Following the growing interest and development of artificial intelligence (AI) across various fields, its use in education has also significantly increased, introducing reflections and debates on its implications. Our contribution offers a pedagogical reflection, based on existing empirical research, on the motivations behind adopting such tools, the risks associated with using AI in education, and the potential of this type of initiative.
University dropout represents a major challenge for global education systems, with significant economic and social repercussions. While some interruptions in academic pathways result from deliberate choices aligned with socio-economic opportunities, involuntary early dropout limits students' prospects for personal and professional growth, hindering the full development of both individual and collective potential. In this context, AI emerges as a strategic tool for identifying risk factors associated with dropout and analysing students' characteristics, thereby facilitating the personalization of learning pathways.
Since the 2017/2018 academic year, we have collaborated as pedagogical supervisors in an experimental project proposed by the department of computer science of a Northern Italian University to develop AI-based software to predict university dropout. The development of the predictive model was based on a dataset containing information on over 10,000 first-year students (e.g., socio-demographic data, academic and university career, and socioeconomic status). Starting from this dataset was developed a completely automated learning process creating various models capable of capturing the context in which dropout occurs through classification algorithms such as LDA, SVM, and Random Forests.
Over the years, we have contributed, as pedagogical team, to the tool's implementation by providing data on students' perceptions regarding their strategic skills in studying and working. Expanding the data provided to the system makes it possible to improve and monitor changes in a student's academic pathway, obtaining additional data useful for designing educational interventions aimed at preventing dropout, such as study incentives, support services for learning strategies, and psychological well-being programs.
The application of predictive models allows the development of more flexible and targeted educational strategies contributing to support institutional decision-making in preventing university dropout. The integration of predictive tools into academic systems has the potential to enhance the quality of educational policies by providing accurate data and more effective intervention methodologies. However, dropout prevention should not be perceived solely as a strategy to mitigate academic failure but as an opportunity to promote an educational model centred on developing students’ capacities. This reflection try to go beyond a purely descriptive analysis of the phenomenon by adopting a theoretical approach with practical implications. It offers a critical perspective aimed at implementing innovative strategies to support students and enhance academic success within higher education institutions.

Keywords: Artificial Intelligence, university dropout, predictive models, pedagogical reflection.