Common-Sense Reasoning

Common-sense reasoning uses background knowledge to interpret situations and revise defeasible conclusions. In Some Expert Systems Need Common Sense, John McCarthy distinguishes specialized competence from reasoning about changing circumstances, other agents, and a system’s limitations.^[1]

This is relevant to Distributional Shift because narrow rules may omit facts needed outside their intended setting. Common sense has no single universally accepted formal definition; this historical account is one research program.

Defeasible planning

Circumscription—A Form of Nonmonotonic Reasoning formalizes assumptions that can be withdrawn when new facts arrive. A boat may be treated as usable when no obstacle is known; that depends on which facts were considered. A planner can seek more facts before acting. This connects practical planning to the reliability of background assumptions, without guaranteeing their completeness.^[2]

Consequences and shared knowledge

McCarthy’s 1984 account separates recommending an action from predicting what follows. An adviser may need to model concurrent events and the knowledge of the person receiving its advice. In the paper’s discussion, he leaves open whether MYCIN requires common sense itself when a physician supplies the missing judgment. This makes the boundary between human and program part of the reliability question.^[1]