AI Glossary / AI and machine learning / Intermediate
Knowledge-based systems
Definition
Knowledge-based systems are computer systems that use explicitly represented knowledge to support reasoning, problem solving or decisions. They typically combine a knowledge base with mechanisms that apply that knowledge to facts or questions.
What it means
A knowledge base can contain facts, rules, concepts and relationships. An inference mechanism uses that representation to derive conclusions, identify relevant information or guide a task. This separates the domain knowledge from at least some of the procedures used to reason with it.
Expert systems are an important example, but the term knowledge-based system is broader. A system may support classification, configuration, explanation or search without attempting to reproduce an entire human expert's judgment.
Explicit knowledge is different from knowledge encoded only in learned model parameters. The distinction is useful, but real systems can combine both approaches. A learned model might identify an object, while a knowledge-based component applies compatibility rules. Neither representation is automatically superior for every problem; their suitability depends on the task and evidence.
How it works
Developers and domain specialists decide what knowledge to represent and in which formalism. Rules may use an if-then structure, while an ontology or semantic network can express categories and relationships. Facts about a particular case are supplied at use time.
An inference engine applies the available reasoning procedures. Forward chaining can start from known facts and derive consequences. Backward chaining can start from a goal and look for facts or rules that support it. Other representations use different inference methods.
Some systems can explain which rules or relationships supported a conclusion. That trace is useful but only as reliable as the underlying knowledge and input facts. Knowledge acquisition, validation and maintenance are therefore central activities. A correct inference from an obsolete rule can still produce an unsuitable real-world answer.
Three practical examples
1. A troubleshooting system applies expert-authored rules to reported symptoms and suggests the next diagnostic question.
2. A product configurator uses component relationships to reject incompatible combinations and explain required dependencies.
3. An information assistant uses a structured knowledge base to connect a person's role, organization and projects when retrieving relevant internal material.
Think of it this way
Think of a reference handbook combined with a disciplined procedure for applying it. The handbook supplies knowledge, and the procedure determines which parts matter for a case. The limit is that a computer representation may omit common sense and exceptions a human reader would supply without being told.
Worked example
Consider two simplified rules. Rule 1 says that if a device is battery-powered and its charge is below 10%, mark it as needing charging. Rule 2 says that if a device needs charging, suggest connecting an approved charger.
For device A, the facts are battery-powered and charge 7%. Since 7 is below 10, rule 1 derives “needs charging,” and rule 2 derives the charger suggestion. For device B at 30%, rule 1 does not fire from these facts.
The example is deliberately limited. Real equipment advice may require additional conditions, such as battery damage or charger compatibility. A visible rule chain explains the system's reasoning without proving that the rule set is complete.
Illustrative case study
A fictional facilities team builds a knowledge-based help tool for common equipment issues. Staff appreciate seeing the rule behind a suggestion, but one procedure changes after a new equipment version is introduced.
The team adds version conditions, assigns an owner to each rule group and keeps a set of representative cases for checking updates. When the system lacks enough information, it asks a targeted question instead of applying a rule outside its scope.
The case illustrates that transparency helps maintenance, but does not replace it. A knowledge-based system needs a process for keeping explicit expertise aligned with the organization's current equipment and practices.
Common misconceptions and mistakes
- A knowledge base is just a large document collection. Knowledge-based reasoning often requires explicit structure and defined relationships.
- An explanation proves a conclusion is correct. It shows a reasoning path whose facts and rules still need validation.
- Knowledge-based systems cannot use learning. Hybrid systems can combine explicit knowledge with statistical models.
Business use
Knowledge-based systems are useful where important rules or relationships can be stated clearly and where consistent application and explanation matter. They can support stable procedures and complement models that handle uncertain perception or prediction.
Begin with a bounded task and identify the knowledge owners. Test normal cases, exceptions, missing facts and contradictory inputs. Provide a clear path for updating or retiring rules. Evaluate the system's conclusions in real operating contexts rather than judging quality only by the number of facts or rules stored.
People also ask
Is every knowledge-based system an expert system?
No. Expert systems are one influential category within a broader family.
Does it learn automatically from use?
Not necessarily. Many systems require explicit knowledge updates, though learning components can be added.
What is an inference engine?
It is the component that applies reasoning procedures to the represented knowledge and case facts.
Check your understanding
A system correctly applies an old compatibility rule and recommends an unsuitable component. Where might the problem lie?
Answer: In outdated or incomplete knowledge, even if the inference engine itself followed the stored rule correctly.
Related terms
- Knowledge base
- Inference engine
- Expert system
- Forward-chaining
- Backward-chaining
- Ontology
- rule-based system
- neuro-symbolic AI
Learn next
Read inference engines and knowledge representation. Turn a small procedure into explicit facts and rules, then identify the exceptions that the first version fails to cover.
More in AI and machine learning
- Knowledge-based recommender system - A knowledge-based recommender system suggests items by reasoning about explicit user requirements and knowledge of how items meet those requirements.
- Knowledge compilation - Knowledge compilation is the transformation of a knowledge representation into another form that supports selected reasoning queries more efficiently.
- Knowledge integration - Knowledge integration is the process of combining knowledge from different sources into a form that can be used together coherently.
- Kolmogorov - Arnold Networks - Kolmogorov - Arnold networks, usually called KANs, are neural-network architectures that learn univariate functions on connections between nodes.
- L 1 regularization - L1 regularization is a penalty added to training that charges the model for the size of its weights, with the distinctive effect of pushing unneeded weights...
- L 2 regularization - L2 regularization is a penalty added to training that charges the model for the square of its weights, shrinking all weights toward zero to keep any single...
Sources
Carnegie Mellon University, The Scone Project, project documentation on knowledge representation, search and inference: https://www.cs.cmu.edu/~sef/scone/. ETH Zürich, Expert Systems and Knowledge-Based Systems teaching material: https://opess.ethz.ch/course/section-17-3/17-3-1-expert-systems-and-knowledge-based-systems/. The charging rules and facilities scenario are fictional teaching examples.
Previous: Knowledge-based recommender system Next: Knowledge citations