logo

Unit 2 Knowledge engineering

Learning outcomes

By the end of this unit you should:

  • know the three main categories of knowledge
  • be able to name examples of different language-related expert systems
  • have considered (briefly) user needs for expert systems related to language learning
Rubik

Activity 1: Review

Work in pairs. Without reading your notes or the course website, describe what an expert system is. Remember to use the technical terms you learnt in the previous unit.

Activity 2: Categories of knowledge

Identify the relationship between the three categories of knowledge described.

Knowledge may be classified by category. The category of knowledge is not the domain, discipline or subject. In this course, we will focus on three categories of knowledge, namely: explicit, tacit and implicit.

  1. Explicit knowledge is the formal knowledge that is easy to describe in words and may be found in books, manuals, databases. This knowledge is readily available.
  2. Tacit knowledge is personal knowledge gained through practice that is difficult to describe in words and is not easily found in books, manuals or databases.
  3. Implicit knowledge describes knowledge is not readily available but can be made explicit by an expert.

Explicit knowledge is written, implicit has not been written but easily be while tactic knowledge is difficult to write down since even the expert may not know how to describe what they know and can do.

Activity 3: Classifying knowledge

Work in pairs. Discuss the category of knowledge for each of the examples below.

Explicit knowledge

Tacit knowledge

Implicit knowledge

Implicit knowledge

Tacit knowledge

Explicit knowledge

Activity 4: Process of knowledge engineering

Read to understand five phases in the knowledge engineering process.

1. Knowledge Acquisition

This is the process of obtaining information from human experts or other credible sources. The data could be about a specific domain, like medical diagnosis, financial forecasting, or weather prediction. This is usually the most time-consuming part of knowledge engineering because it involves extensive interaction with domain experts and often a careful examination of complex domain knowledge.

2. Knowledge Representation

Once the data is obtained, it must be represented in a way that the system can understand. This could be through rule-based systems, semantic networks, frames, or other methods. The choice of representation may affect the ease of learning, the speed of inference, and the system's capabilities.

3. Knowledge Inferencing

It refers to how the expert system applies logic to the knowledge it has been given to arrive at new conclusions or solutions. Depending on the complexity of the problem, this might involve straightforward logical deductions or more complex reasoning methods.

4. Knowledge Validation

This involves testing the system to ensure that it's using its knowledge correctly. This is a critical step for verifying that the system provides correct and reliable outputs.

5. Knowledge Refinement

This is an ongoing process that ensures the knowledge within the system remains accurate and up to date. As new information becomes available or as the field of knowledge progresses, the system should be updated and refined.

Rubik
The process of knowledge engineering (adapted from Turban). Source: Jung et al, 2020

Activity 5: Knowledge acquisition

Read this description to discover the number of methods of knowledge acquisition.

Knowledge acquisition, crucial in the development of expert systems, can be achieved through both non-digital and digital methods. Non-digital methods include interviewing, observation, analysis of documented information, and protocols and scenarios. Interviewing and observation involve direct interaction with domain experts to understand their decision-making process and methodology. Analysis of documented information involves extracting knowledge from existing resources like manuals, documents, or instructional videos. Protocols and scenarios entail presenting experts with hypothetical situations or problems to evaluate their problem-solving approach. On the other hand, the digital method includes machine learning, where systems learn from large datasets rather than relying on explicit programming. The system is trained, evaluated, and tuned based on a relevant dataset, enabling the model to make predictions or decisions without human intervention. These methods, employed based on the specific context and requirements, facilitate capturing knowledge from various sources, subsequently contributing to the creation of effective expert systems.

Activity 6: Knowledge representation

Read the AI-generated descriptions of the following types of knowledge representation. Search online, Use DeepL or refer to ChatGPT to better understand the types that are more difficult to comprehend.

  1. Rule-Based Systems are one of the most common forms of knowledge representation. IF-THEN rules are created to describe the logic used in decision making.
  2. Production Systems use a set of production rules, which are similar to IF-THEN rules, but also include an action part.
  3. Frames are an object-oriented approach, where knowledge is represented in terms of "objects" (or frames), which have attributes and values to represent a real-world entity. Frames subdivide knowledge into manageable segments, illustrating stereotypical scenarios. Each frame is made up of a variety of slots, which can differ in type and size, and these slots contain named values, known as facets. These facets provide specific details about the situation or entity the frame is describing.
  4. First order Logic provides a way to talk about objects and the relationships that can exist between them. First order logic comprises predicate logic and some factual details related to the statements.If you are interested in first-order logic, check out this First order logic tool developed by Matteo Morena, which analyzes first order formula.
  5. Fuzzy Logic a form of many-valued logic in which the truth values of variables may be any real number between 0 and 1 both inclusive. It is used to handle the concept of partial truth, where the truth value may range between completely true and completely false.
  6. Semantic Networks are graph-based structures used to represent knowledge in patterns of interconnected nodes and arcs. The nodes represent concepts and the arcs represent the relationships between these concepts. They are particularly useful for representing knowledge that includes a lot of relationships or categories. Check out ConceptNet, which is a freely-available semantic network.
  7. Conceptual Graphs are based on the existential graphs of Charles Sanders Peirce and the semantic networks of artificial intelligence. It expresses meaning in a form that is logically precise, humanly readable, and computationally tractable.
  8. Bayesian Networks (or belief networks) are a type of probabilistic graphical model that represent a set of variables and their conditional dependencies via a directed acyclic graph (DAG). DAGs comprise vertexes (nodes) and edges (arcs). DAGs start at the source, which is vertex with no incoming edges and end at a sink, which is a vertex without outgoing edges. These are used in expert systems to deal with uncertain or probabilistic knowledge. If interested, enjoy this 8 minute 16 second video explaining the Monty Hall probability problem using Bayes' rule:

Activity 7: Matching examples to type of knowledge representation

Work in pairs. Discuss the type of knowledge representation for each of the examples below.

Rule-based systems

Conceptual graphs

Production systems

First order logic

Fuzzy logic

Frames

Semantic networks

Bayesian networks

Review

Can you:

  1. do this
  2. do that
  3. and do something else.

If you do not, make sure that you do before your next class.

Running count: 30 of 30 concepts covered so far.