Home | Publications | MCM+25

ERP Markers of Visual and Semantic Processing in AI-Generated Images: From Perception to Meaning

MCML Authors

Sven Mayer

Prof. Dr.

Associate

* Former Associate

Abstract

Perceptual similarity assessment plays an important role in processing visual information, which is often employed in Human-AI interaction tasks such as object recognition or content generation. It is important to understand how humans perceive and evaluate visual similarity to iteratively generate outputs that meet the users' expectations better and better. By leveraging physiological signals, systems can rely on users' EEG responses to support the similarity assessment process. We conducted a study (N=20), presenting diverse AI-generated images as stimuli and evaluating their semantic similarity to a target image while recording event-related potentials (ERPs). Our results show that the N400 component distinguishes low, medium, and high similarity of images, while the P2 component showed no significant impact, implying consistent early perceptual processing. Thus, we demonstrate that ERPs allow us to assess the users' perceived visual similarity to support rapid interactions with human-AI systems.

inproceedings