The Ohio State University

Knowledge Bank

kb.osu.edu

Ohio Journal of Science (Ohio Academy of Science)

Ohio Journal of Science: Volume 90, Issue 5 (December, 1990)

1990-12

Research Review: Application of Expert

Systems in the Sciences
Durkin, John
The Ohio Journal of Science. v90, n5 (December, 1990), 171-179
http://hdl.handle.net/1811/23417
Downloaded from the Knowledge Bank, The Ohio State University's institutional repository

RESEARCH REVIEW

Application of Expert Systems in the Sciences 1

JOHN DURKIN, Department of Electrical Engineering, The University of Akron, Akron, OH 44325

Studies from the field of artificial intelligence have given birth to a relatively new but rapidly
growing technology known as expert systems. An expert system is a computer program which captures the
knowledge of a human expert on a given problem, and uses this knowledge to solve problems in a fashion
similar to the expert. The system can assist the expert during problem-solving, or act in the place of the expert
in those situations where the expertise is lacking. Expert systems have been developed in such diverse areas
as science, engineering, business, and medicine. In these areas, they have increased the quality, efficiency,
and competitive leverage of the organizations employing the technology. During the 1980s, scientists and
engineers have used this technology to search for oil, diagnose medical problems, and explore space. This
paper provides an overview of this technology, highlights the major characteristics of expert systems, and
reviews several systems developed for application in the area of science.

ABSTRACT.

OHIO J. SCI. 90 (5): 171-179, 1990

INTRODUCTION
Expert system technology has captured the interest of
professionals in a number of fields in recent years. Systems
have been developed in such diverse areas as science,
engineering, business, and medicine. Almost every professional and computer society currently has a special
interest group for expert systems technology. This widespread interest can be attributed to the ability of the expert
system to aid various organizations in solving practical,
real-world problems. Currently, over two-thirds of the
Fortune 1000 companies have expert system projects
under development (Medsker et al. 1987). Organizations
are looking toward these systems to aid them in increasing
the quality, efficiency, and competitive leverage of their
operations.
During the 1980s, expert systems have been used in a
wide range of applications in the area of science. Scientific
and technology-oriented organizations have applied expert systems underground to find oil or mineral deposits
(Duda et al. 1977), in space to help control various
spacecrafts (Durkin and Tallo 1989), and on earth to help
in diagnosing medical problems (Shortliffe 1976). Expert
systems can aid scientists by interpreting data from an
experiment, interact with a physician to identify a given
disease, or aid an engineer in controlling a particular
process.
This paper provides an overview of this technology,
highlights the major characteristics of expert systems, and
reviews several systems developed for application in the
area of science. The paper also includes a short bibliography on expert systems for the interested reader to
explore further.

EXPERT SYSTEM DEFINITION

Expert systems are an offspring of the more general
area of study known as artificial intelligence (Al). In the
'Manuscript received 24 January 1990 and in revised form 6 September 1990 (#90-2).

simplest sense, Al is the study of developing computer

programs which exhibit human-like intelligence. Early Al
researchers focused on such problems as game theory,
robotic control, and vision systems (Nilsson 1980). Common to each of these problems was research into ways of
representing and reasoning with knowledge, in a computer,
in a fashion similar to humans.
The early studies in Al provided the insight needed to
develop expert systems. In particular, these studies showed
that reasoning alone is not a sufficient measurement of
intelligent behavior, but rather, one had to have a rich set
of knowledge with which to reason. It was also determined that the problem needed to be well-focused, using
only the knowledge relevant to a specific problem. These
two requirements led Al researchers to use human experts
for their source of problem-solving knowledge. By virtue
of being an expert, the human possesses unique talents,
made possible by the human's knowledge and problemsolving skills on a particular subject. Because of the nature
of these intelligent computer programs, they were aptly
called expert systems (Feigenbaum 1977). An expert system is a computer program designed to model the
problem-solving ability of a human expert.
The program models the following characteristics of
the human expert:
Knowledge
Reasoning
Conclusions
Explanations
The expert system models the knowledge of the human
expert, both in terms of content and structure. Reasoning
is modeled by using procedures and control structures
which process the knowledge in a manner similar to the
expert. Conclusions given by the system must be consistent
with the findings of the human expert. The expert system
also provides explanations similar to the human expert.
The system can explain "why" various questions are being
asked, and "how" a given conclusion was obtained.
One of the principal attractions of expert systems is that
they enable computers to assist humans in many fields of

EXPERT SYSTEMS IN THE SCIENCES

172

endeavor with the processes of analyzing and solving

complex problems. They extend the application of computers beyond the conventional mathematical processes
we have customarily assigned computers, to applications
where the computer can carry on a somewhat natural
conversation with the user to arrive at a conclusion or
recommendation that aids the human decision-maker.
This is accomplished by encoding in the expert system the
knowledge and problem-solving skills of a human expert.
This expert computer program can then be used by others
to obtain and use this expertise for solving a current
problem that would have previously required the expert
to be present.

EXPERT SYSTEM STRUCTURE

The structure and operation of an expert system are
modeled after the human expert. Experts use their knowledge about a given domain coupled with specific information about the current problem to arrive at a solution. For
example, a physician would possess knowledge about a
variety of possible diseases and, coupled with specific
information about a given patient, would be able to
diagnose the patient's problem.
Expert systems solve problems using a process which
is very similar to the methods used by the human expert
(see Fig. 1).
Knowledge Base
The knowledge base contains specialized knowledge
on a given subject that makes the human a true expert on
the subject. This knowledge is obtained from the human
expert and encoded in the knowledge base using one of
several knowledge representation techniques. One of the
most common techniques used today for representing the
knowledge in an expert system is rules.
A rule is an IF/THEN type structure which relates some
known information contained in the IF part to other
information. This information can then be concluded to be
contained in the THEN part. For example,
RULE 2
IF Battery voltage is
below 10 volts
THEN Battery is dead

RULE 1
IF Battery is dead
THEN Car will not start

Expert System

KNOWLEDGE BASE
1

General knowledge
INFERENCE
ENGINE
Conclusions
Explanations
WORKING MEMORY
Case facts
Inferred facts

FIGURE 1.

Expert system structure.

VOL. 90

These two rules capture knowledge which represents

natural relationships for automobile diagnostics. The first
rule relates the status of the battery to the status of the car.
The second rule relates the state of the battery to its status.
Using rules like these, one can form a complete knowledge base for diagnosing problems with an automobile.
Representing knowledge in rules has two major advantages. First, each rule is a separate declarative statement
about the problem, allowing one to add rules to the system
as needed. Secondly, rules appear to match the way many
experts formulate their knowledge about a problem in a
natural "cause and effect" manner. Other knowledge
representation techniques used are frames, semantic
networks, and predicate calculus (Barr and Feigenbaum
1981).
Working Memory
Specific information on a current problem is represented as case facts and entered in the expert system's
working memory. The 'working memory contains both the
facts entered by the user from questions asked by the
expert system, and facts inferred by the system. The
working memory could also acquire information from
databases, spreadsheets, or sensors, and be used by the
expert system to conclude additional information about
the problem by using the general knowledge contained in
the knowledge base.
Inference Engine
The analogy of human reasoning is performed in the
expert system with the inference engine. The role of the
inference engine is to work with the available information
contained in the working memory and the general knowledge contained in the knowledge base to derive new
information about the problem. This process is similar to
the way a human reasons with available information to
arrive at a conclusion.
Two principle inference techniques are employed in
the design of an expert system. The first technique relies
upon first establishing a goal or hypothesis, and then
attempting to prove it true. For example, a technician
believes a particular fault exists, then collects data to verify
this hypothesis. This style of reasoning is known as
backward chaining.
The second style of inference first collects information
about the problem and then attempts to infer other
information. For example, a control process engineer
gathers data from senors monitoring some process, and
then uses this information to conclude the present status
of the process. This style of reasoning is known as forward
chaining.
Backward and forward chaining may be integrated to
solve a given problem. For example, a physician may
initially gather information about a patient and use this
information to form a hypothesis of the possible disease.
This hypothesis would then be checked by gathering
additional evidence to support the belief. Both techniques
may also be challenged by a situation where the information is inexact or unknown. In this situation, the inference
technique must have the ability of continuing its reasoning
under the constraint of incomplete information. Inexact

OHIO JOURNAL OF SCIENCE

J. DURKIN

reasoning techniques for expert systems are discussed

further in the reports that are cited.
The inference engine, working either in the backward
or forward chaining mode, will attempt to conclude new
information about the problem from available information
until some goal is reached or the problem is solved.
Explanation Facility
Besides providing final results or conclusions, both
human experts and expert systems can explain how they
arrived at their results. This capability is often important
because the types of problems to which expert systems are
applied require that a justification of the results be
provided to the user. For example, an expert system which
recommends some antibiotic treatment for a patient
would need to explain to the physician how this recommendation was formulated.
Expert systems also have the capability of explaining
why a given question is being asked. When an individual
consults with a human expert, the conversation is highly
interactive, and on occasion, the individual may ask why
a certain line of reasoning is being pursued. The explanation given can make the user feel more comfortable with
the line of questioning and also help to clarify what issues
the expert believes are important for the problem.

CONVENTIONAL PROGRAMS VERSUS

EXPERT SYSTEMS
It is important to understand and appreciate the differences between conventional computer programs and
knowledge processing or expert systems. Knowledge
processing represents an evolution, rather than a revolution, in the way individuals and computers interact to
solve problems.
The most basic difference between the two is that
conventional programs process data, while expert systems
process knowledge. This basic difference influences both
the nature of the processing technique used and the results
obtained. The general differences between expert systems
and conventional programs are characterized in Table 1.
Conventional programs process data which is usually in
TABLE 1
Expert system versus conventional programs.

Conventional Programs

Expert Systems

Numeric

Symbolic

Algorithmic

Heuristic

Precise information

Uncertain information

Command interface

Natural dialogue with explanations

Final solution given

Recommendation with explanation

Optimal solution

Acceptable solution

173

numeric form, while an expert system works with symbolic information. Data are isolated bits of information
about a problem, whereas symbolic information represents statements or facts concerning the problem which
can be used with general knowledge to infer new information.
Conventional programs process data by means of
algorithms, whereas an expert system will use heuristic
reasoning techniques. An algorithm represents a finite set
of well-defined steps to be performed. Heuristic reasoning works with the available information to draw conclusions about the problem, but does not follow a prescribed
sequence of steps.
A conventional program requires complete and precise
information. An expert system can work with the available
information whether it is incomplete or uncertain. In this
sense, an expert system can provide some results even
under the constraints of limited or uncertain information.
A conventional program would be severely limited under
such constraints.
The interface of an expert system permits questions to
be asked and answers given using a natural language style.
This interface is more readily accepted by end-users than
the command interface found with most conventional
programs. Interaction with an expert system also follows
more closely the conversation between one human obtaining advice from another human.
During the conversation, explanations are provided by
the expert to queries as to "why" a question is being asked,
and "how" a given conclusion was reached. This point
makes an expert system considerably unlike a conventional program, which simply provides a final answer.
Conventional programs provide a final solution usually
in the form of a result from a computation. The computation may have involved a complex series of tasks, but the
user will only see the final result and not the intermediate
steps that led to the final result. Expert systems provide a
result in the form of a recommendation, with a justification
in the form of a tracing of its reasoning.
Given the correct information, conventional programs
will provide an exact solution to a problem. It is an "all or
nothing" situation. Expert systems can make mistakes, just
as a human expert might. This point appears to give the
conventional program an advantage over the expert
system. However, this appearance is only an illusion.
Expert systems work on types of problems which are less
structured than conventional programs, and the information available may not be sufficient to obtain an exact
solution. However, the expert system will still be able to
reach some reasonable conclusion, even if it is not
optimal, whereas a conventional program will fail if not
provided with all of the information it needs. This ability
of an expert system to be able to make decisions in the
absence of complete or certain information is the result of
developments in the area of inexact reasoning.

INEXACT REASONING
A number of inexact reasoning techniques have been
developed with three principle ones adopted by expert
system developers: Bayesian, Certainty Theory, and Fuzzy
Logic.

174

EXPERT SYSTEMS IN THE SCIENCES

The Bayesian approach is based in classical probability

theory where data is gathered about the problem, and can
then be used for forming probability statements. This
approach offers a sound mathematical basis for drawing
inexact inference. However, the approach requires the
gathering of data about all possible events of the problem,
a situation which may not be possible or practical for all
problems. One of the most noted expert systems developed using a Bayesian approach for inexact reasoning is
PROSPECTOR (Duda et al. 1977), a system designed for
mineral exploration.
Certainty theory is a product of the work performed
during the development of MYCIN (Shortliffe 1976).
Certainty theory is an ad-hoc approach to inexact reasoning which captures uncertain information and inference in
a number called the certainty factor (CF). If a statement
would be provided such as; "I may go to the game today,"
then this fact would be represented in the system with a
CF value of 0.7, which quantifies the qualitative word may.
This numeric representation of the uncertainty in this fact
can be used by the expert system to form uncertain
inferences.
Certainty theory does not have the formal mathematical
basis such as Bayesian theory, but allows for inexact
reasoning for those problems which lack a wealth of prior
data. This approach to inexact reasoning is also the most
common found in expert systems today.
Fuzzy logic (Zadeh 1964) provides an approach to
inexact reasoning for subjective or vague terms used in our
natural language such as: little; hot; or heavy. In many
problems, an expert will use these vague terms to describe
some issue; fuzzy logic then provides the methods for both
capturing and reasoning with them numerically. Like
certainty theory, fuzzy logic lacks the formal basis of
Bayesian theory; but, expert system developers have
found this inexact reasoning approach to be valuable in a
number of applications. Giarratano and Riley (1989)
provides a detailed discussion of these three and other
approaches to inexact reasoning.

INTEGRATING EXPERT SYSTEMS WITH

CONVENTIONAL PROGRAMS
In an earlier section, the major differences between
conventional programs and expert systems was discussed.
However, though the two play different roles, in many
applications the two are combined to effectively solve a
problem. This section discusses some of the major areas
where expert systems are integrated with conventional
programs.
Intelligent Database Management Systems
In database systems, records of information are maintained which are accessed by the user through the use of
a menu or query interface. To effectively use the system,
the user must have a clear understanding of what information is needed, howto obtain it, and whatdecisions can
be made from the information obtained. These tasks
require the user to make intelligent decisions and to
understand the operation of the database interface.
An expert system acting as an intelligent interface, can

VOL. 90

accommodate the user by assuming the responsibility of

accessing the information and forming decisions. The
expert system can interact with the user to determine the
information which would be of most value, access the
information, analyze it in the context of the problem, and
present its findings to the user. This capability provides
ease of use of existing database systems, makes the
information available to a larger number of people due to
its ease of use, and improves the decision-making process
in using the available information.
One system which integrated these two technologies
successfully is called EP-X (Smith et al. 1985). This system
was designed to aid a user in accessing information in the
Chemical Abstracts database. Like most information retrieval systems, this database contains thousands of references, and the user needs to carefully filter through the
vast amount of records to find relevant information. EP-X
was found to be an effective interface in that it decreased
the time to acquire the information and provided information which best met the needs of the user.
Intelligent database management systems provide the
capability to access and use information for analysis and
decision-making. This capability is not only accommodating to the user, but also allows the expert system to be
embedded into existing database management systems.
Real-Time Control and Monitoring
Another area where the two programming techniques
merge is in real-time control and monitoring applications.
Any control or monitoring system requires the acquisition
of data. This data may be coming in quickly from a number
of sources, and the responsibility of the human or computer system is to form some understanding of the
information and take the appropriate action. At Three Mile
Island, for example, decisions needed to be made immediately to isolate the leak and on whether to evacuate the
area. The problems associated with the large amount of
information, and the time constraints to decide on the
appropriate action, are taxing for not only humans but
conventional computer programs.
Real-time expert system design offers an approach
which can assimilate the information quickly and make
human-like decisions. Several expert system development
packages have been built which incorporate an expert
system module with an interface to external conventional
programs. Together they offer the capability of operating
in real-time. ART by Inference Corp., and G2 by Gensym
Corp., are the two packages which expert system designers
have often turned to for real-time system development.
For example, when NASA wanted a computer system
which would integrate the vast amount of information
they had gathered on the operation of the space shuttle in
order to make navigational decisions, they turned to
Inference Corp. Inference developed NAVEX (Marsh
1984), a real-time expert system which recommends
control actions.
Intelligent Statistical Analysis
A number of statistical analysis programs exist with a
wide range of applications; but, when inexperienced
individuals use them, they often provide poor results.

OHIO JOURNAL OF SCIENCE

Expert system interfaces to these programs can offer the

user advice on how to best use them. The expert system
can suggest what analysis should be done and how to
interpret the results. Consequently, the intelligent interface can ease the use of the statistical package, enable the
package to be used by a wider audience, and avoid most
of the more common errors in their use.
One system which integrated an expert system interface with a standard statistical package was REX, developed at AT&T Bell Laboratories. REX aids in the use of a
commercial regression analysis package called S. The
purpose in developing REX was to increase the productivity in the use of S and increase the number of people who
could effectively use the package. REX interacts with the
user to obtain the data, interfaces with the regression
program, checks all of the assumptions needed for regression analysis, and analyzes the results. If it finds
something wrong, it determines how to correct the
problem and produce a quality regression. AT&T found
that REX eases the use of S and made it available to
individuals who could not previously use the package.
This section discussed some of the more common
applications where expert systems can be integrated with
conventional programs. It provided insight into how an
effective marriage can be formed between two technologies where each perform a function best-suited for their
design. It also highlighted the fact that expert systems can
be embedded into an existing environment to improve
present productivity. Besides the applications discussed in
this section, expert systems have also been successfully
integrated with spreadsheets, hypertext, and CD-ROMS.

DURKIN

175

TABLE 2

Types of problems solved by expert systems.

Paradigm
Control

Description
Interpreting, prediction, repairing, and monitoring
system behaviors

Design

Configuring objects under constraint

Diagnosis

Inferring system malfunctions from observables

Instruction

Diagnosing, debugging, and repairing student

behavior

Interpretation

Inferring situation description from data

Monitoring

Comparing observations to plan vulnerabilities

Planning

Designing actions

Prediction

Inferring likely consequences of given situations

Prescription

Recommending solution to system malfunction

Selection

Identifying best choice from a list of possibilities

DIAGNOSIS

TYPES OF PROBLEMS SOLVED

BY EXPERT SYSTEMS
There are several reasoning strategies used by humans
during problem-solving. However, cognitive studies of
human problem-solving has shown that humans will use
different strategies for different types of problems (HayesRoth et al. 1983). For example, the diagnostic reasoning
used when solving a problem with a carburetor is unlike
the reasoning used for financial planning. Cognitive
studies have also shown that reasoning strategies can be
categorized to reflect common methods used by humans.
The grouping of these common reasoning methods is
referred to as problem-solving paradigms. That is, diagnostics is a problem-solving paradigm, and represents a
style of reasoning which would be similar whether diagnosing a problem with an automobile or a television.
Expert system designers characterize the different styles of
reasoning into various paradigms (Table 2; adapted from
Hayes-Roth et al. 1983).
Each paradigm represents a different style of reasoning
about a given problem. For each paradigm, humans will
collect information about the problem differently and will
process this knowledge with the domain knowledge,
using different methods. The results of a survey of a
number of companies employing expert systems illustrate
the percentage of applications of expert systems for each
of these paradigms (Fig. 2), (SanGiovanni and Romans
1987).
The predominant type of problem addressed by expert

SELECTION
PREDICTION
MONITORING

INSTRUCTION
CONTROL

PLANNING

DESIGN

P-

% Applications

FIGURE 2. Expert system applications.

systems is one of diagnostics. There are two major reasons

for this, one based on the nature of the problem and the
other based on the solution.
The majority of problems involving a human expert are
diagnostic in nature. Experts diagnose problems with
machines, manufacturing processes, or living systems.
These and other types of diagnostic problems can be
found in most organizations. The diagnostic knowledge
needed to solve these problems can generally be captured
in an expert system.
The second reason for the large percentage of total
expert systems which are diagnostic can be traced to their
relative ease of development. Most diagnostic problems
have a finite list of possible solutions, and a limited

176

VOL. 90

EXPERT SYSTEMS IN THE SCIENCES

amount of information is needed to reach a solution.

These bounds provide an environment which is conducive to effective system design.
Another explanation for the large percentage of diagnostic expert systems can be traced to the practical
considerations of introducing a new technology into an
organization. Most organizations prefer to take a low-risk
position when considering new technology. As such,
projects which require minimum resources and have
maximum likelihood of success, are preferred. A wellfocused diagnostic problem within an organization will
usually meet these practical requirements.
There are far fewer applications in other areas such as
planning or design (Fig. 2). The major reason for the
relative absence of expert system applications in these
areas can be traced to the nature of the problem. These
areas are difficult to implement in an expert system
because the knowledge and problem-solving approach
used by the human expert are often difficult to capture.
Although various problem-solving paradigms can be
shown as individual tasks (Fig. 2), for many real-world
problems, a combination of paradigms is used. For
example, a problem on control may involve the task of
data interpretation and diagnostic reasoning, followed by
control of the process.

WHY USE AN EXPERT SYSTEM?

Like any project venture, developing an expert system
must have some justification. Insight for justifying an
expert system can be gained when one compares an
expert system with a human expert. From the comparison
(Table 3), one can formulate several general reasons for
employing an expert system such as:
Replacement of human expert
Assistant to human expert
Transfer of expertise to novice
Using an expert system to replace a human expert is
done primarily to use the system when the expert is not
available. For example, through time constraints, the
human expert may not be available, while an expert
system designed to control some manufacturing process
would be available 24 hours a day. Another expert system,
containing the expertise of a unique expert within a
company, could be made available to company sites
located in other geographic areas. If the expert should
leave or retire from the company, the expertise captured
in the expert system could serve as a replacement for the
expert. Human experts may be scarce, hence expensive.
Expert systems, by contrast, may be inexpensive. Developing an expert system can be a costly venture, but the
finished product would have low operating costs. The
finished system can also be duplicated at low cost and
distributed widely.
In the area of science, justifying an expert system for
replacing a human can be found in such applications as
space exploration (Durkin and Tallo 1989), or providing
the expertise of a geophysicist to some remote oil exploration site (Elf-Aquitaine and Teknowledge 1983)Another example would be to replace the human operator
of a control process.
Assisting a human expert is one of the most commonly

TABLE 3

Comparison between a human expert and an expert system.

Factor

Human Expert

Expert System

Time availability

Workday

Always

Geographic
availability

Local

Anywhere

Perishable

Yes

No

Consistent results

No

Yes

Cost

High

Affordable

Productivity

Variable

Consistent

found applications of expert systems. In this application,

the expert system attempts to aid the human expert in a
routine or mundane task. For example, a physician may
have general knowledge of most diseases, but could use
some additional support in diagnosing a given problem
with a patient. In another example, a bank manager may
be responsible for processing numerous loan applications, but could use help with some of the routine
decisions made. In both applications, the human expert is
fully capable of performing the task, but obtains additional
support from the expert system. In this type of expert
system application, the objective is to improve the overall
productivity of the current practice.
One specialized application of an expert system which
can be used to assist the expert is the ability of the expert
system to learn about a specific problem. The most
common learning method used in expert systems today is
a technique known as induction (Quinlan 1979). The
induction technique works with information contained in
a set of examples to induce a set of rules which capture
the knowledge about the problem. This approach has
particular value for those problems where the expert lacks
the knowledge to form decisions, but has a history of data
on the problem. The induction technique can uncover
classifications in the data which can be used for guiding
the decision process.
The expertise held by a human expert is a valuable
resource. Knowledge is gained by the expert through
years of experience from working on the problem. In
many organizations, it is important that this expertise not
be lost, but transferred to others through training. An
expert system can be developed to accomplish this
training task.

EXPERT SYSTEM APPLICATION AREAS

The application of expert systems has been found to be
of value in many disciplines. Applications of expert
systems vary from aiding a laboratory technician in
interpreting data, to providing guidance in establishing
marketing direction to an executive of an organization.

OHIO JOURNAL OF SCIENCE

This section will review the application of expert systems

in various areas of science. Several applications in each
area will be highlighted and a number of references of
expert system applications for each area will be given.
Agriculture
In the area of agriculture, expert systems have been
applied to such problems as crop management, insect
control, and productivity considerations for raising a given
crop. Farmers and agents from the Department of Agriculture Research Services must make decisions concerning the effective and profitable production of various
crops. Expertise for making these decisions exists, but the
major problem is making this talent available to the large
number of farmers.
PLANT (Boulanger 1983) predicts the damage to corn
caused by the invasion of black cutworms. The system first
obtains information on the current field situation, including such information as the concentration of weeds, soil
condition, and corn variety being grown. This information, coupled with black worm simulation programs, is
used to predict the expected level of damage from this
pest.
COMAX (Baker and Lemmon 1985) incorporates the
knowledge of a model of cotton production to provide
advice on the growing and management of this crop. The
expert system uses the cotton model to predict crop
growth and yield in response to weather conditions, soil
variables, and pest damage. COMAX is capable of providing recommendations for management decisions on a
daily basis to maximize cotton yields.
CROPPRO (Durkin and Godine 1989) was developed
to aid farmers in four major areas of crop production such
as: crop management problems, pest control, financial
considerations, and tutoring on various crop topics. The
system is structured to be applicable for most crops by
addressing common problem areas. Hypertext and interactive graphics are used extensively, which serve to
enhance the system's interface and effectiveness. The
Shiitake mushroom was chosen as a test case for this
expert system.
Chemistry
The majority of expert systems developed in the area of
chemistry have been applied in a laboratory environment.
The major advantage of these systems is the assistance
they provide to the laboratory technician throughout a
given experiment. They can assist in the planning and
monitoring of the experiment, and in interpreting test
data.
One of the first expert systems was developed by
Stanford University upon request by NASA. NASA was
planning to send an unmanned spacecraft to Mars, and
wanted a computer program developed which would
implement chemical analysis of the Martian soil. NASA
would provide information on the soil in the form of mass
spectrograms; and, from this information, the program
would have to determine its molecular structure. To
develop this program, the Stanford team had to encode in
the program the expertise of a chemist with specialized
skills in this area. The resulting program from this effort

J. DURKIN

177

became known as DENDRAL (Buchanan and Feigenbaum

1978). DENDRAL is capable of inferring the molecular
structure of an unknown compound from the mass
spectrogram data.
SPEX (Iwasaki 1982) assists scientists in planning
laboratory experiments in the area of molecular biology.
The scientist defines and describes the various objects to
be used in an experiment, such as the physical environment and the structure of the experiment. The system
assists in developing a plan for achieving the goal of the
experiment.
GA1 (Stefik 1978) determines possible DNA structures
from restriction enzyme segmentation data. The system
uses a model of enzyme digestion analysis of DNA
structures, coupled with knowledge of possible errors in
laboratory test environments, to formulate its result.
Computer Science
Expert system applications in the computer science
area have been concerned with designing or diagnosing
various computer systems. Configuring computer systems
which meet customer defined specifications, or diagnosing faults in a given system, can be a difficult and timeconsuming task.
XCON (McDermott 1980) was developed to configure
VAX computer systems for Digital Equipment Corp. (DEC).
Each of DEC's customers requires computer systems with
certain characteristics and features, and has certain constraints associated with resources and available space. To
accommodate these various characteristics and to make
the configuration recommendation problem more efficient,
DEC attempted to develop a conventional computer
program to help with the configuration task. Unable to
find a satisfactory solution, XCON (originally called Rl)
was, therefore, developed and presently saves DEC a
reported $20 million a year.
DART (Bennett and Hollander 1981) diagnoses faults in
the hardware of computer systems. The system incorporates knowledge of the structure and expected behavior of
a system in order to find design flaws in new computer
systems.
YES (Griesmer et al. 1984) assists computer operators
in controlling the MVS operating system used in large
mainframe IBM computers. This expert system monitors
the operating system, interprets MVS messages, and
makes recommendations to the console operator for
system control. The major tasks of YES are: maintaining
adequate queue space, handling network communications, scheduling batch jobs, responding to system errors,
and responding to system hardware errors.
Engineering
Expert systems have been used in a wide range of
applications in the area of engineering. Design, diagnostics, and control appear to dominate these applications.
Expert system applications in this area can either assist an
engineer in a task or replace the human operator of a
control process.
PEACE (Dincbas 1980) is an expert system developed
to assist engineers in the design of electronic circuits. The
system is a CAD tool which performs both synthesis and

178

EXPERT SYSTEMS IN THE SCIENCES

analysis of passive and digital circuits. The system uses

knowledge on the functional description of the basic
circuit components, coupled with topological structural
constraints. The system can synthesize the circuit in
defined steps which fulfills the design specification under
the problem constraints.
DELTA (Marcus and Steven 1983) aids maintenance
personnel in the identification and correction of faults in
diesel electric locomotives. The system interacts with the
user during the diagnosis process while presenting computer-aided drawings of parts, subsystems, and repair
sequences, using videodisc movies. Through this interaction, a final recommendation is made in the form of a
sequence of repair steps.
The Smidth Cement Kiln Controller (Zadeh 1984) uses
"fuzzy logic" to control the production of cement. The
system determines the needed adjustments in air flow, gas
fuel, raw materials, and rotation speed of the kiln, to
achieve an economical operation of the process. The
fuzzy logic approach allows the system to work with
uncertain control knowledge, which models the approach
taken by the human operator.
Geology
The dominant use of expert systems in the area of
geology has been applied to the problem of exploration.
Expert systems can aid a geophysicist in the interpretation
of survey data or act in their place for those situations
where one is not available.
PROSPECTOR was an expert system developed at the
Stanford Research Institute to aid geologists in the exploration of ore deposits (Duda et al. 1977). PROSPECTOR
uses knowledge based on five different models which
describe various mineral deposits. The system initially
obtains information which characterizes a particular deposit of interest, including information on the geological
environment, structural controls, and the types of minerals
and rocks present or suspected at the site. This information
is then compared with the models and the system notes
any similarities, differences, and missing information.
Then it assesses the potential presence of a given mineral
deposit. In 1980, PROSPECTOR was field tested at a site
near Mount Tolman, WA. The test resulted in the discovery
of a $100 million molybdenum deposit. The dramatic
success of PROSPECTOR inspired a number of other
commercial expert systems which rely on some of the
features used in PROSPECTOR.
DIPMETER (Davis et al. 1981) determines the subsurface geological structure of a given site by interpreting
dipmeter logs. The system uses knowledge about dipmeter data and basic geology to uncover features in the data
that aid in the identification of geological structures. This
capability is of particular importance in oil or mineral
exploration.
The French oil exploration company, Elf Aquitaine,
maintains a number of oil wells and, like many oil
companies, often experiences problems with the drill bits
becoming stuck. It is not unusual for drilling-related
expenses to exceed $100,000 per day, and shutdowns
caused by drilling problems to last for several weeks, until
an expert can be brought to the site. Because of the

VOL. 90

scarcity of experts in this field, Elf Aquitaine contracted

with the California-based company, Teknowledge, to
develop an expert system which would serve in the place
of the human expert. This system was aptly called the
DRILLING ADVISOR (Elf Aquitaine 1983). The system
uses information about the geological formations at the
site, conditions of the current problem, and historical
information about other problems experienced in the
past. The expert system then performs a diagnosis of the
problem, produces a recommendation to correct the
problem, and further provides advice for changes to
current practices to avoid the problem in the future.
Medicine
The most prolific application of expert systems to date
has been in the area of medicine. A possible reason for the
extensive application in this area is that most medical
applications have been diagnostic in nature, an area
where expert systems are very effective. The expert
system can assist a physician in diagnosing medical
problems of a patient or help in the interpretation of
medical test results.
MYCIN was developed to capture the knowledge of an
expert on infectious blood diseases (Shortliffe 1976). This
expert system captured the expertise of individuals on
blood diseases to provide accurate and quick diagnosis of
the present disease and the proper therapeutic recommendation. The system could also work with unknown or
uncertain information which might be all that is available
in an emergency, life-threatening situation. MYCIN was
valuable not only for its ability to diagnose infectious
blood diseases, but for the contributions it made to our
understanding of introducing an expert system into the
workplace. Much of the current usage of rule-based expert
systems is based on the work of MYCIN.
VM (Fagan 1978) monitors a patient in an intensive care
unit and controls the patient's treatments. The system
characterizes the patient's state from sensory data, identifies any alarms, and suggests useful therapies. The system
measures the patient's heart rate, blood pressure, and the
status of operation of a mechanical ventilator that assists
the patient's breathing. Working with this information,
and coupled with information on the medical history of
the patient, the system can then make the needed adjustments to the ventilator.
Space Technology
One of the most recent applications of expert systems
has been in the area of space technology. Systems are now
being built for diagnosing system problems, planning
mission objectives, or controlling spacecraft functions.
ECESIS (Dickey and Toussaint 1984) controls the life
support systems aboard a manned space station. After
interpreting data from sensors, the system decides how to
adjust the life support systems during the transition from
shadow to sun.
NAVEX (Marsh 1984) monitors radar data that estimates
the velocity and position of the space shuttle. The system
detects any errors or predicts if a problem may occur. If an
error is located, the system further recommends the
appropriate actions.

OHIO JOURNAL OF SCIENCE

SATPRO (Durkin 1989) is an expert system for autonomous diagnostics and reconfiguration of a communications satellite. The system models the normal operation of
the satellite system and can detect deviations from normal
behavior and respond by recommending the proper
reconfiguration which will eliminate the fault and maintain proper operation.

SUMMARY
Expert systems technology is an emerging area of
computer science which is finding applications in a
number of diverse areas. Organizations are employing
expert systems to capture the problem-solving skills of
human experts to either assist the expert or use them in
those situations where the expert is not available. This
paper has provided a brief overview of this technology
and has discussed its application in the area of science.
Applications of expert systems in the sciences are expected to increase in the near future. The review of past
systems developed in the various science disciplines
should provide insight into the types of applications which
can be expected.

LITERATURE CITED
Baker, J. and H. Lemmon
1985
Expert Systems For Agriculture,
Computers and Electronics in Agriculture. 1: 31-40.
Barr, A. and E. A. Feigenbaum (eds.) 1981 The Handbook of Artificial
Intelligence, Vol. 1. William Kaufman Publishing.
Bennett, J. and C. Hollander
1981
DART: An Expert System for
Computer Fault Diagnosis. Proceedings IJCAI-81. p. 843-845.
Boulanger, A. G. 1983 The Expert System PLANT/CD: A Case Study
in Applying the General Purpose Inference System ADVISE to
Predicting Black Cutworm Damage in Corn. M. S. Thesis, Computer
Science Dept., Univ. of Illinois at Champaign-Urbana.
Buchanan, B. and E. Feigenbaum
1978
DENDRAL and MetaDENDRAL: Their Applications Dimension. Artificial Intelligence. 11:
5-24.
Davis, R., H. Austin, I. Carlbom, B. Frawley, P. Pruchnik, R. Sneiderman,
and J. Gilreath
1981 The Dipmeter Advisor: Interpretation of
Geological Signals. Proceedings IJCAI-81. p. 846-849.
Dickey, F.J. and A. L.Toussaint 1984 ECESIS: An Application of Expert
Systems to Manned Space Stations. Proceedings of the Fifth Conf. on
Artificial Intelligence, IEEE Computer Society, p. 173-178.
Dincbas, M. 1980 A Knowledge-based Expert System For Automatic
Analysis and Synthesis in CAD. Information Processing 80, IFIPS
Proceedings, p. 705-710.
Duda, R., P. E. Hart, N. J. Nilsson, R. Reboh, J. Slocum, and G. Sutherland
1977 Development of a Computer-Based Consultant for Mineral
Exploration. SRI Report, Stanford Res. Inst., 333 Ravenswood Ave.,
Menlo Park, CA.
Durkin J., R. Godine, and Y. Lu 1989 Expert System for Specialty Crop
Management. The llthlnternat'l. Jnt. Conf. on Artificial Intelligence,
Detroit, MI. p. 312-323.
and D. Tallo 1989 Expert System Diagnostics for a 30/
20 Gigahertz Satellite Transponder. Contract NAG3-923 Semi-annual
report, for period 1 Sept. 1988 through 28 Feb. 1989.
Elf-Aquitaine and Teknowledge
1983
The Drilling Advisor,
Fundamentals of Knowledge Engineering. Teknowledge Report,
Teknowledge Inc., Palo Alto, CA.
Fagan, L. M. 1978 Ventilator Manager: A program to Provide On-line
Consultative Advice In the Intensive Care Unit. Report HPP-78-16,
Computer Sci. Dept., Stanford Univ., Stanford, CA.
Feigenbaum, E. A. 1977 The Art of Artificial Intelligence: Themes and
Case Studies of Knowledge Engineering. IJCAI. 5: 1014-1029.
Giarratano, J. and G. Riley
1989
Expert Systems: Principles and
Programming. PWS-KENT Pub. Co.

J. DURKIN

179

GriesmerJ., S. Hong, M. Karnaugh, J. Kastner, M. Schor, R. Ennis, D.

Klein, K. Milliken, and H. Van Woerkom 1984 YES: A Continuous
Real Time Expert System. Proceedings AAAI-84. p. 122-129.
Hayes-Roth, F., D. A. Waterman, and D. B. Lenat 1983 Building Expert
Systems. Addison-Wesley Pub. Co., Inc., Reading, MA.
Iwasaki, Y.
1982 SPEX: A Second-Generation Experiment Design
System. Proceedings AAAI-82. p. 67-75.
Marcus, S. and J. Steven 1983 Computer Systems Applying Expertise.
The New York Times, Aug. 29- P- 16.
Marsh, A. K.
1984
Pace of Artificial Intelligence Research Shows
Acceleration. Aviation Week & Space Technology, Dec. 10. p. 5.
McDermott, J. 1980 Rl: An Expert System In the Computer Systems
Domain. Proceedings AAAI-80. p. 269-271.
Medsker, L. R., J. H. Morrel, and K. L. Medsker 1987 Expert Systems:
Meeting The Educational Challenge. Collegiate Microcomputer. ~\l{5):
223-229.
Nilsson, N. J. 1980 Principles of Artificial Intelligence. Palo Alto, CA.
Tioga.
Quinlan, J. R.
1979
Discovering Rules From Large Collections of
Examples: A Case Study, Expert Systems in the Micro-Electronic Age,
(D. Michie, ed.). Edinburgh Univ. Press.
SanGiovanni, J. and C. Romans 1987 Expert Systems In Industry: A
Survey. Chemical Engineering Proceedings, Sept. p. 52-59.
Shortliffe, E. H. 1976 Computer-Based Medical Consultation, MYCIN,
New York, Amer. Elsevier.
Smith, P. J., M. H. Chignell, and D. Krawczak 1985 Development of
a Knowledge-Based Bibliographic Information Retrieval System.
Proceedings of the IEEE International Conference on Cybernetics
and Society, Halifax, Nova Scotia.
Stefik, M. 1978 Inferring DNA Structures From Segmentation Data.
Artificial Intelligence. 16: 85-114.
Zadeh, L. A. 1965 Fuzzy Sets. Information and Control, p. 338-353Zadeh, L. 1984 Making Computers Think Like People. IEEE Spectrum.
21: 26-32.

ADDITIONAL READING REGARDING

EXPERT SYSTEMS
Bell, M. 1985 Why Expert Systems Fail. / Opl Res. 36(7): 613-619.
Davis, R. and J. King
1977
An Overview of Production Systems.
Machine Intelligence. 8: 300-322.
, B. Buchanan, and E. Shortliffe 1977 Production Rules
as a Representation for a Knowledge-Based Consultation Program.
Artificial Intelligence. 8: 15-45.
Emrich, M. 1985 Expert Systems Tools and Techniques. Oak Ridge
Nat.'l Lab. report, Aug. ORNL/TM-9555.
Gevarter, W. 1982 An Overview of Expert Systems. NASA report, May.
NBSIR 82-2505.
Harvey, J.
1986 Expert Systems: An Introduction. Electrical Communication. 60(2): 100-108.
Hayes-Roth, F. 1985 Rule-Based Systems. Communications of the
ACM, Sept. 28(9): 921-932.
1984 The Knowledge-Based Expert System: A Tutorial.
IEEE Computer, Sept. p. 11-28.
Mayers, W. 1986 Introduction to Expert Systems. IEEE Expert, Spr.
p. 100-109.
Michie, D. 1980 Expert Systems. The ComputerJournal. 23(4): 369376.
Miller, A. 1985 Expert Systems. IEEE Potentials, Oct. p. 12-15.
Murphy, T. 1985 Setting up an Expert System. The Industrial Control
Magazine, Mar. p. 54-60.
Nau, D. 1983 Expert Computer Systems. IEEE Computer, Feb. 16(2):
63-85.
Quinlan, J. 1980 An Introduction to Knowledge-Based Expert Systems.
The Australian ComputerJournal, May. 12(2): 56-62.
Thompson, J. and W. Thompson 1985 Inside an Expert System. BYTE,
April, p. 315-330.
Teschler, L. 1985 Stripping the Mystery from Expert Systems. Machine
Design, April 25. p. 68-74.
Yaghmai, N. and J. Maxin 1984 Expert Systems: A Tutorial, fournal
of the American Society for Information Science. 35(5): 297-305.
Winfield, M. 1982 Expert Systems: An Introduction for the Layman.

Computer Bulletin, Dec. p. 16-21.