Yumi Iwasaki

Causality in device behavior

Abstract This paper shows how formal characterizations of causality and of the method of comparative statics, long used in econometrics, thermodynamics and other domains, can be applied to clarify and make rigorous the qualitative causal calculus recently proposed by de Kleer and Brown [2]. The formalization shows exactly what assumptions are required to carry out causal analysis of a system of interdependent variables in equilibrium and to propagate disturbances through such a system. The intuitive concepts of causality captured by de Kleer and Brown provide a rough approximation to the standard analytic techniques that are used in the treatment of simultaneous algebraic and differential equations.

The concept and implementation of skeletal plans

A new method for automated planning, refinement of skeletal plans, has been developed for the problem of experiment design in the domain of molecular biology. The method resulted from a study of the problem-solving behavior of scientists which showed that design usually consisted of look-up of abstracted plans followed by hierarchical plan-step refinement. The skeletal plan method has been implemented through two generations of problem-solving systems, the second generation involving a synthesis with the metaplanning approach of Stefik.

Causality and model abstraction

Abstract Much of science and engineering is concerned with characterizing processes by equations describing the relations that hold among parameters of objects and govern their behavior over time. In formal descriptions of processes in terms of parameters and equations, the notion of causality is rarely made explicit. Formal treatments of the foundations of sciences have avoided discussions of causation and spoken only of functional relations among variables. Nevertheless, the notion of causality plays an important role in our understanding of phenomena. Even when we describe the behavior of a system formally in terms of acausal, mathematical relations, we often give an informal, intuitive explanation of why the system behaves the way it does in terms of cause-effect relations. In this paper, we will present an operational definition of causal ordering. The definition allows us to extract causal dependency relations among variables implicit in a model of a system, when a model is represented as a set of acausal, mathematical relations. Our approach is based on the theory of causal ordering first presented by Simon [22]. The paper shows how to use the theory and its extension in reasoning about physical systems. Further, the paper studies the relation of the theory to the problems of model aggregation.

The mathematical bases for qualitative reasoning

The practices of researchers in many fields who use qualitative reasoning are summarized and explained. The goal is to gain an understanding of the formal assumptions and mechanisms that underlie this kind of analysis. The explanations given are based on standard mathematical formalisms, particularly on ordinal properties, continuous differentiable functions, and the mathematics of nonlinear dynamic systems.<<ETX>>

Real-world applications of qualitative reasoning

Qualitative reasoning has attracted much interest from the artificial intelligence research community in the last decade. The emphasis in qualitative reasoning on making inferences about the behavior of physical systems where there is incomplete information makes the technology relevant to many real-world industrial problems. Many applications of qualitative reasoning technology have emerged for such tasks as diagnosis, design, tutoring, real-time monitoring and hazard identification. This article showcases some of these recent applications.

Automated model selection for simulation based on relevance reasoning

Constructing an appropriate model is a crucial step in performing the reasoning required to successfully answer a query about the behavior of a physical situation. In the compositional modeling approach of Falkenhainer and Forbus (1991), a system is provided with a library of composable pieces of knowledge about the physical world called model fragments. The model construction problem involves selecting appropriate model fragments to describe the situation. Model construction can be considered either for static analysis of a single state or for simulation of dynamic behavior over a sequence of states. The latter is significantly more difficult than the former since one must select model fragments without knowing exactly what will happen in the future states. The model construction problem in general can advantageously be formulated as a problem of reasoning about relevance of knowledge that is available to the system using a general framework for reasoning about relevance described by Levy (1993) and Levy and Sagiv (1993). In this paper, we present a model formulation procedure based on that framework for selecting model fragments efficiently for the case of simulation. For such an algorithm to be useful, the generated model must be adequate for answering the given query and, at the same time, as simple as possible. We define formally the concepts of adequacy and simplicity and show that the algorithm in fact generates an adequate and simplest model.

Theories of causal ordering: reply to de Kleer and Brown

Abstract In their reply to our paper, “Causality in Device Behavior,” de Kleer and Brown seek to establish a clear product differentiation between the well-known concepts of causal ordering and comparative statics, on the one side, and their “mythical causality” and qualitative physics, on the other. Most of the differences they see, however, are invisible to our eyes. Contrary to their claim, the earlier notion of causality, quite as much as the later one, is qualitative and “derives from the relationship between the equations and their underlying components which comprise the modeled system.” The concepts of causal ordering and comparative statics offer the advantage of a formal foundation that makes clear exactly what is being postulated. Hence, they can contribute a great deal to the clarification of the causal approaches to system analysis that de Kleer and Brown are seeking to develop. In this brief response to their comments, we discuss the source of the structural equations in the causal ordering approach, and we challenge more generally the claim that there are inherent differences (e.g., in the case of feedback) between the “engineers” and the “economists” approach to the study of system behavior.

Causal functional representation language with behavior-based semantics

Understanding the design of a device requires both knowledge of the general physical principles that determine its behavior and knowledge of its intended functions. However, the majority of work in model-based reasoning has focused on using either one of these types of knowledge alone. In order to use both types of knowledge in understanding a device design, one must represent the functional knowledge in such a way that it has a clear interpretation in terms of observed behavior. We propose a new formalism, causal functional representation language (CFRL),for representing device functions with well-defined semantics in terms of behavior. CFRL allows the specification of conditions that a behavior must satisfy, such as occurrence of temporal sequences of events and causal relations among them and the components. We have used CFRL as the basis for afunctional verification program, which determines whether a behavior achieves an intended function.

Causal ordering, comparative statics, and near decomposability

Abstract The notion of causal ordering has been explicated for systems in the form of sets of linear algebraic equations and also for dynamic models consisting of linear differential equations. Since we often wish to aggregate or disaggregate models of systems or to consider the static equilibrium of a dynamic model, the question arises as to how these transformations from one model of a system to another affect the causal ordering of variables. The present paper provides some answers to this question.

Design Verification through Function- and Behavior-Oriented Representations

This paper focuses on the task of design verification using both knowledge of the structure of a device and its intended functions. In particular, it addresses the question of when one can say a behavior predicted by a prediction system achieves the desired function in the manner intended by the designer. We use Functional Representation (Sembugamoorthy & Chandrasekaran 1986) to represent the function of a device and the expected causal mechanism for achieving it. We present a formal definition of matching between a system trajectory generated by a simulation system and the description of a causal process to achieve a function expressed in Functional Representation. We demonstrate behavior verification based on the definition, using two predicted behaviors of the electrical power system of a satellite. We believe that evaluating a behavior with respect to the expected causal process as well as the function improves the chances of uncovering hidden flaws in a design that may otherwise go undetected at an early stage.