AI-mediated writing and durable learning: a psychological theory of calibration, bypassing and uptake

Abstract

Artificial intelligence (AI)-mediated writing provides a critical case for educational psychology because it reveals how learners regulate cognition, effort, confidence, motivation, and beliefs about competence when fluent external support is available. In English as a foreign language (EFL) writing, such support can correct grammar, reformulate sentences, suggest organization, and sustain revision dialogs with learners who may not have immediate access to human feedback. These affordances invite a tempting inference: better AI-assisted drafts indicate better learning. This Hypothesis and Theory article qualifies that inference by proposing Calibration-Bypassing-Uptake theory, a psychologically grounded account of when AI-mediated writing support strengthens durable competence and when it mainly improves visible performance. The theory links five processes: perceived feedback affordances, dual calibration of source and self, cognitive-affective regulation, strategic uptake or cognitive bypassing, and delayed transfer. Calibration refers to learners’ ability to judge both the reliability of AI feedback and the adequacy of their own understanding. Cognitive bypassing refers to delegating a target learning operation to AI without substantively attempting, evaluating, or reconstructing it during the writing episode; its consequences for learning require separate testing. The article argues that AI supports EFL writing development when learners use feedback to predict, compare, explain, revise, and transfer, rather than merely accept improved text. It derives hypotheses about feedback type, cognitive load, learner proficiency, writing anxiety, self-efficacy, authorship beliefs, hybrid human-AI feedback, alongside a methodological proposition about measuring transfer. The central outcome is therefore not the polish of the assisted draft alone, but the learner’s later capacity to plan, justify, revise, and transfer with reduced support.