Meta-Reasoning about Decisions in Autonomous Semi-Intelligent Systems

Abstract

For intelligent systems to become autonomous in any real sense, they need an ability to make decisions on situations that were not entirely conceived of at compile-time. Machine learning algorithms are excellent in mimicking the behaviour of some gold standard role model, and this can include decision making by the role model. But once out of familiar contexts, the decision making becomes harder and needs an element of more independent probabilistic reasoning and decision making. This paper presents such a method based on a belief mass interpretation of the decision information, where the components are imprecise and thus uncertain by means of intervals.