Methodology of evaluating the quality of artificial intelligence decisions in educational platforms through the lens of pedagogy and development of trust criteria

Abstract

The article analyzes the problem of evaluating the quality of artificial intelligence decisions in educational platforms from the perspective of pedagogical validity and ethical acceptability, identifying the gap between engineering metrics of machine learning and humanistically understood educational outcomes. It is shown that algorithmic optimization based on accuracy and engagement indicators can lead to narrowing the educational horizon, reproduction of institutional biases, and intensification of epistemic opacity , which transforms trust in the digital environment into a matter of responsibility for long-term consequences for personal development. A logic of interdisciplinary expertise is proposed, in which decisions of recommendation and evaluation modules are compared with professional pedagogical strategies; to quantitatively capture consistency, a coefficient of didactic congruence is introduced. Based on adaptive courses, high consistency of difficulty assessments in exact disciplines and a pronounced semantic gap in humanities areas are revealed, where formal text features do not reflect conceptual depth and cognitive load, creating a risk of incorrect simplification or overload of trajectories. Students trust is interpreted as justified confidence depending on the explainability of decisions: transition from a black box to interfaces with pedagogically articulated justification increases acceptance of recommendations, accompanied by growth in academic achievements and reduction in anxiety due to enhanced metacognitive reflection and sense of control. A typology of AI errors by the severity of pedagogical consequences is developed: the most destructive are miscalculations in evaluating creative works and missing fundamental gaps, affecting motivation, appellate behavior, and risk of dropout. Additionally, it is shown that the implementation of AI does not eliminate the role of the teacher but redistributes the workload towards consultations, verification, and preparation of structured content, forming a demand for algorithmic pedagogy and trust criteria oriented towards transparency, fairness, safety, and preservation of human control.