Expert systems

One of the beneficiaries of the spin-off effects of the AI area are expert systems, programs for problem solving in some scientific or technical domain. An expert system can be defined as a computer based system comprising the total sum of knowledge and rules of thumb which represent the highest level of expertise in a given and carefully limited field. Such fields may be medical diagnosis, mineral prospecting, deep-sea diving or operations of a similar kind which sometimes seek the solution of ill-structured, incomplete and probabilistic tasks.

Expert systems are sometimes very useful, but all too often we forget that they are fed with simplified and formalized average knowledge by ordinary human specialists. The fundamental characteristics of such systems are a lack of common sense and of a perspective of their own knowledge. That is, they do not have the capability to analyze their own experience. In order to exercise judgements, an expert system employs certain intellectual rules of thumb, a phenomenon known as heuristics, sometimes described as the art of good guessing. Users of the system are thereby enabled to recognize promising problem-solving approaches and to make educated guesses.

An expert system consists of the following main parts (see Figure 7.3).

  • A knowledge database containing domain facts, the heuristics associated with the problem and the understanding and solving of the same.
  • An inference procedure, or control structure, for the choice of relevant knowledge when using the knowledge database in the solution of the problem.
  • A working memory or global database, keeping track of the problem status, the input data for the actual problem and a history of what has been done.
  • An interface, managing the interaction between man and machine and working in a natural and user-friendly way, preferably with natural speech and images. The interface must permit new data and rules to be implemented without change of the internal structure of the system.

In medicine, diagnostic databases have been constructed which are claimed to be able to challenge the professional skill of an experienced medical specialist. Other well-known expert systems exist within the area of chess-playing, where the capacity of the computer is now said to equal the capacity of a grand master.

A hierarchic model of skill acquisition has been proposed by, among others, Hubert and Stuart Dreyfus (1986). This model, which can be used to explain the deficiencies in the expert-system concept can be illustrated as follows:

  1. Novice
  2. Advanced beginner
  3. Competent person
  4. Proficient person
  5. Expert

The skill of the novice is normally judged on the basis of how well he follows learned rules inasmuch as he lacks a coherent sense of the overall task. The rules employed allow for the accumulation of experience but soon have to be abandoned to make way for a further development. Problem-solving here is entirely of an analytical nature and no special perspective is used.

The advanced beginner has gained a certain amount of experience and can perceive what is similar in previous examples. Actions and judgement may now refer to both new situational and context- free circumstances.

The competent person often registers an overwhelming amount of situational elements, all of which are to be taken into account. To manage the problem and find a solution, a hierarchical procedure of decision- making is used. First a plan is established with which to organize the situation. Then a simplifying procedure is used to focus only on those factors important for the chosen plan. The remaining constellation of facts may generally be referred back to an earlier experience and reasonable conclusions can be drawn, decisions made or expectations investigated.

The proficient person, except for making decisions only in an analytical way, has the qualifications of an expert. Only when we reach the level of the genuine expert are real intuitive decisions possible.

The expert does not follow a general set of rules; instead it is a question of knowing when and how to break the rules and knowing what is relevant or not. If this intuitive, involved methodology has a distinctive feature, it is the very absence of a specific strategy. The expert now has a standard reference case but knows of numerous special cases.

Experts often make their decisions in a condition of flow. This term was coined by the Hungarian Csikszentmihalyi in a book from 1990. Persons doing things with great skill and creativity often enter a mental state without restrictions, where time stops and their concentration is perfect. The need to compulse and supervise oneself disappears due to good preparation and years of exercises. This is a very pleasant state and exceptionally productive.

A closer look at expert systems can give a more realistic understanding of their capacity and usefulness. None of the existing systems will be able to reach the level of a human expert; it is doubtful whether they will qualify as proficient. It is even less probable that the human ability to recognize, synthesize, judge  and make use of intuition can be mimicked by a computer program. None of the semiconscious processes in human thinking that emerge in those superior fuzzy concepts so typical of reality are to be found in expert systems. Human expertise is far too sophisticated to be formalized into a set of rules.

Expert systems can only present what has been stored into them. Often a craftsman makes things absolutely correctly without being able to explain why. Much of the existing experience and learning is what  is called ‘tacit knowledge’, something which the hands, the judgement and the intuition bring about. Notwithstanding that such knowledge is systematically organized, it will have structures that cannot be expressed verbally. Of course, this is extremely difficult to formalize into a computer program.

It must also be kept in mind that even the highest ranked experts within a specific field have sometimes among themselves totally contrary opinions. When critical decisions have to be made, this problem is usually met by requiring the selected expert group to reach a consensus. The human capacity to merge individual irregularities into a collective agreement, a human qualified solution, is scarcely possible to implement in a computer. In other words, when a task is so narrowly defined that it can be performed with much less than the full human capacity of knowledge and judgement, then it is appropriate for an expert system.

In spite of the criticism presented here and, given that their limitations are understood, expert systems of course have their place as advanced intellectual tools. As it belongs to the lower levels of the skill-acquisition hierarchy, an expert system never needs thirty years of experience. It learns the main rules correctly the first time; it never needs practice and it never forgets. The precision and speed compensate for its blindness to situational elements. A good expert system performs about as well as competent human beings.

Source: Skyttner Lars (2006), General Systems Theory: Problems, Perspectives, Practice, Wspc, 2nd Edition.

Leave a Reply

Your email address will not be published. Required fields are marked *