• Children had more communication breakdowns with AI partners but fewer repair attempts • Children’s perceptions of emotional capacity of their partner affect repair behaviors • Perceived homophily influenced repair behaviors specifically in AI interactions • Children adapted to breakdowns with AI using accommodation and seeking external help • Future child-AI interfaces should enhance emotional responsiveness and connection Children’s interactions with artificial intelligence (AI) are growing, yet communication breakdowns—instances where mutual understanding fails—remain a challenge, especially for young children. While generative AI shows promise in engaging children in open-ended conversations, how children navigate and repair these communication breakdowns remains unclear. This study compares how 78 children, aged four to eight years, managed communication breakdowns and repair strategies while co-creating stories with an AI agent (powered by a large language model) versus a human counterpart. Results reveal that the type of conversational partner—human or AI—significantly influenced children’s repair behaviors. Children experienced more communication breakdowns when interacting with the AI partner but attempted repairs more frequently with the human counterpart. Misunderstandings and mishearings are the most frequent causes, with clarification requests as the primary repair strategy in both cases. However, when interacting with the AI, children were more likely to go along with the conversation flow to compensate for AI errors, even when the dialogue deviated from their intended meaning—a pattern not observed with human partners. Additionally, children’s social perceptions of their partner, especially beliefs about emotional capacity and their psychological closeness to their partner, influenced repair attempts. This study expands research on children’s conversational repairs with AI, shedding light on the role of social dynamics in shaping these interactions.

In this study, we explored communication breakdowns and repair strategies among 71 children aged 4-8 years while co-creating stories with a generative AI agent enabled by Large Language Models and a human partner. Analyzing approximately 1420 minutes of video recordings, our findings reveal that children experienced more communication breakdowns when interacting with the AI partner but attempted repairs more frequently with human counterparts. Notably, children who attributed greater mind perception to non-human entities were more proactive in attempting repairs during interactions with both human and AI partners, with this trend being more pronounced when children interacted with AI. This work-in-progress offers theoretical contributions by illuminating the interplay between perception and communication. It also underscores important design considerations for developing LLM-enabled generative AI agents that are socio-cognitively responsible and aligned with children’s perceptions.

Mathematical language is a cornerstone of a child’s mathematical development, and children can effectively acquire this language through storytelling with a knowledgeable and engaging partner. In this study, we leverage the recent advances in large language models to conduct free-form, creative conversations with children. Consequently, we developed Mathemyths, a joint storytelling agent that takes turns co-creating stories with children while integrating mathematical terms into the evolving narrative. This paper details our development process, illustrating how prompt-engineering can optimize LLMs for educational contexts. Through a user study involving 35 children aged 4-8 years, our results suggest that when children interacted with Mathemyths, their learning of mathematical language was comparable to those who co-created stories with a human partner. However, we observed differences in how children engaged with co-creation partners of different natures. Overall, we believe that LLM applications, like Mathemyths, offer children a unique conversational experience pertaining to focused learning objectives.