- The paper finds students using GenAI tools score an average of 6.71 points lower on exams compared to non-users, based on empirical analysis and multivariate regression.
- This negative effect is significantly more pronounced among students with inherently high learning potential, suggesting GenAI use acts as a cognitive shortcut hindering deep engagement.
- Supporting constructivist learning theory, the findings imply current GenAI use may compromise deep comprehension and necessitates strategic policy and pedagogical changes in education.
This paper examines the influence of generative artificial intelligence (GenAI) tools, specifically those like ChatGPT, on students' exam outcomes in higher education settings. The authors use empirical analysis to discern the effects of GenAI usage by leveraging student case paper essays as a mechanism for identifying GenAI users, utilizing detection systems such as ZeroGPT. The authors employ a multivariate regression model, incorporating various control variables to mitigate potential biases and yield insight into the direct impact of GenAI usage on academic performance.
Core Findings
The paper reveals that students who use GenAI tools tend to score on average 6.71 points lower than their peers who do not use such technology, as evidenced by the fixed effects ordinary least square (OLS) regression model employed. A key observation is that this negative effect is notably pronounced among students with inherently high learning potential, suggesting a learning-inhibiting aspect of GenAI use. This implies that reliance on GenAI can act as a cognitive shortcut that detracts from necessary engagement with the learning material. Importantly, the paper employs rigorous checks to confirm the robustness of these findings, using alternate GenAI detection thresholds and considering alternative tools like a German-specific AI detector.
Theoretical Context and Implications
The theoretical underpinning of this research lies predominantly in the interplay between cognitive load theory and constructivist learning theory. Cognitive load theory posits that reducing extraneous cognitive loads can facilitate learning efficiency, implying potential educational benefits of GenAI as a tool to take over mundane information processing tasks. Conversely, constructivist theory emphasizes the necessity for active involvement in learning processes for the development of deep comprehension and knowledge. The findings from this paper lend more support to the latter hypothesis, illustrating how GenAI can thwart deeper learning by allowing students to bypass critical cognitive processes.
The implications of these findings are manifold. Practically, this indicates that despite the superficial efficiency afforded by GenAI tools, deeper cognitive engagement is compromised, thereby impacting learning outcomes at a granular level such as exam performance. For educators, these results stress the need for critical reassessment of GenAI tools within educational curricula. It may encourage institutions to develop frameworks that integrate GenAI in a manner that enhances rather than detracts from learning processes.
Future Directions
Future research should expand on these findings by exploring long-term impacts and different modalities of GenAI’s integration into learning environments. Considering diverse learning contexts, beyond initial accounting courses, is necessary to generalize the effects observed. Moreover, investigating the qualitative aspects of student interaction with GenAI can enrich our understanding of student learning behavior and engagement strategies in the presence of such tools.
In conclusion, the paper provides a comprehensive analysis of how GenAI tools impact student exam performance, revealing potential pitfalls in their current use. Insights from this research prompt an urgent need for strategic policy-making and pedagogical innovations within educational settings, to harness the benefits of GenAI without undermining the foundational engagement critical for student learning.