The Internet Encyclopedia (Volume 3)

P1: IML/FFX P2: IML/FFX QC: IML/FFX T1: IML

WL040-Schalkoff WL040/Bidgolio-Vol I WL040-Sample.cls July 16, 2003 15:52 Char Count= 0

RULE-BASEDEXPERTSYSTEMSDEVELOPMENT 241

The development of expert systems proceeds with the

consulting of an expert (or group of experts) with the

aim of developing a manipulable knowledge base. This

is often referred to as knowledge engineering (KE). The

first phase of the process consists of the formation of

a database of domain specific knowledge. In the expert

interrogation process, the formulation of “good” ques-

tions is paramount. Fortunately, experts often articulate

problem-solving methodologies in terms of if–then struc-

tures. Moreover, an expert may volunteer rationalizations

of the resulting rules (i.e., “I conclude this because... ”).

This type of explanatory production is also desirable.

The development of an expert system is almost al-

ways an iterative task, involving the cycle of expert query,

database formation development of the inference strategy,

verification of system performance, and so on. The gap be-

tween the concept of an ES and a finished, delivered prod-

uct may be enormous. The necessary application-specific

selection of a reasoning structure, interviewing of experts,

development of a prototype, refinement, user training,

and documentation may take several years.

One of the most important aspects of ES develop-

ment is verification of system operation. A set of test

cases is developed and used by both the ES and the hu-

man experts. When responses differ, modifications to the

system and perhaps additional expert consultation are

required.

The ability of an ES to provide the user with an ex-

planation is also important. An expert system response

such as “patient has disease x” is probably insufficient,

even if correct, because no explanation of the inference

process is provided. An explanation may be as simple as

indicating the sequence of rules used or as complicated

as indicating all possible inference paths considered and

the logic that leads to the most appropriate response or

conclusion.

Reasoning With Uncertainty

Unfortunately, knowledge may seldom be put into the

rule-based if–then form without some concern for con-

cepts such as impreciseness, ambiguity, and uncertainty.

Although several techniques (e.g., fuzzy systems) are

treated more fully elsewhere, the incorporation of mea-

sures of uncertainty in the representation as well as the

inference (manipulation) strategy leads to more realistic

ES implementations. A number of significant research

efforts have attempted to incorporate uncertainty in in-

ference techniques. These are the following:

1. Numerical approaches, which attempt to associate a
   quantitative measure of confidence (or certainty) with
   the truth value of facts. This includes fuzzy set and
   probabilistic approaches. For example, “certainty fac-
   tors” have been used to quantify uncertainty in ex-
   pert reasoning. In the expert system MYCIN, a con-
   fidence scale of [−1,1] was used to represent the range
   of confidence associated with a particular fact or as-
   sertion (conclusion). A value of−1 indicates total lack
   of confidence (i.e., complete confidence the assertion is
   FALSE), whereas a measure of 1 represents complete
   certainty the assertion is TRUE. If, after exhausting

all search possibilities, the cumulative confidence mea-

sure associated with a hypothesis is in the interval

[−0.2, 0.2], the hypothesis is regarded as unconfirmed.

Of course, this is an empirically determined range that

is subject to modification or alternate interpretation in

a particular application.

2. Symbolic approaches incorporate uncertainty but in
   a less numerically quantitative manner. Examples are
   linguistic extensions to the connectives of classical
   logic that allow statements using “perhaps,” “may be,”
   and so on. An example rule is the following:

If symptom is believed to be spots,

then diagnosis may be measles.

3. Another approach is the use of multivalued logic (Bolc
   & Borowik, 1992).

Rule-Based Expert Systems

and Intelligent Agents

The notion of agents brings together a number of tech-

nologies and research areas, including artificial intelli-

gence, software engineering, robotics, and distributed

computing. Agents are a powerful, natural metaphor

for conceptualizing, designing, and implementing many

complex, distributed applications. A quantitative defini-

tion is “an agent is an encapsulated computer system

that is situated in some environment, and that is capa-

ble of flexible, autonomous action on behalf of its user (or

owner) in that environment in order to meet prespecified

design objectives.”

The Internet is arguably one of the most complex,

changing, and unpredictable environments with which

software designers must deal work. At the same time, In-

ternet applications are arguably one of the most important

areas from both technical and economic perspectives. The

Internet may offer one of the best opportunities for the

agent paradigm.

The current trend is that of multiple agents; a multi-

agent system is one that consists of a number of agents

that interact with each another as well as the environment.

Characteristics of an agent include the following:

1. An agent is a problem-solving entity with well-defined
   boundaries and interfaces.


2. An agent is embedded in a particular environment.


3. An agent is designed to achieve specific objectives.


4. An agent is autonomous.


5. An agent is flexible and displays (context-dependent)
   problem-solving behavior. In other words, the agent is
   reactive.

One connection between agents and expert and rule-

based systems is straightforward: Expert or intelligent

agents may be implemented using a rule-based paradigm.

Sample applications are shown in Katz (2002) and Grosof

(1995).