Umeå universitets logga

Argumentation is a sub-field of Artificial Intelligence (AI) that gives a way for reasoning with inconsistent, uncertain, and incomplete knowledge. An Argumentation Framework (AF) is an argumentation approach made of a set of arguments and a relation between them called attack. Extension-based semantics and gradual semantics are two ways of reasoning in AFs. The former returns sets of consistent arguments and the latter assigns a value to each argument with the aim of assessing them. Explainable Artificial Intelligence aims to make clear the reasoning of AI systems to the humans (or to other systems) with which they interact. In literature, we can find some approaches for generating explanations for the results of extension-based semantics; however, there is no approach – to the best of our knowledge – for generating explanations for the results of gradual semantics. Thus, this work aims to generate explanations for these types of semantics. Specifically, we will generate contrastive explanations, which are useful and intuitive kind of explanation that explains decisions to lay people by imitating the way in which humans do it. Two types of contrastive explanations will be generated: Property-contrast explanations and Object-contrast ones. The former answers the question why is an argument A in a position p in the resultant ranking rather than in position p'? and the latter answers the question why is an argument E in position p of the ranking whereas argument E' is in position p'?. We use a scenario of decision making for illustrating our approach and make an analysts about when a constrastive explanation can be generated and when it cannot.