When Engagement Misleads: A Dual-Process View of Learning in Generative AI Environments
While Generative Artificial Intelligence (GenAI) has revolutionized the production of personalized, multimodal learning content, the "black box" of how specific AI design features orchestrate learners’ internal psychological states remains largely unexplored. This study investigates the boundary conditions of GenAI-driven instruction by examining how visual styles (reality-based vs. imagination-based) and interaction mechanisms (push vs. pull) influence the nexus between cognitive processing and learning outcomes. Integrating Self-Determination Theory (SDT) with a dual-process neuro-cognitive perspective, we conducted a controlled laboratory experiment involving a valid sample of 30 university students, utilizing Electroencephalography (EEG) to triangulate real-time, unconscious neural responses with self-reported psychological measures. Our findings (shown in Figure 1) reveal that the fulfillment of basic psychological needs significantly mitigates implicit stress (captured via EEG) while simultaneously fostering explicit flow, both of which serve as critical pathways to enhanced learning performance. Crucially, the results identify the pull interaction mechanism as a potent catalyst that positively moderates the transition from need satisfaction to flow; conversely, the moderating effect of visual style was found to be statistically non-significant within this neuro-instructional framework. These results contribute a novel GenAI-based instructional framework to the field, suggesting that the efficacy of AI-mediated learning hinges less on aesthetic realism and more on the agency-driven "pull" of information. This study offers timely implications for designing adaptive, neuro-compatible AI educational systems.