Expert Systems: Architecture and Applications — Beginner-Friendly Notes
1. What an Expert System Is
Expert system = rule-based AI that mimics a human expert's decision-making in a narrow domain (medicine, finance) by encoding expert knowledge as if-then rules and explaining its conclusions.
First successful commercial AI (1980s) — DENDRAL (chemistry), MYCIN (bacterial infection), XCON (computer config).
2. Architecture (diagram to memorise)
User ──► [User Interface / Explanation Facility] ◄── Blackboard / Working Memory
│ ↕ │
└─► [Inference Engine] ──► [Knowledge Base (rules + facts)]
↕
[Knowledge Acquisition Module] (expert → KB)
Parts:
- Knowledge Base (KB): domain rules + facts (often production system, Module 3 topic 11) + certainty factors (Module 4).
- Inference Engine: forward/backward chaining (Module 3) to derive conclusions.
- Working Memory (Blackboard): current case facts (patient symptoms).
- Explanation Facility: "Why/How" — crucial for trust (shows rule chain).
- User Interface: asks questions, shows diagnosis + confidence.
- Knowledge Acquisition: tools for expert to add/update rules.
3. How It Works (trace)
WM: fever, rash, travel
Engine matches R1: IF fever AND rash THEN suspect measles (CF 0.8) → add suspect
Matches R2: IF suspect measles AND travel THEN recommend test → recommend
UI shows: "Recommend test because R1 and R2 fired"
4. Applications (classic + modern view)
| Domain | Classic | Modern successor |
|---|---|---|
| Medical | MYCIN diagnosis | Clinical decision support (still rule + now ML) |
| Finance | Credit approval | Fraud/rule engines + ML scoring |
| Industry | XCON config, fault diagnosis | Predictive maintenance, digital twins |
| Agriculture | Crop advisory | App + ML crop disease (rule + vision) |
5. Strengths vs Limits (exam contrast)
| Strengths | Limits (why they declined) |
|---|---|
| Narrow but expert-level | Brittle outside domain, no learning |
| Explainable (rules) | Knowledge acquisition bottleneck (hard to extract from experts) |
| Modular | Rule explosion, maintenance cost |
| Handles certainty factors | Cannot handle raw perceptual data (needs ML/CV/NLP now) |
Modern hybrid: expert system + ML — rules for known logic, ML for perception/uncertainty.
6. One-Line Summary
An expert system combines a rule-based knowledge base, inference engine and explanation facility over working memory to emulate narrow expert reasoning, providing explainable decisions but limited by brittle, hand-crafted knowledge without learning.
7. Quick Self-Check
- Name the 5 main components.
- Why is explanation facility essential for medical expert systems?