What Are Expert Systems? A KS3 Computing Guide

What are the main components of an expert system?

Every expert system has three core parts:

1. Knowledge base — a structured collection of facts ("A patient with a temperature above 38°C has a fever") and rules ("IF fever AND cough AND shortness of breath THEN suspect pneumonia"). Human experts are interviewed at length to build this; the process is called knowledge acquisition.

2. Inference engine — the reasoning component that applies the rules in the knowledge base to a specific set of inputs. It works through a chain of IF–THEN conditions to arrive at a conclusion. Think of it as the "brain" that connects the dots.

3. User interface — the system asks the user a series of questions and presents its conclusions, along with an explanation of how it reached them. This explanation facility is unusual in AI — most programs cannot say why they gave an answer.

Some expert systems also include a knowledge acquisition module that helps add new rules and facts without requiring a programmer.

How does an inference engine reason?

Inference engines typically use one of two strategies:

Forward chaining example: The system knows the patient has a fever and a cough. It applies the rule "IF fever AND cough THEN suspect respiratory infection." This new fact triggers another rule, and so on, until a diagnosis emerges.

Backward chaining example: The system asks "Could this patient have flu?" It then works backwards: does the patient have a fever? Muscle aches? A sudden onset? It asks questions to confirm or reject each criterion.

What is a classic example of an expert system?

MYCIN (developed at Stanford University in the 1970s) is the most famous medical expert system. It diagnosed bacterial infections and recommended antibiotic treatments for critically ill patients. MYCIN contained roughly 600 rules of the form:

IF   the organism is gram-negative AND
     the organism is rod-shaped AND
     the patient has a urinary tract infection
THEN there is a 85% probability the organism is E. coli

In clinical tests, MYCIN performed comparably to specialist physicians and outperformed junior doctors — a landmark result that helped establish expert systems as a serious field.

Other well-known examples:

• DENDRAL — identified chemical compounds from mass spectrometry data.
• PROSPECTOR — assessed geological data to predict mineral deposits.
• Credit scoring systems — banks still use rule-based systems to decide whether to approve loans.

What are the advantages of expert systems?

• Consistency — the system applies the same rules every time, without fatigue, mood, or distraction.
• Availability — a specialist's knowledge can be made available 24 hours a day, even in remote locations.
• Explanation — unlike a black-box neural network, an expert system can show exactly which rules led to its conclusion, making it auditable and trusted in regulated industries.
• Knowledge preservation — expertise is captured in the knowledge base and does not retire when an expert does.

What are the limitations of expert systems?

• Brittle outside their domain — an expert system for diagnosing respiratory disease knows nothing about fractures. It cannot reason outside its programmed rules.
• Expensive to build — extracting expert knowledge and encoding it as rules is time-consuming and costly. Rules may also be incomplete or contradictory.
• Cannot learn — a traditional expert system does not update itself from new cases; the knowledge base must be maintained manually.
• Struggle with uncertainty — real-world problems often involve ambiguity that rules handle poorly. Machine learning approaches now handle uncertain, unstructured data (images, text) that rules cannot.

How do expert systems differ from modern AI?

Expert systems dominated AI in the 1980s. Machine learning displaced them for many tasks in the 1990s and 2000s. Today, hybrid systems combine rule-based reasoning with statistical learning.

Frequently asked questions

Are expert systems still used today?

Yes — particularly in fields where decisions must be explainable. Credit-scoring systems, fraud detection, medical diagnostic aids, and tax software all incorporate rule-based logic that descends from expert-system research. Pure expert systems are rarer; most modern deployments blend rules with statistical models.

Why can't an expert system teach itself new rules?

A classic expert system stores only the rules explicitly programmed into it. It has no mechanism to observe outcomes, update its knowledge, or discover new patterns. This is by design: in regulated fields like medicine and law, you need to know exactly where every rule came from. Machine learning systems that update automatically are harder to audit, which is why rule-based systems survive in compliance-heavy industries.

What is a fuzzy expert system?

A fuzzy expert system extends the basic IF–THEN framework to handle uncertainty. Instead of "temperature > 38°C" being either true or false, fuzzy logic allows "temperature is somewhat elevated" with a degree of truth between 0 and 1. This makes the system more flexible for cases where boundaries are not sharp — for instance, deciding whether traffic flow is "light", "moderate", or "heavy".

How would I describe an expert system in a KS3 exam answer?

A strong answer explains: what an expert system does (imitates a specialist's reasoning), its two main components (knowledge base and inference engine), one real-world example (medical diagnosis, loan approval), and one limitation (cannot learn, brittle outside its domain). Mentioning that it can explain its reasoning is a useful higher-mark point that distinguishes expert systems from other AI approaches.

---

Want to explore how AI makes decisions — from expert systems to machine learning? Professor Turing at aitutors.me can guide you through the whole landscape.