Human-AI Adversarial Inquiry A Generative AI-Mediated Deep Conceptual Learning Approach in Engineering Education via the Spiral Cognitive Cycle

This paper presents a generative AI (GenAI)supported assignment design based on the Spiral Cognitive Cycle (SCC).Students generate plots with GenAI and then critique and fix them.We distinguish two types of AI hallucinations in this study.The first type is within the taught content and can be checked using course rules.The second type is outside the taught content and cannot be judged using only what students have learned; we call it the AI Illusory Effect.Using an embedded mixed-methods quasi-experiment with 74 first-year undergraduates (experimental: GenAI-assisted exploration, n = 38; control: traditional problem solving, n = 36).The experimental group earned a higher final weighted grades (M = 73.50,SD = 10.10)than the control group (M = 66.19,SD = 9.59), t(72) = 3.19, p = 0.002, d = 0.74.After adjusting for Gaokao score, the experimental group also scored higher on the average of the proctored midterm and final exams (b = 11.06,p = 0.006).Student reflections suggest that finding and correcting AI "pseudo-results" led to repeated checking, clearer links between concepts and plots, and higher autonomy and engagement.These results suggest that adversarial inquiry can turn imperfect GenAI outputs into useful practice for deep concept learning in first-year engineering courses.