Expert Systems

• What is an expert system?

  An expert system is a system that perform a task to a level which can only be achieved by human experts.

  • An expert is commonly accepted to be someone who has specialist knowledge in a domain, in the form of facts and rules, coupled with, most importantly, personal experience.

    Experience is composed of

    • heuristics (rules of thumb)

    • associative experience

    • decision making quality based on personal past experience.

    • Experimental experience.

    And is usually

    • vague

    • difficult to express

Expert systems are computer programs capturing both the factual and the experiental knowledge, and where a problem solving engine (inference engine) is also implemented.

Most expert systems are rule-based systems, consisting of 4 parts:

1. Knowledge Base

   • domain specific knowledge in the form of rules

   • capture from expert by knowledge engineer

2. Inference Engine

   • How deduction is achieved?

     • Forward chaining

     • Backward chaining

     • mixed deduction

   • How heuristics are being applied?

3. Working memory (WM)

   • contain actual facts (database)

   • new facts generated by the inference engine

4. User-interface

   • provide a natural language interface to the user.

   • Ask user questions when necessary

   • Answer WHY and HOW.

An Expert System must have

• expertise

  • exhibit expert performance

  • have high level of skill

  • have adequate robustness

• symbolic reasoning

  • represent knowledge symbolically

  • reformulate symbolic knowledge

• depth

  • handle difficult problem domains

  • use complex rules

• self knowledge

  • examine its own reasoning

  • explain its operation

Why build an expert system?

1. Replacement of Expert:

   • make available expertise after office hours or in other places.

   • Automate a routine task requiring an expert

   • Expert is retiring or leaving

   • Expert is expensive

   • Expert is needed in a hostile environment.

2. Assisting an expert

   • aiding expert in some routine task to improve productivity

   • aiding expert in some difficult task to effectively manage the complexities

   • making available to expert information that is difficult to recall.

Example: 1. cook advisor
2. System configuration: XCON/R1

Expert Systems Problem Solving Paradigm

1. Control

   • Governing system behaviour to meet specifications

     e.g. controlling a manufacturing process
     treatment of patient in hospital

   • need also to perform monitoring and interpretation task to track system behaviour over time.

     Example: VM system û monitoring patient in ICU.

2. Design

   • configuring objects under constraint

     e.g. designing a computer system under user-defined constraints of needed memory, speed etc.

   • system usually perform the task following a series of steps, each with its own specific constraints. These steps will be dependent on other steps making things more complicated.

   Example: PEACE û assists engineers to design electronic circuits.

3. Diagnosis

   • infer system malfunctions from observables

   • knowledge of possible fault condition with means to infer whether the fault exists from information on the system observable behaviour.

   Example: NEAT û assist non-technical staff at a help desk troubleshooting data processing and telecommunication network equipment.

4. Instruction

   • Diagnosing, debugging and repairing student behaviour

   • interact with the student to form a model of the student's understanding on the topic

   • compare this model with an 'ideal' model to find weakness in student's understanding.

   Example: GUIDON û instruct medicine students.
   
   
   
   

5. Interpretation

   • inferring situation description from data

   • translate raw data (e.g. data from sensors, instruments etc) into symbolic form that describe the situation

   example: FXAA û auditing assistance in foreign exchange trading.

6. Monitoring

   • compare observations to expectations

   • when a crucial state is detected, an established sequence of task is performed.

   Example: NAVEX û monitoring radar data and estimates the velocity and position of the space shuttle.

7. Planning

   • Designing actions to achieve a certain goal.

     E.g. planning the different task performed by a robot to accomplish a given work function

   • Sometimes need to backtrack and reject a current line of reasoning in favour of exploring a better one.

   Example: PLANPOWER û financial planning for household

8. Prediction

   • Inferring likely consequences of given situations

   • must be able to reason about time or ordered events

   example: PLANTû predicting the expected damage to a crop from an invading insect.

9. Prescription

   • recommend solution to a given system malfunction

   • may require planning and prediction techniques for tailored remedy, rather than 'canned' prescription.

   Example: BLUEBOX û for depression therapy.

   

10. Selection

    • identify the best choice from a list of possibilities

    • usually employ an inexact reasoning technique or a matching evaluation function

    example: IREX û assists in the selection of industrial robots in a working environment.

11. Simulation

    • model a process or system to permit operational studies under various conditions

    • able to predict operating conditions for the real systems

    example: STEAMER û simulates and explains the operation of the Navy's 1078-class frigate steam propulsion plant.

Characteristics of an Expert System

1. Separate Knowledge from Control

   • task of modifying and maintaining system are easier.

   • Expert system shells can be constructed.

2. Process expert knowledge

3. focus expertise on a narrow area

4. reasoning with symbols

   • e.g. Jack has a fever
     People with fever should take a couple of aspirin

5. Reasoning heuristically

   • rule of thumb
     e.g. I always check whether the power is connected first

6. Inexact reasoning

7. limited to solvable problems.
   
   
   

Expert System vs Conventional Programming

Process of Building an expert system

1. Assessment û feasibility study and justification

   • specifies the important features and scope of the project.

   • Establish the needed resources (including human resources)

   • identify needed knowledge (experts and various reports)

2. Knowledge Acquisition û acquiring, organizing and studying knowledge.

   • Meeting with experts

   • uncover key concepts and general problem-solving methods used by experts

   • bottleneck in expert system development.

3. Design

   • find approaches for representing the expert's knowledge and problem solving strategies.

   • Overall structure and organization of the system's knowledge are defined.

   • Initial prototype built.

4. Testing

   • a continual process throughout the project

5. Documentation

   • include a knowledge dictionary û well organized presentation of the system's knowledge and problem solving procedure.

6. Maintenance.

Qualification Needed by People

1. Domain Expert

   • Has expert knowledge

   • Has efficient problem-solving skills

   • Can communicate the knowledge

   • Can devote time

   • Is not hostile

2. Knowledge Engineer

   • Has knowledge engineering skills

   • Has good communication skills

   • Can match problem to software

   • Has expert system programming skills

3. End-User

   • Can help define interface specification

   • Can aid in knowledge acquisition

   • Can aid in system development.

Expert System Shell

• Tools for building expert systems

• Usually contain an inference engine, a very good user interface and schemes of knowledge representation.

• Most tools are rule-based

• Rapid system development by providing a substantial amount of computer code that would otherwise need to be written, tested, debugged and maintained.

• Provide specific techniques for handling knowledge representation, inference and control that help (or limit) the knowledge engineer to model the salient characteristics of a particular class of problems.

Examples:

EMYCIN û (Empty MYCIN) taking out the rules of MYCIN, leaving the inference engine

KEE û Knowledge Engineering Environment

PC+ û Personal Consultant Plus

M.1

Factors considered

1. Knowledge Representation

   • Rules

   • Frames

   • A-V pairs (attribute-values pairs)

2. Control Strategies in Reasoning

   • Backward Chaining

   • Forward Chaining

3. Reasoning with Uncertainty

   • Fuzzy

   • Probabilistic

   • Certainty Factors (used in MYCIN)/Dempster Shafer Theory.