Abstract
- This study extends previous research on the limitations of human and artificial intelligence (AI) in face recognition by incorporating a broader set of tasks and a larger group of participants. While a prior study focused on 10 super-recognizers (SRs) from PD Chemnitz, evaluating their performance in person identification and lookalike discrimination against AI systems (Delib-based Face Recognition and GhostFaceNet), this study expands the scope by including 21 super-recognizers from multiple police departments in Saxony (PD Dresden, PD Chemnitz, BPD Pirna, PD Leipzig) and additionally tests further AI systems, such as the VGG-Face model. The test design encompasses a diverse range of challenges, including recognizing AI-aged faces, detecting facial changes introduced by social media filters, identifying individuals despite image distortions (e.g., lower resolution or noise filters) and recognizing artificially generated faces using Face Swap techniques. The previous study demonstrated significant individual differences among SRs and highlighted clear limitations of AI in face recognition. The findings of this extended study aim to provide a deeper understanding of human and AI performance across a wider array of facial recognition tasks, further elucidating the strengths and weaknesses of current AI technologies compared to human recognizers.