AI Glossary / AI and machine learning / Intermediate
Commonsense knowledge (artificial intelligence)
Definition
Commonsense knowledge in AI is background knowledge about everyday objects, events and relationships that people often leave unstated. It helps a system interpret situations beyond the literal information explicitly supplied in a sentence or record.
What it means
When someone says they put ice cream in a freezer, most readers need no explanation of why. They connect ordinary knowledge about temperature, melting and food storage. Software may need that background represented or learned before it can make a useful inference.
Commonsense knowledge can concern physical behavior, typical uses, locations, intentions and social expectations. It is not all universally true. A common expectation can have exceptions, vary by context or reflect a cultural assumption.
The distinction between knowledge and reasoning matters. Storing that umbrellas are used for rain does not by itself provide a reliable method for deciding what to do in every weather-related situation.
How it works
One approach represents concepts and relationships explicitly in a knowledge graph. For example, a relation might connect an object with a typical use or location. Resources such as ConceptNet collect many such assertions.
Another approach learns regularities from language and other data. A system may combine learned representations with retrieved knowledge or explicit rules.
The challenge is applying background information appropriately. Systems need to distinguish typical from guaranteed, relevant from distracting and well-supported knowledge from unreliable assertions. Evaluation should include exceptions and cases where the everyday default is overridden by explicit evidence.
Three practical examples
- A search tool connects a query about keeping rain off a person with products used for that purpose.
- A household-planning assistant recognizes that a frozen ingredient needs a preparation step before certain recipes.
- A document system uses ordinary relationships among objects and activities to help interpret short, underspecified descriptions.
Think of it this way
Think of the unwritten background that makes a conversation efficient. People omit details because they expect listeners to fill in familiar connections.
The analogy has limits. Two people may not share the same background, and a computer can reproduce common associations without judging their reliability well. A plausible completion is not necessarily a justified inference about the particular situation.
Worked example
Using a default without ignoring an exception
Consider the invented instruction: “Put the picnic ice cream somewhere it will stay frozen.” A system with relevant background knowledge may suggest a freezer or a suitably maintained insulated container.
Now add: “The freezer is broken and currently warm.” A system that keeps recommending that freezer is applying a typical association while ignoring decisive local evidence.
The reasoning process should separate the general expectation from the observed exception. It should ask about an available working alternative rather than treating the word freezer as proof of a suitable temperature.
This is not a food-storage recommendation; it is a teaching example about knowledge use. The broader point applies to business records: a supplier's usual capability, a department's typical responsibility or a product's common use can be overridden by the specific facts of the case.
Illustrative case study
A fictional office assistant helps staff locate equipment. Its background knowledge suggests that projectors are often kept in meeting rooms. When asked for a particular projector, it confidently names a room without consulting the inventory.
The team changes the workflow so general knowledge helps interpret the request, while the inventory determines the actual location. If the inventory is missing or outdated, the assistant says that the location is uncertain.
The case illustrates the difference between a useful expectation and a fact about a named object. Commonsense can guide a search, but it should not replace authoritative operational records.
Common misconceptions and mistakes
- Treating common knowledge as exception-free: many everyday statements describe tendencies.
- Confusing fluency with reliable reasoning: a natural explanation can still apply the wrong background assumption.
- Assuming one culture's expectations are universal: social defaults need context.
- Using a general association as evidence about a specific entity: a typical location does not establish the current location of a particular item.
Business use
Commonsense knowledge can improve search, dialogue and interpretation of short requests. It is especially relevant when users omit context they consider obvious.
Ask which knowledge sources a system uses and how it handles contradictions with current business data. Include explicit exceptions in testing. A useful assistant should let authoritative records override general expectations and should distinguish a suggestion from a verified fact. This is often more important than the size of its background knowledge collection.
People also ask
Is commonsense knowledge the same as expert knowledge?
No. It usually concerns broad everyday expectations rather than specialized domain facts.
Is a knowledge graph required?
No. Systems can represent or learn background knowledge in several ways.
Does knowing many facts guarantee commonsense reasoning?
No. Selecting and applying the right information remains a separate challenge.
Check your understanding
A system knows that printers are usually in offices. Can it confidently state that a named missing printer is in an office?
Answer: No. That is a general expectation, not evidence of the specific printer's location. It should consult relevant records or state uncertainty.
Related terms
- Commonsense reasoning
- Knowledge representation
- knowledge graph
- default reasoning
- ConceptNet
Learn next
Read knowledge representation. Write three everyday assumptions in your workflow and an exception to each, then test whether an assistant respects the exception.
More in AI and machine learning
- Co-training - Co-training is a semi-supervised method where two models learn the same task from two different views of the data, each labeling examples it is confident...
- Cognitive robotics - Cognitive robotics is the study and engineering of robots that connect perception, learning, reasoning and action to operate adaptively in their surroundings.
- Coherent extrapolated volition - Coherent extrapolated volition, or CEV, is a proposal in AI-alignment thinking that an advanced AI should be guided by what people would collectively want...
- Complex inverse Wishart distribution - The complex inverse Wishart distribution is a probability distribution over complex-valued Hermitian positive-definite matrices.
- Complex Wishart distribution - The complex Wishart distribution describes random matrices formed from sums of outer products of suitable complex Gaussian observation vectors.
- Computational intelligence - Computational intelligence is a broad field of methods for building adaptive computational systems, especially neural networks, fuzzy systems and...
Sources
ConceptNet project: https://conceptnet.io/. MIT Media Lab, ConceptNet overview: https://www.media.mit.edu/projects/conceptnet/overview/. Project API relations documentation: https://github.com/commonsense/conceptnet5/wiki/API. Primary project excerpts consulted; scenarios are original. The knowledge-graph description and typed relation structure were checked against the MIT project overview and official API documentation.
Previous: Commonality analysis Next: Commonsense reasoning