From answer machine to collaborative partner: impact of structured generative AI training on scientific problem solving in STEM education

Abstract Background Generative AI (GenAI) holds great potential for STEM education, especially in the context of problem-solving tasks, but unguided use often leads to metacognitive laziness, where students treat the technology as an answer machine rather than a collaborative partner. On the basis of Zimmerman’s self-regulated learning (SRL) framework and cognitive load theory (CLT), our randomized controlled trial (RCT; N = 95) investigated whether a structured, SRL-aligned scaffolding training session could improve students’ AI-assisted revision performance in electromagnetism problems. The experimental group ( n = 50) received the designed training, which emphasized the nature of GenAI, the SRL-aligned prompting strategy, and the remediation strategy for GenAI deficiencies. The control group ( n = 45) received problem-solving instruction without specific AI guidance. After completing the Conceptual Survey of Electricity and Magnetism (CSEM) independently, both groups subsequently conversed with the large language model Doubao to revise their responses. Results Independent samples t tests revealed that the experimental group showed greater AI-assisted revision performance ( d = 0.81, p < 0.001) than did the control group. The experimental group reported comparable overall cognitive load, despite a marginal elevation in the intrinsic cognitive load. Robust linear models showed that training attenuated the predictive effect of prior GenAI experience on revision performance. Qualitative analysis revealed that trained participants employed more diverse higher-order strategies and deployed systematic error remediation upon identifying AI deficiencies. The trained students achieved an incorrect-to-correct revision rate of 69.3%, compared with 41.3% for the control group. Conclusion Effective GenAI integration requires pedagogical scaffolding rather than mere access to the tool itself. The results suggest that training can enhance students’ revision performance and foster more strategic engagement with GenAI in the short term. These findings not only extend the theoretical applications of SRL and CLT in the human-AI context but also underscore the practical need for student development in AI literacy.