Types of Knowledge and Their Application

TL;DR

In AIML, we classify knowledge into different types to better understand how systems learn and make decisions. Explicit knowledge is clearly stated, implicit knowledge is learned through experience, and meta-knowledge is knowledge about knowledge itself. Understanding these types helps you design more effective AI systems.

1. The Mental Model

Think of knowledge like ingredients in a recipe. Some ingredients are clearly listed (explicit), some you just know from cooking experience (implicit), and some are about how to combine ingredients effectively (meta-knowledge). AI systems use these different "ingredients" to perform tasks.

2. The Core Material

In AIML, knowledge isn't just one thing. It comes in various forms, each useful for different aspects of intelligent behavior. Let's break down the main types:

Declarative Knowledge

Photo by Nadejda Bostanova on Pexels

This is "knowing that". It's factual information, often represented as statements, rules, or objects. Think of it as data that can be directly stated or written down.

• Examples: "The capital of France is Paris." "If a customer is a gold member, then offer a discount."
• Application in AIML: Databases, knowledge bases, rule-based systems, ontologies. A chatbot might use declarative knowledge to answer specific questions from a pre-defined set of facts.

Procedural Knowledge

Photo by Sergei Starostin on Pexels

This is "knowing how". It's about skills, sequences of actions, or steps to achieve a goal. It's often difficult to articulate fully but demonstrated through performance.

• Examples: How to ride a bicycle, how to bake a cake, how to calculate a derivative.
• Application in AIML: Algorithms, control systems, robotic movements, sequential decision-making processes (like reinforcement learning policies). A robot arm picking up an object uses procedural knowledge.

Implicit Knowledge (Tacit Knowledge)

Photo by John Diez on Pexels

This is knowledge we possess but often can't easily explain or articulate. It's gained through experience and practice. It's often intuitive.

• Examples: Recognizing a face in a crowd, the feeling of correctly balancing while walking, understanding nuances in human language.
• Application in AIML: This is where machine learning shines. Neural networks often learn complex patterns that are hard to explicitly define. For instance, a deep learning model classifying images learns implicit features.

Explicit Knowledge

Photo by Markus Winkler on Pexels

This is knowledge that is clearly articulated, codified, and easily shared. It's the opposite of implicit knowledge and can often be converted from implicit knowledge.

• Examples: A user manual, a formal scientific theory, a database of customer preferences.
• Application in AIML: Any system that relies on structured data or clearly defined rules. Expert systems often rely heavily on explicit knowledge provided by human experts.

Meta-Knowledge

This is "knowledge about knowledge". It's understanding how to acquire, manage, use, and update other types of knowledge. It's crucial for adaptive and intelligent systems.

• Examples: Knowing which strategy to use to solve a particular problem, understanding the reliability of a source of information, knowing when you don't know something.
• Application in AIML: Meta-learning, transfer learning, explanation systems, knowledge representation schemes that include confidence levels or source attribution. An AI that can choose the best algorithm for a given dataset uses meta-knowledge.

Here's how these types often relate and flow in AI system design:

graph TD
    A["Human Expert Experience (Implicit)"] --> B["Knowledge Elicitation"]
    B --> C{"Formalized Rules & Facts (Explicit/Declarative)"}
    C --> D["Rule-Based System (Application)"]
    A --> E["Data Collection (Raw Observations)"]
    E --> F["Machine Learning Model Training (Implicit Pattern Recognition)"]
    F --> G["Predictive/Classification System (Application)"]
    C & G --> H["Meta-Knowledge (How to combine, evaluate, learn from these)"]
    H --> I["Adaptive AI System"]

3. Worked Example

Let's consider an AI system designed to recommend movies.

1. Declarative Knowledge: The system knows facts like "Movie X is a sci-fi film," "Actor Y starred in Movie Z," and "User A liked 'Interstellar'." This is stored in a database.
2. Procedural Knowledge: The system has an algorithm for calculating similarity between movies based on genre, actors, and user ratings. It also has a procedure for filtering out movies already watched by the user.
3. Implicit Knowledge: A deep learning recommendation engine, trained on millions of user ratings, learns subtle patterns like "users who like critically acclaimed foreign dramas also tend to enjoy philosophical thrillers," even if those categories aren't explicitly tagged. This knowledge isn't coded as a rule but emerges from the model's weights.
4. Explicit Knowledge: The movie database itself, with all its structured information (genres, directors, actors, synopses), is explicit knowledge.
5. Meta-Knowledge: The system might know that for a new user with limited watch history, it's better to recommend popular, highly-rated movies (using declarative knowledge) rather than relying solely on the deep learning model (which needs more data). It also knows to periodically retrain its implicit model with new user data.

4. Key Takeaways

• Declarative knowledge is about facts and "knowing that," crucial for rule-based reasoning.
• Procedural knowledge is about actions and "knowing how," essential for task execution.
• Implicit knowledge is learned through experience and hard to articulate, often captured by machine learning models.
• Explicit knowledge is clearly stated and easily shareable, forming the basis of many knowledge bases.
• Meta-knowledge is knowledge about knowledge, enabling AI systems to adapt and learn efficiently.
• Different AI tasks benefit from different types of knowledge; effective systems often combine them.
• Understanding knowledge types helps in choosing appropriate AI techniques and representations.

Common Mistakes to Avoid:
- Treating all knowledge as declarative: Not everything can be neatly put into facts or rules; some things are better learned implicitly.
- Underestimating implicit knowledge: Relying only on explicit rules can make systems brittle and unable to handle novel situations.
- Ignoring meta-knowledge: Without knowing when or how to apply knowledge, an AI can make suboptimal decisions or be inefficient.
- Confusing explicit with implicit: Just because you can write down a rule doesn't mean it captures all the underlying intuitive understanding.

5. Now Try It

Think about a self-driving car. For about 15 minutes, list specific examples of how it would use each of the five types of knowledge (declarative, procedural, implicit, explicit, and meta-knowledge) to safely navigate a street.

What success looks like: You should have at least one concrete example for each knowledge type, clearly explaining how the car uses it. For instance, for implicit knowledge, you might think about how it recognizes pedestrians.