Expert Systems and Artificial Intelligence

1
Expert SystemsandArtificial Intelligence
2
The Concept of Categorization
3
Semantic Network
4
Frame Hierarchy
5
Nominal Process in Case-Based Reasoning
6
Common ES Architecture
7
Issues Associated with Better ES Interface Design

• Users should be equal or dominant partners in the
  design of the UI.
• ES dialogues should be flexible enough to allow
  users to volunteer information and allow for
  smooth changes of initiative.
• Explanation facilities must be of high quality
  and detail.
• A natural language interface may not be
  appropriate for an expert system.
• ES dialogues should emphasize graphics over text
  wherever appropriate.

8
Interaction Mechanisms in Expert System Interfaces

• Touch screen
• User-assignable hot keys
• Function keys
• Mouse
• Trackball
• Touch pad
• Light pen
• Voice-activated commands
• Cursor control
• Menuing
• Image mapping

9
Backward Chaining

• Situation You wish to fly from Washington, D.C.
  to San Diego. Unfortunately, all direct flights
  are booked, so will have to accept a connecting
  flight if you are to get to San Diego.
• Backward Chaining You could check the flights
  arriving in San Diego and see which cities they
  are arriving from. Then you could look up the
  flights arriving in those cities and so forth
  until you find Washington, D.C.

10
Forward Chaining

• Situation You wish to fly from Washington, D.C.
  to San Diego. Unfortunately, all direct flights
  are booked, so will have to accept a connecting
  flight if you are to get to San Diego.
• Forward Chaining You could check the flights
  departing Washington D.C. and find their
  destination cities. Then look up the flights
  leaving those cities and so forth until you find
  a flight to San Diego.

11
A Typology of Expert System Tasks

• Interpretation Using sensor data to infer
  situation descriptions and meaning
• Prediction Forecasting from past and present the
  likely consequences of given situations
• Diagnosis Inferring faults and malfunctions
  based on observation and interpretation of data
• Planning Design actions and plans to achieve
  stated goals
• Design Configuring object specifications for
  satisfying particular requirements
• Prescription Prescribing remedies and solutions
  for malfunctions
• Monitoring Comparing signals and observations to
  expected outcomes
• Control Governance of overall system behavior
• Instruction Diagnosis, prescription, and
  guidance of user behavior

12
Candidate Situations for ES Opportunities

• Need for diagnosis of a problem situation or
  variance (audit, troubleshooting, etc.)
• Need to understand the nature of a given
  situation
• Need to predict the outcome of a current or
  future event
• Need to control or govern a particular activity
  or process
• Need to prescribe a solution or course of action
• Need to evaluate and assess a prior event or
  process

13
Criteria for Evaluation of an Expert System

• The system should be responsive and easy to use.
• The design and functionality should conform with
  current diagnostic standards.
• The system should be able to function with
  incomplete knowledge.
• The user should always control the consulting
  process.
• The explanatory facility should be clear and
  user-friendly.
• The knowledge base should contain generally
  accepted knowledge.
• The system should be independent of any
  geographical constraints.
• The system should allow for integration with all
  other current information systems.

14
Limitations Associated with Expert Systems

• The needed knowledge is not always available.
• Experts use common sense. Programming common
  sense is not yet a reality.
• Expertise is difficult to extract and encode.
• Experts can recognize a problem outside the
  knowledge domain much faster than an ES.
• Expert systems cannot eliminate the cognitive
  limitations of the user.
• An ES is functional only within a narrow
  knowledge domain.
• Expert vocabulary is often limited and not easily
  understood by others.
• Human experts adapt to their environments
  naturally while an ES must be explicitly updated.
• An ES has a limited sensory experience compared
  to human experts.

15
Limitations Associated with Common ES Task
Categories

• Task Problems
• Interpretation Data may be noisy or missing
• Data may be inaccurate
• Prediction Must allow for contingencies and
  uncertainties
• Diagnosis Multiple symptoms can confound
  diagnosis
• Planning Multiple alternatives with complex
  scenarios
• Design Conflicting constraints and
  interaction among sub-designs
• Prescription Multiple problems may exist
• Monitoring Error conditions and nominal
  expectations are often context- specifi
  c
• Control Often requires interpretation based
  on common sense