Generativa Between Cognitive Offloading and Thinking: A Theoretical Framework of Metacognitive Tension in the Use of Generative AI

ResumenThe mass adoption of generative artificial intelligence in learning revives a classic question: when does the support of an external tool empower thinking and when does it replace it? This conceptual article, aimed at the construction of theory through theoretical synthesis, integrates three bodies of literature, cognitive download, desirable difficulties and metacognition, to theorize the tension between delegating cognition to AI and developing self-regulation. Its distinctive contribution in the face of close frameworks is to specify boundary conditions. The threshold model of metacognitive tension is proposed, which defines a mediating mechanism, the preservation against the suppression of the diagnostic cues of monitoring and the occupation of the metalevel, three moderators, including the learner's expertise, and a threshold conceived as a transition band in the plane of the locus of discharge and metacognitive posture. It derives a typology of four uses and seven propositions that can be contrasted with their evidential status, and identifies implications for pedagogical, tool, and research design.AbstractThe widespread adoption of generative artificial intelligence in learning revives a classic question: when does support from an external tool enhance thinking, and when does it replace it? This conceptual article, oriented toward theory building through theoretical synthesis, integrates three consolidated bodies of literature, cognitive offloading, desirable difficulties, and metacognition, to theorize the tension between delegating cognition to AI and developing self-regulation. Its distinctive contribution relative to neighboring frameworks is to specify boundary conditions. It proposes the Metacognitive Tension Threshold Model, which defines a mediating mechanism, the preservation versus suppression of the diagnostic cues of monitoring and the occupation of the metalevel, three moderators, including learner expertise, and a threshold located as a transition band within a plane of two determinants, the locus of offloading and metacognitive posture. The model derives a typology of four uses and seven testable propositions with their evidential status, and identifies implications for pedagogical design, tool design, and research.