Judging the Machine: The Impact of AI-Artwork Labels on Perception and Visual Attention

Abstract

In recent years, Generative-AI (GenAI) has entered the public lexicon, growing in popularity via its widespread accessibility and use. In the art domain, the use of GenAI technologies has sparked debates, ranging from copyright concerns to philosophical debates concerning whether GenAI’s creative capabilities can match that of humans. Recent psychological surveys reveal a consistent bias against artwork labeled AI-generated, though these studies fail to capture how such negative preconceptions shape how people visually process these works. Under the neuroaesthetic triad model—which links aesthetic perception to knowledge-meaning, emotion-valuation, and sensory-motor circuits—this study investigates the possibility that labeling artwork as AI-generated not only triggers a cognitive bias, leading to more critical subjective judgments, but further alters viewing behavior. To investigate this research question, a between-subjects eye-tracking experiment was conducted in which participants saw artwork under one of three viewing conditions: 1) all of the artwork was labeled AI-generated, 2) all of the artwork was labeled human-created, and 3) no origin information was provided. After viewing each artwork, subjects answered a series of Likert survey questions to gauge their perceptions, aesthetic, and emotional valuations of the work. The findings suggest that although AI-generated labels lead to more critical, subjective evaluations of the work, such evaluations do not correlate with a generalized change in viewing strategies—measured as various saccade and fixation metrics—compared to human-created and unlabeled artwork. However, there were observable viewing behavioral differences across groups for a small subset of images, suggesting that contextual label effects might be subtle, image-dependent, or reliant on obvious or salient visual anomalies. Preliminary results from a within-subjects pilot study largely replicate these findings, offering additional nuance to the interpretations of the main study and informing future research with considerations of inherent individual variability in eye-movement behavior.