Examining Children’s Scientific Misconceptions from an Educational Video with an Embedded Conversational Agent (Stage 3, 10:26 AM)

• In Event: e-Lightning Ed-Talk Session 16 - Stage 3

Abstract

This study explores how educational media influences children’s science understanding, focusing on molting in Elinor Wonders Why. A randomized trial with 124 children (ages 4–7) compared four conditions: interactive AI agent, non-interactive, pseudo-interactive, and human-interactive. Misconceptions were grouped into five types, with incomplete observations the most common (42.7%) and prior experiences next (25.6%). Children in the AI-interactive condition showed the fewest misconceptions (11.6%), highlighting AI’s effectiveness in clarifying misunderstandings through contingent dialogue. Higher misconception rates in non- and pseudo-interactive groups emphasize the importance of responsive, personalized feedback. These findings suggest AI conversational agents can reduce misconceptions and enhance early science learning by providing tailored interaction that supports children’s cognitive development and improves comprehension of complex concepts.

Authors

• Nikki Nasseri, University of California - Irvine

• Kunlei He, University of California - Irvine

• Mark Warschauer, University of California - Irvine