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

What is an Expert System?

An expert system is a computer program that uses knowledge and inference rules to solve problems in a specific domain, mimicking the decision-making ability of a human expert. Unlike general AI, expert systems focus on narrow, well-defined tasks (e.g., diagnosing diseases, configuring systems, or troubleshooting).

Key Components of an Expert System

How It Works: A Trace Example

Problem: Diagnose a car engine issue (overheating). Knowledge Base (Rules):

1. IF engine temperature > 100°C AND coolant level = low THEN suggest "Check coolant."
2. IF engine temperature > 100°C AND radiator fan = faulty THEN suggest "Replace fan."
3. IF oil pressure = low THEN suggest "Check oil level."

Inference Steps (Forward Chaining):

1. Input: User reports engine temperature = 110°C, coolant level = low.
2. Match: Rule 1 fires → Output: "Check coolant."
3. User Action: User adds radiator fan = faulty.
4. Match: Rule 2 fires → Output: "Replace fan."

---

Types of Expert Systems

---

Expert System Shells

An expert system shell is a pre-built framework that provides the inference engine and tools to build knowledge bases without coding from scratch. Example: CLIPS (C Language Integrated Production System). Advantages:

• Faster development (no need to build inference logic).
• Reusable for multiple domains. Disadvantages:
• Limited flexibility for complex domains.
• May require customization for niche applications.

---

Applications of Expert Systems

1. Medical Diagnosis

Example: MYCIN (1970s) diagnosed bacterial infections by asking symptoms and suggesting antibiotics. Real-World Tie: Nepal’s eSewa Health uses AI to triage symptoms (e.g., "fever + cough" → "Possible COVID-19, consult a doctor").

2. Financial Advisory

Example: Loan approval systems (e.g., NMB Bank’s AI) use rules like:

IF credit_score > 650 AND income > 50,000 AND loan_amount < 5M THEN approve.

Worked Example:

3. Industrial Troubleshooting

Example: NTC’s network fault diagnosis uses expert systems to pinpoint issues in telecom towers. Real-World Trace:

1. Symptom: "No 4G signal in Kathmandu-3."
2. Rule: IF signal drops in a sector AND weather = clear THEN check tower hardware.
3. Action: AI suggests "Inspect antenna alignment."

---

In the Real World

1. Khalti’s Fraud Detection

   • Idea: Rule-based expert system flags transactions using heuristics like:

     IF transaction_amount > 50,000 AND location = "unknown" THEN flag as suspicious.
     

   • Impact: Reduces fraudulent payments by 30%.

2. Pathao’s Driver Routing

   • Idea: Case-based reasoning adapts routes based on past traffic data (e.g., "If it’s 7 PM on a Friday, avoid Thapathali").
   • Visual: Traffic routes as a graph where nodes = locations, edges = travel time.

3. NEPSE’s Stock Advice

   • Idea: Model-based systems simulate market trends (e.g., "If inflation rises 2%, sell tech stocks").
   • Example: An expert system might generate:

     BUY: NMB (PE ratio = 12 < market avg)
     SELL: Global IME (dividend yield = 0.5% < benchmark)
     

---

Building an Expert System: Step-by-Step

Worked Example: Simple Loan Advisor

1. Domain: Personal loans.
2. Rules:
   • IF income > 70,000 AND employment = "permanent" THEN max_loan = 10M.
   • IF credit_history = "poor" THEN max_loan = 2M.
3. Input: User enters income = 80,000, employment = "contract".
4. Output: "Eligible for loan up to 5M (contract employment cap)."

---

Advantages and Limitations

---

Ethical and Societal Impact of AI

Challenges:

1. Bias in Decision-Making
   • Example: A hiring expert system trained on historical data may favor male candidates if past hires were male-dominated.
2. Job Displacement
   • Repetitive roles (e.g., call-center agents, radiologists) are at risk.
3. Accountability
   • Who is responsible if an AI misdiagnoses a disease? The developer? The hospital?

Nepal-Specific Issues:

• Digital Divide: Rural areas lack access to AI tools (e.g., eSewa’s AI requires smartphone connectivity).
• Data Privacy: Expert systems handling health data (e.g., HamroPatri) must comply with Nepal’s Data Privacy Act (2018).

---

Emerging Trends

1. Hybrid AI Systems
   • Combining expert systems with deep learning (e.g., Google’s DeepMind for protein folding).
2. Explainable AI (XAI)
   • Tools like IBM’s AI Fairness 360 ensure transparency in expert system decisions.
3. AI in Agriculture
   • Nepal’s KisanCall uses expert systems to advise farmers on pest control (e.g., "If leaves are yellow, spray neem oil").

---

Exam Tip

1. Define Clearly

   • Distinguish between expert systems (rule-based) and machine learning (data-driven). Example:

     "An expert system uses symbolic reasoning, while ML relies on statistical patterns."

2. Compare with Other AI Techniques

   • Use a table to contrast expert systems, neural networks, and case-based reasoning (as shown earlier).

3. Apply to Real Scenarios

   • Question Type: "How would you design an expert system for NTC’s customer complaint resolution?"
   • Answer Structure:
     1. Domain: Telecom complaints.
     2. Rules: IF complaint = "no signal" AND location = "rural" THEN suggest "Check tower coverage."
     3. Shell: Use CLIPS for rule management.

4. Ethics is Key

   • Expect questions on bias, accountability, and Nepal’s AI policy. Example:

     "Discuss how an expert system for loan approval might discriminate against women in Nepal."

5. Diagrams Save Marks

   • Draw:
     • A rule-chaining trace (as above).
     • A knowledge base structure (facts + rules).
     • A comparison table (expert systems vs. ML).

---