This paper describes a method for reasoning by analogy, which is intended to provide the basis for one of the components of an AI system called ARCHES. This component is to be used whenever the deductive component of ARCHES is unable to solve a given problem. A key part of the method involves the introduction of a hypothesis which states that a certain dependency exists between parts A and B of a known situation P that is to be used as a basis for analogy. Now, if a similarity relationship is established between A and a part C of a new situation T, then the dependency hypothesis can be used to infer from C a new part D in T. This mode of reasoning can be seen as inference by analogy. It depends strongly on the dependency hypothesis, and on the availability of methods for acquiring, modifying, and selecting such hypotheses. This is a promising approach to “approximate reasoning,” and it is being explored currently by several investigators in the area of analogical reasoning.
While the paper discusses in some detail the issue of establishing similarity between parts of two situations, it does not deal with the issue of how dependency hypotheses come into being, and how they are selected by a system in the course of solving a problem. Also, the thorny issue of how a system decides that a problem cannot be solved by conventional deduction methods is not discussed. Some of the notation and formalisms used in the paper tend to obscure the main ideas. However, this is compensated for by a few good examples that illustrate the proposed approach quite effectively.