Two Pathways, One Tool: The Psychological Consequences of Generative AI Use Among University Students

Generative artificial intelligence (GenAI) has been adopted by university students faster than any previous educational technology, yet evidence on its psychological consequences remains fragmented and apparently contradictory. Some studies report gains in motivation, self-efficacy and well-being; others report anxiety, procrastination, cognitive offloading and dependence. This paper reconciles these findings through an integrative review of 55 scholarly sources drawn from educational technology, educational psychology and information systems research. Drawing on selfdetermination theory, cognitive load theory, self-regulated learning theory and the technostress paradigm, a DualPathway Model of GenAI use is proposed. The model holds that psychological consequences are determined not by the volume of use but by the mode of engagement. An augmentation mode, in which GenAI is used as a tutor and thinking partner, initiates an empowerment pathway marked by need satisfaction, academic self-efficacy and eudaimonic wellbeing. A dependence mode, in which GenAI substitutes for the student’s own cognitive effort, initiates an erosion pathway marked by cognitive offloading, integrity anxiety, skill-atrophy perception, procrastination and diminished self-worth. The two modes are specified as continuous, episodic states rather than fixed user types and adaptive cognitive offloading is distinguished from maladaptive cognitive outsourcing. Both pathways are self-reinforcing and their divergence is moderated by AI literacy, self-regulated learning skill and institutional guidance. Eight testable propositions are derived, each annotated for evidential status, together with implications for students, educators, counsellors and policymakers. The model reframes an apparent empirical contradiction as unmeasured moderation and offers a testable structure for future enquiry.