ES Development Stages

1. Identification
2. Conceptualization
3. Formalization
4. Implementation
5. Testing

Identification

• requirements analysis
  • external requirements
  • form of input and output
  • setting where ES will be used
  • user
• must identify
  • participants
  • problems
  • objectives
  • resources
  • costs
  • time frame

Conceptualization

• designing the ES to ensure that specific interactions and relationships in the problem domain are understood and defined
• initial knowledge acquisition
• key concepts, relationships between objects and processes, and control mechanisms are determined
• understanding what expert does:
  • Exactly what decisions do the experts make?
  • What are the decision outcomes?
  • Which outcomes require greater reflection, exploration, or interaction?
  • What resources or inputs are required to reach a decision?
  • What conditions are present when a particular outcome is decided?
  • At what point after exposure to influential inputs is a decision made?
  • How consistently do these conditions predict a given outcome?
  • Given the particulars of a specific case, will the outcome predictions of the knowledge engineering team be consistent with those of the expert?

Formalization

• organizing the key concepts, subproblems, and information flow into formal representations
• design program logic

Implementation

• formalized knowledge is coded into the framework of the development tool or language
• document
• help for the user and explanations
• how will the ES interact with other programs and databases

Testing

• verification of individual relationships
• validation of program performance
• evaluation of utility of the software

Knowledge Acquisition

• major bottleneck to building ES
• 3 areas central to knowledge acquisition
  1. evaluation of domain to determine if the knowledge in the domain is suitable for an ES
  2. source of expertise must be identified and evaluated
  3. if expertise source is a person, the specific knowledge acquisition techniques and practitioners need to be identified

Domain Selection

• must have an expert
• must be general consensus among experts about the accuracy of solutions
• experts must be able to communicate details
• domain should be narrow and well-defined and solutions must not require common sense

Knowledge Acquisition Techniques

• observe expert solving real problems
• identify kinds of data, knowledge, and procedures required to solve different types of problems
• use scenarios with the expert
• have expert solve a series of problems verbally
• develop rules based on interviews
• have expert review rules and general problem solving procedure
• compare responses of outside experts to those of expert and ES
• regularly scheduled interviews
• questionnaires
• decision trees

Representation and Reasoning

Rule-Based Systems

• pattern matching
• parameter driven
• forward chaining
• backward chaining

Pattern Matching in a Forward Chaining System

IF condition
THEN consequent (or action).


IF the volume of a ?product-container is less than a specified
amount,
THEN increase the fill volume.


IF the ?sensor is dead,
THEN inspect ?sensor.

Forward Chaining Example

Facts:
     the volume of item29 is 21
     the volume of item17 is 18
     the temperature of item23 is -5

Rule:
       IF the volume of ?item is < 20,
       THEN the fill amount of ?item should be increased

Parameter Driven Systems

• many parameter driven systems based on MYCIN
• facts are represented by parameter values
• parameters can have various attributes called properties

Decision Trees

• information found at one step in the solution determines what information is needed at the next step
• each item is represented as a node in a tree with possible values forming arcs that emanate from the node

Reasoning With Uncertainty

• much of the information humans use is fuzzy or inexact
• 2 kinds of uncertainty
  • among facts and relationships of the problem (the rules)
  • information supplied by the user

    IF the average number of cockleburs per foot 
    of soybean row is <0.5,
    THEN the application of herbicide to control 
    cockleburs is usually
    not economical.
    
    
    I'm almost certain the weed is Jimson but there 
    is a slight
    chance it is a cocklebur.  I know it is definitely 
    not foxtail.
    

• MYCIN CF
  • numerical
  • range from -100 to +100

Object-Based Systems

Frames

• represent declarative knowledge (answers the question What?)
• represents procedural knowledge (answers the question How?)
  stored as a functional quality of object
• provide a mechanism for inheritance

Inheritance example

Charlotte is a member of the Landshire class.
Landshire is a member of the swine class.
An attribute of swine is the existence of a curly tail.

Through inheritance, Charlotte has a curly tail.

Object-Oriented Programming

• objects represent things
• can also represent generic concepts
• have properties and behaviors
• data are stored within the programs that manipulate them

messages - requests between objects

Truth Maintenance

• new information exists independently from other information in the knowledge base
• no logical links between facts and rules (a fact has no knowledge about how it was derived, what other facts it depends on, or what other facts depend on it)
• useful for hypothetical reasoning, planning, scheduling, and design
• dependency networks, Justification-based Truth Maintenance, Assumption-based Truth Maintenance

Selection of ES Development Tools

Categories of ES Development Tools

• rule-based
• frame-based
• procedure-oriented
• logic-based
• object-oriented
• access-oriented

Categories of ES Applications

• interpretation
• prediction
• diagnosis
• planning
• monitoring
• debugging
• repair
• instruction
• control

Categories of ES Applications and Recommended Tool Category

• interpretation - rule-based (forward chaining), logic-based
• prediction - object-oriented, procedure-oriented, rule-based (forward chaining)
• diagnosis - rule-based (backward chaining)
• planning - rule-based (forward chaining), frame-based, access-oriented
• monitoring - access-oriented, rule-based (forward chaining)
• debugging - rule-based (backward chaining), logic-based, frame-based
• repair - rule-based (backward chaining), logic-based, frame-based
• instruction - rule-based (forward and backward chaining)
• control - access-oriented, object-oriented, rule-based (forward chaining)

ES Tool Features to Consider

• combinations of representation and reasoning methods
• is the tool maintained
• built-in explanation and interaction facilities
• ability to interface with data bases, spread sheets, and other software
• ability to embed in other software
• debugging aids
• uncertain reasoning capabilities
• tool and computer costs
• tool speed
• intended user of tool
• startup cost of the tool (training)

Waterman's questions to ask when selecting an ES development tool

• Does the tool provide the development team with the power and sophistication they need?
• Are the tool's support facilities adequate considering the time frame for development?
• Is the tool reliable?
• Is the tool maintained?
• Does the tool have the features suggested by the needs of the problem?
• Does the tool have the features suggested by the needs of the application?