INFO612 Knowledge-Based Systems Dr. R. Weber. Copyright R. Weber Expert Systems Expert Systems are...

INFO612 Knowledge-Based Systems

Dr. R. Weber

Co

py

rig

ht

R.

We

be

r

Expert Systems Expert Systems are the first successful

knowledge-based methodology uses knowledge (in its knowledge base) &

reasoning Systems that manipulate knowledge and reasoning

to solve problems rationally.

KBS, Knowledge engineering, ES

Co

py

rig

ht

R.

We

be

r

Expert Systems, what and what not ES use knowledge and inference procedures to

solve problems that are difficult enough to require human expertise to solve (Feigenbaum, 82)

ES is a methodology to develop computer programs that manipulates expertise to solve problems that require human expertise in restricted domains (Weber 02)

Co

py

rig

ht

R.

We

be

r

Expert Systems: history (i)

Co

py

rig

ht

R.

We

be

r

Expert Systems: history (ii)

Co

py

rig

ht

R.

We

be

r

Expert Systems: domain areas

agriculture, business, chemistry, communications, computer systems, education, electronics, engineering, environment, geology, law, manufacturing, mathematics, medicine, simulation, transportation, etc.

Co

py

rig

ht

R.

We

be

r

Expert Systems: when

ES are indicated to solve expert problems in restricted domains without an efficient algorithmic solution

is there an alternative method? ill-structured problems is the domain well-bounded? how available is the source of knowledge? is the approach to the problem it trial-and-error? is the approach to the problem heuristic?

Co

py

rig

ht

R.

We

be

r

Expert Systems: tasks

analysis, configuration, control, design, diagnosis, instruction, interpretation, monitoring, planning, prediction, prescription, prognosis, remedy, selection and simulation.

Co

py

rig

ht

R.

We

be

r

ES and AI tasks

From: Durkin, J. (1994). Expert Systems: design and development. Prentice-Hall, Inc., New Jersey.

Co

py

rig

ht

R.

We

be

r

Types of AI tasks

mundane: face recognition argumentation shopping planning

expert: diet prescription medical diagnosis legal argumentation legal, military, business planning

Co

py

rig

ht

R.

We

be

r

the concept

knowledgebase

(e.g.,framesand methods)

knowledgebase

(e.g.,framesand methods)

expertproblemexpert

problem

inferenceengine

(agenda)

inferenceengine

(agenda) expertsolutionexpert

solution

knowledge

reasoning

Co

py

rig

ht

R.

We

be

r

expertsolutionexpert

solution

The complete methodology

knowledgebase

(e.g.,framesand methods)

knowledgebase

(e.g.,framesand methods)

explanationexplanation

generalknowledgegeneral

knowledge

userInterface

userInterface

expertproblemexpert

probleminferenceengine

(agenda)

inferenceengine

(agenda)

working memory(short-term mem/information)

working memory(short-term mem/information)

Knowledge acquisitionKnowledge acquisition

Co

py

rig

ht

R.

We

be

r

What are rule-based expert systems?

Co

py

rig

ht

R.

We

be

r

Represent knowledge?

How ES represent knowledge? Knowledge representation formalisms

associated to ES:• rules• semantic networks• frames• logic

Co

py

rig

ht

R.

We

be

r

(Production) Rules

A logic sequence of an antecedent (premise, condition) and a consequence (conclusion, action).

Both antecedent and conclusion are, in essence, facts.

Co

py

rig

ht

R.

We

be

r

(Production) Rules (ii)

The antecedent attempts to verify if the fact is true or false, when the fact composing the antecedent is true, the conclusion is triggered.

The antecedent can be composed of several facts connected through operators such as and, or, and not.

Conclusions usually change or assign values to attributes of an object, call methods or trigger other rules.

Co

py

rig

ht

R.

We

be

r

Semantic Networks (SN)• commonly used in logic-based expert systems• directed graphs where:

• nodes represent objects and concepts• arcs represent relationships between objects and

attributes

• Quillian, 1968• used to represent static elements of a

representation such as the class, the instances and its features

Co

py

rig

ht

R.

We

be

r

Characteristics of SN

• cannot represent all magnitude of data (meal varying from sandwich to 20 course meal)

• very restricted in terms of inferencing• only inheritance through instance and subclass

• convenient when mathematical algorithms are applied over knowledge because graphs also provide a formal and precise representation model

Co

py

rig

ht

R.

We

be

r

Example of Semantic Networks

Co

py

rig

ht

R.

We

be

r

Frames

representation formalism commonly used in expert systems

represents declarative, structural and procedural knowledge

Co

py

rig

ht

R.

We

be

r

Frames

Introduced by M. Minski in 1975, “A frame is a data-structure for representing a

stereotyped situation, like being in a certain kind of living room, or going to a child's birthday party. Attached to each frame are several kinds of information. Some of this information is about how to use the frame. Some is about what one can expect to happen next. Some is about what to do if these expectations are not confirmed”.

Co

py

rig

ht

R.

We

be

r

Concepts, Objects and Facts

An object is a basic entity that can be instantiated.

A concept tells something about the object.

A concept can be represented as an abstraction of an object when several objects can be grouped under the same concept (e.g., client 1, client 2, all clients);

Co

py

rig

ht

R.

We

be

r

Concepts, Objects and Facts (ii)

or a concept can be an attribute, when it tells something exclusively about this object or due to the analysis it is not worthy to represent it as an abstraction.

When an object is associated to a valued attribute, it is a fact. A fact is a statement that can be either true or false (Durkin, 1994).

Concepts can be described in a computer program via Y/N or T/F statements.

Co

py

rig

ht

R.

We

be

r

Characteristics of frames

support inheritance (subclasses and instances)

support methodswhen neededafter changedbefore changed

easy to implement in different programming paradigms, logic-based or not

Co

py

rig

ht

R.

We

be

r

Rules combined with frames

advantages faster inferences, increases inferential efficiency

rules with variables in its antecedents and conclusions

Co

py

rig

ht

R.

We

be

r

Decision trees

Knowledge representation formalism Represent mutually exclusive rules (disjunction) A way of breaking up a data set into classes or

categories

Co

py

rig

ht

R.

We

be

r

Decision treesconsist of:-leaf nodes (classes)

- decision nodes (tests on attribute values)

-from decision nodes branches grow for each possible outcome of the test

From Cawsey, 1997

Co

py

rig

ht

R.

We

be

r

Knowledge representation formalisms

rules logic concepts, frames semantic nets decision trees

representational adequacyinferential adequacyinferential efficiency

clear syntax and semanticsnaturalness

Co

py

rig

ht

R.

We

be

r

Inference engine forward chaining backward chaining logic theory

Co

py

rig

ht

R.

We

be

r

Investment Advisor

Frames Concepts, Objects, Facts Rules Backward Chaining

Co

py

rig

ht

R.

We

be

r

Heart Attack Triage

Facts/Predicates Rules Forward Chaining

Co

py

rig

ht

R.

We

be

r

Clips

Rule-based Forward chaining Logic-based

Co

py

rig

ht

R.

We

be

r

Clips rules or productions

Co

py

rig

ht

R.

We

be

r

Compound productions

Co

py

rig

ht

R.

We

be

r

Expert Systems: requirements

high quality: system must perform equally or better than a human expert

response time should be adequate to the problem it solves

reliable: not prone to crashes & errors

explanation capability should be present with the purpose of justification and verification of correctness (p. 9,10 for explanation styles)

flexible: supported by good maintenance methods

Co

py

rig

ht

R.

We

be

r

Advantages (i) Permanence of knowledge: Expert systems do not forget or retire or quit, but human experts may

Breadth: One ES can entail knowledge learned from an unlimited number of human experts.

Reproducibility: Many copies of an expert system can be made, but training new human experts is time-consuming and expensive.

Efficiency: can increase throughput and decrease personnel costs

Differentiation: In some cases, an expert system can differentiate a product or can be related to the focus of the firm

Cost: Although expert systems are expensive to build and maintain, they are inexpensive to operate. Development and maintenance costs can be spread over many users. Cost savings, e.g., wages, minimize loan loss, reduce customer support effort. The overall cost can be quite reasonable when compared to expensive and scarce human experts

Co

py

rig

ht

R.

We

be

r

Advantages (ii) Documentation - An expert system can provide permanent documentation of the decision process

Increased availability: the mass production of expertise

Completeness - An expert system can review all the transactions, a human expert can only review a sample; an ES solution will always be complete and deterministic

Timeliness - Fraud and/or errors can be prevented. Information is available sooner for decision making

Consistency - With expert systems similar transactions handled in the same way. Humans are influenced by recency effects and primacy effects (early information dominates the judgment).

Entry barriers - Expert systems can help a firm create entry barriers for potential competitors

Co

py

rig

ht

R.

We

be

r

Advantages (iii)

Computer programs are best in those situations where there is a structure that is noted as previously existing or can be elicited

Reduced danger: ES can be used in any environment

Reliability: ES will keep working properly regardless of of external conditions that may cause stress to humans

Explanation: ES can trace back their reasoning providing justification, increasing the confidence that the correct decision was made

Domain analysis: Indirect advantage is that the development of an ES requires that knowledge and processes are verified for correctness, completeness, and consistency. If there is a maze of rules (e.g. tax and auditing or laws), then the expert system can "unravel" the maze

Maintenance: only knowledge base can be modified, without interference to other modules of the program

Co

py

rig

ht

R.

We

be

r

Disadvantages of Rule-based ES (i)

Common sense - In addition to a great deal of technical knowledge, human experts have common sense. To program common sense in an ES, you must acquire and represent rules.

Creativity - Human experts can respond creatively to unusual situations, expert systems cannot.

Learning - Human experts automatically adapt to changing environments; expert systems must be explicitly updated.

Complexity and interrelations of rules grow exponentially as more rules are added.

Co

py

rig

ht

R.

We

be

r

Degradation - Expert systems are not good at recognizing when no answer exists or when the problem is outside their area of expertise. So, ES may provide a solution that is not optimal like one that is optimal

High knowledge engineering requirements: In many real world domains, the amount of knowledge necessary to cover an expert problem is abundant making ES development time-consuming and complex

Knowledge acquisition bottleneck

Difficulty to deal with imprecision (I.e., incompleteness, , uncertainty, ignorance, ambiguity) poses an extra engineering requirement; treatments of imprecision also have to be represented

Disadvantages of Rule-based ES (ii)

Co

py

rig

ht

R.

We

be

r

Necessary grounds for computer understanding

Ability to represent knowledge and reason with it.

Perceive equivalences and analogies between two different representations of the same entity/situation.

Learning and reorganizing new knowledge.

From Peter Jackson (1998) Introduction to Expert systems. Addison-Wesley third edition. Chapter 2, page 27.