Knowledge Based Systems

Knowledge Based Systems

An important impetus for AI research is to construct knowledge based systems to
automatically solve difficult problems. Ever since the 1980’s, knowledge
engineering has become the most remarkable characteristic of AI applications.
Knowledge based systems (KBS) include expert system, knowledge base system,
intelligent decision support system, etc. In 1965, DENDRAL, which was
designed to illustrate organic chemistry structures, developed to a series of expert
system programs. Such systems mainly include two parts: one is the knowledge
base, which represents and stores the set of task-related specific domain
knowledge, including not only facts about the related domain, but also heuristic
knowledge in expert level; the other is the inference engine, which includes
series of inference methodologies to retrieve the reasoning path, and thus to form
premises, satisfy objectives, solve problems, etc. As different mechanisms and
concepts can be adopted, the inference engines have multiple patterns.

In knowledge based systems knowledge will be stored in the computer in
defined structure for knowledge management, problem solving and knowledge
sharing. Projects and softwares of “Knowledge Based Management System
(KBMS)” have been initiated and developed all over the world, such as in
America, in Japan (the NTT Company), as well as in China. Remarkable
characteristic of KBMS is the integration of inference and query, which improves
the maintenance of the knowledge base, and provides useful development
environment for specific domain knowledge based systems.

Decision Support System (DSS) is evolved from the Management
Information System (MIS), with its concept initiated in the early 1970’s. It
developed fast as an important tool to improve the competitiveness and
productivity of companies, as well as to decide on the successfulness of a
company. DSS has been adopted by various levels of decision makers in abroad,
and attracted great focuses in China. Decision support techniques are critical to
support scientific decision making. Early DSS is based on MIS and includes
some standard models, such as the operational research model and the
econometric model. In 1980, Ralph Sprague proposed a DSS structure based on
data base, model base and dialog generation and management software, which
has a great impact on later research and applications. In recent years, AI
technologies have been gradually applied to DSS, and thus came in to being the
intelligent decision support system (IDSS). In 1986, the author proposed the
intelligent decision system composed of data base, model base, and knowledge
base (Shi, 1988b), which improved the level of scientific management by
providing an effective means to solve semi-structured and ill-structured decision
problems. Characteristics of IDSS include the application of AI techniques to
DSS, and the integration of database and information retrieval techniques with
model based qualitative analysis techniques. In the 1990’s, we developed the
Group DSS (GDSS) based on MAS technologies, which attracted enormous
research interests.

Building intelligent systems can imitate, extend and augment human
intelligence to achieve certain “machine intelligence”, which has great theoretical
meanings and practical values. Intelligent systems can be roughly classified into
four categories according to the knowledge contained and the paradigms
processed: single-domain single-paradigm intelligent system, multi-domain
single-paradigm intelligent system, single-domain multi-paradigm intelligent
system, and multi-domain multi-paradigm intelligent system.

1. Single-domain single-paradigm intelligent system

Such systems contain knowledge about a single domain, and process only
problems of a single paradigm. Examples of such systems include the first and
second generation of expert systems, as well as the intelligent control system.
Introduction 27
Expert systems apply domain-specific knowledge and reasoning methods to
solve complex and specific problems usually settled only by human experts, so
that to construct intelligent computer programs with similar problem solving
capabilities as experts. They can make explanations about decision making
procedure and learn to acquire related problem solving knowledge. The first
generation of expert systems (such as DENDRAL, MACSYMA, etc.) had highly
professional and specific problem solving capabilities, yet they lacked
completeness and portability in architecture, and were weak in problem solving.
The second generation of expert systems (such as MYCIN, CASNET,
PROSPECTOR, HEARSAY, etc.) was subject-specific professional application
system. They were complete in architecture with better portability, and were
improved in aspects such as human-machine interface, explanation mechanisms,
knowledge acquisition, uncertain reasoning, enhanced expert system knowledge
representation, heuristics and generality of reasoning, etc.

2. Multi-domain single-paradigm intelligent system

Such systems contain knowledge about multiple domains, yet only process
problems of a certain paradigm. Examples include most distributed problem
solving system and multi-expert system. Generally, expert system development
tools and environments are used to construct such large-scale synthetical
intelligent systems.

Since intelligent systems are widely applied to various domains such as
engineering technology, social economics, national defense affairs and ecological
environment, several requirements are put forward for intelligent systems. To
solve the many real-world problems such as medical diagnosis, economic
planning, military commanding, financial projects, crop planting and
environment protection, expert knowledge and experience of multiple domains
might be involved. Many existing expert systems are single-subject, specific
micro expert systems, which might not satisfy the users’ practical demands. To
construct multi-domain single-paradigm intelligent systems might be an
approach to meet the users’ requirements in certain degrees. Characteristics of
such systems include:

  1. solve the user’s real-world complex problems;
  2. adopt knowledge and experience of multiple domains, disciplines and
    professionals for cooperative problem solving;
  3. based on distributed open software, hardware and network environment;
  4. constructed with expert system development tools and environments;
  5. achieve knowledge sharing and knowledge reuse.

3. Single-domain multi-paradigm intelligent system

Such systems contain knowledge of only a single domain, yet process problems
of multiple paradigms. Examples include compound intelligent system.
Generally, knowledge can be acquired through neural network training, and then
transformed into production rules to be used in problem solving by inference
engines.
Multiple mechanisms can be used to process a single problem in problem
solving. Take an illness diagnosis system as an example, both symbolic
reasoning and artificial neural networks can be used. Then, compare and
integrate the results of different methods processing the same problem, through
which correct results might be obtained and unilateralism can be avoided.

4. Multi-domain multi-paradigm intelligent system

Fig. 1.4 illustrates the sketch map of such systems, which contain knowledge of
multiple domains and process problems of different paradigms. Collective
intelligence in the figure means that when processing multiple paradigms,
different processing mechanisms work separately, accomplish different duties,
and cooperate with each other, so that to represent collective intelligent
behaviors.

Leave a Reply

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