Epistemic Autonomy in the Age of Generative Artificial Intelligence: Foundations for a Reverse Scaffolding Model

Ramos-Zaga F.A. · Revista de Educacion (2026) · DOI: 10.66757/re-2026-414-119824

Resumen

The integration of generative artificial intelligence into higher education has reconfigured how knowledge is accessed, processed, and produced by introducing an algorithmic intermediary capable of generating complete academic outputs before students develop an autonomous conceptual understanding. Although these tools can streamline tasks and expand access to information, their adoption raises concerns about preserving intellectual agency and students’ ability to justify, evaluate, and generate knowledge independently. This article examines the transformation of epistemic autonomy associated with the use of generative artificial intelligence as cognitive scaffolding in higher education. Using a qualitative theoretical-analytical approach grounded in a critical narrative review and an integrative conceptual analysis, the study brings together insights from sociocultural psychology, cognitive load theory, metacognition, virtue epistemology, and the literature on the educational integration of these technologies. The analysis shows that algorithmic mediation alters core properties of classical scaffolding by reversing the sequence between understanding and production, reducing desirable difficulties, and weakening the gradual withdrawal of external support. The article’s original contribution lies in linking the concepts of cognitive debt and reverse scaffolding to explain, respectively, the processes of procedural substitution that erode epistemic agency and the metacognitive operations through which students evaluate, revise, and redirect algorithmically generated outputs. Building on this conceptual integration, the article argues that the educational value of generative artificial intelligence depends less on its technical capabilities than on the pedagogical architecture governing its use. Accordingly, its effective integration requires process traceability, prior independent effort, gradual withdrawal of support, critical AI literacy among faculty, and institutional frameworks capable of preserving self-regulation, reflective assessment, and the autonomous construction of knowledge.

Instituciones: Universidad Privada del Norte