— Ch. 1 · Origins And Definition —
Frame problem.
~4 min read · Ch. 1 of 6
John McCarthy and Patrick J. Hayes published a paper in 1969 titled Some Philosophical Problems from the Standpoint of Artificial Intelligence. This document introduced the frame problem as a core issue within artificial intelligence research. The authors described how traditional first-order logic requires many axioms to express facts about a robot in the world. These axioms simply imply that things in the environment do not change arbitrarily. For example, a block cannot change position unless it is physically moved. The formal mathematical problem became a starting point for more general discussions of knowledge representation difficulties. Issues such as providing rational default assumptions emerged alongside this technical challenge. What humans consider common sense in a virtual environment also came under scrutiny during this period.
Logical Representation Challenges
A scenario with a door and a light illustrates the difficulty clearly. The door can be open or closed while the light can be on or off. These conditions are statically represented by two propositions at time zero. If these conditions change over time, they require predicates called fluents that depend on time. A domain where the door opens at time one but the light stays off needs specific logical formulae. Three initial formulae represent the situation and the effect of opening the door. Yet these three formulae alone do not suffice to correctly draw consequences. Another set of conditions remains consistent with those same three formulae yet contradicts expectations. Specifying only which conditions change does not entail that all other conditions remain unchanged. One frame axiom becomes necessary for every pair of action and condition where the action does not affect the condition. Formalizing a dynamical domain without explicitly specifying these axioms proves impossible using standard logic alone.