This project addresses an aspect of problem solving that is key for both machines and humans: reasoning by analogy with relevant precedents. The goal of the proposed research is a computational model of precedent-based analogical reasoning using a simple but expressive and cognitively plausible suite of representations that capture knowledge about entities, classes, causes, values, and motion along real and abstract trajectories. The project's target domain is political science, where in areas such as intelligence analysis, international relations, and government policy, experts use this type of case-based reasoning to analyze, assess, and make recommendations about new situations based on relevant past experience. The project, which builds on analogy research begun in the NSF-sponsored Bridge System, will make use of collections of well-documented political scenarios and analogies that have been analyzed by experts. While this project focuses primarily on development of a computational theory about reasoning by analogy from experience, a prototype tool will be built that is intended to assist analysts charged with evaluating courses of action by addressing an important aspect of analytical expertise: the ability to detect unintended consequences with precedential reasoning. This "blunder stopper" component of the prototype will monitor situations and proposed actions, match these against a data base of precedents, and, if appropriate, advise the analyst to examine a relevant precedent for possibilities suggested by the precedent that might otherwise be overlooked. In addressing a problem of interest to both computer science and social science, this project will contribute to theories of precedent-based reasoning in general, and course of action analysis in political science in particular.
