Dual-pathway processing of AI-generated Chinese ink painting: evidence from eye-tracking, EEG, and artistic expertise

Abstract

AI-driven generative art is changing the scope of artistic creation, but the principles behind its aesthetics and the perception of its outcomes remain relatively unexplored, especially concerning culturally specific art forms. The present study investigates how viewers cognitively, affectively, and evaluatively process AI-generated Chinese ink paintings, and whether artistic expertise moderates these responses. We introduce the concept of expressive expansion as a viewer-perceived psychological effect — the tendency for AI-generated ink works to elicit higher ratings of expressiveness and dynamic tension than traditional counterparts — and examine its neural and behavioral correlates using a multimodal paradigm. Stimuli generated using a pre-trained Stable Diffusion architecture were assessed according to three aesthetic metrics along with traditional and digital non-AI artworks by 120 respondents equally split into professional artists and non-artists. Multidimensional measurement based on eye-tracking, semantic differential rating, and EEG showed that, under blind viewing conditions, AI-generated stimuli yielded significantly greater scan-path entropy and beta band suppression than traditional and digital non-AI works, suggesting that they engaged broader attentional exploration and increased attentional engagement as indexed by implicit behavioral and neural measures. Professional and non-professional evaluators agreed on expressiveness evaluation scores but differed considerably in their cultural authenticity perception, suggesting a dissociation between aesthetic appeal and cultural validity. These findings are consistent with a dual-pathway interpretive framework in which perceptual-affective and cultural-evaluative responses can dissociate as a function of artistic expertise — an account advanced as consistent with the observed dissociation rather than as a mechanism established by the present data — and contribute to the emerging empirical literature on audience responses to AI-generated art in culturally specific domains. All conclusions are restricted to the selected stimulus sets under the present experimental conditions. The source code, LoRA model weights, and experimental datasets will be made publicly available at: https://github.com/15100376058/ink-painting-ai-expressive-expansion (DOI: 10.5281/zenodo.20540420).