🧩 Symbolic AI · Lesson 2
🔧 The Tools: How we build a "White Box" machine
Recap: The "White Box" Approach
In the previous lesson, we established that Symbolic AI is a top-down approach where humans explicitly write the rules and logic.
Because you can literally trace every logical step the machine takes, we call it a "White Box"—meaning the reasoning is fully transparent and easy to audit, unlike the "Black Box" of Machine Learning.
But how do we actually build one of these systems? There are three core tools that every Symbolic AI uses:
Knowledge Representation (how it stores facts), Search Algorithms (how it navigates those facts), and Inference Engines (how it applies logic to combine them).
🛠️ Tool 1: Knowledge Representation (KR)
Knowledge Representation is the way we structure and store facts and rules in a format a computer can understand.
Instead of vague text, we use strict logic.
The most common form of KR is the IF-THEN rule. These rules form a Tree of Logic.
If you have a chain of linear rules, it forms a straight line, but often there are branching points.
🌳 The "Tree of Logic" Example: "My car won't start."
IF (Car won't start)
├── THEN (Check battery)
│ ├── IF (Battery is dead) → THEN (Jump start it)
│ └── IF (Battery is fine) → THEN (Check fuel gauge)
│ ├── IF (Fuel is empty) → THEN (Refuel)
│ └── IF (Fuel is not empty) → THEN (Call a mechanic)
Why this is a tree: At the "Check battery" step, the logic branches off. If the battery is fine, you don't jump start it. Instead, you move to the next branch (checking the fuel).
📦 Other forms of KR:
- Decision Trees: A visual, tree-like model of decisions and their possible consequences.
- Semantic Networks: A graph showing concepts as nodes and relationships as connecting lines (e.g., Dog → "is a" → Mammal).
Why B is correct: Knowledge Representation is about explicitly defining facts and rules. Writing down an "IF-THEN" medical rule is a perfect example of structuring knowledge.
A) is a Search problem, C) is a Search problem with an opponent, and D) is Machine Learning.
🛠️ Tool 2: Search Algorithms
Once we have a massive library of rules and facts, we need a way to navigate through them to find the one we need. This is where Search Algorithms come in.
🔍 Uninformed Search (Blind)
How it works: The computer tries every path without knowing which one is better. It explores blindly until it finds the goal.
Example: Walking through a maze blindfolded, trying every hallway until you find the exit.
🧭 Informed Search (Heuristic)
How it works: The computer uses a "heuristic" (an educated guess) to prioritize the most promising paths and avoid wasting time.
Example: Using a Map/GPS app. It uses the "straight-line distance" as a heuristic to avoid roads that take you further away from your destination.
📱 The Map Application & Constraints:
When a GPS finds the fastest route, it applies a constraint: "Am I getting closer to my destination?" If a road takes it further away (raising the heuristic score), it prunes that branch and doesn't waste time exploring it. This makes it vastly more efficient than blind search.
Why B is correct: Finding the shortest path in a map is the textbook definition of a Search Algorithm. By using a heuristic (the straight-line distance) to avoid moving further away, it becomes an informed search.
A) is KR, C) is an Inference Engine, and D) is KR (Semantic Network).
🛠️ Tool 3: Inference Engines
The Inference Engine is the "brain" of the system. It takes the rules (from Knowledge Representation) and uses Search to find the relevant ones, then chains them together to generate new conclusions that weren't explicitly written down.
💡 The Logic Chain in Action:
- Rule 1: IF (CPU usage > 90%) THEN (Cause is a runaway process).
- Rule 2: IF (Cause is a runaway process) AND (Process is Antivirus) THEN (Recommend scheduling the scan).
The Inference Engine doesn't just "find" Rule 1. It actively combines Rule 1 and Rule 2 to deduce: "The user must schedule the antivirus scan." This is the process of logical deduction.
Why C is correct: The core purpose of an Inference Engine is to chain rules together to derive new, unstated facts. A) is a Semantic Network (KR). B) is a Search Algorithm. D) is just raw data storage.
Why A is correct: A Semantic Network (or Knowledge Graph) is a specific method of Knowledge Representation. It is purely about storing the relationships between concepts. It doesn't search or infer anything on its own.
Why B is correct: Calculating future moves in a game is a classic Search Algorithm problem, specifically known as Adversarial Search (which we will explore in the next lesson).
Why C is correct: Combining two separate rules to produce a brand new, previously unstated conclusion is the exact function of an Inference Engine. It is the "logic brain" that chains rules together.
🎉 You've mastered the Tools of Symbolic AI!
Recap:
• Knowledge Representation: Storing facts and rules (like a Tree of Logic).
• Search Algorithms: Navigating through those facts (like a GPS using heuristics).
• Inference Engines: Combining rules to deduce new conclusions (the logical brain).
In Lesson 3, we will see these three tools combined into real-world applications like Game-Playing AI, GPS Navigation, and the ultimate Symbolic AI product: Expert Systems.