Generative Artificial Intelligence (AI), particularly large language models, has changed the conditions under which individuals judge their own cognitive performance. While AI-assisted tools can improve task outcomes, such improvements do not necessarily lead to more accurate self-evaluation. This article develops an integrative conceptual review of calibration and miscalibration in AI-assisted cognition. Drawing on research on metacognitive monitoring, self-regulated learning, judgment calibration, cognitive offloading, cognitive engagement, and trust in AI, the article identifies a central gap in the literature: the lack of an explanatory framework showing how AI-supported performance becomes a cue for users’ judgments of their own competence. To address this gap, the article proposes an eight-axis explanatory framework organized around the functional position of AI in the task, reflective support versus cognitive substitution, metacognitive engagement, effort redistribution, cognitive engagement, the distinction between assisted performance and actual learning, trust regulation and attribution of success, and self-evaluation accuracy. The framework is presented through qualitative relational expressions and a synthetic conceptual figure, not as an empirically estimated model. Its main contribution is to explain why AI may support calibration when it sustains reflection, verification, and learning, but may contribute to miscalibration when it promotes cognitive substitution, effort reduction, overreliance, or erroneous attribution of success. The article offers a conceptual basis for future empirical research on self-evaluation accuracy in human–AI interaction.
Monica Maier (Tue,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: