integrating Marker Passing and Problem Solving: A Spreading Activation Approach To Improved Choice in Planning

One standard problem in artificial intelligence is that of making choices during the planning process. Planning required traversing a large space of partially ordered subtasks; thus, search limitations often prohibit a planner from taking advantage of existing information to make an optimal choice. The usual method of coping with this is to keep an agenda and to provide backtracking or do replanning when a plan fails. We describe how the technique known as "marker-passing," a parallel, nondeductive, activation-spreading algorithm, can be used to enhance the choice mechanism in a problem-solving system. We discuss the integration of a marker-passer into a problem-solving system and the design of such a combined system: SCRAPS. We present details about the organization of SCRAPS and how the marker-passer and problem-solver interact. We discuss the computational implementation of such systems, including issues relating to parallel computation and concurrency. How the program handles several examples is shown in detail. We also describe how this work might be of use to the cognitive scientist in its relation to spreading-activation models of memory and propose some possible psychological experiments to explore the role of spreading activation in the planning process.