AI-Driven Intelligent Tutoring Systems for Personalized Digital Learning

Abstract

Intelligent Tutoring Systems (ITS) have evolved from rule-based expert systems to AI-driven adaptive platforms capable of personalising learning pathways, diagnosing knowledge gaps, and providing real-time formative feedback at scale. Recent advances in large language models (LLMs), knowledge tracing algorithms, and reinforcement learning for pedagogical policy optimisation have dramatically expanded ITS capabilities beyond structured STEM domains to open-ended writing, computational thinking, and project-based learning. This paper proposes the AI Tutoring System Evaluation Framework (ATSEF), a systematic assessment of six AI-driven ITS architectures -- knowledge tracing (BKT, DKT), LLM-based tutoring, reinforcement learning pedagogical agents, multimodal adaptive systems, and collaborative AI tutoring -- across 18 learning outcome benchmarks in K-12 and higher education contexts. ATSEF introduces the Personalised Learning Effectiveness Index (PLEI) and evaluates systems on 4,200 learner sessions across mathematics, programming, and science domains. Key results: LLM-based tutoring achieves the highest PLEI (0.884) with 34.2% learning gain improvement over non-adaptive baselines; RL pedagogical agents reduce time-to-mastery by 28.6%; knowledge tracing achieves 91.4% next-response accuracy. The framework provides ITS design guidance and an open benchmark suite for AI-driven personalised learning.