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

What is an Expert System?

An expert system is an AI program that uses knowledge and inference rules to solve problems in a specific domain, mimicking a human expert’s decision-making. It consists of three core components:

classDiagram
    class KnowledgeBase {
        +Facts (declarative knowledge)
        +Rules (if-then statements)
    }
    class InferenceEngine {
        +Forward Chaining (data-driven)
        +Backward Chaining (goal-driven)
        +Certainty Factors (uncertainty handling)
    }
    class UserInterface {
        +Explanation Facility
        +Query Handling
    }
    KnowledgeBase --> InferenceEngine : "Uses"
    InferenceEngine --> UserInterface : "Drives"

Key Definitions

• Knowledge Base (KB): Stores domain-specific facts and rules (e.g., "If fever > 38°C and cough then possible flu").
• Inference Engine: Applies logical rules to derive conclusions (forward/backward chaining).
• Explanation Facility: Justifies decisions to users (critical for trust).

How Expert Systems Work: Forward vs. Backward Chaining

1. Forward Chaining (Data-Driven)

Starts with known facts and applies rules to reach conclusions. Used when the goal is unknown but data is abundant. Example: Medical diagnosis system.

graph TD
    A["Facts: Fever=38.5°C, Cough=Yes"] --> B["Rule 1: If Fever>38°C then PossibleInfection=True"]
    B --> C["Rule 2: If PossibleInfection=True and Cough=Yes then PossibleFlu=True"]
    C --> D["Conclusion: PossibleFlu=True"]

Worked Example: Facts:

• Patient temperature = 39°C
• Patient has headache
• Patient has fatigue

Rules:

1. If temperature > 38°C then possible_infection = True
2. If possible_infection = True and headache = True then possible_flu = True
3. If possible_flu = True and fatigue = True then diagnose = "Flu"

Trace:

1. Rule 1 fires → possible_infection = True
2. Rule 2 fires → possible_flu = True
3. Rule 3 fires → diagnose = "Flu"

2. Backward Chaining (Goal-Driven)

Starts with a hypothesis and works backward to verify it. Used when the goal is known but data is incomplete. Example: Loan approval system.

graph TD
    A["Goal: ApproveLoan?"] --> B["Rule: If CreditScore>700 and Income>50000 then ApproveLoan=True"]
    B --> C["Check CreditScore"]
    B --> D["Check Income"]
    C --> E["CreditScore=750"]
    D --> F["Income=60000"]
    E & F --> G["ApproveLoan=True"]

Worked Example: Goal: Can the customer get a loan? Rules:

1. If credit_score > 700 and income > 50,000 then approve_loan = True
2. If income < 30,000 then reject_loan = True

Trace:

1. Start with goal: approve_loan?
2. Check Rule 1: Need credit_score and income.
   • Query credit_score → returns 750 (satisfies >700).
   • Query income → returns 60,000 (satisfies >50,000).
3. Conclusion: approve_loan = True.

Handling Uncertainty: Certainty Factors

Real-world data is often uncertain. Certainty Factors (CF) assign confidence levels (0–1) to rules and conclusions. Formula: $$CF(A|B)=\{\begin{array}{ll}\text{max}(0,CF(B)+(1-CF(B))\times W(A|B))& \text{if }CF(B)\ge 0\\ \frac{CF(B)}{CF(B)-(1-W(A|B))}\times CF(B)& \text{if }CF(B)<0\end{array}$$ Where:

• $W(A|B)$ = weight of rule (e.g., 0.8 for "If cough then flu").
• $CF(B)$ = certainty of premise (e.g., 0.7 for "Patient has cough").

Worked Example: Facts:

• Cough (CF = 0.7)
• Fever (CF = 0.6)

Rules:

1. If cough then flu (CF = 0.8)
2. If fever then flu (CF = 0.7)

Calculation:

1. For Rule 1: $CF(\text{flu}|\text{cough})=0.7+(1-0.7)\times 0.8=0.94$
2. For Rule 2: $CF(\text{flu}|\text{fever})=0.6+(1-0.6)\times 0.7=0.88$
3. Combine using OR (max): $CF(\text{flu})=\text{max}(0.94,0.88)=0.94$

Conclusion: Flu diagnosis with 94% confidence.

Applications of Expert Systems

In Nepal:

1. eSewa (Nepal Government):

   • Use: Automated tax filing and business registration.
   • AI Idea: Rule-based validation of documents (e.g., "If PAN card expired then reject application").
   • Impact: Reduces human error in 10,000+ daily submissions.

2. Ncell Customer Support Chatbot:

   • Use: Troubleshooting network issues.
   • AI Idea: Decision tree for diagnostics (e.g., "If signal weak and area rural then suggest tower upgrade").
   • Impact: Handles 30% of tier-1 queries without human agents.

3. Nepal Rastra Bank (NRB) Loan Approval:

   • Use: Credit scoring for SMEs.
   • AI Idea: Backward chaining to verify loan eligibility (e.g., "If turnover > 5M and repayment history good then approve").

Globally:

1. IBM Watson (Healthcare):

   • Use: Oncology treatment recommendations.
   • AI Idea: Forward chaining over medical literature to suggest therapies.

2. SAP’s AI for Supply Chain (Daraz, Amazon):

   • Use: Predictive inventory management.
   • AI Idea: Certainty factors for demand forecasting (e.g., "If monsoon delay then rice demand CF=0.9").

3. WhatsApp Business API (Pathao, Food Delivery):

   • Use: Automated order routing.
   • AI Idea: Rule-based dispatch (e.g., "If rider distance < 5km and order urgent then prioritize").

Advantages and Limitations

Real-World Example: Kathmandu Traffic Management

Problem: Congestion on Ring Road during peak hours (7–9 AM). Expert System Solution:

1. Knowledge Base:
   • Rule 1: If traffic density > 80 vehicles/km then activate signal priority.
   • Rule 2: If accident reported and ambulance en route then clear lane 3.
2. Inference Engine: Forward chaining to trigger actions in real-time.
3. Impact: Reduced delays by 25% in pilot tests (used by NTC’s smart traffic project).

Exam Tip

1. Diagrams are mandatory: Always draw the 3-component architecture (KB, inference engine, UI) and chaining traces (forward/backward).
2. Certainty factors: Memorize the formula and practice combining rules (e.g., "If A or B then C").
3. Case studies: Link to eSewa, Ncell, or NRB for application questions. Example:

   "How would you design an expert system for Daraz’s order fulfillment?" Answer: Use backward chaining to verify stock → shipping rules → delivery constraints.

4. Weaknesses: Expect questions on why expert systems fail for unstructured data (e.g., social media sentiment).
5. Maths: Show step-by-step CF calculations for uncertainty questions.