Dynamics of Argumentation Frameworks

Argumentation The purpose of argument is to change the nature of truth. Frank Herbert – Children of Dune Argumentation is a field of research particulary interesting for the Artificial Intelligence community. It concerns the ability, for an agent (a human being, a machine, a piece of software,. . . ), to understand a topic, to reason about it, taking into account every (possibly contradictory) piece of information on this topic. Of course, argumentation is also a way to convince another agent to change her mind. In both cases, argumentation can be a tool for decision making. More generally, argumentation can be used to model any kind of reasoning, as soon as there is some notion of incompatibility between the data of the problem. In this chapter, we introduce the basic notions of abstract argumentation as presented by [Dun95]. We explain the different ways to deduce some information from an argumentation framework, defined by acceptability semantics. These semantics allow to generate some sets of arguments, called extensions, which can be accepted together. Extensions can be refined through the notion of labellings [Cam06], which map each argument in the framework to a label indicating if it has to be accepted, rejected, or if the agent cannot decide between the first two options. We explain then the approach from [BD04] to map each pair of an argumentation framework and a semantics to a propositional formula. This kind of encoding is at the core of some of our contributions. Lastly, we present some examples of applications of argumentation frameworks, like the case of decision making [ADM08] and goal-oriented persuasion [BMM14]. To conclude, we present an example of application of Dung’s framework which concerns an unusual scenario: resource allocation.