A team of researchers, led by roboticist Hod Lipson at Columbia University, has utilized artificial intelligence (AI) to analyze fingerprints and determine whether prints from different fingers belong to the same person. The study challenges the traditional assumption that each finger's print is unique and explores the surprising similarity between fingerprints on different digits.
The researchers trained a neural network on synthetic fingerprint images and real fingerprint data from 927 individuals. The AI was presented with trios of images, consisting of an "anchor" print from one individual, a "positive" print from a different finger of the same individual, and a "negative" print from another person. The AI achieved a 77% success rate in identifying prints from the same person, exceeding the 50% expected by random chance.
Hod Lipson suggests that this AI system could have practical applications in forensic investigations. For example, if fingerprint sets from different crime scenes are analyzed, the system could identify whether they originate from the same person, even if the fingerprints are from different fingers.
The study also highlighted a surprising finding: the ridge orientation at the center of a person's fingerprints is similar across different fingers. This discovery showcases AI's potential to unveil new scientific insights.
While the research team envisions applications in comparing latent prints (oily traces left on objects) from different crime scenes, critics, such as Simon Cole from the University of California, Irvine, express skepticism. Cole argues that there is no foolproof way to conclude that two prints come from the same skin, especially in complex scenarios involved in latent fingerprint analysis.
The study raises questions about the potential utility and reliability of AI in fingerprint analysis, particularly in forensic contexts.
