|
The
Intuitive Algorithm Applications for Expert Systems
Artificial
Intelligence for accessing data. The use of the
personal computer has become a world-wide phenomenon,
enabling people everywhere to improve the quality of
their work. Initial applications focused on financial
accounting, word processing and spread sheets. Recently,
the Internet opened opportunities to access information
from computer databases in a wide variety of fields. The
use of key-words now enable people to locate topics of
interest. But, in fields where specialized words are
used, the user needs to know the exact word to locate a
subject. Expertise is essentially the knowledge of the
exact word that defines a problem - such as the name of a
disease which exhibits a group of symptoms. Expert
systems can locate a problem from a description of such
symptoms. They can play a major role in assisting people
to locate vitally needed information. But, expert systems
should be fast and they should avoid stupid questions.
A
wide field of possibilities. Expert systems can
assist millions of users to access key information
regarding computer software, which grows more complex by
the day. The legal aspects of commercial activities cover
taxation, company law and constitutional law. Speedy
access to particular case laws is a vital need for the
legal profession. Computer diagnosis of diseases can
assist hospitals, general practitioners and students to
find vital information in specialized fields. Expert
systems can guide staff in large organizations which have
thousands of pages of manuals concerning complex
procedures. Diagnostics can assist in problems related to
machinery and equipment. In all these fields, existing
manuals can be entered into expert systems if only the
process was fairly simple and straightforward.
Simplification
of procedures. Traditional expert systems
require knowledge engineers, who understand the logical
reasoning in a diagnostic session and can encode this
logic into "If, then, else" rules. When the
database is large, questioning priorities may need to be
supported by probability estimates of likely questions or
heuristic assessment of enquiry directions. Such rule
based systems also become complex and intractable when
the size of the knowledge base expands. This section
describes an Expert System Shell based on the Intuitive
Algorithm (IA). The IA shell requires merely the
categorized entry of data and the design of questions
which can identify these categories. The shell isolates
categories, taking uncertainty into account - a question
may or may not identify a particular category. The shell
avoids the perennial AI problem of asking stupid
questions. The shell prioritises questions and produces
answers based on the IA elimination process.
General
terminology. The Shell follows a certain
terminology in its diagnostic processes. There are:
Objects. Objects have Properties. Properties suffer
Alterations. Alterations are induced by Causes. The
Relationship between Causes and Alterations form
Patterns. Causes, Alterations and the Patterns of their
Relationships are stored in the memory of an Expert
System.
Typical
Applications
Object:
Person. Property: Health. Alteration: Symptom. Cause:
Disease. Objective: Recognize Disease from an evaluation
of Symptoms, using the Pattern of their Relationships.
Object:
Legal Entity. Property: Freedom. Alteration: Civil
Activity. Cause: Legislation, or Case Laws. Objective:
Identify Legislation or Case Laws, from an evaluation of
Civil Activities, using the Pattern of their
Relationships.
The
Shell Program. An Expert inputs Knowledge into
the Shell Program to create an User Program. The User
inputs Y/N answers to onscreen Alteration Queries which
help to identify Causes. The general functions are as
follows:
Type
Names. A 40 Character Alteration Type Name and
Cause Type Name for data entry reference. For a Medical
Program: Alteration Type Name = Symptom. Cause Type Name
= Disease. Further references in the Program will be to
Symptom and Disease.
Alterations.
A 20 Character Alteration Name. An 80 Character Question
to User. Each screen holds 64 Alteration Entries, so that
the Expert can have a global view of the questioning
process. A 4000 Character description screen
permits the end user to obtain details concerning the
question covered by the Alteration. All data entry can be
edited.
Causes.
A 20 Character Cause Name. An 80 Character Identifying
Statement. Each screen holds 64 Cause entries. A 4000
Character description screen permits the end user to
obtain details of the Cause. All data entry can be
edited.
Hypertext.
The Shell allows the Expert to create hypertext
links between Causes, allowing the User to search through
the database, by clicking on highlighted words.
Relationships.
The Shell screen permits the entry of the Relationship
between an Alteration and a Cause. Yes/No/Maybe entries
can be entered with a single keystroke. "Yes"
is entered when the Alteration is positively present for
the Cause and absence of the Alteration clearly indicates
absence of the Cause. "No" is entered when the
Alteration is absent for the Cause and presence of the
Alteration indicates that this Cause can be eliminated
from further consideration. "Maybe" is entered
when presence, or absence of the Alteration does not
indicate presence or absence of the Cause.
Preparation
of the expert system. The Shell is designed to
enable the Expert to view the global range of Causes and
design Alteration questions which efficiently slice the
matrix of Causes in multiple directions. Other inputs
include the Title of the Expert System, Introductory
opening screens and Menu screens. Data in the completed
program is compressed and the program is compiled
producing a .EXE file.
User
interaction. The User is presented with
the option to carry out a word search, a menu search, or
an expert system search. The expert system choice
presents the User with a sequence of questions, with
Yes/No/Skip options to arrive at a list of Probable
Causes. The User can get further details of each selected
Cause to verify the diagnosis. The User can also
backtrack the questioning process and alter the Y/N/S
entries.
The
process. An "Yes" answer eliminates
all Causes which have been entered with a "No"
relationship to the Alteration question. A "No"
answer eliminates all Causes which have been entered with
a "Yes" relationship to the Alteration
question. The program chooses questioning priority by
selecting Alteration with highest number of "Y"
relationships. The program also eliminates all Alteration
questions, which have "Y" relationships only to
eliminated Causes. When there are less than 4 remaining
Causes, the program presents a list of Probable Causes.
Unlimited
rules. Since it is not necessary to design
complex reasoning chains, there is no theoretical limit
to the size of the database which can be handled by the
IA system. Each Cause is eliminated based on a logical
relationship. Such logical relationships are entered as
"rules" in the traditional expert system. While
such systems will be prone to error when the number of
rules exceed a thousand, the IA system can accurately
work with even a hundred thousand rules. This opens the
possibility of using AI in voluminous subjects which have
never been attempted because of the complexity of rule
based expert systems.
Uncertainty.
An extremely powerful part of the program is its ability
to handle questions which may or may not have an impact
on the outcome. A particular symptom may or may not be
present for a disease. The program will still eliminate
those diseases which have a positive or negative impact,
depending on the answer. In spite of the uncertainty, the
elimination proceeds with power. The ability to deal
logically with uncertainty is an exceptional feature,
which is not present in any other type of computer based
logic.
Stupid
questions. If an answer clearly indicates the
absence of a related disease, a further question which
indicates the disease is called a "stupid
question". Traditional expert systems struggle with
the problem of trying to avoid stupid questions. In the
IA system, when a Cause is eliminated, the program also
eliminates any Alteration question which has a
"Y" relationship only to the Cause. So, the
program will never ask a stupid question and the expert
does not need to design the program to cover this
eventuality.
Commercial
value and optimal size. Speedy access to data
has a commercial value in all those areas where people
routinely use computers. Expert systems which use IA can
provide a third level of help for commercial computer
programs. The experience of the author is that expert
systems, which solve problems in other areas, require an
optimal size to be of value. They should not appear to be
toys. Speedy access to all the information in a 400 page
manual may not enthuse users. They may consider such
information to be basic. A 3000 page data base may be
considered more useful. In India, Constitutional Law can
be summed up in about 400 pages. Related case laws may
cover 10,000 pages. A law practitioner may consider the
extraction of a Constitutional Law Provision as too
basic, but would value the extraction of a related Case
Law. An expert system may be planned only for areas of
commercial value and should be of optimal size.
Unrealized
potential. AI researchers have tended to focus
on the need for codification of knowledge from experts.
But in all commercially viable fields in today's world,
expertise is already recorded in research papers,
reference books and manuals. It is more practical to
design an expert system from published data and use the
expert only to verify the accuracy of the data and the
acceptability of the questions. The lack of availability
of a wide range of expert systems for public use is a
clear indication of the rule size limitations, complexity
and impracticality of current rule based expert systems.
There is an urgent need for the use of practical AI
solutions in thousands of areas for problems which people
encounter in their daily lives.
|