Generative artificial intelligence (AI) systems are transforming how humans think, learn, and make decisions. Yet, the behavioral mechanisms through which learners adapt to cognitive collaboration with AI remain poorly understood. This study investigates how AI-assisted tools influence the depth and quality of learning in higher education through the Behavioral Adaptation Framework for AI-Assisted Learning (BAF-AIL), which integrates three mechanisms: cognitive outsourcing, trust calibration, and motivation regulation. A randomized experiment involving 300 university students compared AI-assisted and traditional learning conditions. Quantitative analyses demonstrated that AI support increased efficiency but significantly reduced conceptual integration and transfer, confirming an efficiency–understanding trade-off. Structural-equation modelling revealed that cognitive outsourcing mediated this relationship, while trust calibration and motivation regulation moderated it: over-trust and motivational decline amplified the loss of learning depth, whereas calibrated trust and sustained motivation preserved deep engagement. Qualitative interviews corroborated these dynamics, showing that learners who treated AI as a reflective partner maintained agency, while those who over-relied on automation reported disengagement and dependency. The findings suggest that behavioral outcomes in AI-mediated learning depend less on access to intelligent systems and more on how humans regulate trust, effort, and autonomy. This work contributes to the behavioral sciences by providing insight into the mechanisms of cognitive delegation in algorithmic environments, offering broader insight into how humans sustain meaningful thought in an age of intelligent automation. • Developed the Behavioral Adaptation Framework for AI-Assisted Learning (BAF-AIL) integrating cognitive outsourcing, trust calibration, and motivation regulation. • Found an efficiency–understanding trade-off: AI use improved task efficiency but reduced conceptual depth. • Cognitive outsourcing mediated the relationship between AI assistance and learning depth. • Trust calibration and motivation regulation moderated this effect, defining adaptive vs. maladaptive outcomes. • Results reveal behavioral mechanisms of human–AI co-learning, highlighting how trust and motivation sustain deep engagement in automated environments.
Abbas et al. (Wed,) studied this question.