How Children Learn with Artificial Intelligence: A Study of Dialogic Story Listening

Abstract

As AI becomes more common, its effects on the development of children come into question. The goal of this study is to examine how children learn with artificial intelligence (AI), specifically with an AI voice agent that functions similarly to other popular smart speakers like Amazon's Alexa. Given that dialogic reading has been shown to increase story reading comprehension in children using guided questions, we employed a similar variation of this technique, called dialogic listening, to build story listening comprehension. We were interested in exploring how the benefits of this interactive technique might differ between listening with a human versus with an AI counterpart. To evaluate this, we asked 60 children aged 7-13 to listen to a third-grade reading level story, Henry Huggins, and answer questions asked by either a human conversational partner or AI voice agent. The children's responses were evaluated by analyzing five subcomponents of verbal engagement: language productivity, lexical diversity, topical relevance, the accuracy of the response, and intelligibility of the child's utterances. While our analysis failed to demonstrate similarities in overall story listening comprehension across human and AI conditions, children exhibited comparable levels of verbal engagement and even responded to the AI voice agent with fewer errors. In sum, our findings suggest that while AI is not yet as effective in scaffolding story comprehension as a human, it is able to engage children despite limitations to its current design.