Expert Systems & AI Applications: Rules, Shells, and Real-World Impact

What is an Expert System?

An expert system is an AI program that encodes human expertise in a specific domain (e.g., diagnosing diseases, configuring hardware) to solve complex problems. It consists of:

• Knowledge Base (KB): Facts and rules (e.g., "If fever > 38°C and cough, then likely flu").
• Inference Engine: Applies rules to derive conclusions (forward or backward chaining).
• User Interface: Lets users input data and receive explanations.

How It Works: Forward vs. Backward Chaining

flowchart LR
    A["User Input: Symptoms = {fever, cough}"] --> B["Forward Chaining"]
    B --> C["Apply all rules matching symptoms"]
    C --> D["Conclude: Likely flu"]
    E["User Input: Goal = Diagnose flu"] --> F["Backward Chaining"]
    F --> G["Check if flu rules are satisfied"]
    G --> H["Request missing data: fever temp?"]
    H --> I["Conclude: flu"]

Key Difference:

• Forward Chaining: Starts with data, fires rules to reach conclusions (data-driven).
• Backward Chaining: Starts with a goal, works backward to find supporting data (goal-driven).

Knowledge Representation in Expert Systems

1. Production Rules (If-Then)

Rules are written as: IF <conditions> THEN <action/conclusion> [CONFIDENCE: x%] Example (Medical Diagnosis):

IF (temperature > 38 AND cough) THEN (diagnosis = flu) [CONFIDENCE: 85%]
IF (temperature > 38 AND rash) THEN (diagnosis = measles) [CONFIDENCE: 90%]

Worked Example: Loan Approval (Nepal’s Nabil Bank) Assume rules:

1. IF (income ≥ 50,000 AND credit_score ≥ 700) THEN (approve_loan = YES)
2. IF (income < 50,000 AND savings ≥ 200,000) THEN (approve_loan = YES)
3. ELSE (approve_loan = NO)

Input: Income = 45,000, credit_score = 720, savings = 250,000. Trace:

• Rule 1 fails (income < 50,000).
• Rule 2 fires (savings ≥ 200,000) → Output: approve_loan = YES.

2. Frames and Objects

Frames group related data (e.g., a Patient frame):

Patient = {
    "name": "Ramesh",
    "symptoms": ["fever", "headache"],
    "tests": {"blood_pressure": 120/80}
}

Advantage: Organizes complex data hierarchically (e.g., Patient → symptoms → fever).

Blackboard Architecture

For problems requiring multiple experts (e.g., medical diagnosis), the blackboard model uses:

• Blackboard: Shared memory where partial solutions are posted.
• Knowledge Sources (KS): Independent experts (e.g., "Cardiology KS," "Radiology KS").
• Control Module: Decides which KS to activate next.

graph TD
    A["Blackboard"] -->|"Posts"| B["Hypothesis: Pneumonia"]
    C["Cardiology KS"] -->|"Reads"| A
    D["Radiology KS"] -->|"Reads"| A
    E["Control Module"] -->|"Triggers"| C & D

Real-World Use: Ncell’s fraud detection system uses a blackboard to combine signals from call logs, transaction history, and device fingerprints.

Applications of Expert Systems

1. Nepal-Specific Examples

2. Global Examples

In the Real World

1. eSewa’s Fraud Detection

   • Idea Used: Rule-based expert system with backward chaining.
   • How: If a transaction flagged as "unusual" (e.g., sudden large payment), the system checks rules like:
     • IF (amount > 50,000 AND location ≠ user’s usual) THEN request OTP.
   • Impact: Blocks 90% of fraudulent transactions without human review.

2. Pathao’s Driver Routing

   • Idea Used: Constraint satisfaction (a type of expert system).
   • How: Rules like:
     • IF (driver_location = A AND destination = B AND traffic_alert = HIGH) THEN reroute via C.
   • Impact: Reduces delivery time by 20% during peak hours.

3. NTC’s Network Fault Diagnosis

   • Idea Used: Blackboard model.
   • How: Multiple "experts" (e.g., "Cable KS," "Hardware KS") post hypotheses (e.g., "Fault in tower X") on a shared blackboard until the root cause is identified.
   • Impact: Cuts repair time from hours to minutes.

Exam Tip

1. Define Clearly: Start answers with:
   • "An expert system is a rule-based AI program that..."
   • "Forward chaining vs. backward chaining differ in..."
2. Draw Diagrams: For blackboard architecture or rule-firing traces, sketch:
   • A blackboard with arrows from KSs.
   • A rule-chaining flowchart (like the loan example above).
3. Compare Tables: Memorize the Nepal vs. Global application table for short-answer questions.
4. Worked Examples: Always show step-by-step traces (like the loan approval) for numerical problems.
5. Real-World Links: Examiners love connections to eSewa, Ncell, or NTC. Mention these in explanations.

Limitations and Challenges

IMAGE: "expert system blackboard architecture diagram" | Blackboard model showing Knowledge Sources (KS) posting to shared memory

IMAGE: "ibm watson interface screenshot" | IBM Watson’s medical diagnosis dashboard

IMAGE: "ncell fraud detection alert" | SMS alert from Ncell showing a blocked fraudulent transaction

IMAGE: "pathao driver app route optimization" | Pathao’s app screen showing rerouted path due to traffic