135

7.1 What is an expert system?

7 Expert systems

Before starting this chapter you should:

+ be familiar with the terms ‘software’, ‘database’, ‘hypothesis’ and

‘information’.

7.1 What is an expert system?

Expert systems are computer-based systems used to solve problems, such as

providing a diagnosis of something. They make use of the knowledge of a

number of human experts within a specific field. They are unlike other problem-

solving software in that they use reasoning to produce possible solutions. Data,

often referred to as a database of facts is stored within the computer and this is

the knowledge part of a knowledge base. The system uses a set of rules, which

represents the reasoning.

Expert systems can only be used in those areas where human experts already

operate. Clearly defined rules and subject facts must already be in existence.

Expert systems cannot be applied to problems in which there are missing facts,

or in which rules can be interpreted in a variety of ways.

One way of approaching this topic is to think of the internet. The world wide web

contains a huge amount of information or knowledge. This knowledge is used to

solve a problem by you asking questions (making searches). You use the answers to

these questions (searches), which we could argue are possible solutions. You then

consider the possibilities before making a decision on the most appropriate solution.

Let us look at a use of an expert system in the field of medicine. It is used to

help a doctor to make a diagnosis. The doctor types in the symptoms and the

expert system produces a list of possible solutions, each usually with a percentage

probability. The doctor then uses their experience to make a decision, choosing

which of the possibilities produced by the system is likely to be correct.

Expert systems are not like most problem-solving programs. Problem-solving

programs are produced to solve a specific problem. They can only solve that

particular problem and it is difficult to adapt the program to alternative

scenarios. The facts needed to solve the problem are an integral part of the

programming. If the facts change, then the whole program may need to be

rewritten. Compare that with expert systems, which are used to collect the

knowledge of experts in order to help create a

knowledge base.

In this chapter you will learn:

+ what an expert system is

+ whatthecomponentsofanexpertsystem

are

+ howexpertsystemsareusedtoproduce

possible solutions for different scenarios

+ whattheconceptsofforwardandbackward

chaining involve

136

7 EXPERT SYSTEMS

There are alternative ways of looking at the make-up of an expert system, but for

the purposes of our text we shall consider the system to be made up of a

shell of

the major components and a knowledge base which comprises the

database of

facts and the rules base (see Figure 7.1).

Explanation system

Inference engine

Knowledge

base editor

User

interface

Shell Knowledge base

Database

of facts

Rules base

V Figure 7.1 A graphic representation of our expert system

As you can see, many of the components are housed within a shell. The shell

often incorporates the knowledge base editor, which is basically software

that allows the knowledge engineer to edit the rules and facts within the

knowledge base to bring them up to date. It also contains the explanation

system

. This is the part of the system which explains the logical reasoning that

the expert system has used to come to its conclusions.

The explanation system causes the expert system to explain to a user how a

particular conclusion was reached and an alternative conclusion was not. It also must

be able to explain why a specific fact was needed, or not, as the case may be. An

expert system must be able to explain its reasoning and justify its recommendations.

The shell of the expert system is very important. A good expert system has a

shell which can be used for solving different problems just by changing the

knowledge base to one that matches the type of problem. The shell of an expert

system, technically speaking, is an expert system with an empty knowledge base.

It can be used for the development of other expert systems by adding a new

knowledge base. It is possible that a diagnostic expert system could be used to

diagnose car faults or medical illnesses just by changing the knowledge base.

Another important feature of an expert system is the user interface. This is the

component that allows the user to interact with the computer which stores

the expert system. It usually consists of a keyboard and mouse together with a

monitor. The questions which need answering are displayed on the monitor. The

user can then manoeuvre round the screen of the monitor using the mouse and

can use the keyboard to type in their responses to the questions. Supplementary

questions are often asked by the system and appear on the screen, which the user

then answers using the keyboard. Validation checks are carried out to ensure the

minimum number of invalid responses are entered by the user.

The

inference engine is another important component of an expert system.

It performs the reasoning of an expert system. It uses a series of IF…THEN

statements within the rules base and interrogates the database of facts in the

knowledge base. It uses inputs from the user to try and match them with the

contents of the database of facts. It processes the rules to try and reach possible

solutions. It can also generate questions via the user interface based on the

inputs it has received so that it can use the answers to narrow down the number

of possible solutions. In short, it produces a set of possible solutions comparing

the inputs with the facts, using the rules contained in the knowledge base.

137

7.2 Different scenarios where expert systems are used

When an expert system is created, a knowledge engineer needs to be employed.

Human experts or specialists have deep knowledge of their subject area and

the engineer makes use of this by collecting information from them. This will

consist of the experts’ knowledge and any reasoning strategies that are used by

these experts. The knowledge engineer must also find out from the experts the

type of system they want. In addition, the engineer gathers data from databases

that may exist for the subject area the expert system is being created for. The

knowledge engineer must then design the knowledge base and the rules base.

During the whole process the knowledge engineer and subject expert work

together in order to define the problem the expert system will solve. It is the job

of the engineer to integrate the knowledge into the knowledge base. Then the

inference engine has to be created to provide the reasoning aspect.

Activity 7a

Describe the components of an expert system.

7.2 Different scenarios where expert

systems are used

There are many different uses to which expert systems can be put. In the

following scenarios reference is made to many of the components of these

systems. You must also try to see how the components which are not mentioned

would apply. The knowledge base editor software, for example, is used in much

the same way regardless of the scenario. The same applies to explanation systems

and knowledge engineers.

7.2.1 Mineral prospecting

Expert systems are used to help in making decisions on where to drill for a

specific mineral. They help geologists evaluate whether particular sites are

likely to contain the mineral or ore being searched for. The system produces

probabilities of minerals being found in particular locations. As well as

producing a conclusion about the existence of a particular ore deposit, these

types of system often provide explanations regarding their conclusions. The

expert system’s knowledge base will store models of successful drilling sites and

their associated minerals together with a rules base. First of all, the geologist

enters data about the types of rocks and geological structure as well as the

minerals to be found at the site under consideration. The inference engine

then compares this data with the models in the knowledge base using the

IF…THEN rules in the rules base. The system may well ask the geologist

further questions through the user interface and the resulting percentage

probabilities of finding the mineral at this site are produced.

7.2.2 Investment analysis

Expert systems have been developed which act as assistants to financial

advisors at many banks. The financial advisor still makes the final decision,

but the expert system proves to be a very useful tool in helping the advisor

make recommendations to would-be investors. When making a decision, it is

important for the financial advisor to know four things about the investor:

» The amount the client wishes to invest

» What level of risk the investor is willing to take

138

7 EXPERT SYSTEMS

» The desired rate of return

» How long the investor is willing to make the investment for.

The user interface asks questions in the everyday language of the investor based

on the above factors. Depending on the answers to the questions, the system

may ask further questions through the user interface. The database of facts will

consist of all the information the bank has gathered over the years about the

performance and stability of different companies. The rules in the rules base

will have been originally created by an expert investor. The expert system will,

however, allow the financial advisor to add rules to the rules base as they see fit,

based on their experience of personally dealing with the investor. The inference

engine will search the database of facts using the rules from the rules base.

The expert system will then produce recommendations for purchasing stocks

or shares in a company, together with the reasons for recommending these

particular stocks and shares.

7.2.3 Financial planning

Today, expert systems in personal financial planning are needed more than ever.

Personal financial planning has become so complex that it is extremely difficult

for one human expert to deal with it. The result is a need for computerised

expert systems to assist in the process. An expert system is different to a

normal computer program in that the problems it addresses tend not to have

a unique solution. There are so many variables to deal with that only a few

can be considered by one person in the time usually available. Expert systems

can be used to create financial plans for individuals. Such plans include

recommendations as to how individuals can manage their debts, reduce the

amount of income tax they pay, and organise investments. These plans can also

recommend the amount of insurance individuals should have and how they

can plan for their retirement. They can also suggest short-term personal saving

targets. Systems often involve the client completing a detailed questionnaire

regarding their personal and economic situation, including income, taxes paid

or due, investments, employee benefits, marital status, number of dependants,

as well as any future obligations such as retirement plans or funding their

children’s university education. The results of the questionnaire are entered

using the user interface. The inference engine analyses this information using

the database of facts and a very large number of rules in the system. The

database uses the results of books and articles which have been published

relating to personal financial planning as well as information gathered from

interviews with experts, such as professional financial advisors. A report would

then be produced by the financial advisor as to the client’s recommended

financial plans. The report would be based on a combination of a human

expert’s insights and the use of the results of the expert system. The report

would serve as a financial action checklist for the client in order to achieve their

financial plans.

7.2.4 Insurance planning

Insurance involves the use of underwriters. Underwriters are usually banks,

insurance companies or investment companies. They guarantee to pay a certain

amount of money to the person who has taken out the insurance if a particular

event happens. Individuals or organisations can take out insurance for any

number of circumstances. Life insurance is taken out against an individual’s

life. Should the person die then his or her relatives receive a fixed amount of

money. Car insurance is when a person insures their car against damage or theft.

139

7.2 Different scenarios where expert systems are used

Property insurance involves safeguarding against damage or destruction to

property.

Insurance can involve the payment of one large sum of money or regular,

usually monthly, payments. Some insurance policies are for a fixed amount of

time. A car is usually insured for a year at a time. At the end of the year the

driver has to renew their insurance. Obviously, if there has been no damage

to the vehicle in that time the underwriter has made some money. The same

applies to life insurance. A life insurance policy usually lasts for a fixed period

of time, such as 10 or 20 years. If the insured person dies within that time the

underwriter has to pay out to their relatives. If the person is still alive then the

underwriter will be able to keep all the monthly payments as profit. There are

many other types of insurance, for example health insurance. With any type of

insurance, the underwriters have to assess the risk involved. How likely is it that

they will have to pay out?

Expert systems are frequently used to assess the risk. In order to set up an expert

system, insurance experts would be interviewed by the knowledge engineer and

all the policies that the underwriter currently offers would be examined. The

results would then be used to create the knowledge base, including the rules

based on those used by the experts when arriving at their decision.

The process of calculating life or health insurance premiums (monthly

payments) involves the consideration of a number of factors, such as the

individual’s age, gender, whether or not they smoke, and their general health.

In general terms, the older a person is, the greater amount they have to pay

for their insurance. Men tend to pay more than women as their life expectancy

tends to be lower. Smokers pay a lot more than non-smokers, as again their

life expectancy is shorter. If you are not in good health, then you are a bad

risk to the underwriter and so will pay a lot more than a healthy person. An

insurance expert system’s user interface will ask questions of the user and based

on their responses will, through the use of the inference engine, compare

these inputs with the facts in the knowledge base. Using the rules which have

been created, it will produce a recommendation for the cost of the individual’s

premium. It may, of course, recommend that the application be rejected. Many

underwriters use human insurance consultants who will use the expert system’s

recommendations to make a final decision.

7.2.5Carenginefaultdiagnosis

Like any type of mechanical equipment, motor vehicle engines can develop

faults. When this happens, the driver has to seek the help of a mechanic (expert)

who knows a lot about vehicle engines. Modern-day mechanics use expert

systems to help them diagnose the fault. In order to create such an expert

system, many expert mechanics have to be interviewed to gather together their

knowledge and rules (systems) for diagnosing.

Expert systems are now an everyday aid to a car mechanic’s ability to diagnose

engine faults. The inclusion of complex electronic technologies into a typical

car engine has led to the need for highly complex expert systems. A systematic

approach to the diagnosis of faults is required, as the amount of knowledge

required is so vast and there are many steps which need to be followed in order

to identify the fault.

The use of expert systems is often based on the expert system’s user interface

asking questions regarding the problems with the car and the mechanic typing

in the symptoms. The expert system would then supply a list of possible causes

140

7 EXPERT SYSTEMS

of the problem and the mechanic would decide which of the suggestions are

most likely.

A second type of expert fault diagnosis system is a device which is plugged into

the car engine’s computer and questions it. We often think these are true expert

systems, in that mechanics can read the results of testing from the expert system

and identify the exact problem that has occurred.

What often happens, in reality, is that true expert systems inform the mechanic

of a range of readings from sensors but do not detail the cause of the problems.

The mechanic has to use their experience and expertise to actually diagnose

the fault. However, new true expert systems have been developed in which the

readings from the sensors are used by an inference engine, after comparison

with those in the knowledge base, to produce a list of possible faults. Although,

again, with this system, the mechanic makes the final decision.

7.2.6 Medical diagnosis

Since early computers were developed, it became apparent that they would

be able to assist medical practitioners in their day-to-day work. Applications

such as drug development and controlling the administration of drugs are

commonplace. Computers are also used as aids when diagnosing illnesses.

Because enormous amounts of data are required, it was originally thought that

expert systems could only be used with large, expensive computers. However,

with the development of more powerful, cheaper PCs and laptops, expert

systems are now readily available to many medical practitioners.

The creation of a medical diagnosis expert system starts with the production

of the shell followed by the knowledge base. The knowledge base is created by

a knowledge engineer, who interviews a number of expert doctors and then

shapes the knowledge into a form that can be easily accessed by the system. The

components of an expert system are used to produce a list of possible diagnoses.

The user interface allows the patient, or doctor, to enter symptoms and, at

the end of the process, displays the output. The knowledge base provides the

database of existing illnesses as well as the rules from the rules base, often IF…

THEN rules. The IF part is the symptoms part and the THEN part is the

illness or disease that is suggested. The inference engine searches the knowledge

base for similar symptoms to those entered by comparing those typed in with

those in the knowledge base. This is the reasoning part of the system. The

knowledge base editor allows for the knowledge base to be increased by adding

further rules and data. The explanation system clarifies how the expert system

has produced its possible diagnoses.

Other types of medical diagnosis expert systems have been developed which

allow the patient to interact with the system directly. However, these are only

really suitable for minor complaints and such systems, after considering the

symptoms, will often recommend the patient should visit a human doctor.

7.2.7Routeschedulingfordeliveryvehicles

For most manufacturing companies, the cost of transport is a major part of their

outgoings. Therefore, it is essential that companies use their transport vehicles

as efficiently as possible. It is important that they develop delivery schedules

which prevent vehicles from travelling further than necessary. One of the most

efficient ways of ensuring this is to use an expert system. Human experts have

developed skills which enable them to solve such problems. It is important,

therefore, to incorporate these schedulers’ skills in an expert system. A major

141

7.2 Different scenarios where expert systems are used

problem, however, is that there tends to be little agreement about the facts or

rules required in such a knowledge base.

For example, a company which manufactures washing machines might have to

deliver these washing machines to distribution points at the cheapest cost. If

the company is situated in a country which occupies a large area, this makes

the problem rather difficult. It could be that it has several distribution points in

different parts of the country, each with several delivery vehicles, each of which

may be a different size and may be available at different times. The problem

then becomes, how do you ensure that the available vehicles provide coverage

of all the distribution points but travel the least number of miles? The problem

might also need to be broken down into the number of vehicles, size of vehicles,

and loading of the vehicles. Vehicles have to be loaded so that each distribution

point has quick access to its order. The vehicle has to be loaded so that when

it arrives at a distribution point, machines do not have to be climbed over, or

temporarily unloaded, to get to the order for that distribution point. In order to

achieve this, the route for each vehicle has to be planned in advance.

An expert system can have the locations of each distribution point in its

database of facts, together with the type and speed of the vehicle being used,

the total available time and perhaps other constraints such as the topology or

terrain of each road, whether it is mountainous or flat for example. This aspect

of the problem is often referred to as the ‘travelling salesman’ problem. This

refers to the times when a person was employed to travel to different places, to

try to persuade people to buy their company’s goods. The company would pay

them commission, that is, they would receive a small percentage of each sale

they made. It was, therefore, in their own interest to visit as many customers

over a large area in as short a time as possible.

Although there are several linear programming algorithms available to solve

such problems, they tend to take more time and more computer processing

power to get to a solution. The expert system would examine the orders for

the day, total the weights for the order for each location and the inference

engine would match those against the available types of vehicle. The expert

system would suggest an allocation of orders to each vehicle and suggest a list

of machines in reverse order so that each vehicle had the first order loaded on to

the vehicle last. The company’s vehicle scheduler would take these suggestions

from the expert system and decide on the number of vehicles needed and would

know the total time to be taken.

7.2.8Plantandanimalidentification

There are many hundreds of thousands of different species of plants and the

number of different species of animals runs into millions. Identification of

either plants or animals can prove difficult, particularly with rare species, but

it has become extremely important with climate change affecting the existence

of certain species. Experts are required to master a number of techniques and

gather a great deal of knowledge in order to be able to identify plants and/

or animals. They need additional tools to help them identify all the different

species.

When considering plant or animal identification then, an expert system would

be helpful. The inference engine would use a series of IF…THEN rules. Plants

can be possibly divided into whether they are woody or herbaceous. IF they

are woody the system, through the user interface, could ask another question,

such as whether they have a single trunk, which would lead in turn to other

questions based on the shape/type of leaf and so on. Eventually, the possibilities

142

7 EXPERT SYSTEMS

would be output through the user interface. The database of facts, which would

be interrogated by the inference engine, would include the data regarding all

knowledge about as many species as possible. This type of system would ask

questions rather than inviting the expert to type in details about the plant.

Animal classification or identification follows a similar pattern.

Activity 7b

1 State the inputs to an investment analysis expert system.

2 State the outputs from an insurance planning expert system.

3 Describe the difficulties involved in producing an expert system for

scheduling routes for delivery vehicles.

7.3 Chaining

When an inference engine interrogates the knowledge base, it imitates the way a

human would use reasoning. It uses what is known as chaining. The reasoning

that the inference engine uses consists of

backward chaining or forward

chaining. Often it uses a combination of both. Here we are going to see how

both forms of reasoning work.

7.3.1Backwardchaining

Backward chaining is often referred to as goal-driven. This is because it starts

with a goal or set of goals that basically establish which rules are to be followed.

It works backwards from the goals. As we said earlier, the rules in an expert

system are based on the IF…THEN construct. When an inference engine uses

backward chaining, it explores the rules the system has been given until it finds

one which has a THEN part that matches a required goal. If the IF part of that

rule is known to be true, then it is added to the list of goals.

Let us look at an example of the use of backward chaining.

Example 7.1 Backward chaining

Suppose a zoo has just received a new animal from Africa. We shall call her

Edwina and she has been delivered in an enclosed trailer. We cannot see

inside the trailer, but we are given two facts about Edwina:

Fact 1: Edwina has hooves.

Fact 2: Edwina can run fast.

We want to know whether Edwina has a striped coat. In terms of using an

expert system, the goal is to decide whether Edwina has a striped coat. We

will use a rules base containing the following four rules:

Rule 1: IF X has hooves and can run fast, THEN X is a zebra.

Rule 2: IF X cannot fly and X is a bird, THEN X is an ostrich.

Rule 3: IF X is a zebra, THEN X has a striped coat.

Rule 4: IF X is a giraffe, THEN X has irregular shaped spots.

We call the part that begins with IF the premise (what we are supposing

could be true). We will call the part that is based on the result of whether the

premise is true, the THEN part, the conclusion.

143

7.3 Chaining

Because the selection and use of rules is based on the list of goals, this method

is often called goal-driven.

7.3.2 Forward chaining

We are now going to compare the above method of reasoning with forward

chaining. Forward chaining is often referred to as data-driven, because the data

entered into the system determines which rules are selected and used. When an

inference engine uses forward chaining, it explores the rules the system has been

given until it finds one which has an IF part that is true. It takes the THEN

part to add new facts. This process is repeated until a goal is achieved.

Let us take a different approach to our problem regarding Edwina the zebra.

Using backward chaining, an inference engine can decide whether Edwina has

stripes in four steps. The goal that has to be proved is ‘Edwina has a striped coat’.

Step 1: ‘Edwina’ replaces X in Rule 3 to see if its conclusion matches the

goal, so Rule 3 becomes:

IF Edwina is a zebra, THEN Edwina has a striped coat.

Since the conclusion matches our goal ‘Edwina has a striped coat’, we now

need to see if the premise – Edwina is a zebra – can be proved. So, the

premise now becomes our new goal: ‘Edwina is a zebra’.

Step 2: Again, replacing X with ‘Edwina’, in Rule 1, the rule becomes:

IF Edwina has hooves and can run fast, THEN Edwina is a zebra.

Since the conclusion matches the current goal ‘Edwina is a zebra’, the

inference engine now needs to see if the premise – Edwina has hooves and

can run fast – can be proved. The premise therefore becomes the new goal:

‘Edwina has hooves AND can run fast’.

Step 3: Since this goal is a combination of two statements, the inference

engine breaks it down into two sub-goals, both of which must be proved:

‘Edwina has hooves’ and ‘Edwina can run fast’.

Step 4: To prove both of these sub-goals, the inference engine sees that both

of these sub-goals were given as Fact 1 and Fact 2 at the start of this problem.

Therefore, the statement is true: ‘Edwina has hooves AND can run fast’.

The premise of Rule 1 is true and the conclusion must be true: Edwina is a zebra.

It follows that the premise of Rule 3 is true and the conclusion must be true:

Edwina has a striped coat.

Notice that Rules 2 and 4 were not used.

Example 7.2 Forward chaining

Suppose that the goal is to decide the pattern of the coat of Edwina. Again,

we can assume that Edwina has hooves and can run fast and that the rules

base contains the following four rules:

Rule 1: IF X has hooves and can run fast, THEN X is a zebra.

Rule 2: IF X cannot fly and X is a bird, THEN X is an ostrich.

Rule 3: IF X is a zebra, THEN X has a striped coat.

Rule 4: IF X is a giraffe, THEN X has irregular shaped spots.

Again, we have two known facts:

144

7 EXPERT SYSTEMS

The difference in approaches is that backward chaining starts with us having a

goal to achieve or an assertion we are trying to prove. We think Edwina has a

striped coat, but we need to prove it. With forward chaining we do not know

what type of coat Edwina has but we want to find out.

Activity 7c

Explain, by referring to the IF…THEN construct, the use of the words ‘forward’

and ‘backward’ when applied to methods of chaining.

7.3.3Applicationsofforwardandbackwardchaining

Generally, the choice of chaining strategy depends on the type of problem.

Forward chaining, we know, is data-driven. The inference engine starts with

the facts of the case and progresses to a goal or conclusion. This approach is

thus determined by the facts available and by the premises that can be met. The

inference engine attempts to match the condition (IF) part of each rule in the

rules base with the facts which have been input. Forward-chaining systems are

commonly used to solve more open-ended problems of a design or planning

nature, such as establishing the configuration of a computer system. Backward

chaining on the other hand is goal-driven. If there are very few possible goals or

conclusions, this is a better approach to take. Backward chaining is best suited

to problems where the possible conclusions are few and well defined.

Diagnoses

Expert systems have been associated with diagnosis applications for many years.

Classification systems or diagnosis applications are best suited to backward

chaining as there are few possible conclusions which need to be checked against

the data. Backward chaining is the best choice if the goal can be sensibly

guessed at the beginning of a consultation. Forward chaining is the best choice

if all the facts are provided at the outset or there is not any sensible way to

guess what the goal is at the beginning of a consultation. A backward-chaining

system is more likely to cause questions to be asked which seem focused and in

an order that appears logical to the user. In an expert system used for diagnosis,

it is important that the system should seem to behave like a human expert. In

this case, backward chaining would usually be the most appropriate method.

Diagnosis applications tend to have a fixed number of possible solutions which

makes backward chaining more suitable.

Fact 1: Edwina has hooves.

Fact 2: Edwina can run fast.

With forward chaining, the inference engine can deduce that Edwina has a

striped coat by using the following steps:

Step 1: Since the facts we are given show that Edwina has hooves and

Edwina can run fast, the premise of Rule 1 is matched by replacing X with

‘Edwina’ (IF Edwina has hooves and can run fast). The inference engine

comes to the conclusion: Edwina is a zebra.

Step 2: We can now use Rule 3 by replacing X with ‘Edwina’ (IF Edwina is a

zebra). The inference engine comes to the conclusion: Edwina has a striped

coat.

145

7.3 Chaining

To sum up, the choice of chaining method depends on the problem. If you

have clear hypotheses, backward chaining is likely to be a better solution. It

can be used for classification systems such as animals and plants. Forward

chaining would be preferred if you have no clear hypothesis and want to see

what can be concluded from the current state of affairs. Despite this, however,

most expert system applications tend to use a mixture of forward and backward

chaining. A number of diagnosis expert systems actually begin by using forward

chaining and this is why this approach was described in such detail in the

medical diagnosis section on page 140. It is often the case that in a face to face

consultation the doctor would be aware of all the facts of the case.

Gaming

How does an expert system work within computer gaming? Let us consider

the most popular type of game, which involves the use of non-player characters

(NPCs). The goal of the expert system in this case is to select which weapon

is to be used during the course of the game by an NPC. Every game is a form

of animation which consists of individual pictures called frames, in a sequence

of moving images. The choice of weapon in a game will depend on a number

of different variables that are changing in every frame of the game. The expert

system is required to make a decision. The input variables may be the distance to

the target, the type of target and the range of the weapon. The system will need

to output the type of weapon, why it was chosen, as well as the characteristics

of the weapon. When setting up the system the knowledge engineer would

have gathered from the game expert known facts such as individual weapon

characteristics and their range, as well as which weapon is best for achieving

the goal of eliminating the target. As the contents of the frame are constantly

changing, the NPC will constantly have to re-evaluate the situation, much like a

human player. Creation of an expert system will improve game play and provide

a challenge, with the NPC making its own decisions on weapon selection.

The inference engine in such a system tends to use forward chaining. As we have

seen, this method uses the data or facts which are input to the system and uses

rules to determine a goal. Using the forward-chaining method, a series of IF

statements are executed that sort through the data to THEN find a goal. The

goal in this case is to select an appropriate weapon. These IF questions could,

for example, be based on whether the target is a missile of some sort or another

player. The THEN part will be the recommended type of weapon.

A good example of backward chaining in computer games is found in the

game of chess. In chess, there are a fixed number of moves to reach a winning

position. To compute all the possible ways of reaching a winning position,

the system chains backwards and builds up a list of possible ways to get to the

winning position. Backward chaining involves using all the possible premises

which lead to getting a ‘checkmate’ conclusion. Backward chaining is used in

gaming with games that have fewer rules.

Artificial intelligence

Artificial intelligence (AI) can be thought of as a piece of software that

simulates or mimics the behaviour and judgement of a human being. It involves

the system being able to learn, reason and assess its own actions. In AI, forward

chaining can be used to solve logic problems using rules and previous learning

to find solutions to problems. An AI might use forward chaining to look at the

available information, answer a question or solve a problem. Forward chaining

works through the logic of a problem from beginning to end.

146

7 EXPERT SYSTEMS

There are countless applications of AI: in the healthcare industry for dosing

drugs (calculating the dose of drugs to be given to a patient) and suggesting

different treatment for patients; in surgical procedures in operating theatres in

hospitals; in computers that play chess and in self-driving cars. Each of these

machines must be able to consider each action it takes, as each action will

have an impact on the end result. In chess, its goal is to get its opponent in

checkmate; in self-driving cars, its goal is to get from A to B without crashing.

AI can also be used in the financial sector, where it can alert fraud departments

to unusual spending patterns on a bank card or large amounts of money being

deposited in bank accounts.

Another example of AI is in the field of robotics where basic expert systems

are incorporated. These tend to use forward chaining where there is a clearly

defined goal, such as moving from one spot to another, although the variables

and obstacles may be numerous. Natural language processing (NLP) is another

application of AI. The goal is for people to be able to converse with computers

and computers being able to understand. Such systems tend to use a mixture of

forward and backward chaining.

Examination-style questions

1 Doctors often use expert systems to help them diagnose illnesses.

Explain how an expert system would work in this situation.

[6]

An expert system consists of many components. Describe the role of a

knowledge base in such a system.

[4]

Expert systems often use forward chaining or backward chaining when

producing suggestions. Explain what is meant by:

a forward chaining [3]

backward chaining. [3]