Augmentation or Erosion? Regulating Cognitive Offloading in Generative AI-Supported STEM Learning

Generative artificial intelligence (GenAI) confronts educational psychology with sharply heterogeneous evidence: meta-analyses report positive average effects on learning alongside prediction intervals spanning harm to benefit, and experiments document gains in some configurations and learning penalties in others. This article develops an augmentation–erosion framework that accounts for part of this heterogeneity by treating the educational effect of GenAI not as a property of the tool but as a property of the partition of cognitive labor between learner and machine. Integrating research on cognitive offloading, distributed cognition, cognitive load theory, and self-regulated learning, the framework holds that offloading augments learning when it discharges operations peripheral to the current learning goal, and risks erosion, whether of acquisition or of maintenance, when it bypasses the processing constitutive of that goal, unless the learner substantively processes the output or later reconstructs it. It supplies criteria for locating operations along a graded peripheral–constitutive continuum, specifies the comparison condition against which augmentation and erosion are defined, and states seven conditioned, testable propositions, each illustrated with a common example and paired with supporting and challenging patterns. On this basis, offloading regulation is proposed as a learning-oriented dimension of AI literacy with domain-general and discipline-specific components: a competence comprising partition discernment, calibrated monitoring, disciplinary verification, and re-internalization, enacted through metacognitive control and conditioned by learners’ goals and motivation. The article closes with an intervention research program, a measurement strategy that triangulates interaction logs with process measures, and equity implications, using STEM learning as a strategically revealing case.