Abstract:

Generative Artificial intelligence (AI) is an available tool for university students whose ethical and judicious usage is subject to ongoing debate. Traditionally, education has consisted of a controlled introduction to problems of increasing complexity, which is now compromised by instant responses that deliver syntactically polished, conceptually packaged and emotionally calibrated content. However, the deprivation of the time and effort needed to address problems is generating what some authors have called 'cognitive debt,' as the shortcut of consulting AI does not leave a sense of mastery in reasoning.

This study presents a case study designed to highlight the problems associated with the unrestricted use of generative AI in the field of complex physical process modelling. To this end, second-year undergraduate students of Energy Engineering were invited to a workshop on heat transfer simulation using MATLAB. Participants were divided into three groups. The first (supervised GenAI group) received a prior in-person introduction to the case study and was allowed to use GenAI during the test. The second (unsupervised GenAI group) had unrestricted access to GenAI but no prior supervised introduction. The third (solo learning group) worked exclusively with course materials, without access to GenAI. After the workshop, all students took a test in which the use of AI was not permitted, in order to assess effective concept retention.

The results suggested that the supervised GenAI group achieved the highest performance on conceptual questions and on knowledge retention, while the unsupervised GenAI group performed best only on the most numerically demanding question and underperformed on conceptual items, consistent with a cognitive-debt effect. The solo learning group showed greater resilience, no participant gave up the test, but lower overall accuracy, reflecting a missed learning opportunity. This study demonstrates that updating university practices to integrate digital resources with supervised teaching help in order to mitigate the cognitive debt that students in vulnerable situations may face and improves academic performance, as it takes advantage of the real opportunities offered by the use of AI.

Keywords:

GenAI, Cognitive debt , Agency, Programming.