AI-human interaction through explanations

We model the interaction between two agents, an AI system and a human explainee, in which the former conveys personalized explanations to the latter. Our starting point is the framework recently provided by Luo, Studer and Dastani, which combines resources from both epistemic and justification logics. We modify such a framework by allowing for: (i) a sharper distinction between the types of agent involved, (ii) the possibility for the explainee to accept information they do not yet justify, and (iii) the possibility for the explainee to have false beliefs. We introduce a framework that accommodates (i)-(iii) and prove a convergence result that enables us to identify the conditions under which the interaction between the two agents succeeds.