Griffith University

School of Computing and Information Technology

Domain: Expert Systems Design and Development

The Evolution of Expert Systems

available electronically at:

http://www.cit.gu.edu.au/~noran

© Ovidiu S. Noran
Table of Contents.

1 Introduction....................................................................................................1
1.1 Artificial Intelligence and Expert Systems. ........................................................1
1.1.1 Artificial Intelligence. .......................................................................................... 1
1.1.2 Expert System. .................................................................................................. 1
1.2 Motivation.........................................................................................................2
1.3 The objective of this paper. ..............................................................................2

2 Expert Systems Primer. .................................................................................4

2.1 General concepts. ............................................................................................4
2.1.1 Rule-based expert systems. .............................................................................. 4
2.1.2 Development of an expert system. .................................................................... 4
2.1.3 Limitations of expert systems. ........................................................................... 5
2.2 Advanced concepts. .........................................................................................7
2.2.1 Languages, tools and shells. ............................................................................. 7
2.2.2 Components of a rule-based expert system...................................................... 7
2.2.3 The production rules. ......................................................................................... 8
2.3 Features...........................................................................................................9
2.3.1 Essential features. ............................................................................................. 9
2.3.2 Advanced features............................................................................................. 9
2.4 Application domains. ......................................................................................10

3 Expert Systems Evolution............................................................................11

3.1 Timeline. ........................................................................................................11
3.2 Milestones. .....................................................................................................12
3.2.1 The Beginning.................................................................................................. 12
3.2.2 The human problem solving model. ................................................................ 12
A. A change of concept. ............................................................................................. 13
3.2.3 A new trend...................................................................................................... 15
3.2.4 Artificial Neural Systems.................................................................................. 15
3.2.4.1 Basic concepts......................................................................................... 15
3.2.4.2 ANS advantages...................................................................................... 16
3.2.4.3 ANS timeline. ........................................................................................... 16
3.2.5 The future......................................................................................................... 17

4 Knowledge Representation in Expert Systems............................................19

4.1 What is knowledge ? ......................................................................................19
4.2 Knowledge representation techniques............................................................20
4.2.1 Productions. ..................................................................................................... 20
4.2.2 Semantic Nets. ................................................................................................ 20
4.2.2.1 Concepts.................................................................................................. 20
4.2.2.2 Semantic net problems............................................................................ 20
4.2.3 Schemata......................................................................................................... 21
4.2.4 Frames............................................................................................................. 21
4.2.5 Logic and Sets ................................................................................................. 21

5 Conclusions. ................................................................................................23

6 References. .................................................................................................23
The Evolution of Expert Systems Introduction

1 Introduction.

1.1 Artificial Intelligence and Expert Systems.

Mankind has always been asking questions. It is a way to relate to the world and to
evolve: 'I think, therefore I am'. Some questions though are difficult to answer. Take
for example: 'What is the nature of knowledge ?', 'Can knowledge be represented ?'.
In trying to answer these questions, Artificial Intelligence (A.I.) is an important
contributor.

1.1.1 Artificial Intelligence.

But, what is A.I. ? One of the many definitions may be:
'The branch of computer science concerned with the automation of intelligent behaviour'.
As a computer science, A.I. is based on computer science specific principles, such as
data structures, algorithms, languages and programming techniques. In contrast with
these well defined terms, questions such as 'what is intelligence ?' or 'may intelligence
be achieved on a computer or just within a biological existence?' are yet unanswered.
An important fact though is that A.I., like any other science, is a human endeavour and
therefore may also be defined as 'the collection of problems and methodologies
studied by artificial intelligence researchers' (Ref.1, p.2).
A.I. has many areas of interest: robotics, vision, natural language understanding and
semantic modelling, speech, artificial neural systems and parallel processing,
automated reasoning and theorem proving, expert systems, game playing and human
performance modelling. While each and every A.I. domain is both challenging and
fascinating, we will have to limit the topic of this paper to expert systems.
One of the major roots of the expert systems is in the cognitive science: the way
humans process information (i.e., the way they think and solve problems).

1.1.2 Expert System.

So, what is an expert system ?
An expert system is a computer system that emulates the decision-making ability of a
human expert, i.e. it acts in all respects as its human counterpart.
The term expert may be misleading. In the early days expert systems only contained
expert knowledge. Presently however, any system using expert system technology
(even if not containing highly specialised expertise in a certain domain) is called an
expert system. Therefore, the term knowledge-based system is more appropriate,
although most people use the term expert system because it is shorter.
Expert systems have emerged from early work in problem solving, mainly because of
the importance of domain-specific knowledge. A human expert's knowledge is specific
to a problem domain. In much the same way, expert systems are designed to address
a specific domain, called the knowledge domain.
Fig. 0. 1 shows the concept of a knowledge-based expert system. The expert system
receives facts from the user and provides expertise in return. The main components of
the expert system (invisible from the outside) are the knowledge base and the
inference engine. The inference engine may infer conclusions (solutions) from the
knowledge base, based on the 'facts' supplied by the user.

1
The Evolution of Expert Systems Introduction

Fig. 0. 1 Basic concept of an expert system (Ref.1, p.3)

Expert systems are suitable for solving the ill-structured (or not entirely defined) type
problems, which are usually difficult or impossible to solve by 'conventional' methods
(e.g. procedural programming).

1.2 Motivation.
In the development of any tool, method, etc, a question will inevitably be asked: Why
develop and use it ? What is it good for ?
In order to determine the advantages of a generic expert system, we will have to
evaluate it with the only comparison term available: the human expert.
availability: On any suitable computer hardware;
cost: The per-user cost is greatly lowered;
danger: May be used in environments hazardous for humans;
permanence: will last indefinitely (depending on the hardware);
multiplicity: more expert systems may co-operate to increase the level of
expertise;
reliability: provide a second opinion to human experts, mediate opinions;
explanation: the expert system may always explain how it reached the
conclusion;
response: may provide fast or real-time response for critical applications;
emotional: in real-time or emergency situations, the expert system will
always provide an unemotional and complete response;
database: may be used to intelligently access a database (data mining);
tutoring: may be used as an intelligent tutor: explain reasoning, etc.

The most obvious shortcomings are that an expert system has to be developed, trained,
updated and that it cannot capture nor explain 'deep' knowledge (understanding of
underlying causes). Also, unless carefully designed, an expert system may entirely
reflect a single human's expertise, which would defeat most of the advantages
enumerated above. The limitations of the expert systems will be discussed later on in
this paper.

1.3 The objective of this paper.

This paper is intended as a primer in the area of Artificial Intelligence and especially
Expert Systems. We have started by giving a motivation, some definitions and basic
concepts. In the following sections we will detail some expert systems features and
trace their development from the early days through to present day and future trends.

2
The Evolution of Expert Systems Introduction

Back to Table Of Contents.

3
The Evolution of Expert Systems Expert Systems Primer

2 Expert Systems Primer.

In order to understand the development of the expert systems, the reader needs to
build a minimum of domain knowledge in the area of expert systems. Therefore we will
start by presenting some basic and advanced concepts, followed by desired features
and application domains of expert systems. Armed with these notions, we will then be
ready to follow the evolution of the expert systems.

2.1 General concepts.

2.1.1 Rule-based expert systems.

Knowledge representation in expert systems may be rule-based or encapsulated in
objects. The rule-based approach uses IF-THEN type rules and it is the method
currently used in constructing expert systems. IF-THEN rules take the following form:
IF there is a flame THEN there is a fire.
The modern rule-based expert systems are based on the Newel and Simon model of
human problem solving in terms of long-term memory (rules), short-term memory
(working memory) and cognitive processor (inference engine).
Elaborate expert systems may be based on thousands of rules (e.g. XCON/R1 system
from Digital Equipment Corporation, used for configuring computers) and surpass a
human expert in a particular field. However, even smaller sized expert systems,
based on several hundred rules may be extremely efficient in very specialised areas.
While a knowledge-based system may rely on knowledge commonly available, a true
'expert' system will be based on unwritten expertise, acquired from a human expert.
In the conditions where no algorithm is available to solve a particular problem, a
reasonable solution is the best we can expect from an expert (system or human). The
expert system will infer a solution from the facts provided by the user and the rules in
the knowledge base. Therefore, it should be able to explain the reasoning employed
to achieve the solution. The explanation facility is an important feature of the rule-
based expert systems, since it provides a mechanism for a human to follow and check
the correctness of the solution achieved by the expert system. A further enhancement
to this facility is the availability of what-if scenarios (employing hypothetical reasoning
questions), where the user may examine the outcomes of several possible situations.

2.1.2 Development of an expert system.

The process of building an expert system is commonly known as knowledge
engineering. This implies knowledge acquisition from a human or other source and
coding it into the knowledge base of the expert system (refer Fig. 1. 1).
The main phases in the knowledge engineering process are:
• The dialog process represented in Fig. 1. 1 is similar to the task of a system
designer discussing the requirements of the program with the client, in
conventional programming. After acquiring knowledge from the human expert,
the knowledge engineer has to explicitly code it into the expert system
knowledge base.
• After the coding stage, the human expert evaluates the expert system and gives
feedback/critique to the knowledge engineer.

4
The Evolution of Expert Systems Expert Systems Primer

• The knowledge engineer alters the knowledge base in order to reflect the
human expert's comments.

Fig. 1. 1 Basic concept of an expert system (Ref.1, p.7)

This process will iterate until the human expert finds the expert system satisfactory for
the envisaged purpose.
Some rule-based systems may display the capability to learn rules by example (rule
induction), creating rules from tables of data. Learning is a complex process, hindered
by ambiguities, duplications and inconsistencies in the human expert knowledge.

2.1.3 Limitations of expert systems.

Although the expert system technology has progressed substantially, it would be a
mistake to overestimate the abilities of the technology. Deficiencies of the current
expert systems include:
• the lack of causal (or deep) knowledge. The expert systems do not have a real
understanding of the causes and effects in a system, mainly because it is much
easier to program expert systems with shallow knowledge, based on empirical
and heuristic knowledge. Designing an expert system based on basic structures,
functions and behaviours of objects takes a lot more effort and the resulting
system is a lot more complex (and more difficult to maintain / upgrade)
• dealing with uncertainty. Human experts recognise the limits of their knowledge
and can qualify their expertise when the problem reaches their so-called limit of
ignorance. An expert system, unless this problem is explicitly catered for, will
make confident recommendations even in the presence of incomplete and/or
inaccurate data. Conclusion: the expert system's solution must degrade
gracefully, just like in the case of a human expert.
• the current expert systems cannot make analogies, i.e. cannot generalise their
knowledge in order to reason about new situations in the way that people can.
Rule induction may help the system achieve only some types of new knowledge.
But there’s also good news. Expert systems, even limited in the present form, have still
been able to solve problems that conventional programming methodologies were

5
The Evolution of Expert Systems Expert Systems Primer

unable to scope with. By fully understanding the capabilities and limitations of the
expert system technology, we can make appropriate use of it.

6
The Evolution of Expert Systems Expert Systems Primer

2.2 Advanced concepts.

2.2.1 Languages, tools and shells.

In defining a problem, a very important decision is the way to model it. If the problem
may be modelled by conventional programming, that paradigm is usually preferred
because of the vast amount of experience available in that domain. However some
problems can only be solved via a non-procedural programming paradigm. In this
case, expert system languages will have to be used.
Expert system languages usually address a smaller range of problems in a more
efficient and natural way, however they are not generally suitable for general-purpose
programming. This is particularly true for the specialised AI languages, such as IPL-II,
SAIL, CONNIVER, Smalltalk, etc. The commonly used AI languages may have
extensions that make them usable for a wide variety of problems. For example, one of
the current Prolog versions, Visual Prolog, can be used for expert systems, but also
for database and Web programming.
The main difference between procedural languages and expert system languages is the
focus of representation. Procedural languages provide flexible and robust ways of
representing data, (arrays, records, linked lists, etc) and even provide for data
abstraction (modules, packages). However, the data and methods to manipulate it
however are tightly interlaced. Expert systems languages by contrast allow two levels
of abstraction: data abstraction and knowledge abstraction. For example, in the rule-
based expert systems, the facts (data abstraction) are separated from the rules
(knowledge abstraction). The data – knowledge separation allows for less control of
the execution sequence (e.g. typically a separate piece of code, the inference engine,
applies knowledge to the data).
In order to avoid confusion, the following terms have to be defined:
• language: translator of command written in a specific syntax. An expert system
language provides an inference engine to execute statements of the language;
• tool: language plus associated utility programs for debugging, development,
deploying applications. Examples: text editors, (cross) compilers, code
generators (which can generate code from tables). Some tools may be
integrated with all its utility programs under a single integrated user interface;
• shell: a special purpose tool comprising a complete expert system, however
without the knowledge base. The user must only supply the knowledge base.
Example: EMYCIN, obtained by removing the medical database from the
MYCIN expert system.

2.2.2 Components of a rule-based expert system.

The structure of a rule-based expert system is shown in Fig. 1. 2. In this case, the
domain knowledge needed to solve problems is contained in the knowledge base in
the form of rules.
Components of the rule-based expert system:
• user interface: it is the user – expert system means of communication;
• explanation facility: explains the expert system reasoning to the user;
• working memory: a global database of facts used by the rules;
• inference engine: decides which rule is satisfied by the supplied facts, prioritise
the rules and execute the highest priority rule;

7
The Evolution of Expert Systems Expert Systems Primer

• agenda: the prioritised list of rules whose patterns are satisfied by facts (or
objects) in the working memory;
• (optional) knowledge acquisition facility: a way for the user to directly enter
knowledge in the system, without the need for explicit knowledge coding.

Fig. 1. 2 Structure of a Rule based expert system (Ref.1, p.24)

A rule whose patterns are satisfied by the facts are said to be activated, or instantiated.
Given more rules are instantiated at the same time, the inference engine must select
one rule for firing (termed by analogy with a nerve cell - neuron). Like a neuron, once
the rule has fired, it may not fire again for a certain period of time. This behaviour is
aimed at preventing trivial loops and is called refraction.

2.2.3 The production rules.

In rule-based expert systems, the knowledge base is also known as the production
memory. The rules contained within the knowledge base are called production rules
and seen as a way of transforming a string of symbols into another string of symbols
(refer Chapter 2).
Rules may be expressed in an IF…THEN… pseudocode. The section between the IF
and THEN parts is called the antecedent, conditional part, pattern part, of LHS (left
hand side). The THEN part of the rule is a list of actions to be performed when the
rule fires (i.e. LHS is satisfied and the rule has enough priority to be executed by the
inference engine). This part of the rule is called the consequent or the RHS (right
hand side).
When multiple rules are selected for firing, they are activated and put on the agenda in
descending priorities order. However, some rules' firing may determine a change in
system status, so that other rules waiting on the agenda no longer have their LHS
(pattern) satisfied. In this case, they are simply removed from the agenda.

8
The Evolution of Expert Systems Expert Systems Primer

2.3 Features.
In order to perform within the intended scope, expert systems must contain a set of
desired features. These features must ensure that the expert system is appropriately
accepted and utilised by the intended end users. An expert system that is not applied
and / or acknowledged by the users is a useless exercise even if it is a very good and
comprehensive tool.

2.3.1 Essential features.

A set of essential features expected in an expert system would be:
• high performance. The quality of the advice given by the system must be very
high – comparable or better than a human expert;
• adequate response time. Again, the expert system must respond in an amount
of time comparable or better than a human expert. In case of real time
processes, the time constraint becomes especially important;
• reliability. The expert system must be immune to crashes and stable, or its
usability and acceptance will be greatly reduced;
• understandable. The system must be able to provide an explanation for the
conclusions / solutions it reaches. This feature is extremely important for the
following reasons:
⇒ solution justification – needed due to the great potential for harm (human
life and property may depend on the answers of an expert system);
⇒ expert system development – debugging misunderstandings between
the knowledge engineer and the expert;
⇒ behaviour checking – debug any unforeseen interactions in the expert
system, seen as a parallel program with the rules acting as independent
knowledge processors;
• flexible. The system must be easy to maintain: add, modify, and delete change.

2.3.2 Advanced features.

The explanation facility within the expert system may be simple or sophisticated,
according to the size / intended use of the system. The explanation facility may
provide one or more of the following:
• list all the pro and con reasons for a particular hypothesis;
• list all hypothesis that may explain the provided fact;
• explain all consequences of a particular hypothesis;
• use a what-if approach (prognosis on what will occur if a hypothesis is true);
• justify the questions asked by the expert system – maybe provide a way for the
user to guide the system towards achieving a solution more efficiently. The user
input has to be very carefully balanced though;
• justify the knowledge of the program on user request; a particular case is
quoting a meta-rule for the justification of a particular rule. Meta-rules are
knowledge about rules.

9
The Evolution of Expert Systems Expert Systems Primer

2.4 Application domains.

Expert systems can only be effective in certain domains. Therefore selecting the
appropriate paradigm for the problem is essential. In many cases conventional
programming will solve the problem. The appropriate domain for an expert system
however depends on a few critical factors:
• If the problem be solved reasonably through conventional programming, then an
expert system is not the best choice. However for situations where there is no
efficient algorithmic solution (ill-structured), expert systems are best suited.
When dealing with ill-structured problems, there is a possibility that the system may
follow an algorithmic solution after all. This may happen if the knowledge engineer
sets the priorities of too many rules, so the systems unknowingly mirrors an
algorithmic solution. An expert system with strong control structure (displaying
opportunistic behaviour - favouring certain paths) often indicates a concealed
algorithm that could be represented through conventional programming.
• The expert system must be clearly delimited and is expected to 'know' its own
capabilities. By adding domains to the knowledge base, the expert system
becomes more and more knowledgeable, however its complexity increases too.
Coordinating multiple expert systems (in the HEARSAY II and III systems, Ref 1 p.19)
has proven to be a complex task, presently suited more for research than for a
deliverable (commercial) product.
• There must be a need and desire for an expert system. As previously
mentioned, it is meaningless to build an expert system if there is no one willing
to make use of it. The expert system needs to be accepted as a helping tool, or
it will be rejected even if it is needed.
Expert systems are new technology, so gaining management support is crucial. This is
mainly because of uncertainty of the achievements and the lack of widespread expert
system knowledge.
• There must be a human expert willing to co-operate. At least one, but not too
many, since the system may become a repository of contradictory rules (as we
know experts rarely agree in opinions);
• The human expert knowledge must be understood by the knowledge engineer.
The knowledge engineer must understand what the human expert is talking
about, because he is the middle tier between human expert and expert system.
He has to explicitly code the knowledge into the system.
• Expert systems are especially suitable when the human expert's knowledge is
mainly heuristic (from experience) and uncertain. The expert's knowledge may
be trial-and-error approach rather than based on logic and algorithms (which
may be better modelled by conventional programming).

Back to Table Of Contents.

10
The Evolution of Expert Systems Expert Systems Evolution

3 Expert Systems Evolution.

3.1 Timeline.
The expert systems technology is based on a wide background. Fig. 2. 1 presents the
major milestones in the development of the expert systems.

11
The Evolution of Expert Systems Expert Systems Evolution

Fig. 2. 1 Milestones in the Expert Systems history. (Adapted from Ref.1, p.11)
As one can see, the Expert Systems history starts at around 1943 with the Post
production rules. The initial trend was towards intelligent systems that relied on
powerful methods of reasoning rather than making use of domain knowledge. They
were also designed to cover any domain rather than a specialised area.
The shortcoming of this approach was that the machine had to discover everything from
first principles in every new domain. In contrast, a human expert would rely on domain
knowledge for high performance.

3.2 Milestones.

3.2.1 The Beginning.

It all started in the late 50s and early 60s, when several programs aimed at general
problem solving were written. Post (Ref 3) was the first to use the production systems
in symbolic logic, proving that any system of mathematics or logic could be written as
a certain type of production rule system. The basic idea was that any mathematic
/logic system is just a set of rules specifying how to change a string of symbols into
another set of symbols.
The major difference between a Post production system and a human is that the
manipulation of strings is only based on syntax and not on any semantics or
understanding of the underlying knowledge. A Post production system consists of a
group of production rules. The limitation of the Post production system is that, if two or
more rules apply, there is no control specifying whether to apply one, some or all
rules, and in what order. The Post production rules are not adequate for writing
practical programs because of the lack of a control strategy.
At the same time (around 1943), McCulloch and Pitts developed a mathematical
modelling of neurons, which later gave birth to the connectionist programming
paradigm. Hebb gave an explanation of the learning process of the neurons in 1949.
The next step happened around 1954, with the introduction of a control structure for the
Post production system, by Markov. A so-called Markov algorithm is an ordered group
of productions, applied in order of priority to an input string. The termination of the
Markov algorithm is triggered by either a production not applicable to the string or by
the application of a rule terminated with a period. The Markov algorithm as a definite
control strategy. The higher priority rules are used first. If not applicable, the algorithm
tries the lower priority rules. Although applicable to a real expert system, the Markov
algorithm may take a long time to execute on a system with many rules. If the
response time is too long, the system will not be used no matter how good it is.
The possible improvement to Markov's approach of always trying all the rules, was of
course to be able to apply any rule, in a non-sequential order. This would be possible
if the algorithm already knew about all the rules. This was to happen later on, around
1977 - 1979.

3.2.2 The human problem solving model.

An outstanding example of a problem solving program was to be the General Problem
Solver, described in the published work Human Problem Solving (Ref 2), by Newell
and Simon. Their work demonstrated that much of the human problem-solving
(cognition) could be expressed as IF..THEN production rules.
Newell and Simon regarded the knowledge as modular, each rule corresponding to a
small module of knowledge called a chunk. In this model, the human memory is

12
The Evolution of Expert Systems Expert Systems Evolution

organised chunks loosely arranged, with links to other related chunks. Sensory input
would provide stimuli to the brain and trigger appropriate rules of long term memory,
which then produce the suitable response. The long term memory is where the
knowledge is stored; the short term memory however, is used for temporary
knowledge storing during problem solving.
Human problem solving must also include a so-called cognitive processor. This is the
equivalent of the inference engine in a modern expert system. The cognitive
processor will try to find the rules to be activated by the appropriate stimuli. If more
than one rule is to be activated, then a conflict resolution must be performed in order
to decide which rule will be activated first (priority assignment).
Newell and Simon have not only managed to create a credible model of human problem
solving, but have also set the basis of the modern rule-based expert systems. The
concept of knowledge chunks brings with it the problem of granularity in the design of
expert systems. More granularity means several chunks of knowledge are blended
together in one rule; less granularity makes an isolated rule difficult to understand.
A. A change of concept.
The General Problem Solver reflected the current trend of the sixties to attempt
producing general-purpose systems that heavily relied on reasoning and not on
domain knowledge. Both approaches have their advantages and disadvantages
though. Domain knowledge may be powerful, but it is limited to a particular domain.
Applying it to another domain involves skill rather than genuine expertise.
In the early seventies it was already obvious that the only way to achieve machines
comparable in performance with human counterparts was to use domain knowledge.
In many cases, human experts rely less on reasoning and a lot more on a vast
knowledge of heuristics (practical knowledge) and experience. When a human expert
is confronted with a substantially different problem, he/she must use reasoning from
the first principles and they are no better in this respect than any other person.
Therefore, expert systems relying solely on reasoning are doomed to failure. The
factors that contributed to the creation of the modern rule-based systems are
presented in Fig. 2. 2.

Fig. 2. 2 Creation of the modern rule-based expert system (Ref.1, p.15)

In 1965, work has begun on DENDRAL, the first expert system, aimed at identifying
chemical constituents. However, DENDRAL was an unique system in which the
knowledge base was mixed together with the inference engine. This made the task of
reusing it to build a new expert system almost impossible.
1972-73 brought the development of the MYCIN system, intended to diagnose
illnesses. MYCIN was a step forward because:

13
The Evolution of Expert Systems Expert Systems Evolution

• it has demonstrated that AI may be used in real-life situations;

• MYCIN was used to test and prototype the explanation facility, the automatic
knowledge acquisition (bypassing the explicit knowledge coding) and even
intelligent tutoring (found in the GUIDON system);
• the concept of expert system shell was demonstrated to be viable.
MYCIN explicitly separated the knowledge base from the inference engine, making the
core of the system reusable. Emptying out the knowledge base of the MYCIN system
for example, and reusing the inference and explanation modules could therefore build
a new expert system. The shell produced by removing the medical knowledge from
the MYCIN system produced EMYCIN (Empty MYCIN).
Back to the improvements in the control algorithms. In 1979 Charles R. Forgy, at the
Carnegie-Mellon University developed a fast pattern matcher algorithm, which stores
information about rules in a network. The RETE algorithm is fast because it does not
match facts against every rule on every recognise-act cycle. Instead, it looks for
changes in matches. Because the static data does not change, it can be ignored and
the processing speed increases dramatically. This algorithm was used to develop the
OPS (Official Production System) language and shell.
The foundations of the modern rule-based expert systems are shown in Fig. 2. 3.
A variation of the DENDRAL system was developed in 1978, called META-DENDRAL.
This program used induction to infer new rules of chemical structure. META-
DENDRAL has been one of the most important attempts to defeat the acquisition
bottleneck, where the knowledge has to be extracted from a human expert and
explicitly coded into the knowledge database.
Another example of meta-reasoning was the TEIRESIAS knowledge acquisition
program for MYCIN. In case of an erroneous diagnosis, TEIRESIAS would
interactively lead an expert through the reasoning MYCIN has accomplished, until the
point where the mistake was made is discovered. The acquired rule would first be
tested for compatibility with the existing rules, and if no conflicts found, added to the
knowledge base. Otherwise, TEIRESIAS would query the user about the conflict.
Meta-knowledge in TEIRESIAS can be either a control strategy (on how the rules
must be applied), or validation-and-verification meta-knowledge (whether the rule is in
the correct form so it can to be added to the knowledge base).

14
The Evolution of Expert Systems Expert Systems Evolution

Fig. 2. 3 The foundation of the rule based expert systems (Ref.1, p.32)

3.2.3 A new trend.

Around 1980, companies start producing commercial expert systems, accompanied by
powerful software for expert system development. An example is the Automated
Reasoning Tool (ART, released later on in 1985). Also in 1980, new specialised
hardware was developed, optimised to run the expert system software faster. The
LISP machines were specially designed to run the most used expert system
development base language of the period, LISP. The native assembly language,
operating system, etc were actually created in LISP (LISt Processing language).
As one can imagine, this type of dedicated technology came to a price. While being
more efficient and faster at running LISP code, the dedicated machines had price tags
around $100000. Obviously there was a need for more affordable software that could
run on common machines.
1985 is another milestone to the development of expert systems, with the introduction
of CLIPS by NASA. Written in C, CLIPS is portable (i.e. suitable for a large number of
platforms), fast (it uses the RETE pattern matching algorithm developed by Forgy in
1979) and cheap or free for some users. CLIPS can run on any machine that can run
the standards C language. CLIPS had actually made expert system development and
study accessible to large groups of potential users. This is always vital in promoting
the acceptance of new technologies.
1990 sees the revival of the neural networks with the Artificial Neural Systems (ANS,
refer section E, part 1). In 1992, 11 shell programs were available for Macintosh, 29
for MS-DOS programs, 4 for UNIX platforms and 12 for dedicated mainframe
applications (Ref.7).
By 1994, commercial ANS are being offered (refer section E, part 3) by various
companies at different levels of price and features. The number of expert systems
patents in the US has grown steadily from 20 in 1998 to over 900 in 1996.(Ref.7).
The GUESS system (Generically Used Expert Scheduling System) , by Liebovitz et al,
was devised in 1997, using an object oriented, constraint-based AI toolkit approach to
perform scheduling.
The current (1997-) trend cannot escape the influence of the Internet. Envisaged
applications are:
• intelligent customer and vendor agents for use on the web in E-commerce;
• intelligent search engines;
• interagent communications methods relating to AI in E-commerce.

3.2.4 Artificial Neural Systems

3.2.4.1 Basic concepts.

Artificial Neural Systems (ANS) have tried to solve problems by modelling the human
brain information processing. They represent a different programming paradigm, also
known as connectionism. Connectionism models solutions to problems by training
simulated neurons connected in a network. ANS may well be the ideal front end to
expert systems that require massive amounts of sensor inputs and fast response.

15
The Evolution of Expert Systems Expert Systems Evolution

Fig. 2. 4 A neuron processing element (Ref.1, p.45)

ANS uses simple processing elements connected in parallel, called neurons. A neuron
may have multiple inputs but only one output. The key to the functioning of an ANS is
the weights associated with each element (refer Fig. 2. 4). Each element (neuron) has
an associated threshold value θ, subtracted from the input I. The input signals of the
neuron are multiplied by the weights and summed to yield the total neuron input,
equal to 1. The output is given by the formula in Fig. 2. 4 and it is called an activation
function.

3.2.4.2 ANS advantages.

Since ANSes are modelled after current brain theories, information is represented by
weights. The distributed representation of information is similar to a hologram, where
the lines of the hologram diffract a laser beam passing through in order to reconstruct
the stored image.
Advantages of ANS over conventional computers:
• fault tolerant storage. Removing portions of the net will only determine a
degradation of the quality in the stored data;
• graceful degradation of the quality of the stored image, in proportion to the amount
of net removed;
• data is stored in the form of associative memory (i.e. where partial data is sufficient
to recall the complete stored information);
• nets can extrapolate and interpolate from their stored information. A net can be
trained to seek significant features / relationships in the data. Subsequently, the net
may suggest relationships with the new data.
• nets display plasticity. After the removal of a part of the neurons in the net, the net
can be retrained to its original skill level if sufficient neurons are left;
• ANSes are simple to build and cheap to maintain, mainly because of plasticity.
However, for number intensive tasks or algorithmic solutions, ANSes are not a good
choice.

3.2.4.3 ANS timeline.

The origin of artificial neural systems dates back to 1943 (refer paragraph A). However,
a milestone in the connectionist AI approach was 1961, with Rosenblatt's definition of

16
The Evolution of Expert Systems Expert Systems Evolution

perceptron, a new type of artificial neuron. The perceptron consisted of two layers of
neurons, and a basic learning algorithm. It displayed remarkable capabilities for
learning and pattern recognition. However in his model, the weights had to be set
manually.
In 1969, a new milestone occurs with the publication of the book Perceptrons by Minksy
and Papert. The book showed the limitations of the perceptrons as general computing
machines, proving that a perceptron could not recognise the exclusive OR logic
function. The book has had a negative impact on the ANS research. Eventually, new
methods of representing symbolic AI information by frames became popular in the
1970s, but research only continued on a small scale.

Fig. 2. 5 A back propagation net (Ref.1, p.46)

In 1982 though, Hopfield's work brought a revival of ANS by introducing the two-layer
Hopfield Net and demonstrating that ANS could solve a wide variety of problems. The
back-propagation net, commonly implemented as a three-layer net, is an example of
ANS that can solve the XOR problem (refer Fig. 2. 5).
Other types of ANS emerged, such as the counter-propagation net, invented by Hecht-
Nielsen in 1986. It was also proven that a three-layer network with n inputs and 2n+1
neurons in the hidden layer could map any continuous function.
The present trend in expert systems is towards the artificial neural nets. Also, there are
masses of commercially available expert systems and shells.
New companies and existing firms have been organised to develop ANS technology
and products. For example, NestorWrite is a program by Nestor Corp., which can
recognise handwritten input and convert it to text.
ANS simulators and hardware accelerator boards (to speed up the learning process)
are also marketed by a number of companies such as Texas Instruments, Synaptics,
Neural Tech, etc.

3.2.5 The future.

The future of expert systems is bright. Expert systems are being already used in almost
all aspects of our life, from space travel to agriculture, Internet to underwater devices.
Wherever there is a need to solve ill-structured problems in real-time, to be 99.9%
reliable and unemotional and to have an unlimited capacity for learning and
remembering, the expert systems will do nicely.

17
The Evolution of Expert Systems Expert Systems Evolution

Artificial Intelligence itself is advancing rapidly, and with the latest progress in genetics,
the first human druids may not be that far away in time. Such developments may
create totally new social and ethical problems, which may also need to be
investigated.

18
The Evolution of Expert Systems Knowledge Representation in Expert Systems

4 Knowledge Representation in Expert Systems.

This chapter will try to present the meaning of knowledge and some commonly used
representations of knowledge for expert systems.

4.1 What is knowledge ?

Knowledge is a very hard to define term and it has many meanings. Data, facts,
information are terms also being used with the meaning of knowledge.
The study of knowledge is epistemology. It contains:
• philosophic theories (Aristotle, Plato, Descartes, Kant, etc)
• a priori knowledge (considered to be universally true);
• a posteriori knowledge (derived from the senses).
Knowledge may be classified in:
• procedural knowledge (know-how);
• declarative knowledge (declarative statements);
• tacit knowledge (unconscious knowledge) - cannot be expressed by language. A
good example is how to contract a muscle.
Knowledge is of central importance to expert systems. Fig. 3. 1 shows a possible
hierarchy of knowledge. The of this structure foundation is noise, meaning items of
little interest and obscure data. Items of potential interest make up the next level, the
data. Processed data represents information, and sits on top of the data. The
knowledge layer, consisting of specialised information, is next. The last and most
interesting layer is the meta-knowledge, (meta meaning above).
Meta knowledge is knowledge about knowledge. For example in an expert system with
knowledge in several domains, the meta-knowledge would specify which knowledge
is applicable for a particular domain. Within a particular domain, meta-knowledge may
be used to determine which set of rules is applicable for a specific problem.

Fig. 3. 1 The hierarchy of knowledge (Ref.1, p.59)

19
The Evolution of Expert Systems Knowledge Representation in Expert Systems

4.2 Knowledge representation techniques.

4.2.1 Productions.
Production rules are commonly used in the knowledge bases of expert systems. A
formal notation for defining productions is the Backus-Naur form (BNF). BNF
represents metalanguage for defining the syntax of a language. While syntax defines
the form, the semantics refers to the meaning of the language. A grammar is a
complete set of production rules that uniquely defines a language.
There is a wide variety of languages that BNF can define, such as natural, logic,
mathematical, computer.

4.2.2 Semantic Nets.

Semantic network (nets) are also known as propositional nets. A proposition is a
statement that can either be true or false, and is a form of declarative knowledge (they
state facts).

4.2.2.1 Concepts.

The structure of a semantic net may be shown graphically in terms of nodes (objects)
and arcs (links or edges) connecting them. Nodes represent objects, concepts or
situations. The links show the relationships between nodes. If the links are directed
arrows, then the net becomes a directed graph (refer Fig. 3. 2). Relationships give the
knowledge contained in the nodes a cohesive structure about which other knowledge
may be inferred.

Fig. 3. 2 A type of semantic net (directed graph) (Ref.1, p.64)

4.2.2.2 Semantic net problems.

Unfortunately, semantic nets have some limitations in representing knowledge:

• Lack of link name standards and naming rules for the nodes. Only if the link and
nodes are unambiguously defined may the semantic net represent definitive
knowledge (knowledge that can be defined).
• Semantic nets cannot define knowledge in a similar way to logic;
• Heuristics (rules of thumb) on how to efficiently search the net cannot be
embedded in a semantic net. Heuristics play a major role in expert systems.

20
The Evolution of Expert Systems Knowledge Representation in Expert Systems

Although a number of approaches have been tried to correct the above-mentioned

problems, little improvement has been achieved at the expense of the natural
expressiveness of the semantic nets.

4.2.3 Schemata.
All the knowledge in a semantic net is contained in the links and nodes. Therefore, a
semantic net is an example of shallow knowledge structure. In contrast, a deep
knowledge structure may explain why something occurs via causal knowledge. In a
human expert's case, deep knowledge is usually called upon when causal knowledge
fails to solve the problem.
Schemas are used to represent more complex knowledge structures. Conceptual
schemas are abstractions in which specific objects are classified by their general
properties. Schemas are used to focus on the general properties of an object without
being distracted by irrelevant details.
Schemas have an internal structure to their nodes, in contrast with the semantic nets. A
semantic net is like a data structure in computer science. The schema is more like a
data structure in which nodes contain records that can further contain records, data or
pointers to other nodes.

4.2.4 Frames.
Two types of schemas have been used in many AI applications: the frame and the
script (a time-ordered sequence of frames)
Frames are very suitable for representing objects typical to a particular situation
(stereotypes). Commonsense knowledge is very difficult to master by the computers,
and frames can be of great help. While semantic nets are used to represent broad
knowledge, frames are efficient at representing a narrow subject containing much
default knowledge. Special purpose language have been designed for frames, such
as FRL, SRL, KRL, etc
An analogy can be made between a frame and a record in Pascal or an atom in LISP.
The correspondent frame elements to the fields and values of the record are the slots
and slot fillers of a frame. The fillers may be values (e.g. a property) or a range of
values (a type slot). Slots may contain attached procedures which can be if-needed
(when a filler value is needed but none present), default (expectations of a situation),
or if-added (when a value is added to a slot).
Sophisticated frame systems have been used in discovering mathematical concepts
and describing mathematical understanding in linear algebra.
Frame systems that allow unrestricted alteration or cancellation of slots may however
display major problems. Most frame systems do not provide a way to define
unchangeable slots. This leads to a system where nothing is really certain and no
universal statements can be made. Building composite elements from simpler frames
is also restricted.

4.2.5 Logic and Sets

Knowledge may also be represented by symbols of logic (the study of exact reasoning).
Logic is extremely important in expert systems, as the inference engine reasons from
facts to conclusions. The term of automated reasoning systems would include both
logic programming and expert systems.
Two common formal logic methods of representing knowledge are:

21
The Evolution of Expert Systems Knowledge Representation in Expert Systems

• syllogisms. Invented by Aristotle in the fourth century B.C., they have two
premises and one conclusion, which is inferred from the premises. The
premises provide the evidence from which the conclusion must follow. The
classical example is (Ref.1, p.64):
o Premise: All men are mortal
o Premise: Socrates is a man
o Conclusion: Socrates is mortal
• Venn Diagrams. In this model, a circle represents a set, and the objects
contained in the set are the elements (refer Fig. 3. 3). Socrates is an element,
therefore is represented as a dot and not a circle.

Fig. 3. 3 A Venn diagram for the example (Adapted from Ref.1, p.79)
The detailed descriptions of the knowledge representation methods previously
presented are beyond the scope of this paper. The interested reader is encouraged to
refer to Ref.1, pp. 57-91.

22
The Evolution of Expert Systems Conclusions and References

5 Conclusions.
This paper aimed to achieve two objectives:
• give a basic coverage on some important areas of expert systems, and maybe
incite the reader into further individual expert systems technology reading and
research;
• allow the author to train its own natural neural net (pun intended) with knowledge
in the field of AI and expert systems.
The history of Artificial Intelligence and expert systems started in fact very early, with
the ancient philosophers and then Renaissance scientists. The foundations of modern
expert systems are quite recent though (i.e. 1943 onwards). In terms of computer
history however, this is equivalent to hundreds of years of human history.
While Artificial Intelligence is far from being perfect and expert systems still have a long
way to go to fully model a human expert, the progress achieved in such a short period
is astonishing. And better still, the development pace is accelerating constantly, in
step with computer hardware technology.

Back to Table Of Contents.

6 References.

1. Giarratano & Riley (1998) Expert Systems: Principles and Programming, 3rd Ed.,
PWS Publishing Company, Boston, MA, ISBN 0 534 95053 1
2. Allen Newell and Herbert A. Simon, (1972) Human Problem Solving, Prentice-Hall.
3. Emil L. Post (1943), Formal Reductions of the General Combinatorial Decision
Problem, American Journal of Mathematics
4. George F. Luger, William A. Stubblefield (1993) Artificial Intelligence - Structures and
Strategies for Complex Problem Solving, Benjamin-Cummings, Albuquerque, ISBN
0-8053-4780-1
5. Stephen L. Gallant, Connectionist Expert Systems, Comm of the ACM, Feb 1998
6. Buchanan, Feigenbaum, Dendral and META-Dendral: Their Applications Dimension,
Artificial Intelligence, 1978.
7. AI Magazine, IEEE Journal, AI Expert publications, AAAI web site papers.

Back to Table Of Contents.

23