The Role of AI-Powered Tutoring Systems in Personalized Learning and Academic Performance in Modern Classrooms

Abstract

The given qualitative multiple-case study will focus on the way AI-based tutoring systems influence individual learning and perceived academic success in the classroom in the present-day. Over 6-10 weeks in four different classrooms (math, science, language arts), we observed lessons based on AI (2-3 lessons per classroom), talked to teachers and students, gathered de-identified learning artifacts, coded platform feedback traces descriptively, and analyzed teacher reflective journals. Reflexive thematic analysis with triangulation and cross-case synthesis provided us with the findings that AI can be most useful in cases where the technical affordances of adaptive pacing, step-by-step hints, localized explanations and mastery dashboards are realized through constrained autonomy and routine agency. In good classrooms, the use of cycle of commit- compare- revise and touch upon reveal after commit hints combined with teacher revoicing/withholding AI feedback and dashboard triaging to targeted conferencing.

Three processes are associated with AI use and perceived performance: (1) timely, specific feedback, making the next actionable; (2) cognitive scaffolding, which changes as one becomes more independent, and results in self-explanations; (3) metacognitive activation, and the transition to the AI says to I decided. The rewards (on-demand assistance, multilingual assistance) were based on reliability of the infrastructure, clear data procedures, and the advanced learner challenge paths. Limitations: There are context-boundedness, descriptive traces and short duration. We suggest a pragmatic logic model that relates features to practices to mechanisms to outcomes and provides implementation recommendations to the teachers, designers, and leaders.