Cognitive Scaffold or Neural Crutch? An fNIRS Analysis of Self-Reported Generative AI Dependency and Cognitive Performance from Controlled Tasks and GAI-Mediated Reading Comprehension: How Interaction Strategies Mediate Prefrontal Oxygenation and Performance

As Generative Artificial Intelligence integrates into cognitive workflows, understanding its neurobehavioral impact is critical. This exploratory study investigated how self-reported GAI dependency associates with cognitive performance and prefrontal/temporal hemodynamics. Experiment was conducted with 82 young adults across two experimental phases involving unassisted and GAI-mediated tasks. During unassisted tasks, high and low dependency groups showed no substantial behavioral differences, except for marginal Phase 1 category fluency trend (p = 0.049). However, uncorrected fNIRS data tracked elevated oxygenated hemoglobin (ΔHbO) in high-dependency group, tentatively suggesting localized patterns of compensatory neural recruitment. During GAI-mediated task, unassisted low-dependency group served as organic control, and GAI granted high-dependency group spontaneously bifurcated into non-users, intermittent users, and complete offloaders. While offloaders completed tasks faster, they showed reduced localised neural engagement, whereas intermittent GAI users maintained a neuro-hemodynamic profile similar to controls. These findings are interpreted strictly as provisional, hypothesis-generating, and require independent, systemically balanced empirical validation.