Neurocognitive and affective responses to narrative and divergent AI-generated content in human-AI co-creation: An fNIRS study

This study evaluates the neurocognitive responses of different user groups during design co-creation tasks involving Generative AI (Gen-AI). We propose a neural dynamics-based method to assess user cognitive preferences and a brain network-based method to characterize neurocognitive coordination during human-AI co-creation (HAC) using functional near-infrared spectroscopy (fNIRS). The participant pool is divided into a professional design group and a non-design group. The study evaluates the objective neurocognitive data and subjective psychological responses of both groups during narrative and divergent HAC tasks. 41 students from a single university participate in the experiment. The HAC experimental task is divided into three stages: the image generation stage, the image screening stage, and the creation description stage. An fNIRS scanner is used to monitor participants' neurocognitive activity, including brain activation (BA) and functional connectivity (FC), while they performed both HAC tasks. The Self-Assessment Manikin (SAM) scale collects users' self-ratings. Evaluation of user cognitive preferences based on neural dynamics reveals that design group show higher BA across all three stages of the divergent Gen-AI task compared to the narrative task. In contrast, the non-design group exhibit higher BA during the image generation and screening stages of narrative Gen-AI, but show the opposite pattern during the creation description stage. Regarding brain network patterns during HAC, both groups show higher FC during the divergent HAC task. This result supports our prediction for the design group but does not support our prediction for the non-design group. The SAM scales provided complementary subjective evidence about participants’ affective experiences during the two HAC tasks.