“Circumscription - A Form of Non-Monotonic Reasoning” was written by John McCarthy of the Stanford Computer Science Department and published in the journal Artificial Intelligence, volume 13, in 1980 (pages 27 to 39). The copy McCarthy kept on his Stanford site is a 1986 retypeset version with the same content. The paper tackles a gap that ordinary formal logic leaves open: classical logic is monotonic, meaning that adding new facts can never cancel a conclusion you already drew, yet common-sense reasoning constantly does exactly that.
McCarthy’s example, drawn from the start of the paper, is using a boat to cross a river. Stating every condition that must hold (the oars are present and unbroken, the rowlocks fit, and so on) is hopeless, because one can always think of another qualification. He calls this the qualification problem. Circumscription is a rule of conjecture for “jumping to certain conclusions”: the objects that can be shown to have a property by reasoning from the known facts are assumed to be all the objects that have it. In effect the reasoner circumscribes, or minimizes, the set of exceptional cases, and is prepared to retract that assumption if a later fact reveals an exception.
This idea launched a major research line. Because the conclusions can be withdrawn when new information arrives, circumscription is “non-monotonic,” and it sits alongside Raymond Reiter’s default logic and Jon Doyle’s truth maintenance work as one of the founding formalisms for reasoning under incomplete knowledge. It also gave logicians a precise handle on closed-world assumptions, where everything not known to be true is taken to be false.
Why a business reader should care: any system that must act on incomplete information, from a diagnostic tool to a planning agent, faces McCarthy’s qualification problem, and circumscription is the classic statement of how a rule-based system can make reasonable default assumptions while staying ready to revise them.