AI-feedback in education: user experience analysis
Artem Feshchenko, Irina Yurjevna Malkova, Kristina Igorevna Buyakova
Abstract
The subject of the study is the subjective experience and perception of students' feedback system based on generative artificial intelligence (AI) in real conditions of the educational process. The aim of the work was to identify the key advantages and disadvantages of the AI-assessor technology and to determine the factors affecting the trust and satisfaction of its users. The research methodology is based on a qualitative approach using combined data analysis. The empirical base of the study: 69 spontaneous user feedbacks obtained during the pilot implementation of the “AI-evaluator” system in the programs of additional professional education. The methods of quantitative (tone analysis, frequency analysis) and qualitative (thematic analysis, argumentation analysis) analysis were used. The scientific novelty consists in the qualitative analysis of the subjective experience gained by students in the course of direct interaction with educational content within the framework of domestic educational programs. For the first time the influence of the unique aspect of AI “characters” design on students' perception of the technology has been considered in detail. The study revealed polarized user attitudes toward the AI assistant. On the one hand, responsiveness, 24/7 availability and innovative elements, particularly the use of different AI ‘characters’, were highly valued, increasing engagement and creating a safe educational environment. On the other hand, significant problems are recorded: inconsistency and errors in assessment, formalism, lack of understanding of context and technical glitches that undermine trust in the system and cause frustration. The data analysis showed that the polarization of opinions is due to the conflict between the high innovation potential of the technology and its current functional limitations. The need is emphasized not only for technical improvement of algorithms (including dealing with fairness and bias), but also for the development of hybrid interaction models that combine automation with human control and expertise to build epistemic trust in users. The results can be applied by developers of educational AI tools as well as educators and methodologists to improve practices of implementation and use of such system in work.