AI in Education: Personalized Learning through Intelligent Tutors

Abstract

One such approach is personalized learning using intelligent tutoring systems (ITS) that provide appropriate content based on student’s learning needs, pace of study, feedback and assessment by harnessing artificial intelligence. This paper introduces and assesses an intelligent tutor architecture that integrates transformer based knowledge tracing, adaptive curriculum sequencing and a conversational feedback module to aid scaffold learning in introductory algebra. We combine a transformer-augmented knowledge tracing model which predicts moment-by-moment mastery, a reinforcement-learning curriculum manager that decides the next activities to present in order to maximize expected learning gain and, an LLM-powered conversational tutor for contextual hints and formative feedback. A pilot study with 240 secondary-school students in four schools used a mixed-methods design: participants were randomized to ITS (n = 120) or a control adaptive-practice alternative (n = 120) for an 8-week intervention. Pre- and post-tests evaluated learning gains; engagement, time-on-task, and perceived usability were obtained from analytics and surveys. The ITS group achieved a mean normalized learning improvement of 0.32 (Cohen’s d = 0.48) relative to control which had a value of 0.18 (d = 0.26; p < 0.01). The knowledge-tracing prediction accuracy (AUC) was 0.89, and the curriculum manager improved mastery acquisition rate by 22% over a fixed-sequence baseline. It was found that persons' qualitative responses indicated that medical advice-giving for the conversational module appeared to be more relevant and useful. It describes the technical design, assessment metrics, and limitations, and implications for scaling ITS in-classroom. Results demonstrate that the use of an advanced student model in synergy with adaptive sequencing and conversational feedback can lead to significant gains for personalized learning experiences.