Impasse-Driven Processing

An impasse in cognitive architectures is defined to be an inability to proceed with problem solving because of a lack of knowledge. During the course of normal problem solving, an agent will perceive the environment, examine its knowledge and goals, and formulate a response. However, if it is unable to generate a response, impasse-driven architectures automatically cast this lack of knowledge as a subgoal to be resolved, and attempts to find knowledge relevant to this subgoal. It is possible to again fail to progress on this subgoal, leading to another impasse and another subgoal. As more impasses occur in subgoals, processing becomes a stack of subgoals, with each goal waiting for its subgoal to resolve its impasse. If the same goal generates several subgoals over time, the problem solving trace attains a tree structure.

The ability to automatically generate subgoals improves the flexibility of the architecture and affords a natural opportunity to learn. An example of an architecture which is unable to impasse is Ralph. Partially as of result of this limitation, Ralph's only search mechanism is forward chaining, and it is unable to learn. The following is a list of architectures that solve problems by resolving impasses:


Back to List of Other Properties