242 Part II • Information Technology
and parameters involved in the decision; thus, the same
inference engine can be used for many different expert
systems, each with a different knowledge base. The user
interface is the module used by the end user—for example,
an inexperienced estate planner. Ideally, the interface is
very user-friendly. The other modules include an explana-
tion subsystem to explain the reasoning that the system
followed in arriving at a decision, a knowledge acquisition
subsystem to assist the knowledge engineer in recording
inference rules and parameters in the knowledge base, and
a workspace for the computer to use as the decision is
being made.
Obtaining an Expert System
Is it necessary to build all these pieces each time your
organization wants to develop and use an expert system?
Absolutely not. There are three general approaches to
obtaining an expert system, and only one of them requires
construction of all these pieces. First, an organization can
buy a fully developed system that has been created for a
specific application. For example, in the late 1980s,
Syntelligence, Inc., developed an expert system called
Lending Advisor to assist in making commercial lending
decisions for banks and other financial institutions.
Lending Advisor incorporated the many factors involved
in approving or rejecting a commercial loan, and it was
installed in several banks. In general, however, the
circumstances leading to the desire for an expert system
are unique to the organization, and in most cases, this
“off-the-shelf” expert system option is not viable.
Second, an organization can develop an expert
system itself using an artificial intelligence shell(also
called an expert systems shell). The shell, which can be
purchased from a software company, provides the basic
framework illustrated in Figure 6.6 and a limited but
user-friendly special language with which to develop the
expert system. With the basic expert system functions
already in place in the shell, the system builder can
concentrate on the details of the business decision being
modeled and the development of the knowledge base.
Third, an organization can have internal or external
knowledge engineers custom-build the expert system. In
this case, the system is usually programmed in a special-
purpose language such as Prolog or Lisp. This final
approach is clearly the most expensive, and it can be
justified only if the potential payoff from the expert system
is quite high and no other way is possible.
Examples of Expert Systems
Perhaps the classic example of an expert system is
MYCIN, which was developed at Stanford University in
the mid-1970s to diagnose and prescribe treatment for
meningitis and blood diseases. General Electric Co.
created an expert system called CATS-1 to diagnose
mechanical problems in diesel locomotives, and AT&T
developed ACE to locate faults in telephone cables.
Schlumberger, Ltd., an international oil company,
developed an expert system named Dipmeter to give
advice when a drill bit gets stuck while drilling a well.
These examples and others are concerned with diagnos-
ing problem situations and prescribing appropriate
actions, because experts are not always present when a
problem occurs.
Diagnosis of a different sort is accomplished by an
expert system at the American Stock Exchange that has
been built to help detect insider trading on the exchange.
This expert system, named Market Surveillance, is
designed to support analysts in making recommendations
on whether to open an investigation of suspected insider
trading. The relevant database of stock price activity is
entered into the expert system, and the analyst responds to
a series of questions from the system. The output consists
of two scores—the first is the probability that an investiga-
tion should be opened and the second is the probability that
an investigation should not be opened (Exsys Inc., 2010).
Earlier we mentioned that expert systems were used
to assist in making commercial lending decisions as early
as the 1980s. Today, over one-third of the top 100 commer-
cial banks in the United States and Canada use FAST
(Financial Analysis Support Techniques) software for
credit analysis. The FAST expert system gives a credit
analyst access to the expertise of more experienced
analysts, speeding up the training process and increasing
productivity. FAST also provides a complete range of
traditional analytical reports on both a historical and a pro
formabasis (Exsys Inc., 2010).
Expert systems often serve in an advisory role to
decision makers of all kinds. For example, the IDP
(individual development plan) Goal Advisor is an expert
system that assists a supervisor and an employee in setting
short-range and long-range employee career goals and the
developmental objectives to reach these goals. Nestlé
Foods has developed an expert system to provide informa-
tion to employees on their pension fund status. Using the
expert system, an employee can conduct a private “inter-
view” with a pension fund expert and ask what-if questions
about benefits. The expert system enables the employee to
make more knowledgeable personal financial planning
decisions without requiring extensive personnel depart-
ment consultation. EXNUT is an expert system developed
by the National Peanut Research Laboratory and the U.S.
Department of Agriculture to help peanut farmers manage
irrigated peanut production. Based on extensive data