Introduction
A new artificial intelligence (AI) tool has challenged the long-held assumption that fingerprints from different fingers of the same person are unique and unmatchable. The tool, developed by researchers from Columbia University and the University at Buffalo, SUNY, can identify with high accuracy whether prints from different fingers come from one person by analyzing the angles and curvatures of the swirls at the center of the fingerprint.
Methodology and Accuracy
The AI tool is based on a deep contrastive network, a type of machine learning model that learns to compare pairs of inputs and outputs. The researchers trained the tool on a public database of approximately 60,000 fingerprints, belonging to different fingers and different people. The tool learned to correlate fingerprints from the same person by focusing on the orientation of the ridges in the center of the fingerprint, rather than the minutiae, or endpoints and bifurcations in fingerprint ridges, which are traditionally used by forensic experts. The researchers claim that the tool can achieve an accuracy of 77% for a single pair of fingerprints, and up to 90% when multiple pairs are presented.
Implications
The AI tool could have significant implications for both biometrics and forensic science. For biometrics, it could mean that using one particular finger to unlock a device or provide identification may not be secure enough, as other fingers from the same person could also be used. For forensic science, it could mean that prints from different fingers found at different crime scenes could be linked to the same person, generating new leads or evidence in investigations. However, the researchers caution that more research is needed to validate their findings using larger and more diverse datasets, and to understand how the AI tool works and what features it is using to match fingerprints.
Controversy
The AI tool is controversial because it contradicts decades of forensic practice and belief that fingerprints from different fingers of one person are unique and therefore unmatchable. The researchers faced multiple rejections from journals before their paper was accepted for publication in Nature Communications. Some reviewers and editors dismissed their findings as impossible or well-known, while others questioned their methodology and data quality. The researchers argue that their AI tool reveals a new kind of forensic marker that has been overlooked by human experts, and that their findings could have important implications for criminal justice.