A class of inferential processes in which adding new premises can invalidate previously drawn conclusions; inference is defeasible rather than strictly monotonic.

Principle

Inference is guided by defeasible rules and preferences so that conclusions may be withdrawn when additional information conflicts with earlier assumptions; minimal change and specificity often govern which conclusions survive.

Demonstration

An agent assumes "birds fly" by default and concludes that a particular bird b can fly; after learning b is a penguin (an exception to the default), the agent withdraws the earlier conclusion that b flies.

Misapplication

Treating nonmonotonic reasoning as mere uncertainty quantification and using it to justify persistent contradictions, or applying default withdrawals without respecting priorities and context, producing unstable or oscillating conclusions.

Consequence

Supports flexible, context-sensitive inference appropriate for common-sense reasoning and many AI applications, enabling systems to make tentative conclusions that are retractable when new facts arrive.

Reversal

Monotonic reasoning: conclusions accumulate monotonically and are never invalidated by adding further premises; once entailed, a conclusion remains entailed.

Boundary

Applies to logics and systems that permit defeasible entailment (default logic, defeasible logics, circumscription etc.); does not cover purely monotonic deductive systems, though it may interact with probabilistic update mechanisms.

Semantic Tension

Tension with probabilistic or Bayesian reasoning: both handle uncertain knowledge but differ in mechanism (defeasible logical rules versus degrees of belief); tension also with belief revision which focuses on change of an agent's belief set rather than local defeasible inferences.

Synthesis

Nonmonotonic reasoning is the family of rule-based, defeasible inference procedures that permit provisional conclusions to be withdrawn under new information, organized by priorities, specificity, and minimal-change principles to model commonsense defeasible inference.