Expert systems (Development)

Requirements for creation

That development of ES was possible for the specific application, simultaneous accomplishment of the following requirements is minimum necessary:

• there are experts in the field who solve a problem much better, than the beginning specialists
• experts meet in assessment of proposed solution, otherwise it will be impossible to estimate quality of the developed ES
• experts are capable to explain the methods used by them in a system, otherwise knowledge of experts will be difficult to be entered to ES
• the solution of a task requires only reasonings, but not actions
• the task should not be too difficult (i.e. its solution should take the expert several hours or days, no more)
• the task should belong to rather "clear" and structured area, i.e. the basic concepts, the relations and the known algorithms of the solution of a task should be selected
• abstract thinking should not use the solution of a task substantially

Correctness of use

There are cases when use of expert system is possible, but is not justified. And as ES justified use can be considered in the presence of the following factors:

• the solution of a task will bring considerable effect, for example economic
• use of the person expert is impossible or because of the insufficient number of experts, or because of need to execute examination at the same time in different places
• at information transfer to the expert there is an unacceptable dead time or information
• there is a need to solve a problem in the environment hostile for the person

Technology of creation of expert system

Compliance of the application to the ES methods

The created application corresponds to the ES methods if the solvable task has set of some characteristics, namely:

• the task can be naturally solved using character reasonings, but not manipulations with numbers, as in mathematical techniques and in traditional programming
• the task should have the heuristic, but not algorithmic nature. Tasks which can be guaranteed solved at observance of the set restrictions using some formal procedures are not suitable for application of ES
• the task should be difficult so to justify costs for creation of ES. However it should not be so difficult that the expert system could not solve it in acceptable time
• the task should be rather narrow and almost significant

Concept of a fast prototype

The concept of a fast prototype is, as a rule, used during creation of expert systems. The essence of the concept consists that developers do not aim to construct an end product at once. At first there is a creation of one or several prototypes of ES. The prototype should meet two contradictory requirements: on the one hand, they should solve typical problems of the specific application, on the other hand, time and labor input of their development should be very insignificant. The last requirement is necessary that it was possible to combine as much as possible the process of accumulation and debugging of knowledge performed by the expert with development process of software tools. For observance of these requirements during creation of a prototype the various means accelerating creation process are used.

Methods of a prototype of expert system, besides, should be suitable for the specific program. In case of success of knowledge of a prototype of a problem area are exposed to expansion. At failure development of a new prototype can be required, or the decision on unfitness of the ES methods for this application is made. In process of expansion of the knowledge base the prototype can reach such status when all problems of this application are solved successfully. Conversion of a prototype to an end product is usually connected with reprogramming of a system in the languages of the low level providing both increase in high-speed performance of ES, and reduction of required memory. The labor input and time of creation of ES substantially depend on type of instruments of creation.

Development stages

During long-term works on creation of ES there was a certain technology of their development including six consecutive stages:

• identification
• conceptualization
• formalization
• accomplishment
• testing
• trial operation

At a stage of identification define the tasks which are subject to the solution, the development purposes come to light, experts and types of users are defined.

During conceptualization the detailed analysis of a problem area is carried out, concepts and their interrelations come to light, methods of solving of tasks are defined.

Formalization means the choice of tools and methods of representation of all types of knowledge. At this stage the basic concepts are formalized, methods of interpretation of knowledge are defined, system operation is modelled, the adequacy to the purposes of a system of the recorded concepts, methods of solutions, means of representation of knowledge and management is estimated by them.

At a stage of accomplishment there is a filling of a system knowledge. Because a basis of ES is knowledge, this stage exceeds all others on the importance and labor input. Process of filling of the knowledge base is divided into "extraction of knowledge" from the expert, their organization ensuring effective system operation, and interpretation of knowledge in clear ES a type. Process of acquisition of knowledge is performed by the knowledge engineer on the basis of the analysis of activity of the expert in the solution of real tasks.

Connected subjects

• Expert systems (Architecture)
• Expert systems
• Expert systems (representation of knowledge)

Links

• Consulting and trainings of Moscow
• Online Magazine