Abstract

This study presents an approach to organizing problem-solving instruction in Natural Science with fading generative AI support and reports preliminary findings from the Grade 9 topic of mechanical energy. The investigation was conducted over two lessons at Phan Dinh Phung Lower Secondary School in Da Nang, involving 161 students from four classes under two conditions: constant and fading AI support. Of these students, 159 completed individual pre- and post-lesson assessments without using AI. Mean scores increased from 7.41 to 8.10 under constant support and from 7.67 to 8.96 under fading support on a 0–15 scale. Some indicators of students’ independent decision-making improved across the instructional stages, although the changes were inconsistent. Given the small number of classes, short duration, and insufficiently validated assessment instruments, the findings do not provide adequate evidence to establish the effectiveness of fading support. Nevertheless, the procedure offers a means of tracking the tasks students perform independently while using AI for learning.

Keywords: Autonomy, fading support, generative AI, Natural Science, problem-solving instruction.