Knowledge acquisition in expert system

Knowledge acquisition

Knowledge acquisition is the process of extracting, structuring and organising knowledge from human experts or other sources (books, manuals, databases) and transferring it into the knowledge base of an expert system. It is the biggest bottleneck in building expert systems.

How it is performed

1. The knowledge engineer studies the domain and interviews the domain expert (structured/unstructured interviews, questionnaires, observation, protocol analysis).
2. The knowledge is analysed, and the concepts, facts and heuristics are organised.
3. It is coded in a representation such as IF-THEN rules, using a knowledge acquisition tool or editor.
4. The rules are tested with cases and refined with the expert.
5. Automatic methods may also be used: machine learning (e.g. ID3 from examples).

Example: a doctor says, "If a patient has fever and a rash on the face and the cough is dry, I suspect measles." The engineer writes: IF fever AND facial_rash AND dry_cough THEN disease = measles.

Architecture of an expert system

 User <--> [User interface] <--> [Inference engine] <--> [Knowledge base]
                |                      |                      ^
          [Explanation]         [Working memory]     [Knowledge acquisition]
                                                      (expert, engineer)

• Knowledge base: facts and rules of the domain.
• Inference engine: applies rules by forward or backward chaining.
• Working memory: facts of the current case.
• User interface: question-answer dialogue.
• Explanation facility: shows how and why.
• Knowledge acquisition module: to build and update the knowledge base.

Real example: MYCIN

MYCIN (Stanford, 1970s) diagnoses bacterial blood infections and recommends antibiotics. Its knowledge base holds about 500 IF-THEN rules with certainty factors; the inference engine uses backward chaining; the user answers questions about symptoms and lab tests; the explanation facility answers "Why do you ask?" and "How did you conclude?". Its performance matched that of specialists.