Modelling with ContextDependent Causality Maria Lee.pdfVIP

Modelling with Context-Dependent Causality Maria Lee* Paul Compton† Bob Jansen* * CSIRO Division of Information Technology, P.O. Box 1599, North Ryde, AUSTRALIA, fax: +61-2-888-7787, e-mail: lee@syd.dit.csiro.au † Artificial Intelligence Lab, School of Computer Science and Engineering, University of New South Wales, AUSTRALIA, e-mail: compton@spectrum.cs.unsw.oz.au A paper presented at JKAW92, 12-13 November, 1992, Japan Abstract Decision support systems may require the use of existing complex mathematical models. It is desirable to reduce the equations of such a model to an explanatory causal form to support decision making. It is argued here that a combination of extracting information from the user and heuristics to reduce the equations to a suitable form are sufficient to produce a reasonable causal explanation from equations. The method identifies causality in a physical system with or without feedback. The method also allows consistency checking between equations and user generated causal reasoning. This method has been implemented in Hypercard on a Macintosh to provide an accessible graphically oriented decision support tool. 1. Introduction Experts are often requested to make judgements and give advice in the context of adoption or moderation of future change. For example, political decisions are needed on the human-induced greenhouse warming of the Earth, and an expert adviser is required because of the complexity of the problems. Large mathematical models can be used to simulate the greenhouse effect, but these cannot directly support decision making for reasons to be specified. The need to produce a