YOU ARE DOWNLOADING DOCUMENT

Please tick the box to continue:

Transcript
Page 1: Lecture5 Expert Systems And Artificial Intelligence

1

Expert Systems and Artificial Intelligence

Modified from Marakas 2003

Reading: Chapter 7“the expert is the one who has made the most mistakes in the domain of interest”

Page 2: Lecture5 Expert Systems And Artificial Intelligence

2

The Concept of Expertise

Expertise: extensive knowledge in a narrow field– Gained by making mistakes– Arrive at solutions through logic– Logic establishes the True/Falsity of assertions– IF-THEN to arrive at conclusions– Derive new information from existing to arrive at

conclusions– Deductions and inference are used to establish

how well facts “fit” a scenario– Experts use this “information fit” to arrive at their

decisions

Page 3: Lecture5 Expert Systems And Artificial Intelligence

3

Example of “information fit”

Known information– John is Sam’s son– John is the eldest child– Mary is Sam’s daughter– John and Mary’s mother is called Anna– Sam has been married to Anna for 50 years

Derived information– If John is Sam’s son, THEN John must be a boy– If Sam and Anna have been married for 50 years,

THEN John and Mary are their children by either birth or adoption

Page 4: Lecture5 Expert Systems And Artificial Intelligence

4

Expert Sytems & Artificial Intelligence

Expert systems: a computer application that employs a set of rules based on human knowledge to solve problems that require human expertise– Imitates reasoning of experts on “information fit”– A non-expert simulates a dialog with an expert to solve

complex problems Artificial Intelligence: practical mechanisms that

enable computers to simulate the human reasoning process– Interface of compute science/cognitive psychology– “the study of how to make computers do things which

humans do better” – the Turing Test?

Page 5: Lecture5 Expert Systems And Artificial Intelligence

5

The Intelligence of Artificial Intelligence

How do people reason? (the realm of cog. psych.)

– Categorization Can reason at levels of abstraction Rules derived from relationships

among categories

– Specific Rules– Heuristics

Less rigorous rules of thumb

– Past Experience Use a similar scenario to model new

one (benefit of 20/20 hindsight)

– Expectations Frequency, pattern recognition

So can computers simulate these processes of reasoning?

car bike

personal

Motor bike

commercial

Land transport

bus taxi

Page 6: Lecture5 Expert Systems And Artificial Intelligence

6

How Do Computers Reason?

Rule-based reasoning: IF-THEN statements represent knowledge encoded as rules– If TRUE rule is “instantiated”, otherwise ignored

Pattern recognition: detecting sounds, shapes or long sequences– Analogous to “Expectation”, similar conditions

Frames: Object-oriented approach of creating hierarchical data structures analogous to “categorisation” – databasey!!

Case-based reasoning: adapting previous solutions to a current problem

Page 7: Lecture5 Expert Systems And Artificial Intelligence

7

Types of IF……… Then

Inferential– If premise THEN conclusion– If snowing THEN drive with caution

Procedural– If situation THEN action– If Average grade =A THEN award 1st class degree

Declarative– If Antecedent THEN consequent– If student has mitigating circumstances THEN award

‘incomplete’ grade

Page 8: Lecture5 Expert Systems And Artificial Intelligence

8

The CBR Cycle

N ew p ro b lem

A M e m oryK now le dge

B a s e

N ew p ro b lem

R etr iev e f r o mm em o r y

R etr iev a l

S u g g es ts o lu tio n

C o n f ir m s o lu tio n

R eu s e

R ev is e

N ew s o lu t io n

R eta in

Page 9: Lecture5 Expert Systems And Artificial Intelligence

9

Other Forms of AI

Machine learning – neural networks and genetic algorithms (“learning” mechanisms)

Automatic programming – mechanisms that generate a program to do a specific task (allows non-programmers to “program”)– User describes inputs and generates a program

Artificial life – attempts to recreate biological phenomena within computer-based systems– As opposed to dissecting frogs!– Transfer to design of engineering projects

(software, spacecraft, robotics etc)

Page 10: Lecture5 Expert Systems And Artificial Intelligence

10

Neural Network

Train it? – supervisedOr just let it get on with it? - unsupervised

Page 11: Lecture5 Expert Systems And Artificial Intelligence

11

Genetic Algorithm

mutationcrossover

Used in scheduling (timetabling?), design, marketing

Page 12: Lecture5 Expert Systems And Artificial Intelligence

12

The Concept and Structure of Expert Systems

Basic structure of an ES follows the generic structure of a DSS– User interface, Knowledge base, inference engine

The knowledge base is specific to a particular problem domain associated with the ES

The main difference between an ES and DSS is that the ES contains knowledge acquired from experts in the application domain

Page 13: Lecture5 Expert Systems And Artificial Intelligence

13

Common Expert System Architecture

UserKnowledge Engineer

UserInterface

Inference Engine

Knowledge Base

User Environment

KE

Tool Kit

KEInterface

Development Environment

OrganizationSystemsInterface

Page 14: Lecture5 Expert Systems And Artificial Intelligence

14

The User Interface in an ES

Design of the UI focuses on human concerns such as ease of use, reliability and reduction of fatigue– Critical to its success– Balance with storage capacity / hardware

constraints

Design should allow for a variety of methods of interaction (input, control and query)

UI should allow for a variety of interactive mechanisms:– touch screen, keypad, light pens, voice

command, hot keys

Page 15: Lecture5 Expert Systems And Artificial Intelligence

15

The Knowledge Base – the “Brains”

Contains the domain-specific knowledge acquired from the domain experts

Can consist of object descriptions, problem-solving behaviors, constraints, heuristics and uncertainties

The success of an ES relies on the completeness and accuracy of its knowledge base– Distinguish a database (data facts) from a

knowledge base (experts’ rules, cases, etc)

Page 16: Lecture5 Expert Systems And Artificial Intelligence

16

The Inference Engine – “the brawn”!

Here, the knowledge is put to use to produce solutions

The engine is capable of performing deduction or inference based on rules or facts

Also capable of using inexact or fuzzy reasoning based on probability or pattern matching

Cycle consists of:1. Match rules with given facts

2. Select the rule that is to be executed

3. Execute the rule by adding the deduced fact to the working memory

Page 17: Lecture5 Expert Systems And Artificial Intelligence

17

Chaining

Simple methods used by most inference engines to produce a line of reasoning

Two methods are possible depending on the direction of reasoning

Forward chaining: the engine begins with the initial content of the workspace and proceeds toward a final conclusion

Backward chaining: the engine starts with a goal and finds knowledge to support that goal

Page 18: Lecture5 Expert Systems And Artificial Intelligence

18

Forward Chaining Example

Suppose we have three rules:

R1: If A and B then D

R2: If B then C

R3: If C and D then E

If facts A and B are present, we infer D from R1 and infer C from R2. With D and C inferred, we now infer E from R3.

Page 19: Lecture5 Expert Systems And Artificial Intelligence

19

Backward Chaining Example

The same three rules:

R1: If A and B then D

R2: If B then C

R3: If C and D then E

If E is known, then R3 implies C and D are true. R2 thus implies B is true (from C) and R1 implies A and B are true (from D).

Page 20: Lecture5 Expert Systems And Artificial Intelligence

20

Real Chaining Example

Which method to choose? Backward or forward? Situation

– You wish to fly from Warsaw to Glasgow but all direct flights are full

Backward chain– Check which cities flights arriving in Glasgow are from

and keep going backwards

Forward chaining– Check destinations from Warsaw and then check their

destinations until you find a flight going to Glasgow

Page 21: Lecture5 Expert Systems And Artificial Intelligence

21

Expert Systems Shells

Expert System Shells: generic systems that contain reasoning mechanisms but not the problem-specific knowledge

Early shells were cumbersome but still allowed the user to avoid having to completely program the system from scratch

Modern shells contain two primary modules: a rule set builder (to construct the initial knowledge base) and an inference engine (as the vehicle for arriving at conclusions)

Page 22: Lecture5 Expert Systems And Artificial Intelligence

22

Building an Expert System

An early step is to identify the type of tasks – Interpretation, predicting likely consequences– Diagnosis of faults, monitoring expected outcomes– User guidance

Key awareness is to recognise opportunities– E.g. to replace scarce expert resources– Provide an expert assistant– To free up experts onto other tasks– Look for activities where experts are:

Overburdened Undersupplied Expensive Unavailable

Candidates– Diagnosis, prediction, control, evaluation, prescription of solution

Page 23: Lecture5 Expert Systems And Artificial Intelligence

23

Pre-design Activities

Another important step is choosing the experts who will contribute knowledge– Involved throughout the development– Not be threatened, rather enhanced

Development team should have a functional understanding of the knowledge domain

Unlike more general information systems design projects, the software tools and hardware platform are selected very early

Page 24: Lecture5 Expert Systems And Artificial Intelligence

24

Benefits of Expert Systems

Increased timeliness in decision making• ES available on a 24 hour service

Increased productivity of experts• By taking some of the lesser important decisions• Replication at many sites

Improved consistency in decisions• Eliminates cognitive bias/limitation, memory loss

Improved understanding and explanation Improved management of uncertainty

• Fuzzy logic – e.g. how tall is tall? Formalization of knowledge within organisation

• And little amnesia! – knowledge is codified.

Page 25: Lecture5 Expert Systems And Artificial Intelligence

25

Limitations Associated With ES’s

How do you code common sense? expertise is difficult to extract and encode. Expert “errors” transferred to model Another is that human experts adapt

naturally but an ES must be recoded. human experts better recognize when a

problem is outside the knowledge domain, but an ES may just keep working

ES’s can’t eliminate the cognitive limitations of the user

An ES is functional only in a narrow domain

Page 26: Lecture5 Expert Systems And Artificial Intelligence

26

Key Point Summary

The Concept of Expertise Use of logic Methods of human reasoning

– Categorisation etc

AI methods of mimicking human reasoning– Case-based reasoning, pattern recognition etc

ES architecture– User interface, Knowledge base, inferencing

Designing and building ES’s Benefits and problems with ES’s

Page 27: Lecture5 Expert Systems And Artificial Intelligence

27

Area of Research for Coursework

Case Based Reasoning– http://en.wikipedia.org/wiki/Case_based_reasoning

Neural Networks– http://www.cs.stir.ac.uk/~lss/NNIntro/InvSlides.html

Genetic Algorithms (http://cs.felk.cvut.cz/~xobitko/ga) Rule-based systems

– http://www.cee.hw.ac.uk/~alison/ai3notes/section2_4_4.html Pattern recognition

– http://cgm.cs.mcgill.ca/~godfried/teaching/pr-web.html Artificial life

– http://www.wisegeek.com/what-is-artificial-life.htm?referrer=adwords_campaign=artificiallife_ad=016171&_search_kw=what%20is%20artificial%20life

Automatic programming– http://www.cs.utexas.edu/users/novak/autop.html

Fuzzy Logic– http://www.seattlerobotics.org/encoder/mar98/fuz/flindex.html

Data Mining – Adriaans and Zantinge (library) Google found these with “What is …….?” queries You might find better ones to get you going


Related Documents