A casual reasoning approach to behaviour-oriented design (original) (raw)

1994

Abstract

This thesis describes a basis for applying causal reasoning to the problem domain of design. Causal reasoning covers one of the fundamental aspects of reasoning humans perform when dealing with dynamic change. Such temporal features prevail in design problems which aim to provide artefacts based on some desired behaviour. This thesis first investigates causal reasoning to develop a language with expressiveness for representation and reasoning about change and for application to design. The second half of the thesis describes how this can be used for simple design tasks.The language for causal reasoning is based on the logic CI introduced by Yoav Shoham. CI is a nonmonotonic temporal logic for prediction, which prefers those models which are chronologically more ignorant, i.e., if changes were to take place they should happen as late as possible. It was presented as a method for dealing with frame problems. In this thesis, a framework for abduction is suggested for CI. It assumes that certain states persist not only forward, but also backward. This thesis argues that it is often the case that reasoning about the past events is in fact reasoning about the events between two or more time points. Problems of such nature are referred to as interpolation problems, and this thesis describes a way of dealing with this class of problems by bidirectional sweeping, which performs reasoning forwards and backwards over the time range. The outcome of this operation is a plausible sequence of events that took place in that time range.This method of causal reasoning is applied to design problems; by focusing on how de¬ vices work, we develop the notion of behaviour oriented design which aims to achieve the desired behaviour of devices and their interactions with their surroundings. Es¬ sentially, the specification is provided as the sequences of events which are desired to happen by functioning of the device in the given environment. All the design knowl¬ edge involved is represented by causal theories of the language CI, which provides a mechanism to predict the performance of the current design. The comparison between the behavioural specification and the predicted model allows verification of designs and suggestions for modifications to them. The proposed design framework is useful for domains which involve dynamic change and in which design requirements can readily be expressed as temporal sequences of events

Keiichi Nakata hasn't uploaded this paper.

Let Keiichi know you want this paper to be uploaded.

Ask for this paper to be uploaded.