Creating Causal Models

The Problem The task of developing qualitative and causal reasoning systems to perform problem solving on physical systems has two aspects : (1) Designing representations for structure, behavior, and causality within which to describe the physical systems of interest and their constituent objects and processes, and (2) Developing algorithms which operate on the chosen representation to efficiently draw inferences to perform the given problem solving task-simulation, troubleshooting, etc. Embedded in this methodology is an assumption that complete .models for the physical systems of interest are known and can be rendered in the chosen representation. My research concerns the problem which arises when this assumption is removed: How to create the models of physical systems on which problem solving can be performed. A more specific statement of my research problem is: Given (1) a description of the externally observable structure and behavior of a physical system, and (2) a'vocabulary of primitive mechanism. schemata which provide causal explanations for specific kinds of behavior, the task is to hypothesize configurations of mechanisms which provide a complete causal account of the observable behavior of the physical system. This causal modelling problem can be cast as theory formation-the physical system being modelled is considered static and the task is to construct and incrementally refine a description of the device which explains an ever-growing set of examples of the behavior of the device. Further more, each hypothesis is a proposed design-the observable behavior serves as-a set of functional specifications and the task is to construct a configuration of mechanisms from a given vocabulary which realizes those specifications. Clearly, there are many possible designs for a given device and some means of constraining hypotheses must be employed in order to make the problem tractable. This issue is the main focus of my work .