The Promise and Peril of Intelligent Tutoring Systems
Current Intelligent Tutoring Systems (ITS) excel at content personalization but require careful guardrails when applied in educational contexts. Research has raised significant concerns regarding the long-term impact of over reliance on these systems, revealing a troubling pattern: while AI tutoring can improve immediate performance, it may undermine the very skills education seeks to develop.
A 2025 study by Michael Gerlich at SBS Swiss Business School demonstrated that increased reliance on AI was associated with the erosion of critical thinking skills, largely due to cognitive off-loading, a process in which individuals reduce their mental effort by depending on AI tools. This finding was reinforced by controlled research at the University of Pennsylvania, which found that students performed better while using an AI tutor, yet once the tutor was removed, those same students performed worse compared with peers who had never relied on an AI system. These findings suggest that students may come to use AI as a crutch, thereby missing opportunities to engage in independent thinking and problem-solving.
The Need for a Framework
The effective implementation of Intelligent Tutoring Systems in education requires frameworks that not only support learning but also promote critical thinking and align with classroom pedagogy. However, existing approaches reveal important limitations. Systems that assess student knowledge through binary mastery classifications, indicating only whether a concept is understood or not, may provide surface-level diagnostics and immediate remediation, but they do not offer systematic guidance on the depth of remediation needed. Moreover, research has shown that AI systems face systematic challenges in logical reasoning, limiting their capacity to diagnose foundational learning deficiencies without a structured instructional framework.
Framework-based AI tutors that impose externally designed taxonomic structures risk misalignment with classroom pedagogies. Additionally, external taxonomic structures may risk Simplicity Bias, caused when the AI uses more complex terminology than necessary. Current educational technologies more broadly lack integration with robust measurement and psychometric modeling. Psychometric data provides educators with clear interpretable insight on student learning and challenges, making it all the more important.
The Conceptual Regression Depth (CReD) Framework
To address these challenges, we introduce Conceptual Regression Depth (CReD), a framework that quantifies prerequisite learning distances while simultaneously generating psychometric profiles from conversational learning data. Part of this work was influenced by Aaron Hu who stated:
While the journey towards intelligent adaptive learning systems is complex and uncharted, the destination is one worth pursuing. By thoughtfully leveraging AI to enhance psychometric assessments and personalized support, we have the potential to revolutionize education, enabling every student – regardless of their learning difficulties – to thrive and reach their full potential. It is a grand challenge, but one that promises profound benefits for individual learners, educational equity and society at large.
Unlike traditional applications of psychometric approaches that rely on formal assessments, CReD-integrated ITS transforms patterns of student interaction, including mistakes, questions, and problem-solving attempts, into Item Response Theory parameters that capture individual ability estimates and concept difficulty calibrations. The proposed system draws topic progression, content guidelines and item banks from teacher-supplied artifacts that align with a student’s classroom learning environment.
This approach provides teachers with precise diagnostic information about student progress and enables evidence-based refinement of learning trajectories and remediation pathways. In doing so, the framework strengthens foundational concepts by providing attentive remediation and promoting critical thinking through the ITS, rather than serving as a crutch.
Our Motivation
Beyond technological considerations, the framework also responds to pressing challenges in the educational workforce. Public schools in the United States entered the 2024–25 academic year with an average of six teaching vacancies, and seventy-four percent reported difficulty filling one or more of those positions with fully certified teachers. A review conducted by the Learning Policy Institute in July 2025 reported that one in eight teaching positions was either unfilled or staffed by under-qualified teachers.
Simultaneously, educational outcomes reveal urgent needs for individualized support. The Nation’s Report Card in 2024 revealed that fewer than one-third of students nationwide were performing at the National Assessment of Educational Progress (NAEP) Proficient level in reading at grades four and twelve. The demand for individualized student attention is therefore critical.
By introducing a human-in-the-loop design, the CReD framework augments rather than replaces teachers, offering guided support that mirrors classroom pedagogy while equipping educators with diagnostics that would otherwise require dramatically smaller class sizes. Our framework extends the reach of quality instruction beyond traditional classroom constraints while preserving the central role of educators in the learning process.