Multi-modal Cognitive Architectures: A Partial Solution to the Frame Problem

Since its definition by McCarthy in 1969, the Frame Problem (FP) has been one of the more heavily debated problems in AI. Part of the debate has been on the exact definition of what the FP really is. The computational aspect of the FP can be thought of as reasoning about what changes and what doesn’t change in a dynamic world. The “sleeping dog strategy” is considered to be a viable solution to this aspect of the FP. We intend to show that this strategy has a weakness that can be partially solved using diagrammatic reasoning, under certain conditions. A related and equally important problem, called the Ramification Problem, is to be able to reason about the indirect effects of an action in the world. Our proposal provides a more efficient solution to the Ramification Problem when reasoning about spatial relations. To illustrate our solution, we introduce a problem solving architecture based on Soar that is augmented with a diagrammatic reasoning component. A problem state in this augmented Soar is bi-modal in nature, one part being symbolic and the other diagrammatic. We describe its use in certain problems and show how the use of diagrams can handle the frame and ramification problems with respect to spatial relations.