Problem Spaces

All cognitive tasks in Soar are formulated in terms of problem spaces. The decision to treat all problems as search in a problem space was driven by the problem space hypothesis. A problem space, according to the original definition, is a state search space including an initial state, a set of final (or goal) states, and operators for transforming states to other states. The primitive functions in Soar are all concerned with the manipulation of the problem space and its components. These primitive functions are:

Select a Problem Space;
Select a State from those Directly Available;
Select an Operator; and
Apply an Operator to Obtain a New State.

These functions correspond to the decisions made at the conclusion of each iteration of Soar's decision cycle.

The search in the problem space is distinguished from knowledge search. Knowledge search occurs within the long-term memory and refers to knowledge that can be brought to bear in implementing operators in the problem space and directing the state-space search. The use of this knowledge search in the search for a solution in the problem space results in great flexibility. When little task-specific knowledge is available, Soar's default knowledge enables a methodological search of the state space. As more knowledge is available (perhaps through the experience of successfully completing the search and "chunking" its result) and is brought to bear, the system behaves routinely, proceeding directly to the goal with no problem space search.

The use of problem spaces is an example of Soar's use of uniform, orthogonal mechanisms. Problem spaces enable two specific properties of the Soar system:

To return, press HOME. To go to the next document, press NEXT.