#660: Examining the Efficacy of AI-based Scaffolds for Multilingual Learners’ Science Inquiry

Traditional STEM instruction for multilingual learners (MLs) often reflects a deficit-oriented approach that focuses on the rote memorization of vocabulary and decontextualized facts, which restricts MLs from meaningful engagement in authentic STEM tasks. Consequently, research has found persistent disparities between MLs and non-MLs’ STEM achievement. This study investigated the efficacy of AI-based scaffolds designed to lower barriers for MLs’ participation in NGSS-aligned science inquiry practices. Using fine-grained log data from an Intelligent Tutoring System (ITS), we analyzed how scaffolds supported MLs’ development of science inquiry competencies. Results indicated that the automated scaffolds supported MLs’ development in forming a testable hypothesis, testing their articulated hypothesis, and drawing correct inferences about their data, even on a different topic and at a later time when scaffolds were not available. These findings suggest that AI-based scaffolds can promote equitable access to authentic STEM learning for MLs while supporting their conceptual and language development.