The Frame Problem

The frame problem is the challenge, identified by John McCarthy and Patrick Hayes in 1969, of specifying what does not change when an action is performed in a formal system. In a logical representation of a world state, changing one fact ("the robot moved the block") requires explicitly asserting that everything else remained unchanged ("the block's color did not change," "the room's temperature did not change," "the robot's battery level did not change..."). The number of non-changes that must be explicitly represented grows combinatorially with the complexity of the world, making complete formalization computationally intractable.

The frame problem is not merely a technical obstacle to better knowledge representation. It is a diagnostic: it reveals that the representational paradigm — the assumption that intelligence requires explicit, propositionally structured knowledge about the world — must either be abandoned or supplemented with mechanisms that handle change implicitly. Connectionist approaches avoid the frame problem by not representing world states as discrete propositions at all; dynamic systems approaches handle it by treating change as continuous rather than discrete. The problem persists, however, in any system that must reason about action and change using symbolic representations.

In modern AI, the frame problem manifests in robotic systems that must plan sequences of actions without explicitly reasoning about every possible side effect, and in large language models that generate text without formal world models but still struggle with consistency across long contexts. The frame problem has not been solved. It has been bypassed — by systems that do not represent the world in the way that made the problem appear.