Understanding methods of problem solving is a main goal of both Cognitive Science and Artificial Intelligence. Although general problem solvers that use weak methods have been developed, they are not sufficient for reasoning about complex problems in complicated domains. For such tasks, the search is generally intractable: the height of the search tree is determined by the complexity of the solution, and the branching factor at each level is determined by the large number of applicable operators. The principle problem is that weak methods operating on large domain theories provide the problem solver with only a very limited notion of which operators might be relevant to which goals. What is needed to solve this problem, then, is a set of learning methods that can select and retrieve past experiences that are relevant to the current goal.
Jan-11-2006, 07:50:09 GMT