Engage AI and Child in Explanatory Dialogue on Commonsense Reasoning

Abstract

Human-level commonsense reasoning capability is vital for human-AI interaction, enabling AI to understand, anticipate, and respond to human’s thoughts, feelings, and behaviors. Despite the recent advancements in AI commonsense reasoning due to generative language models, a young child is often more rational than state-of-the-art AIs in terms of commonsense reasoning. The field of cognitive science, child development, and explainable AI have long recognized the importance of explanations for sharing knowledge and resolving contradictions. We, therefore, raise the question: can AIs leverage the power of explanations to learn human-level commonsense reasoning? More specifically, can explanatory dialogue with children help AIs to develop commonsense reasoning capabilities? As a first step in this line of research, we aim to engage children in explanatory dialogue with AIs during story reading. We present our novel explanatory dialogue interface based on a state-of-the-art multi-step commonsense reasoning engine and discuss our upcoming pilot study.