Expert Systems Overview

Tags: #◇Computer Science #◇CSINTSY #◇Expert Systems Overview #◇Problem Solving #◇M5

---

◇Overview

Expert Systems are a type of traditional, logic-based artificial intelligence system designed to mimic the decision-making capabilities of a human expert within a specific, well-defined domain (e.g., MYCIN for bacterial infections). Instead of learning from data (like ML), they solve problems by reasoning through a pre-defined body of knowledge.

Core Components:

1. Knowledge Base: Stores facts and rules about the world, typically encoded in a formal logic language.
2. Inference Engine: An algorithm that processes information in the knowledge base to draw conclusions or deduce new facts.
3. User Interface: Allows users to input queries and view conclusions.

Inference Strategies: Forward Chaining: Starts with known facts and applies rules to generate all possible deductions until a goal is reached. Backward Chaining: Starts with a target goal or query and works backward, checking if the prerequisites can be proven by the known facts.

Strengths and Limitations: Strengths: Highly efficient for specific domains; interpretable and explainable since rules are explicitly hand-crafted. Limitations: Hard to apply to ill-structured domains (e.g., image recognition); maintaining a large knowledge base is difficult; rules must be manually crafted for every new domain, which is costly.

◇Important Formulas

Propositional and First-Order Logic Notations: Conjunction (AND): $\land$ Disjunction (OR): $\lor$ Implication (If-Then): $\rightarrow$ Negation (NOT): $\neg$ Universal Quantifier (For all): $\forall x$ Existential Quantifier (There exists): $\exists x$

◇Worked Examples

Forward Chaining Example:

1. Facts: $\text{American(West)}$, $\text{Missile(M1)}$, $\text{Owns(Nono, M1)}$, $\text{Enemy(Nono, America)}$
2. Rules: An American selling weapons to hostile nations is a criminal.
3. Steps: Combine facts to deduce $\text{Weapon(M1)}$, $\text{Sells(West, M1, Nono)}$, $\text{Hostile(Nono)}$, ultimately deducing $\text{Criminal(West)}$.

Backward Chaining Example:

1. Query: Is the plant a pine?
2. The system checks prerequisites: "Is the leaf shape needlelike?"
3. It asks the user or checks facts until the condition is proven true or false.

◇References & Navigation

• Previous Topic: ◇K-Means Clustering
• Next Topic: ◇Markov Decision Processes
• Related Notes: ◇Artificial Intelligence Overview