Analysis of the use of generative artificial intelligence to strengthen algorithmic honesty through a mixed methodological design applied to university students

Abstract

This study examines the integration of generative artificial intelligence into university self-directed learning, aiming to understand the gap between its perceived high utility and actual verification practices, promoting algorithmic honesty in the face of risks of dependency and conceptual delusion. A cross-sectional descriptive design with a mixed-methods approach was adopted, applying online questionnaires and interviews to 179 students of exact and natural sciences, medicine, and engineering at the University of Cartagena during the 2026-I semester. The findings reveal frequent delusions—factual errors and fabricated references—along with barriers such as excessive dependency and ethical dilemmas regarding academic integrity. Furthermore, a partial epistemic delegation to AI and the opacity of the “black box” are noted, factors that can erode analytical skills if pedagogical intervention is not implemented. In response, four lines of action are proposed: critical algorithmic literacy, algorithmic honesty guidelines, argument reconstruction tasks, and collaborative verification, ensuring that AI expands reflective autonomy only through formative strategies that prioritize transparency and rigorous validation.