Expert Systems: Rules, Inference, and Applications

What is an Expert System?

An expert system (ES) is an AI program that mimics human expertise in a specific domain by using knowledge bases (facts/rules) and an inference engine to solve complex problems. Unlike general AI, ES focuses on narrow, well-defined tasks where human experts exist.

Key Components

classDiagram
    class KnowledgeBase {
        +Facts (declarative knowledge)
        +Rules (if-then conditions)
    }
    class InferenceEngine {
        +Forward Chaining (data-driven)
        +Backward Chaining (goal-driven)
        +Conflict Resolution
    }
    class UserInterface {
        +Query Input
        +Explanation Facility
    }
    KnowledgeBase --> InferenceEngine : "Uses"
    InferenceEngine --> UserInterface : "Processes"

Knowledge Representation

1. Facts (Declarative Knowledge)

• Stored as propositions (e.g., Patient(Fever, John)).
• Represented in predicate logic or frames (structured objects).

2. Rules (Procedural Knowledge)

• If-Then format:

  IF (Condition1 AND Condition2) THEN (Conclusion)
  

• Example (Medical Diagnosis):

  IF (Symptom(Fever) AND Symptom(Cough)) THEN (Disease(CommonCold))
  

3. Heuristics

• Rules of thumb derived from expert experience (e.g., "If blood pressure > 140, check for hypertension").

Inference Engines

Forward Chaining (Data-Driven)

• Starts with known facts and applies rules to derive new conclusions.
• Used in monitoring systems (e.g., alarm triggers).

Example: Fire Alarm System

1. Facts: SmokeDetected(TRUE), Temperature(High)
2. Rule: IF SmokeDetected(TRUE) AND Temperature(High) THEN Alert(Fire)
3. Conclusion: Alert(Fire) → Triggers siren.

Backward Chaining (Goal-Driven)

• Starts with a hypothesis and works backward to verify conditions.
• Used in diagnostic systems (e.g., medical tests).

Example: Medical Diagnosis

1. Goal: Disease(Diabetes)
2. Rule: IF BloodSugar(High) AND Thirst(Excessive) THEN Disease(Diabetes)
3. Check: Verify BloodSugar(High) and Thirst(Excessive) from patient data.

Types of Expert Systems

Real-World Applications in Nepal

1. eSewa (Nepal Government)

• Idea Used: Rule-Based Decision Making
• How: eSewa’s vehicle tax calculation uses predefined rules (e.g., "If vehicle age > 5 years, apply 20% surcharge").
• Example:
  • Input: Vehicle(Age=6, Type=Car)
  • Rule: IF Age > 5 AND Type=Car THEN Tax=BaseTax + 20%
  • Output: Tax=₹5,000 + 20% = ₹6,000

2. Khalti (Digital Payments)

• Idea Used: Fraud Detection (Backward Chaining)
• How: Khalti’s system checks transactions against fraud rules:
  • Rule: IF Transaction(Amount>₹50,000) AND Location(HighRisk) THEN Flag(Fraud)
  • Example: A ₹60,000 transfer from Kathmandu to Dubai triggers a manual review.

3. NTC (Telecom Billing)

• Idea Used: Forward Chaining for Billing
• How: NTC’s billing system applies rules sequentially:
  • Facts: Plan(UnlimitedData), Usage(100GB)
  • Rule: IF Plan=UnlimitedData AND Usage>80GB THEN Warn(SlowSpeed)
  • Output: SMS alert to user.

Worked Example: Loan Approval System (Bank of Kathmandu)

Scenario: A bank uses an expert system to approve loans based on customer data.

Step 1: Define Rules

Rule 1: IF Income > ₹50,000 AND CreditScore > 700 THEN Approve(Loan)
Rule 2: IF Income ≤ ₹50,000 AND Savings > ₹200,000 THEN Approve(Loan)
Rule 3: IF Income ≤ ₹50,000 AND Savings ≤ ₹200,000 THEN Reject(Loan)

Step 2: Input Data

• Customer A: Income=₹45,000, CreditScore=650, Savings=₹250,000
• Customer B: Income=₹60,000, CreditScore=680, Savings=₹100,000

Step 3: Apply Inference

Visualization of Rule Application:

flowchart TD
    A["Customer A: Income=₹45K"] -->|"Savings=₹250K"| B["Rule 2"]
    B -->|"Savings > ₹200K"| C["Approve Loan"]
    D["Customer B: Income=₹60K"] -->|"CreditScore=680"| E["Rule 1"]
    E -->|"Income > ₹50K"| C

Advantages and Limitations

✅ Advantages

• Consistency: Rules ensure uniform decisions (no human bias).
• Explainability: Systems can justify conclusions (e.g., "Loan rejected because savings were low").
• Speed: Faster than human experts for repetitive tasks.
• Scalability: Handles large volumes (e.g., NTC billing millions of users).

❌ Limitations

• Brittleness: Fails outside predefined rules (e.g., a new disease symptom).
• Knowledge Acquisition Bottleneck: Requires experts to encode rules.
• No Common Sense: Cannot handle ambiguous or novel situations.

Expert Systems vs. Traditional AI

Exam Tip

1. Define Clearly: Always start with the components (knowledge base + inference engine) and types of expert systems.
2. Show Work: For numerical examples (e.g., loan approval), list rules step-by-step and mark the applied rule.
3. Compare: Questions often ask to contrast expert systems with ML/AI. Use the table above.
4. Real-World Tie: Link examples to Nepali contexts (e.g., eSewa rules, Khalti fraud detection).
5. Diagrams: Draw rule-chaining flows or system architectures to visualize inference steps.