EXPERT SYSTEM AND MACHINE LEARNING

(COM423)
LECTURE NOTES
Explain the concept, history and application of expert systems

1.0 Definition of Expert Systems

An expert system (ES) is a type of artificial intelligence (AI) that emulates the decision-
making abilities of a human expert (e.g. Medical Doctors, Lawyers etc.). It uses knowledge
and inference rules to solve complex problems that usually require human expertise. Expert
systems are designed to provide solutions, explanations, and advice in specific domains such
as medical diagnosis, engineering, finance, and more.

Expert System Representation:

Problem input Problem output

User Interface

Working memory Interface Engine

Knowledge from Expert

Knowledge Base

Figure 1: Architecture of an Expert System

1.1 History of Expert Systems

a) 1960s: The concept of expert systems began to take shape during the early days of AI
research. The first expert system, Dendral, was developed at Stanford University to help
chemists identify unknown organic molecules.
b) 1970s: MYCIN, another notable expert system, was created to diagnose bacterial
infections and recommend antibiotics. Its success demonstrated the potential of expert
systems in medical diagnosis.
c) 1980s: The commercial application of expert systems grew rapidly. Industries began
adopting them for various purposes, such as troubleshooting, decision support, and
automation of routine tasks. Tools like the R1/XCON system by Digital Equipment
Corporation were used for configuring computer systems.
d) 1990s: Expert systems faced challenges due to the limitations in knowledge
representation, maintenance, and scalability. The rise of machine learning and neural
networks started to overshadow traditional rule-based systems.
e) 2000s and Beyond: Expert systems are still used in specific applications but are often
integrated with other AI technologies to enhance their capabilities. Modern AI
developments, such as deep learning and natural language processing, have broadened
the scope and efficiency of intelligent systems.
Examples of ES and their functions

Here are ten examples of expert systems, showcasing their application across various domains:

1. MYCIN:
o Domain: Medical Diagnosis
o Function: Assists in diagnosing bacterial infections and recommending
appropriate antibiotics based on patient data.
2. Dendral:
o Domain: Chemistry
o Function: Helps chemists identify the structure of organic molecules using mass
spectrometry data.
3. R1/XCON:
o Domain: Computer Configuration
o Function: Configures VAX computer systems by determining the optimal
hardware and software configurations.
4. CADUCEUS:
o Domain: Medical Diagnosis
o Function: Provides medical diagnostic support by analyzing symptoms and
medical history to suggest possible diseases.
5. CLIPS (C Language Integrated Production System):
o Domain: General Purpose
o Function: A tool for building expert systems that can be applied in various fields
like manufacturing, aerospace, and defense.
6. Prospector:
o Domain: Geology
o Function: Assists geologists in identifying potential locations for mineral and ore
deposits.
7. DART (Dynamic Analysis and Replanning Tool):
o Domain: Military Logistics
o Function: Used by the U.S. military for planning and optimizing logistics
operations, such as transportation and supply chain management.
8. SMARTS (System for Managing Automated Reasoning and Theorem-proving
Systems):
o Domain: Telecommunications
o Function: Manages and diagnoses network issues, optimizing network
performance and reliability.
9. PIE (Portable Intelligent Executive):
o Domain: Space Exploration
o Function: Used by NASA to support decision-making in space missions,
including the management of spacecraft and rover operations.
10. TEIRESIAS:
o Domain: Rule-based Systems
o Function: An expert system used for validating and refining the rules in other
expert systems, ensuring their accuracy and reliability.
Characteristics of Expert Systems in AI

1. High performance

The first and foremost characteristic of an expert system is to deliver high performance 24×7

2. Understandable

The expert system should be easy to comprehend for all the people using it.

3. Reliable

An expert system must be reliable in the sense that it is error-free so that it is trustable.

4. Highly Responsive

An expert system must be proactive and provide responses for each and every detail of the
problem.

Components of Expert Systems

An expert system comprises several key components that work together to mimic the decision-
making abilities of a human expert. These components include:

1. User Interface

The user interface allows users to interact with the expert system. It enables users to input data,
ask questions, and receive explanations or solutions. A user-friendly interface is crucial for the
effective use of an expert system, especially for users who may not be experts in the domain.

2. Inference Engine

The inference engine is the reasoning component of an expert system. It applies logical rules to
the knowledge base to draw conclusions or solve problems. The inference engine can operate in
two main modes:
Forward Chaining:

Forward chaining is a method of reasoning in expert systems that starts with the known facts and
applies inference rules to extract more data or reach a conclusion.

This process is typically used for several types of prediction, including:

A. Diagnosis Prediction:
o Example: Medical expert systems predict possible diseases based on the observed
symptoms and patient data.
B. System Configuration Prediction:
o Example: In computer systems, expert systems predict the optimal configuration
of hardware and software based on the current requirements and constraints.
C. Process Control Prediction:
o Example: In manufacturing, expert systems predict the adjustments needed in the
production process to maintain quality and efficiency.
D. Fault Detection and Troubleshooting Prediction:
o Example: Expert systems predict the possible causes of machinery or system
failures and suggest troubleshooting steps.
E. Financial Forecasting Prediction:
o Example: Financial expert systems predict market trends, stock prices, and
investment opportunities based on current and historical data.
F. Resource Management Prediction:
o Example: In project management, expert systems predict resource allocation and
scheduling needs to ensure timely project completion.
G. Agricultural Yield Prediction:
o Example: Expert systems predict crop yields based on soil conditions, weather
patterns, and farming practices.
H. Environmental Impact Prediction:
o Example: Expert systems predict the environmental impact of industrial
activities, construction projects, or policy changes.
I. Supply Chain Optimization Prediction:
o Example: Expert systems predict supply chain demands, inventory levels, and
logistics needs to optimize operations.
J. Customer Behavior Prediction:
 o Example: In marketing, expert systems predict customer preferences and
purchasing behavior based on demographic and transactional data.

Backward Chaining:

Backward chaining is a reasoning method used in expert systems that works backward from the
goal or conclusion to determine the conditions or facts that support it. This approach is
particularly useful for certain types of predictions and problem-solving scenarios.

Here are the main types of predictions done by backward chaining:

A. Diagnostic Predictions

Example: Medical diagnosis systems

• Goal: Diagnose a disease based on observed symptoms.

• Process: Start with a potential diagnosis (e.g., a specific disease) and work backward to check if
the observed symptoms (facts) match the conditions (rules) associated with that disease.

B. Causal Predictions

Example: Fault diagnosis in engineering systems

• Goal: Identify the root cause of a problem or failure.

• Process: Begin with the observed problem or malfunction and trace back through the system to
find potential causes based on established rules and dependencies.

C. Goal-oriented Predictions

Example: Planning and decision-making systems

• Goal: Achieve a specific objective or outcome.

• Process: Start with the desired goal and work backward to determine the necessary steps,
conditions, or prerequisites required to achieve that goal.
D. Legal and Regulatory Predictions

Example: Legal advisory systems

• Goal: Determine the applicability of laws or regulations to a specific case.

• Process: Begin with a legal outcome or decision and work backward to identify the legal
conditions, precedents, or rules that justify or support that outcome.

E. Troubleshooting and Repair Predictions

Example: Technical support systems for electronic devices

• Goal: Identify the steps needed to fix an issue or malfunction.

• Process: Start with the desired state (e.g., a fully functional device) and work backward to
identify the necessary diagnostic steps, parts, and repairs needed to achieve that state.

F. Configuration and Design Predictions

Example: Computer configuration systems (e.g., R1/XCON)

• Goal: Configure a system or product to meet specific requirements.

• Process: Begin with the desired configuration or specifications and work backward to determine
the components, settings, and arrangements needed to achieve those specifications.

G. Risk Assessment and Mitigation Predictions

Example: Financial risk assessment systems

• Goal: Assess and mitigate risks in investment portfolios or projects.

• Process: Start with potential risk outcomes (e.g., financial loss) and work backward to identify
the factors and conditions that could lead to those risks, along with mitigation strategies.

H. Behavioral Predictions

Example: Customer behavior analysis systems

• Goal: Predict customer actions or preferences.

• Process: Begin with the desired customer behavior (e.g., purchase of a product) and work
backward to identify the factors, incentives, and conditions that could influence that behavior.

I. Security and Intrusion Detection Predictions

Example: Cybersecurity systems

• Goal: Detect and prevent security breaches or intrusions.

 • Process: Start with potential security breaches and work backward to identify the vulnerabilities,
attack vectors, and conditions that could lead to those breaches.

J. Compliance and Audit Predictions

Example: Financial audit systems

• Goal: Ensure compliance with regulations and standards.

• Process: Begin with the desired compliance state (e.g., regulatory compliance) and work
backward to determine the necessary documentation, processes, and controls needed to achieve
and maintain that state.

3. Knowledge Base

The knowledge base is the core component of an expert system, containing domain-specific
knowledge. This knowledge is represented in the form of:

• Facts: Basic units of information about the domain, such as specific data points or conditions.
• Rules: Conditional statements that link facts together to derive new information or make
decisions. Rules are typically in the form of "IF-THEN" statements.

Types of Knowledge

i) Declarative Knowledge:
o Facts and information about objects, events, and their properties.
o Example: "Paris is the capital of France."
ii) Procedural Knowledge:
o Instructions or rules for performing tasks or solving problems.
o Example: Steps to solve a quadratic equation.
iii) Meta-Knowledge:
o Knowledge about knowledge, including strategies for acquiring, organizing, and
using knowledge.
o Example: Knowing which problem-solving method to apply in each situation.
iv) Heuristic Knowledge:
o Rules of thumb or educated guesses based on experience.
o Example: "If a plant's leaves are wilting, it probably needs water."

4. Explanation Facility

The explanation facility provides users with insights into how the expert system arrived at a
particular conclusion or recommendation. This component enhances the transparency and
trustworthiness of the system by explaining the reasoning process and the rules applied.
5. Knowledge Acquisition Module

The knowledge acquisition module is responsible for gathering and updating the knowledge
base. This process involves:

• Manual Input: Domain experts input their knowledge directly into the system.
• Automated Tools: Software tools and algorithms that extract knowledge from databases,
documents, and other sources.
• Learning Capabilities: Some expert systems incorporate machine learning techniques to
automatically update and refine the knowledge base based on new data.

6. Knowledge Representation

Knowledge representation involves the methods used to encode the knowledge in the knowledge
base. Common representation techniques include:

• Production Rules: "IF-THEN" rules that define the relationships between facts and actions.
• Frames: Data structures that represent stereotyped situations, organizing knowledge into
hierarchies of related concepts.
• Semantic Networks: Graph-based structures that depict relationships between concepts.

7. Working Memory

Working memory is a temporary storage area where the current state of the problem-solving
process is maintained. It holds the facts and intermediate results generated during the inference
process, allowing the system to keep track of progress and make decisions based on the most
recent information.

8. Agenda

The agenda is a control structure that manages the sequence in which rules are applied by the
inference engine. It prioritizes and schedules the execution of rules based on their relevance and
the current state of the working memory.

9. Knowledge Refinement

Knowledge refinement involves the continuous improvement and updating of the knowledge
base to ensure the expert system remains accurate and relevant. This process may involve:

• Reviewing and Updating Rules: Regularly checking and modifying rules to reflect new
knowledge and changing conditions.
• Incorporating User Feedback: Using feedback from users to identify and correct inaccuracies or
gaps in the knowledge base.
10. Integration Interface

The integration interface allows the expert system to interact with other software systems,
databases, and external data sources. This capability is essential for accessing real-time data and
integrating the expert system into larger information systems and workflows.

Applications of Expert Systems

• Expert systems are being used in designing and manufacturing domain for the production
of vehicles and gadgets like cameras.
• In the knowledge domain, Expert Systems are used for delivering the required knowledge
to the client. The knowledge can be legal advice, tax advice, or something other than that.
• In the banking and finance sector, expert systems are widely used for the detection of
frauds.
• Expert Systems can also use in the diagnosis and troubleshooting of medical equipment.
• Apart from this, Expert Systems can also have use cases in Planning and Scheduling
tasks.

Advantages of Expert Systems

• Expert Systems are easily available as they are not so difficult to develop and are thus
easier to reproduce.
• Increased accuracy is a prominent advantage of Expert Systems.
• They can be of use at workstations where there is a risk to human lives.
• They can be made to work 24×7 without the need for any human intervention.
• Expert Systems offer a very speedy decision-making process which in most cases is
error-free.

Limitations of Expert Systems

Expert systems, which are a type of artificial intelligence designed to emulate the decision-
making abilities of a human expert, have several limitations:

1. Knowledge Acquisition Bottleneck: Capturing and formalizing the knowledge of

human experts into a format usable by an expert system can be time-consuming and
challenging. This process, known as knowledge engineering, often requires extensive
interviews and observations of experts.
2. Limited Scope: Expert systems are usually designed for very specific domains. They
struggle to generalize or adapt to new problems outside their predefined knowledge base.
3. Lack of Common Sense: These systems lack the ability to apply common sense or
general world knowledge, making them less effective in situations requiring broader
understanding or context.
4. Rigidity: Once built, expert systems can be difficult to update or modify. Changes in the
domain knowledge or new discoveries may necessitate significant overhauls of the
system.
5. Maintenance and Scalability Issues: Keeping the knowledge base up to date can be
resource intensive. As the system grows, maintaining and ensuring the accuracy of its
knowledge base becomes more complex.
6. Inference Engine Limitations: The inference mechanisms (rules and algorithms) used to
draw conclusions can be computationally intensive, leading to slow performance,
especially with large knowledge bases.
7. User Interface and Usability: Expert systems often have user interfaces that are not
user-friendly, making them difficult for non-experts to use effectively.
8. Uncertainty Handling: Many expert systems struggle to handle uncertainty or
incomplete information. They often rely on probabilistic or heuristic approaches, which
may not always provide reliable results.
9. Bias and Subjectivity: The knowledge encoded in an expert system reflects the biases
and perspectives of the human experts who contributed to its development, which can
limit the system's objectivity and accuracy.
10. Lack of Learning: Traditional expert systems do not learn from new data or experiences.
They need manual updates to incorporate new knowledge, unlike modern machine
learning systems that can improve over time with additional data.