Task and System Overview

Our system's task domain is disaster response planning. Disaster response planning is the initial strategic planning used to determine how to assess damage, evacuate victims, etc., in response to natural and man-made disasters such as earthquakes and chemical spills. Disaster response planning is an ill-structured real-world domain, and is a domain for which humans have been shown to depend heavily on prior experiences when they address new problem situations [Rosenthal et al.1989], but the lessons of those experiences must often be revised to fit the new experience. For example, when a earthquake occurs in Liwa, Indonesia, the prevention of looting is a relevant problem. Fixing the problem using a prior earthquake in Los Angeles does not work as the national guard was dispatched to quell potential looting in Los Angeles and Indonesia has no national guard. The solution is to adapt the old solution to use the Indonesian army, rather than the national guard.

Our testbed system, DIAL, processes a conceptual representation of a news story describing the initial events in a disaster, and proposes a response plan by retrieving and adapting the response plan for a similar prior disaster. After an initial story is input, DIAL's basic processing sequence is as follows:

• Candidate response plan cases for similar problem situations are retrieved using coarse-grained static similarity assessment techniques.
• A finer-grained similarity assessment process uses learned information about adaptation to select the response plan case with the smallest expected adaptation cost.
• Applicability problems in the response plan of the selected case are repaired by case adaptation. DIAL learns by storing traces of its case adaptation and memory search processes used to find the needed information.
• The resulting response plan case is stored for future reuse.