Expert system architecture

An expert system (ES) is a computer program that holds the knowledge of human experts in a narrow domain and uses it to give advice or solve problems at expert level (e.g. MYCIN for diagnosing infections, DENDRAL, PROSPECTOR).

Architecture

   User
    |  ^
    v  |
 +--------------------+      +-----------------------+
 |   User interface   |<---->|  Explanation          |
 +---------+----------+      |  facility             |
           |                 +-----------+-----------+
           v                             |
 +--------------------+      +-----------v-----------+
 |  Inference engine  |<---->|   Knowledge base      |
 | (reasoning control)|      |  (facts + rules)      |
 +---------+----------+      +-----------^-----------+
           |                             |
           v                 +-----------+-----------+
 +--------------------+      | Knowledge acquisition |<-- Expert /
 |  Working memory    |      | facility              |    knowledge
 | (case facts)       |      +-----------------------+    engineer
 +--------------------+

Blocks

1. Knowledge base: stores domain knowledge as facts and, mostly, IF-THEN production rules (also frames or semantic nets). It is kept separate from the reasoning mechanism.
2. Inference engine: the reasoning part. It matches rules against known facts and derives new conclusions using forward chaining (data-driven) or backward chaining (goal-driven). It also resolves conflicts when several rules apply.
3. Working memory (global database): holds the facts of the current case, user inputs and intermediate conclusions.
4. User interface: lets the user enter data and see results, usually through questions and answers or menus in language close to natural language.
5. Explanation facility: explains how a conclusion was reached ("How?") and why a question is asked ("Why?"), which builds trust and helps debugging.
6. Knowledge acquisition facility: helps the knowledge engineer and domain expert enter, edit and check knowledge in the knowledge base.

Roles of people

• Domain expert: source of knowledge.
• Knowledge engineer: encodes the expert's knowledge.
• End user: consults the system.

Example rule: IF fever AND rash AND cough THEN disease = measles (CF 0.8).