SAsSy—scrutable autonomous systems.

Abstract

An autonomous system consists of physical or virtual systems that can perform tasks without continuous human guidance. Autonomous systems are becoming increasingly ubiquitous, ranging from unmanned vehicles, to robotic surgery devices, to virtual agents which collate and process information on the internet. Existing autonomous systems are opaque, limiting their usefulness in many situations. In order to realise their promise, techniques for making such autonomous systems scrutable are therefore required. We believe that the creation of such scrutable autonomous systems rests on four foundations, namely an appropriate planning representation; the use of a human understandable reasoning mechanism, such as argumentation theory; appropriate natural language generation tools to translate logical statements into natural ones; and information presentation techniques to enable the user to cope with the deluge of information that autonomous systems can provide. Each of these foundations has its own unique challenges, as does the integration of all of these into a single system.