The frame problem is a difficulty in getting a logic-based AI to reason about a changing world, first named by John McCarthy and Patrick Hayes in their 1969 paper “Some Philosophical Problems from the Standpoint of Artificial Intelligence,” hosted on McCarthy’s Stanford site. The paper introduced the situation calculus, a formal way of describing how the state of the world changes when actions are performed, and the frame problem fell out of it.
The puzzle is this: when an action changes one thing, almost everything else stays the same, but a logical system has no built-in way to know that. If a robot moves a block, the system can deduce the block’s new position - but it must also somehow conclude that the color of the walls, the location of every other object, and countless other facts did not change. Writing out a “frame axiom” for every fact that each action leaves untouched is hopeless, because the number of such non-effects is enormous. The technical frame problem is how to represent what does not change without stating it all explicitly.
In its narrow form the frame problem is a representation issue that AI researchers later addressed with techniques such as nonmonotonic reasoning and the commonsense law of inertia, which assumes facts persist unless an action is known to change them. Philosophers, notably Daniel Dennett, broadened the term to a deeper worry: how does any mind decide which of the limitless facts about a situation are even relevant to consider, without checking them all. In that broad sense the frame problem became emblematic of why common-sense reasoning is so hard to mechanize.
Why business readers should care: the frame problem is one concrete reason the rule-based “good old-fashioned AI” of the expert-systems era struggled with the open, messy real world, and it helps explain why robust common-sense reasoning remains a hard frontier even for today’s systems.