Scaffolding third graders’ computational thinking in an artificial intelligence course: Concepts, practice, and perspectives

Scaffolding type and timing are key factors in designing strategies to develop students’ computational thinking (CT). Although these factors have been studied separately, their combined effects remain underexplored. To address this gap, a 2×2 factorial experiment was conducted in this study to examine the effect of scaffolding type (conceptual versus problem-solving) and timing (immediate versus delayed) on CT concepts, practices, and perspectives. A total of 136 third graders from four classes were assigned to four experimental conditions: delayed conceptual scaffolding, immediate conceptual scaffolding, immediate problem-solving scaffolding, and delayed problem-solving scaffolding. The study was conducted over 6 weeks using the machine learning unit of an artificial intelligence (AI) course. The results, as measured by knowledge tests and final worksheets, showed that problem-solving scaffolding and immediate scaffolding were more effective in developing students’ CT concepts than conceptual scaffolding and delayed scaffolding. Lag sequential analysis indicated that students guided by problem-solving scaffolding and delayed scaffolding developed more complete problem-solving paths and achieved higher levels of CT practice. Additionally, an analysis of students’ learning reflections revealed that most students could explain the functions of machine learning techniques. However, students guided by problem-solving scaffolding and immediate scaffolding were better at understanding and reflecting on the problem-solving process. These findings provide valuable insights for researchers and practical guidance for teachers. By highlighting effective scaffolding methods and their timing, this study advances CT education and helps prepare students for a technology-driven world.