What is Nonmonotonic Logic?
Nonmonotonic logic (NML) is a family of formal systems designed to model defeasible reasoning—the everyday practice of drawing tentative conclusions from incomplete or uncertain information, and revising them when new evidence arrives. Unlike classical logic, which guarantees that once a conclusion is derived, it remains valid regardless of additional premises, nonmonotonic logics allow conclusions to be retracted when exceptions or contradictions emerge. Here are some examples:
- You read of a bird called Tweety. You imagine it to be able to fly. But after learning that it is a Kiwi, you retract the inference.
- In the morning, you look out of the windows and see that the streets are wet. You infer that it rained over the night. Afterwards your partner tells you that the streets are cleaned in the early morning on Monday. It is Monday. You retract your inference that it rained.
- You get stranded on an island. It is inhabited by a type of bird you’ve never seen before. As the years go by, you see more and more of these birds and they are all blue, which is why you call them bluebirds, for the lack of a better name. You infer, that bluebirds are always blue. However, one day you see green and very old bluebird, which is when you revise your statement to “bluebirds are blue, unless they are very old.”
Why Classical Logic Isn’t Enough
Classical logic is strictly monotonic and truth-preserving: adding information never invalidates prior results. This works beautifully in mathematics, but fails in real-world contexts where knowledge is partial, rules admit exceptions, and conflicts are common. For example, classical logic cannot naturally represent “Birds typically fly”, unless it lists every possible exception (penguins, injured birds, etc.) which is not natural, and might in many cases of defeasible inference even be impossible. Nonmonotonic logics solve this by treating rules as defaults that hold unless defeated by more specific or overriding information.
Purpose & Applications
NML formalizes how agents—humans or other animals(?) or AI—reason adaptively under uncertainty. It is foundational in philosophy, cognitive science, and AI, powering applications like medical diagnosis, legal reasoning, knowledge representation, planning, and belief revision. Modern NML frameworks typically operate through three complementary approaches: formal argumentation (building and resolving conflicting arguments), consistent accumulation (stepwise adding plausible information while avoiding contradictions), and semantic methods (ranking possible worlds by normality or plausibility).
Literature
If you want to find out more about this topic you can check out Christian’s entry in the stanford encyclopedia or his recent book on the Cambridge Element series.
