Is default logic probabilistic?

Not inherently. Consistency-based permission to assume a statement is different from assigning it a numerical probability.

Does a more specific default always override a general one?

Not automatically in the basic framework. The representation or an additional prioritization scheme must support the intended behavior.

Is a default conclusion certain?

It is supported under the selected default theory and acceptance policy, but it may be defeated by new information.

AI Glossary / Graphs and symbolic AI / Advanced

Default logic

Definition

Default logic is a formal approach to reasoning with assumptions that may be withdrawn when additional information becomes available. In Reiter's framework, ordinary facts are combined with default rules whose use depends on consistency conditions.

What it means

People often reason from what normally holds while recognizing exceptions. A library item might normally be available for borrowing unless there is evidence that it is restricted. This is different from an absolute rule saying that every library item can be borrowed.

Default logic makes that distinction explicit. A conclusion can be supported provisionally without being a classical logical consequence of the facts alone. Learning an exception can remove that support.

This property is called nonmonotonicity: adding information can reduce the conclusions accepted by the default reasoning system. It does not mean that ordinary deduction has become invalid or that a conclusion can be changed arbitrarily.

How it works

A default theory contains background facts and default rules. A typical default has a prerequisite, a justification that must remain consistent, and a conclusion. A normal default uses the same formula for its justification and conclusion.

The accepted beliefs are described through extensions. An extension includes the background theory, its deductive consequences and the conclusions of defaults applicable under the framework's conditions. Consistency checks concern the candidate resulting belief set, not simply a temporary list of facts before rules are applied.

There can be several extensions, and general default theories can have none. Credulous reasoning accepts a statement supported in at least one extension; skeptical reasoning requires support in every extension. When no extension exists, an implementation needs an explicit handling policy rather than treating the situation as an ordinary successful inference.

Three practical examples

Knowledge representation: A system expresses typical properties of objects while retaining explicit exceptions.
Workflow assistance: A reasoning layer proposes a usual processing route when no known condition blocks that assumption.
Model comparison: Researchers examine alternative belief sets produced by conflicting defaults to understand which conclusions are stable across interpretations.

Think of it this way

Think of penciling an expected arrangement into a shared calendar before all details are known. The penciled entry is usable but revisable when contradictory information arrives. The analogy captures provisional commitment; default logic adds precise conditions for justified belief and does not prescribe a human scheduling policy.

Worked example

Use one object, report R. The background fact says Ordinary(R). Define a normal default: if R is ordinary and Shareable(R) is consistent with the accepted beliefs, conclude Shareable(R).

With no other relevant facts or rules, the extension contains Ordinary(R), Shareable(R), and their classical consequences. Now add the explicit fact ¬Shareable(R). The default's justification is no longer consistent, so it does not establish Shareable(R). This is a withdrawal of a default conclusion after new information arrives.

For a second illustration, start with no facts about proposition P and two prerequisite-free normal defaults: assume P when P is consistent, and assume not-P when not-P is consistent.

There are two extensions: the deductive closure of {P} and the deductive closure of {not-P}. The first blocks the contrary default, and the second blocks the first. Neither extension contains both P and not-P.

P is credulously supported because it belongs to one extension, but it is not skeptically supported because it is absent from the other. The same is true of not-P. The system therefore exposes alternative coherent positions rather than combining contradictory conclusions into one supposedly settled answer.

Illustrative case study

Fictional case: Alder Catalogue maintains an assistant that suggests whether reference materials can enter an internal sharing queue. Its model distinguishes explicit sharing restrictions from assumptions used when records are incomplete.

A review finds that two defaults can support incompatible recommendations for a particular material type. The team stops presenting whichever result happened to be computed first as uniquely justified. It records the competing outcomes and routes unresolved cases for a policy decision.

When a restriction is later added, the system recomputes affected conclusions and explains which assumption lost support. The team also maintains a separate action policy: a provisional logical conclusion does not itself authorize release of a document. This case illustrates modeling and interface choices, not a production compliance system.

Common misconceptions and mistakes

A default is an ordinary implication with softer wording. Its consistency condition changes the inference behavior.
Absence of evidence for a statement proves its negation. Default assumptions require an explicit rule; ignorance alone is not classical proof.
Every default theory has one preferred extension. Multiple or absent extensions are possible.
Contradictory defaults automatically produce an inconsistent belief set. They can instead lead to separate extensions, as in the worked example.

Business use

Default logic is useful when a knowledge model needs to distinguish hard facts from defeasible assumptions. It can clarify what an automated recommendation relies on and which new evidence would change it.

For a business application, define how conflicting extensions are handled and whether recommendations use skeptical or credulous acceptance. Preserve the provenance of assumptions and recompute affected results when facts change. Priority schemes, exception specificity and action permissions require explicit design; their intended behavior should not be assumed merely because defaults are present.

Check your understanding

A theory has two extensions. Statement Q belongs to both; statement P belongs to only one. Which statement is skeptically supported?

Answer: Q is skeptically supported. P is only credulously supported, assuming those are the extensions under consideration.

Related terms

Default rules
Closed-world assumption
Belief revision
Nonmonotonic reasoning
default rule
extension
defeasible reasoning
skeptical reasoning
credulous reasoning

Learn next

Add a hard fact P to the conflicting-default example and determine which extension remains. Then explain why replacing both defaults with unconditional assertions would create a different problem. Finally, choose an acceptance policy for a low-stakes recommendation and justify it.

Sources

Reiter's original paper introduces the default-logic framework: https://www.sciencedirect.com/science/article/pii/0004370280900144

Rychlik's research paper states Reiter's extension conditions before examining variations: https://cdn.aaai.org/AAAI/1991/AAAI91-058.pdf

Reiter and Criscuolo discuss representation issues in default reasoning: https://www.cs.ubc.ca/tr/1980/tr-80-07

The report examples and Alder Catalogue case are original illustrations of the stated framework.

Previous: Deepfake Next: Default rules