An Argumentation-Based Approach for Generating Explanations in Activity Reasoning

Abstract

Human-aware Artificial Intelligent systems are goal-directed autonomous systems that are capable of interacting, collaborating, and teaming with humans. Some relevant tasks of these systems are recognizing human’s desires and intentions and exhibiting explicable behavior, giving cogent explanations on demand and engendering trust. This article tackles the problems of reasoning about activities a human is performing and generating explanations about the recognized activities. For the activity reasoning, our approach is divided in two steps: a local selection and a global selection. The former aims to distinguish possible performed activities and the latter aims to determine the status of the recognized activities. For local selection, from a set of observations, a model of the world and the human is constructed in form of hypothetical fragments of activities, which are goal-oriented actions and may be conflicting. Such conflicts indicate that they belongto different activities. In order to deal with conflicts, we base on formal argumentation; thus, we use argumentation semantics for identifying possible different activities from conflicting hypothetical fragments. The result will be consistent sets of hypothetical fragments that are part of an activity or are part of a set of non-conflicting activities. For global selection, we base on the consistent sets of hypothetical fragments to determine if an activity is achieved, partially achieved, or not achieved at all. Besides, we determine the degrees of fulfillment of the recognized activities. Regarding the explanations generation, we generate two types of explanations based on the outputs of the global selection. We apply our proposal to a scenario where a man performs different activities. Finally, we make a theoretical evaluation of the explanation generation.

Supported by organization CAPES/Brazil and CNPq Proc. 409523/2021-6.