Currently, much of machine learning is opaque, just like a "black box." However, in order for humans to understand, trust and effectively manage the emerging AI systems, an AI needs to be able to explain its decisions and conclusions. In this paper, I propose an argumentation-based approach to explainable AI, which has the potential to generate more comprehensive explanations than existing approaches.

Autonomous agents operating in a dynamic environment must be able to reason and make decisions about actions in pursuit of their goals. In addition, in a normative environment an agent's actions are not only directed by the agent's goals, but also by the norms imposed on the agent. Practical reasoning is reasoning about what to do in a given situation, particularly in the presence of conflicts between the agent's practical attitude such as goals, plans and norms. In this thesis we aim: (i) to introduce a model for normative practical reasoning that allows the agents to plan for multiple and potentially conflicting goals and norms at the same time (ii) to implement the model both formally and computationally, (iii) to identify the best plan for the agent to execute by means of argumentation framework and grounded semantics, (iv) to justify the best plan via argumentation-based persuasion dialogue for grounded semantics.

Our work aims to support decision making in situations where the source of the information on which decisions are based is of varying trustworthiness. Our approach uses formal argumentation to capture the relationships between such information sources and conclusions drawn from them. This allows the decision maker to explore how information from particular sources impacts the decisions they have to make. We describe the formal system that underlies our work, and a prototype implementation of that system, applied to a problem from military decision making.

To encompass them, several extensions of Dung's argumentation framework (AF) [Dung, This extended research abstract describes an 1995] have been proposed, but the most general, as shown in argumentation-based approach to modelling articulated [Baroni et al., 2011], is the Argumentation Framework with decision making contexts. The approach Recursive Attacks (AF RA) formalism [Baroni et al., 2009b; encompasses a variety of argument and attack 2011]. In[Baroni et al., 2009a; 2010b] we showed how to organise schemes aimed at representing basic knowledge arguments that are instances of argument schemes in and reasoning patterns for decision support.