have on our daily lives will be determined by what happens in the field of Artificial Intelligence both in the next few years and in the next several decades. Intelligent computers and expert systems are bound to have a great effect on the way we do business. However, cognitive styles will have an even greater effect on how we design and implement such systems.

References

[1.] "Artificial Intelligence: Expert Systems Moving From Glamour Technology to Workhorse." Infosystems, September 1986, p. 14.

[2.] Bentley, D., I. Ho, and J. Whitten. Systems Analysis and Design Methods. St. Louis: Times Mirror/Mosby, 1986.

[3.] Blaylock, B.K. and E.P. Winkofsky. "An Explanation of R&D Decision Process Through Individual Information Processing Preference." R&D Management, Vol. 13, No. 3, July 1983.

[4.] Bonnet, A. Artificial Intelligence: Promise and Performance. Englewood Cliffs, New Jersey: Prentice Hall, 1984.

[5.] Cross, T.B. "Expert Systems and Intelligent Building." Journal of Property Management, July-August 1986, pp. 70-72.

[6.] Dreyfus, H.L. What Computer Can't Do: Critique of Artificial Intelligence. New York: Harper and Row, 1972.

[7.] "Expert Systems in the Financial Services Industry." Coopers and Lybrand Survey Report.

[8.] "Financial Services: Slow Start, Huge Potential." Dun's Business Monthly, October 1986.

[9.] George, F.H. Artificial Intelligence, Its Philosophy and Neural Content. New York: Gordon and Breach, 1984.

[10.] Guterl, F.V. "Computers Think for Business." Dun's Business Monthly, Vol. 128, Issue 4, October 1986, pp. 30-37.

[11.] Harmon, P. and D. King. Expert Systems. New York: John Wiley, 1985.

[12.] Haugeland, J. Artificial Intelligence: The Very Idea. Cambridge, Massachusetts: MIT Press, 1985.

[13.] Hickox, F. "Learning About Artificial Intelligence." Institutional Investor, July 1986.

[14.] Hobbes, T. Elements of Philosophy. Moleworth, Vol. 1, 1839-45.

[15.] Jacobs, S. and R.T. Keim. "Expert Systems: The DDS of the Future?" Journal of Systems Management, Vol. 37, No. 12, December 1986, pp. 6-14.

[16.] Jung, C. Psychological Types. New Jersey: Princeton Press, Reprinted 1971.

[17.] Leonard-Barton, D. and J.J. Sviokla. "Putting Expert Systems to Work." Harvard Business Review, Vol. 66, No. 2, March-April 1988, pp. 91-98.

[18.] Liebowitz, J. An Introduction to Expert Systems. Santa Cruz, California: Mitchell Publishing, 1988, pp. 37-38.

[19.] McClellan, E. Personal interview with Helen Vassallo, 1988.

[20.] McCulloch W. and W. Pitts." A Logical Calculus of the Ideas Imminent in Nervous Activity." Biological Math & Biophysics, Vol. 5., 1943, pp. 115-133.

[21.] Rohm, W.G. "Artificial Intelligence: A Remote Promise." Infosystems, September 1986, pp. 52-54.

[22.] Schank, R.C. and P. Childers. The Cognitive Computer: On Language, Learning, and Artificial Intelligence. Reading, Massachusetts: Addision Wesley, 1984.

[23.] Shpilberg, D., J. DeSalvo, and S. Michalski. "Tomorrow's Expert Systems." Coopers and Lybrand Publication, Reprinted from Best's Review, May 1987.

[24.] Sowa, J.F. Conceptual Structures: Information Processing in Mind and Machine. Reading, Massachusetts: Addison Wesley, 1984.

[25.] Stefik, M., R. Aikins, J. Balzer, L. Benoit, L. Birnbaum, F. Hayes-Roth, and E. Sacerdoti, "Basic Concepts for Building Expert Systems." In Waterman, D.A., ed. A Guide to Expert Systems. Reading, Massachusetts: Addison Wesley, 1986.

Helen G. Vassallo is Assistant Professor of Management at Worcester Polytechnic Institute in Worcester, Massachussetts; John M. Lanasa is Assistant Professor of Marketing and Management Information Systems at Duquesne University In Pittsburgh, Pennsylvania.
COPYRIGHT 1990 St. John's University, College of Business Administration
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1990 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:simulation of human intelligence in the design of expert systems or computers
Author:Vassallo, Helen G.; Lanasa, John M.
Publication:Review of Business
Date:Dec 22, 1990
Words:3876
Previous Article:Approaching convertibility in Eastern Europe and the Soviet Union.
Next Article:Community banks and the importance of lending.
Topics:


Related Articles
Computer smarts changing business.
Artificial intelligence and natural confusion.
The AI factory; how artificial intelligence will create 'smart plants.' (Cover Story)
Better decision making through expert systems for management.
Expert systems: application to inventory control and production management.
The role of expert systems in improving the management of processes in total quality management organizations.
Understanding our differences: there are many things that set us apart as individuals--fingerprints, DNA, and the ways in which we think and learn....
Mind-expanding machines: artificial intelligence meets good old-fashioned human thought.
The future of simulation technology for law enforcement: diverse experience with realistic simulated humans.
Using AI to learn about algorithms.

Terms of use | Copyright © 2016 Farlex, Inc. | Feedback | For webmasters