INVESTIGATING HUMAN PERCEPTIONS OF AUTHENTICITY IN AI-GENERATED VISUAL CONTENT

Abstract

This study investigates human perceptions of authenticity in AI-generated visual content, specifically examining participants’ ability to distinguish AI-generated images from real photographs. A mixed-methods approach was implemented for this study by combining both quantitative data from 125 participants’ performance in identifying AI-generated images with qualitative insights gathered from two focus group discussions involving 16 participants.

The results revealed that participants correctly identified an overall 59.74% of images, with notable differences across categories. Animal images were the easiest to identify, with 71.12% accuracy, while portraits proved the most challenging, having an accuracy of 53.76%. Participants reported a moderate to high level of familiarity with AI-generated content. However, this familiarity did not correlate directly with improved accuracy in identification. When examined by age group, participants under the age of 18 outperformed all other age groups, having the highest average accuracy rate of 64.88%. Focus group discussions revealed that participants relied on visual cues like texture, lighting, and facial proportions found on the image but remained uncertain.

These findings suggest that while AI-generated images can be identified, they are heavily influenced by how we process information and prior exposure to digital media. The study highlights the growing challenge of distinguishing authentic images and the importance of digital media literacy to effectively assess AI-generated content. The implications of this study are significant for improving educational practices related to media literacy and for developing tools that can support more accurate identification of AI-generated visual content.