From Content Adaptation to Learning-Strategy Adaptation: The Conceptual Shift of Personalized Learning in the Age of Artificial Intelligence

From Content Adaptation to Learning-Strategy Adaptation: The Conceptual Shift of Personalized Learning in the Age of Artificial Intelligence

sepideh sarayloo1

1) PhD Student, Computer Engineering (Software Specialization), Islamic Azad University, Tehran North Branch, Iran

IOI Code : AAHD-ACZEAE
Publication : National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
    Abstract For four decades, personalized learning has been operationalized as content adaptation: the automatic tuning of what material is presented, in what order, at what difficulty and in what format, inferred from the learner s pattern of correct and incorrect responses. The diffusion of large language models has made the generation and re-arrangement of instructional content abundant and nearly costless, and the bottleneck of personalization has consequently moved from content to learning strategy. Combining a systematic review of evidence published between 2015 and 2026 with conceptual analysis, this paper shows that the content-adaptation paradigm faces an empirical anomaly: intelligent systems sharply raise assisted performance without a comparable gain in independent competence. Recent meta-analytic work reports an effect on observable learning behaviour more than twice as large as the effect on perceived competence, and a field trial with nearly a thousand students found that unrestricted access to a generative assistant improved practice performance by 48 percent while depressing unassisted exam performance by 17 percent, an effect fully neutralized when the same model was wrapped in instructional scaffolding. On this basis the paper formalizes a performance-competence gap index and proposes a four-layer strategy-adaptation framework that moves the unit of adaptation from subject matter to the learner s cognitive and metacognitive strategy and progressively hands regulation back to the learner. The findings indicate that the design of guardrails and scaffolds, rather than raw model capability, is the principal determinant of educational effect, and that evaluation must rest on delayed, unassisted performance.

Keywords : personalized learning generative artificial intelligence self regulated learning metacognitive scaffolding hybrid human ai intelligence