Comparative analysis of eye-tracking and AI tools for predicting visual attention in virtual human perception

Abstract

Visual attention plays an important role in how users perceive and engage with digital content, particularly in areas such as human–computer inter action and interface design. Eye-tracking is commonly used to study these processes, as it provides detailed and reliable data on visual attention. However, it requires specialized equipment and controlled experimental conditions, which may limit its practical application.
In recent years, artificial intelligence (AI) tools have been developed to predict visual attention based on previously collected eye-tracking data. While these tools are fast and easy to use, it remains unclear how closely their predictions reflect actual human behavior, particularly when observ ing virtual humans.
This paper presents a comparative study of eye-tracking results and AI based attention predictions. The study involved 300 participants, whose visual attention was recorded while viewing a set of virtual human stimuli. The same stimuli were then analyzed using the AI tool across several avail able modes. The comparison was performed using SSIM, MSE, and Pearson correlation, as well as through the analysis of predefined areas of interest.
The results show a strong overall agreement between the two approaches, especially in identifying key facial regions such as the eyes and lips. At the same time, noticeable differences were observed in less prominent areas of the face, as well as during shorter observation periods. AI-generated maps also tend to appear more diffuse compared to eye-tracking data.
These findings suggest that AI tools can provide a useful approximation of visual attention patterns, particularly in the early stages of design. How ever, they do not fully capture the precision of eye-tracking and should therefore be used as a complementary method rather than a replacement.