Columbia engineers have introduced an artificial intelligence (AI) system that challenges the conventional belief in the unique nature of fingerprints from different fingers of the same individual. Developed by undergraduate Gabe Guo, the AI employs a deep contrastive network and analyzes 60,000 fingerprints from a U.S. government database.
Contrary to traditional forensic assumptions, the AI reveals that intra-person fingerprints share greater similarities than previously thought, achieving 77% accuracy for a single pair and even higher accuracy for multiple pairs over time. The collaborative project between Columbia Engineering and the University at Buffalo, SUNY, published its findings in Science Advances, facing initial skepticism and rejection before being accepted, highlighting the significant shift in understanding fingerprint uniqueness.
Credit: Gabe Guo, Columbia Engineering; Midjourney generated silhouette.
The AI focuses on fingerprint center angles and curvatures, departing from traditional minutiae, and while not yet case-decision ready, it aims to enhance forensic efficiency by prioritizing leads. The study, part of the NSF AI Institute for Dynamical Systems involving the University of Washington, Columbia, and Harvard, emphasizes the transformative potential of AI, challenging established beliefs and underscoring the importance of broader datasets for validation.