Model-based reasoning of device behavior with causal ordering
Yumi Iwasaki · 1988
Much of sciences and engineering is concerned with characterizations of processes by equations that describe the relations that hold among parameters of objects and that govern their behavior over time. Formal treatment of the foundations of sciences have avoided notations of causation and spoke only of functional relations among variables. Nevertheless, the notion of causality plays an important role in our understanding of phenomena. This thesis describes a computational approach, based on the theory of causal ordering, for inferring causality from an acausal, formal description of a phenomena. Causal ordering, first proposed by Simon, is an asymmetric relation among the variables in a self-contained equilibrium or dynamic model, which reflects people's intuitive notion of causal dependency relations among variables. The thesis extends the theory to cover models consisting of a mixture of dynamic and equilibrium equations. When people's intuitive causal understanding of a situation is based on a mixed description, the causal ordering produced by the extension reflects this intuitive understanding better than that of an equilibrium description. The thesis explores the use of the theory of causal ordering and its extension in reasoning about the behavior of physical systems. As the number of variables involved in the system under study gets larger and the system becomes more complex, abstraction becomes essential in reasoning about its behavior. Aggregation of a nearly decomposable dynamic systems is an abstraction technique proposed by Ando and Simon. The technique provides a formal justification for commonsense abstraction whose application is easily observable in everyday life. The thesis generalizes the method of aggregation and also shows a close connection between aggregation and causal ordering. For correct application of the method of causal ordering, the equations comprising the model of the device must be such that each of them stands for a conceptually distinct mechanism. The thesis also discusses the issue of building a model that meets this requirement and presents our solution of automatically generating equations from an explicit, network representation of processes taking place in the device.