A Unifying Semantics for Causal Reasoning about Action

Abstract

        One of the principal concerns in the research area of Reasoning about Action and Change is determining the ramifications of actions in changing environments. A particular tendency emerging in recent literature endorses the explicit incorporation of causal knowledge in logic-based action theories. It is argued that causal extensions not only enhance the expressive power of theories of action, but may also provide more concise and intuitive representations.

This dissertation investigates semantics for causal reasoning about action and change. It does so by exploring the role of several fundamental underlying principles, such as the Principle of Minimal Change and the Principle of Causal Change. The development of this work culminates in a general unifying semantics for a class of action theories represented by a number of recent and influential approaches.

We focus on three of the most prominent causal frameworks in the Reasoning about Actions literature: the causal systems with fixed-points suggested by McCain and Turner, the causal relationship approach of Thielscher, and Sandewall's causal propagation semantics (also known as the transition cascade semantics). Each is studied via a semantics which includes a preferential component augmented with a causal relation.

The foregoing results are used to develop a general augmented preferential-style semantics that subsumes the causal systems with fixed-points and the causal relationship approach. The causal propagation semantics of Sandewall is shown to be a special case as well, characterised under certain uniformity assumptions.

The unifying general augmented preferential semantics, emerging as a result of this study, captures both Principles of Change and shows their clear and distinct roles - they are not inter-reducible but go hand-in-hand in causal action theories. Furthermore, the general semantics emphasises the role of contextual information affecting both minimality and causality, and provides a means for balancing different contributing factors. It is argued that hidden or less immediate forces shaping our motivating approaches become transparent with the help of the general semantics. In addition, it is hoped that the unifying semantics would provide further insights into views on causation and minimality, shared by these and other approaches.

Contact:
ict@csiro.au