Expert Systems: Guidelines for Managers
| Published date | 01 April 1992 |
| DOI | https://doi.org/10.1108/02635579210012269 |
| Pages | 23-25 |
| Date | 01 April 1992 |
| Author | Suresh Subramoniam |
| Subject Matter | Economics,Information & knowledge management,Management science & operations |
EXPERT SYSTEMS: GUIDELINES
FOR
MANAGERS
23
Expert
Systems:
Guidelines for Managers
Suresh Subramoniam
Industrial Management & Data Systems, Vol. 92 No. 4, 1992, pp. 23-25,
©MCB University Press Limited, 0263-5577
F
uzziness plays a vital role when a
particular scenario is in place.
Introduction
Expert systems (ESs) constitute a branch of artificial
intelligence which deals with intelligent problem solving.
An
ES can be defined, in simple terms, as a software that
captures and manipulates knowledge and strategies to
solve a problem at
hand.
The ES can be rule based, frame
based or
logic
based. The guidelines to be followed happen
to be the same irrespective of the paradigm followed. A
rule based ES consists of two parts, a knowledge base
or
rule
base and an inference engine. The knowledge base
or rule base is a collection of rules in the form of if-then
statements which contains rules of thumb, uncertainty
handling, fact-to-fact relationship and reasoning for why
and
how
certain rules are fired or conclusions are reached.
The knowledge in the form of rules in a rule-based ES
is a function of
the
quality or expertise of
the
ES at hand.
The more heuristics rules the knowledge contains, the
better the performance and quality of the ES at problem
solving. An inference engine is an aid for the ES which
helps it in arriving at conclusions about a problem or a
scenario at hand. There are two types of inferencing
mechanism, namely forward chaining or data-driven
reasoning and backward chaining or goal-driven reasoning.
Some inference engines use a combination of these
methodologies as in the case of ES shells. The main
difference between backward and forward chaining
methodology is the approach used in getting input from
the user. In a forward chaining engine,
all
the facts which
describe the problem are input to the system and the
system tries to reach useful conclusions by firing rules
in the knowledge base with the information fed. In a
backward chaining engine, the system first scans the
conclusions and tries to get input about its premisses,
in order to prove them. In this case, the inference engine
runs backward from conclusion to premisses.
The questions posed by the inference engine are chosen
by the software in an intelligent way. For example, the
facts which are already present in the memory of the
system will never be asked repeatedly. Moreover, by
positioning the rules carefully in the knowledge base, the
system can be made to prompt for input regarding a
variable in a logical fashion for which answers cannot be
deduced
by
the system
by
firing some of
the
existing rules
with the facts at hand. The fundamental differences
between a decision support system (DSS) and an ES
should be given due importance while developing an
ES[1].
A DSS is suitable for what-if analysis, optimization, risk
analysis and numerical
calculations.
Procedural languages
which are fast in number crunching like Fortran, Pascal,
etc.
are ideally suited for this purpose. On the other hand,
an ES should be good at symbolic manipulation,
recommending solutions, use of heuristics in problem
solving and description of behaviour through pattern
matching. Languages like Prolog and Lisp, known for its
string-handling capabilities, are more suitable for ES
development. In the case of
Prolog,
it is having a built-in
chaining mechanism based on backward principle.
Expert System can also be developed in Fortran or any
other procedural language for which chaining algorithm
also needs to be developed. An
ES
shell provides
a
user-
friendly interface for adding, deleting and modifying the
rule-base. Users need not have knowledge of symbolic
processing and experience with Lisp or Prolog syntax.
The ES shells have an inbuilt inference engine with pop-
up menus for easy maintenance of
the
rule base and user-
friendly interaction for problem solving. Other features
are consistency checking for rules, uncertainty-handling
capability, rule prioritization, security provisions and
interaction with external databases and programmes.
Evaluation of Problem Suitability for ES
There are some basic characteristics[2] which need to be
studied about the problem for analysing whether the
domain is ideally suited for an ES solution. They are
causality, temporality, uncertainty, fuzziness[3] and
structure. The causality defines the relationship between
the situation and the phenomena. If certain symptoms
exist then the cause can be predicted in these cases with
a very high degree of accuracy. Temporality deals with
the time precedence of different scenarios. Scenarios
follow a certain sequence and this needs to be studied
for efficient ES implementation. Uncertainty is the result
when strict adherence to logic is impossible due to lack
To continue reading
Request your trial