Use of artificial intelligence in the learning process of higher education students

Abstract

The article examines the characteristics and consequences of the integration of artificial intelligence tools into the everyday learning practices of higher education students under conditions of accelerated digital transformation of the didactic environment. It analyzes the shift in the functional role of AI from an auxiliary aid to a resource that partially assumes cognitive operations, thereby intensifying concerns regarding the validity of traditional assessment formats, the boundaries of authorship, and the maintenance of academic integrity amid the blurring of distinctions between independent knowledge production and algorithmic generation. Based on a dataset obtained from 1,845 students across various fields of study and supplemented by an analysis of digital traces in learning management systems, expert evaluation of student texts, and focus group materials, the study demonstrates the heterogeneity of AI usage strategies and their relationship with educational outcomes. An inverse relationship is identified between academic performance and the daily intensity of engagement with generative models: the most frequent use is concentrated among low-achieving students, whereas high-performing students more often opt for an episodic, instrumental mode of use. A pronounced disciplinary specificity of delegated tasks is demonstrated: students in the humanities predominantly use AI for generating coherent texts, IT students for writing code, engineering and technical specialties for solving mathematical problems, while the natural sciences show a more balanced distribution of queries. The dynamics of critical attitudes toward machine-generated output are described separately: as students progress along their educational trajectories, the share of verified content increases and unconditional trust in AI responses declines, accompanied by more demanding assessments of accuracy. Cluster analysis identifies a risk group combining high frequency of AI use, low academic performance, and inflated trust in generated content, while correlation estimates reveal differing directions of association with academic performance depending on the type of delegated tasks: metacognitive applications (structuring and search) are associated with better outcomes, whereas the generation of ready-made solutions correlates with poorer results. The findings allow for an assessment of the article’s relevance to the development of university policies for AI regulation and the updating of methodological support with regard to fostering epistemic vigilance and preserving student agency.