Abstract
The rapid emergence of Generative Artificial Intelligence (GAI) has transformed the landscape of higher education, influencing pedagogy, assessment, and student learning experiences. Despite its widespread adoption, a significant research gap persists regarding the empirical measurement of its impact on specific learning outcomes. While GAI tools are widely adopted, existing assessment frameworks often fail to distinguish between machine-generated efficiency and genuine cognitive development. This study addresses this gap by developing the Robust Assessment Metrics Framework (RAMF), evaluated through a mixed-methods approach involving students and faculty (N=295) at McPherson University. Quantitative findings reveal a significant "Efficiency-Cognition Trade-off": while frequent GAI usage significantly enhances task efficiency (p < 0.001), it correlates with a statistically significant decline in critical thinking (p < 0.01) and self-reported originality (p < 0.01). Interestingly, regression analysis shows that AI literacy and institutional policy clarity are stronger predictors of academic confidence than usage frequency. This suggests a psychological "confidence paradox" where students feel more capable despite lower cognitive engagement. Qualitatively, thematic analysis highlights a shift toward "shortcut learning" that necessitates a move from product-oriented to process-oriented evaluation. The RAMF introduces expert-validated protocols such as the ‘30/70 Synthesis Rule’ and "Process Logs," to safeguard academic rigor. This research provides institutional leaders with an expert-validated framework proposed for institutional trial to shift from product-oriented to process-oriented assessment in the AI era. By focusing on the interplay between human agency and algorithmic assistance, this research offers broader implications for pedagogical redesign in an AI-saturated academic environment.