ARTIFICIAL INTELLIGENCE: A GUIDE TO INFORMATION TECHNOLOGY
by Andreas Sofroniou
December 9, 2000

Artificial Intelligence (AI)

 

Expert Systems

Until recently, computers have been concerned with handling numbers and data. Today, many fields of human endeavor express our ideas and problems in non-numeric (symbolic) terms. An Expert System is concerned with concepts and methods of symbolic inference and representation of knowledge. Special computer languages, principally LISP and Prolog, have been developed to process symbolic information. Computer programs written in these languages can be much shorter and clearer than their conventional equivalents. The penalty which is paid, is that on conventional computers the programs are much less efficient and a new generation of computers and languages have and still are being developed to improve their execution.

In October, 1981, the Japanese announced their intention to embark on a research program to develop fifth generation computer systems. This program, jointly undertaken by universities, independent institutes and major Japanese companies, aims to move the whole basis of computing from data to knowledge processing. What they hope to achieve, is simply the development of knowledge-based systems, which will be the future driving force of their economy.

Faced with this threat, most of the technologically advanced countries have launched vigorous research programs of their own. In the U.S.A., both the public (via the Defense Research Projects Agency) and the private sector are investing heavily in artificial intelligence research tools.

The UK's response was the Alvey Program, a collaborative research project involving academic and industrial effort. What all activity shows, is that the future of computing lies in Artificial Intelligence (AI) systems. Today, many products have computer systems as integral components; these systems can only benefit from artificial intelligence, so products with ‘added AI value' will give a considerable market advantage to companies which produce them.

Many companies in the fields of defense, electronics, communications and manufacturing are already planning AI programs for this reason. In the financial sector, as well, market pressures will leave organizations with little choice. Without the power and versatility of AI, survival will be difficult.

The areas of expert systems is fashionable at the moment. Over the past fifteen years, interest in expert systems has increased dramatically, especially in the commercial field. In practice, it must be understood that expert systems are not threatening traditional computer skills.

The objective in constructing an expert system is to program into the computer a representation of human knowledge. There is an area of controversy over what exactly constitutes the definition of an expert system. Basically, this rests on artificial intelligence, or computing science, to form the base on which such systems are founded. That is, whether the primary stress should be put on knowledge, or logic.

Expert systems deal with knowledge which is accepted and certified, rather than working at a point where new knowledge can then be refined and checked pragmatically, against a recognized specialist.

Problems occur where there is controversy over what constitutes knowledge. If statistics are involved, there can be more than one interpretation of any results of which are obtained.

The applications of expert systems are wider in scope, than mathematical techniques, because heuristics can be used in situations which are not precisely defined, such as decision making.

 

Applications Of Expert Systems

There has been steady growth in the commercial adoption of expert systems, although the number of firms employing them is still small, compared to the total number of companies using computers. Those organizations using these systems have been surprised at the progress that can be made in this area, if the requirements are not excessively ambitious at the outset. Companies that have introduced expert systems, have done so, not to replace the work of existing computer systems, but to add to the functions that can be done automatically for the firm.

Expert systems can be employed for a range of relatively simple operations, such as fault diagnosis, where they can marginally improve on the performance of the average fitter. Other areas of applications which are being currently used are for sales advice, customer order handling and some aspects of training.

Although such systems would benefit smaller firms, it is the larger companies which employ their own research groups and have larger budgets to accomplish the latest technical advances.

Much of the early development work was done through academic research. The business community has been particularly active in investigating knowledge communication systems, a branch of expert systems which concentrates on eliciting and distributing knowledge for human consumption other than using it for programming.

Research concentrates on defining that part of human knowledge and skill which passes on verbally, or is taken to be tacit knowledge by the practitioners. The technology is being used to write down, check and define this type of knowledge. In some cases, the resulting programs can be used for training purposes.

Applications of expert systems to medicine have already been given much publicity. Although there are not many systems in routine use, as yet. The potential of prototypes has been demonstrated and are now close to the flowering of expert systems as an aid to doctors and physicians.

There are limitations which are under investigation at the moment and it is clear that applications of this technology will continue to develop rapidly. Expert systems are potentially an extremely useful aid to many human activities, not least, because they aim to utilize nothing more elaborate than plain common sense.

 

History Of Artificial Intelligence

Artificial Intelligence (AI) has been a recognized, although somewhat dispersed field of endeavor since the mid-fifties; the term ‘expert system' is quite recent. Its main force for popularity is derived from two major sources. First, the recognized inflexibility of established commercial computing and second, the promise given by new insight that intelligent machine behavior could be achieved. The solution proposed to eliminate many evolving difficulties, is to refocus attention on the ‘knowledge', rather than the ‘data'.

Among the outstanding problems of knowledge engineering and expert systems development, are those relating to the difficulty of transferring their ‘knowledge' from the human expert to the machine. It has been suggested, that one solution to the difficulty, lies in the use of inductive method for transfer, which would allow the expert to input examples from which the program can infer the rules, i.e. the inductive method should allow some automation of the knowledge engineer's task.

The resulting expert system must, of course, conform to the:

  · agreed expert systems requirements,
  · extend over a reasonable domain of expertise,
  · give reliable decisions,
  · its ‘reasoning' must be accessible, both for the expert and non-expert user.

A further constraint, could require the expert system to deliver new knowledge, recognizable by the human expert, as such.

When faced with a complex task, the experienced programmer splits it into sub-problems. Each sub-problem is then programmed as a procedure and procedures can be nested to any level. The precise choice and hierarchical order of sub-problem in the province of top-down design is called ‘structured induction'.

 

The Fifth Generation Language

The next generation of computer systems - the fifth generation language (5GL) - seeks to develop artificial intelligence (AI), which will greatly expand the applicability of computer systems from data processing into knowledge processing.

Since the fifties, computer researchers have believed that it is possible to make computers which can, to a limited extend, be considered to be capable of reasoning. The technical problems are enormous and it is only the last ten years, or so, that commercial organizations have begun to exploit artificial intelligence technology.

In its simplest form, artificial intelligence represents the ability of a computer to encompass the knowledge and problem-solving skills of experts from a range of disciplines; it helps the computer user to interpret information from several viewpoints, presenting that information in a form which allow a well informed decision about a particular problem to be made.

 

Techniques

Artificial intelligence techniques play a major part in improving the interface between humans and computers - image and speech understanding and robotics have all benefited from AI research.

By the end of the year 2000, it is forecast that more than thirty percent of existing computer applications will use artificial intelligence techniques and an even greater volume of new applications of computers will have arisen, addressing complex problems, which cannot currently be solved using today's computers.

Many of the world's largest companies are already developing AI-based applications. These are frequently called ‘expert', or ‘knowledge-based' systems. They seek to assist, or even replace ‘experts', thus releasing manpower for more creative tasks.

To develop these complex systems quickly and efficiently, a new type of computer is needed. This is because the computer language normally used for AI applications requires much more processing power, than most conventional computers and more importantly, the computer power is used in a different way.

The principal difficulty lay in the fact that the fifth generation languages, which are most suitable for AI applications are very processor intensive and require novel computer architectures for efficient execution. Since commercial AI applications need, almost without exception, to interact with ‘conventional' data processing systems, this imposes a serious limitation in their use. One solution is to network these ‘machines' to a more general purpose computer.

 

Meeting The Needs Of Industry

Applying artificial intelligence techniques to industrial applications is not very easy. Some of the most useful AI-based solutions, such as continuous speech recognition, will not be practicable until later in the decade. However, a number of useful applications have already been developed to the extend that they show a substantial commercial benefit. The most commonly discussed application is the expert system.

The expert system seeks to encapsulate the knowledge and skills of highly qualified individuals (usually a scarce resource) and increase productivity by applying these skills equally and consistently throughout the organization. Expert systems are particularly applicable where the task is repetitive, but low level. For example, configuring large computer systems is a good application for an expert system, but designing computers is not.

One of the most frequent mistakes which is made in designing expert systems is to under-estimate the size of the knowledge base which is required.

The major use of expert systems is for the rapid prototyping and development of complex software systems. Programmer productivity using the window-oriented graphics work-station and the high level software tools shows a ten-fold productivity increase over conventional programming language on a good quality super-personal computer.

Although 5GLs are generally associated with symbolic computing, many AI applications, also, require good performance for non-symbolic processing. Image-processing, model-based expert systems and systems accessing conventional databases are some examples.

In recent years, shell-based application packages have increased in many areas of systems. As a result, many companies explain how they have been using shells as tools to create complete ‘virtual' models of their products and manufacturing processes and the benefits in cost, money and quality that shells have given them.

The installation of CAD\CAM\CAE\CASE technology, within a framework of concurrent engineering, has resulted in improvements in the design-to-manufacture lead-times for the producers of goods at large. The world of manufacturing and computing technology continues to advance at breakneck speed and keeping up can be difficult for those technology is intended. This year has seen the launch of numerous new products for expert systems in engineering.

 

A Guide To Information Technology
© Copyrights.
ISBN: 0 9527956 4 7
Copyright © Andreas Sofroniou