Isabella Chang
Research Mentor: Sunhyo Oh
Mentor Department: Combined Program in Education and Psychology, Education
Author(s): Isabella Chang, Sunhyo Oh
Session: Session 1 (9:00 AM – 9:50 AM)
Presentation Type: Poster 97
Abstract
In the age of generative AI, children are growing up in a new technological landscape, of which the long-term pedagogical and psychological impacts have not yet been studied. AI has greatly expanded children’s access to information by supporting their information-seeking in a conversational manner with various modalities such as text, audio, and images. At the same time, it delivers massive amounts of information, often falsified and biased, in a plausible way that mimics human conversation, putting children at risk of placing undue trust in AI and the information it generates. To this end, this study aims to take the first step toward understanding how children search for information from AI and to investigate how AI may impact children’s curiosity and natural world building. We collected the conversation data between a child-friendly AI chatbot, Curio (Oh et al, 2025), and 52 children aged 8-10. They were given access to Curio for two weeks at home and prompted to ask questions when they became curious in their everyday lives. This project will analyze this conversation log data to specifically investigate how children unfold their curiosity by asking a series of relevant questions in sequence. One possible way to understand this is to analyze the relationship between the questions they ask and the responses they receive from AI. We will calculate a semantic coherence score between them, using cosine similarity with TF-IDF, an NLP-based metric. This metric analyzes each chain of conversation and uses a point system to calculate how coherent the chain is overall. The coherence score is calculated by weighting follow-up questions of the same topic or intent differently from questions anchored in an AI response. These results will show how similarities and disparities in the language and content of children’s questions change over the course of interaction in response to their evaluation of the quality of AI-generated information. Such understanding of how children unleash their curiosity with AI can yield practical design implications for developing child-facing AI apps that align with children’s developmental needs and capacities. Also, this study will enable educators and caregivers to develop protocols for guiding children’s use of generative AI across contexts.
