AI-supported virtual laboratory for flipped physics learning: effects on middle school students' achievement, motivation, and cognitive load

This study examined the effects of an AI-supported virtual laboratory (AI-VL) on middle school students' flipped physics learning. A pretest-posttest control-group quasi-experimental design was adopted with 90 Grade 9 students from a middle school in central China. Three intact classes were assigned to the AI-VL, virtual laboratory (VL), and physical laboratory (PL) groups, with 30 students in each group. ANCOVA revealed large group effects on physics learning achievement, F(2, 86) = 113.38, p < 0.001, partial η² = 0.73, and overall learning motivation, F(2, 86) = 113.37, p < 0.001, partial η² = 0.73. The AI-VL group outperformed the VL and PL groups, while the VL group also outperformed the PL group. The AI-VL group reported higher intrinsic motivation, whereas the group difference in extrinsic motivation was not statistically significant (p = 0.054). Significant group differences were also found across the three cognitive load dimensions, with F(2, 87) values ranging from 9.07 to 72.94 and partial η² values ranging from 0.17 to 0.63. The AI-VL group reported the lowest perceived intrinsic cognitive load and the highest germane cognitive load, while both AI-VL and VL resulted in lower extraneous cognitive load than PL. These findings extend research on virtual laboratories and AI-supported learning by demonstrating that AI can be embedded within a VL to help students manage cognitive demands and engage in meaningful learning. Practically, AI-VL can complement PL in flipped physics learning, particularly for pre-class preparation, review, and topics involving abstract or difficult-to-observe processes, with AI agents designed as learning guides rather than answer providers.