Expert Systems

CIT 720

Computers as Cognitive Tools Tinus Kühn July 2007

Modelling with Expert Systems Definition Origin Usefulness Components of Expert Systems

User

Current problem User interface

IF Conditions

Inference engine Types of inference engine Solution/Advice

Problem-solving Anatomy of an Expert System Planning the Expert System Limitations

Expert editor

Knowledge base

Definition A computer program that simulates expert reasoning with a view to facilitating decision-making for problemsolving. Examples • Doctors diagnosing strange bacterial infections. • Geologists deciding where to drill for oil. • Firefighters deciding how to extinguish a fire.

Consists of a knowledge base of facts and rules.

Origin Artificial intelligence [AI]: Computer sciences that develop programming techniques enabling computers to perform tasks regarded as intelligent when done by people. Artificial = simulated [imitated]. Intelligence = learning, reasoning, understanding. Artificial intelligence programs simulate human thinking such as decision-making. They only imitate our idea of mental activity. They cannot reason like humans. Human reasoning is transferable or generalisable; computer intelligence is not.

Usefulness For problem-solving that requires decision-making or inferential thinking about implications of findings. The knowledge base of an expert system represents causal, procedural knowledge about content domains. For supporting learners’ causal reasoning when solving problems. Building one requires formal, logical reasoning. Reflecting on and representing the thinking involved in solving a problem is metacognitive. Propagation of inferential and implicational reasoning. The learner has to go beyond existing information.

Components of Expert Systems User

Provides a need/problem

User interface

For communication between user and expert system

Current problem

Data about the problem are entered

Knowledge base

Facts and rules leading to a decision

Expert editor

For entering facts and rules into the knowledge base

Inference engine

Queries the knowledge base

Solution/Advice

Feedback that solves the problem

User

User with a need or problem makes an input. Example Client asks novice bank official for a loan. Decision depends on i.a. client’s age, income, credit record, etc. An expert system can help the bank official.

Current Problem Information

Data about the current situation are collected from the user and entered into the computer memory to help the expert system find a solution. Answers to questions about the current situation are integrated within a knowledge base of information.

User Interface

This makes it possible for the expert system to communicate with the user. It asks questions and provides explanations about questions and decisions made by the expert system.

Knowledge Base This is analogous to one’s long-term memory. It consists of facts about objects and rules about relationships among objects; these represent knowledge structures used by experts. Facts state given conditions; rules consist of conditions and decisions. The latter state that IF conditions exist, THEN a decision is reached. Conditions can be combined into sets. Sets of IF conditions can be used using conjunctions, disjunctions and negations.

IF Conditions

Conjunctions

Condition 1 AND condition 2 must exist …

Disjunctions

Condition 1 OR condition 2 must exist …

Negations

Condition 1 but NOT condition 2 must exist …

… for a decision to be reached.

Knowledge Base (Continued)

A decision may be an action or it may state another condition. This is then combined with other conditions to reach a decision. A decision may be an action or it may state another condition. This is then combined with other conditions to reach a decision.

Expert Editor This is usually provided by the expert system. It enables the expert/knowledge engineer to enter information into the knowledge base. Editors consist of text editors and parsers. The text editor makes it possible to input facts and rules into the knowledge base in a specified format. The parser checks the syntax of information entered as well as the validity or logic.

Inference Engine The inference engine queries the knowledge base and is constructed of AI programming techniques. It works after the user has posed a problem and has entered current problem information. It contains the logical programming to examine the information provided by the user as well as the facts and rules specified in the knowledge base. It evaluates the current problem situation and looks for advice about the situation.

Types of Inference Engine Backward chaining Goal-driven

Start

Forward chaining Data-driven

Start

Solution/Decision

Matches data with condition[s] stated in rules.

Searches knowledge base for rules containing the conditions to fulfil the solution.

Examines knowledge base to see if solution is viable with only that information.

Solution/Advice

The expert system finally presents a solution to the problem. The solution is generated by the inference engine that is based on the permanent knowledge base and current problem information.

Problem-solving Solving a problem requires causal reasoning. The more complex the problem, the more sophisticated the causal reasoning must be. An expert system models the causal relationships in problem-solving.

Anatomy of an Expert System An expert system uses a knowledge base of facts and rules. It compares the information supplied by the user to the knowledge base and presents a decision. It deals with changing conditions in a problem situation; therefore the user is asked questions. The answers provided define the conditions that will be evaluated by the rules in the knowledge base.

Anatomy of an Expert System (Continued) The expert system retrieves information from the knowledge base and displays it for the user.

Planning the Expert System Identify the goals and then state the decisions. Write the rules by using Boolean [IF-THEN] logic. The rules relate the decisions to the questions. Identify the questions [called decision factors] asked of the user.

Limitations Difficult to build an expert system – intellectually demanding. Best suited for learners of 13+ years. Appeals to learners with abstract reasoning ability. Dependent on high level of intellectual development and motivation.