Conceptual Regression Depth (CReD): A Framework for Psychometric-Integrated Tutoring Systems that Preserve Critical Thinking

Abstract

Current Intelligent Tutoring Systems (ITS) face a critical paradox: while they can improve immediate performance, they often promote cognitive off-loading that fosters dependency rather than independent learning. This paper introduces Conceptual Regression Depth (CReD), a framework that leverages personalized and statistically validated prerequisite learning paths to guide remediation as an alternative to cognitive off-loading. CReD functions by parsing educator supplied instructional content into discrete concept units broken down by Bloom’s taxonomy cognitive levels representing nodes in a directed acyclic graph. The resulting knowledge tree guides learning, which is pedagogically aligned with the classroom, through the ITS. As students interact with the ITS, proficiency for each node is calculated through mastery probability estimation, which is represented as a temporal sparse matrix. This data is suitable for Item Response Theory (IRT) modeling which makes detailed psychometric analysis possible, offering several benefits; teachers obtain student level diagnostics, learning and remediation pathways can be validated, and educators can effectively design targeted intervention. CReD contributes a systematic method for bridging conversational AI with well-established psychometric methods, extracting insights that facilitate the responsible integration of generative AI in education.

Keywords: Measurement, Assessment, Artificial Intelligence (AI) Tutor, Intelligent Tutoring System (ITS), Bloom’s Taxonomy,