Problems in Representing Knowledge

Definition

Problems in representing knowledge refer to the difficulties encountered when converting real-world facts, relationships, rules, and common-sense understanding into a formal structure that a computer can store, interpret, and reason with correctly and efficiently.

Main Content

1. Nature of Knowledge and Its Complexity

Knowledge is not uniform or simple

Human knowledge includes facts, rules, procedures, beliefs, expectations, exceptions, and tacit understanding. For example, knowing that “birds fly” is a general rule, but “penguins do not fly” is an exception. A representation system must handle both generality and exceptions without losing meaning.

Knowledge exists in many forms and levels

Some knowledge is declarative, such as “Paris is the capital of France,” while some is procedural, such as “how to solve a math problem.” There is also heuristic knowledge, like expert shortcuts, and common-sense knowledge, like understanding that water is wet and objects fall downward. Each type of knowledge may require a different representation technique.

Different granularity and abstraction levels

A fact may be represented at a very high level (“A vehicle transports people”) or at a very detailed level (“A car has four wheels, an engine, fuel system, steering system”). If the representation is too abstract, important details are lost; if it is too detailed, the system becomes inefficient and hard to manage.

Context matters

The same statement can mean different things in different situations. For example, “bank” may refer to a financial institution or a river bank. Representing contextual meaning is difficult because a single symbol may not be sufficient.

Example

Consider the statement “A doctor treats patients.” In reality, a doctor may treat patients in a hospital, a clinic, or during emergency care. The representation must capture roles, settings, and possible variations.

2. Uncertainty, Incompleteness, and Inconsistency

Real-world knowledge is often incomplete

A system may not know all facts about a situation. For example, if it knows “Ali is a student” but not his age, address, or enrolled course, the knowledge base is incomplete. Traditional logic-based systems often assume complete information, which is unrealistic.

Knowledge may be uncertain

Many statements are not absolutely true or false. For example, “It will probably rain tomorrow” is a probabilistic statement. Human experts often reason with degrees of belief rather than certainty, but representing such uncertainty formally is difficult.

Contradictory information can occur

One source may say a machine is faulty, while another says it is working properly. Inconsistencies arise when data is collected from multiple sources or when new facts conflict with old ones. A representation scheme must decide how to detect, manage, or resolve contradictions.

Non-monotonic reasoning challenge

In classical logic, once something is proven, it remains true even if more information arrives. In real life, conclusions may need revision. For example, “Tweety is a bird, so Tweety can fly” may later be invalidated when we learn Tweety is a penguin. This makes reasoning with changing knowledge difficult.

Example

In medical diagnosis, a symptom may suggest several diseases. If a patient has fever, cough, and fatigue, the system cannot conclude one exact disease with certainty without additional tests.

3. Expressiveness, Efficiency, and Reasoning Difficulty

Trade-off between expressiveness and efficiency

A knowledge representation language must be expressive enough to capture meaningful relationships, but not so complex that reasoning becomes computationally expensive. For example, first-order logic is expressive, but reasoning over it can be difficult or undecidable in many cases.

Reasoning can become computationally hard

As the number of rules, objects, and relationships grows, inference may take too long. This is a serious problem in large systems such as semantic web applications, intelligent tutoring systems, and diagnostic expert systems.

Representation affects inference quality

If knowledge is encoded poorly, the system may draw wrong conclusions or miss valid ones. For example, if “all mammals are animals” is represented incorrectly, inheritance-based reasoning will fail.

Need for balance in design

A good representation must support efficient operations such as retrieval, matching, inheritance, and classification while still remaining sufficiently rich to model the target domain.

Example diagram for the trade-off

  More Expressive --------------------> More Difficult Reasoning
         |                                     |
         |                                     |
         v                                     v
  Simpler Representation ---------------> Faster Computation

Practical implication

Systems often use a combination of representations, such as rules for decision-making, frames for structured objects, and semantic networks for relationships, to balance readability and performance.

Working / Process

1. Identify the domain knowledge

First, analyze what kind of knowledge needs to be represented: facts, rules, exceptions, procedures, or uncertain beliefs.
Example: In a hospital system, the knowledge may include diseases, symptoms, medications, treatment procedures, and patient history.
This step is important because different kinds of knowledge require different representation forms.

2. Choose a suitable representation scheme

Select a method such as logic, semantic networks, frames, production rules, ontologies, or probabilistic models.
The choice depends on the nature of the problem. For instance, rules are useful for decision-making, while frames are useful for structured objects with attributes.
The selected scheme should support the necessary reasoning tasks, such as deduction, inheritance, or pattern matching.

3. Encode, test, and revise the knowledge

Convert the domain knowledge into formal statements, then test whether the system can reason correctly.
If the system fails to capture exceptions, uncertainty, or context, revise the representation.
This process is iterative because knowledge representation often needs refinement as new facts, new exceptions, or new use cases appear.

Advantages / Applications

Improves intelligent decision-making

A good representation enables systems to reason from known facts to new conclusions, which is essential in expert systems, diagnosis tools, and planning systems.

Supports knowledge sharing and reuse

Formal representation makes it easier to store, transfer, and reuse knowledge across applications, such as medical ontologies, engineering databases, and educational systems.

Helps in building expert and AI systems

Knowledge representation is foundational for robotics, natural language understanding, semantic search, recommendation systems, and automated reasoning.

Makes hidden relationships explicit

By structuring knowledge, systems can reveal relationships among concepts, such as subclass, part-of, cause-effect, and temporal order.

Example applications

Medical diagnosis, legal reasoning, intelligent tutoring, robotics navigation, semantic web, and decision support systems.

Summary

Knowledge representation is difficult because real-world information is complex, uncertain, and context-dependent.
A good representation must balance expressiveness with efficient reasoning.
Common issues include incomplete facts, conflicting information, and exceptions to general rules.
Important terms to remember: knowledge representation, uncertainty, inconsistency, expressiveness, inference.