Cognitive engagement and students' perceptions of ChatGPT during discipline-specific problem solving in mining engineering graduate education
Abstract
As generative artificial intelligence becomes increasingly integrated into higher education, concerns have emerged regarding its influence on students’ cognitive engagement, evaluative judgment, and reliance on AI-generated outputs. Despite the growing interest in ChatGPT in STEM education, limited research has examined students’ perceptions of ChatGPT, and the cognitive engagement reflected in their interactions with the tool in graduate engineering education. This mixed-methods study investigated graduate students’ perceptions of ChatGPT while interacting with the tool in three different mining engineering tasks. Twelve students participated in the study and multiple data sources, including screen recordings, surveys, written reflections, and interviews, were collected to examine their interactions with the tool. Qualitative analysis identified five distinct interaction types: (1) reliance with minimal verification, (2) reliance with critical evaluation, (3) collaborative engagement, (4) skeptical or limited use, and (5) learning-driven use. The interaction patterns differed in agency, metacognitive engagement, planning, verification, and evaluative judgment. Students’ interaction types also shifted dynamically across tasks, with later and more open-ended tasks encouraging more verification, reflection, and independent reasoning. Interaction time and the number of follow-up questions further differentiated the interaction types. Quantitative comparisons indicated descriptively higher post-survey means across credibility, believability, satisfaction, and productivity, but none reached statistical significance. Fair/Unfair and Accurate/Inaccurate yielded larger effect-size estimates (Cohen’s d = 0.72 and 0.70); however, these exploratory estimates require cautious interpretation given the small sample. Overall, the findings suggest that graduate engineering students engage with the tool in nuanced and adaptive ways that extend beyond simple dependence or acceptance of AI-generated outputs. The study highlights the importance of instructional and prompting strategies that foster reflective, verification-oriented, and responsible AI-assisted learning practices in engineering education.