Geoscience Reference
In-Depth Information
11.1 INTRODUCTION
Expert systems (ES), also known as knowledge-based systems or knowledge-based ES, are one of
the earliest and most successful applications of artificial intelligence (AI) (Giarratano and Riley
2005; Padhy 2005), where AI is a branch of computer science concerned with the development of
systems that exhibit human-like intelligence (Russell and Norvig 2010). A key element of AI is the
use of heuristics, that is, procedures that involve rules of thumb, tricks, strategies, simplifications or
any other method that may facilitate the solution to a narrow domain problem. Heuristics are often
used by humans to make decisions or to solve complex problems. ES are therefore an attempt to
loosely emulate the powerful cognitive reasoning and decision-making abilities of a human expert
in implementing a specific task using encapsulated knowledge (Medsker and Liebowitz 1994;
Giarratano and Riley 2005; Negnevitsky 2005). ES are defined as computer systems that are able
to represent and reason with knowledge in order to solve specific problems that would ordinarily
require human expertise (Turban 1995; Jackson 1999). Moreover, ES have been very successful in
modelling real-world decision-making problems which conventional programming techniques are
unable to handle (Giarratano and Riley 2005). Some of the types of tasks that have been addressed
using ES include interpretation, prediction, diagnosis, design, planning, monitoring, control and
repair across many different fields including medicine, engineering and financial analysis (Liao
2005; Tyler 2007).
ES have been shown to provide a number of tangible benefits which include increased productiv-
ity since they can work faster than humans and in hazardous environments; increased quality, reli-
ability and transparency in providing solutions; the ability to capture scarce and expensive expertise
that can be easily transferred to non-experts; flexibility in terms of the range of potential application
domains of ES as long as the knowledge can be harvested; the ability to work with incomplete or
uncertain information; the ability to interpret the decisions made through the explanation facility
which can also serve as a training tool; the ability to integrate ES with other systems such as geo-
graphic information systems (GIS), which has led to the development of intelligent hybrid systems
(Goonatilake and Khebbal 1995; Birkin et al. 1996); and the ability to solve complex unstructured
or semi-structured problems (i.e. difficult problems that do not have a straightforward algorith-
mic solution) that involve single or multiple human decision-making (Turban 1995; Giarratano and
Riley 2005; Negnevitsky 2005; Padhy 2005).
The foundations of ES can be traced back to the 1960s with the development of the first ES
application called DENDRAL (Barr and Feigenbaum 1982). This system was used to determine the
chemical structure of molecules based on a series of rules derived from expert chemists in combi-
nation with mass spectrometry data and formed a framework for future ES development. The next
major advance occurred in 1970 with the development of META-DENDRAL, a data mining tool
for generating new rules relating mass spectrometry data to chemical structure, where these rules
were then fed back into DENDRAL. This system was followed by the construction of MYCIN, a
diagnosis tool for doctors, in particular for infectious diseases (Buchanan and Shortcliff 1985) and
PROSPECTOR (Duda et al. 1977), a system to aid in the discovery of mineral deposits. The 1970s
saw ES development on high-end workstations using languages such as Prolog and Lisp. However,
with a shift to personal computers in the 1980s, accessibility to ES was opened up and many ES
were then built using ES shells, which are specialised tools for the creation of ES. The success of
these systems led to increased commercial interest in ES, with a significant boom in the 1980s to
early 1990s in which the technology spread to a wide range of domains such as management, eco-
nomics, medicine, agriculture, engineering and planning. Reviews undertaken by Durkin (1996)
and Liao (2005) provide an extensive list of application domains.
Despite the positive success of ES, there have also been many critical views of this technology.
For example, Openshaw and Openshaw (1997: p.109) stated that ES are 'a flawed technology [and]
that the idea [that] lies behind it may be of greater value than the current achievements of the tech-
nology itself'. This critical view is also acknowledged by Keyes (1989) who surveyed a large number
