AI Disrupts Established Forensic Fingerprint Analysis—Not Every Fingerprint Is Unique
by Jo Ellen Knott
On January 10, 2024, Forensic Mag delivered astonishing news: Research out of Columbia University and the University at Buffalo radically challenged the long-held belief that fingerprints from different fingers of the same person are always unique and unmatchable.
The research team, led by Columbia Engineering undergraduate senior Gabe Guo, developed an AI-based system that has shown a remarkable ability to correlate fingerprints from different fingers to the same individual with high accuracy. The team used a public U.S. government database of approximately 60,000 fingerprints to train their artificial intelligence system.
Guo and his colleagues, with no background in forensic analysis, fed the fingerprint data into a neural network. At times, they fed pairs from the same person, other times prints from two different people. They trained “twin deep neural networks to predict whether two fingerprint samples (not necessarily from the same finger) were from the same person.” The neural network learned to correlate a person’s unique fingerprints with a high degree of accuracy. According to the researchers, it does this by analyzing the curvature of the swirls at the center of the fingerprint rather than the minutiae, or endpoints in fingerprint ridges.
From the paper published in Science Advances Guo writes: “our main discovery is that fingerprints from different fingers of the same person share strong similarities; these results hold across all combinations of fingers, even from different hands of the same person. These similarities can mostly be explained by fingerprint ridge orientation.”
Guo and his team have identified different fingerprints belonging to the same person—or intra-person prints—with a success rate of up to 77 percent for a single pair of prints. The accuracy rate improved with multiple pairs. They suggest that the intra-person fingerprint similarities are important because it can help investigators find leads when the fingerprints lifted at the crime scene are from different fingers than the fingerprints already on file.
Guo hopes that the additional information his research provides to forensic analysis can “help prioritize leads when many possibilities exist, help exonerate innocent suspects, or even help create leads for cold cases.”
As important as Guo’s research is, the journey to publish it was difficult. The paper was rejected by a forensic journal and even Science Advances, which eventually published it. The peer reviewers were deeply skeptical of Guo’s findings because of the widely accepted belief in fingerprint uniqueness. The researchers, determined to dispel the skepticism, refined their AI model with additional data.
The paper, titled “Unveiling intra-person fingerprint similarity via deep contrastive learning,” was finally published after Professor Hod Lipson of the Makerspace Facility at Columbia emphasized that its findings can potentially reopen cold cases and ensure justice for innocent individuals.
The study’s implications extend beyond forensic science and highlight the transformative power of artificial intelligence in challenging established principles. Lipson believes in the potential for AI-led scientific discovery by non-experts, predicting an era of innovation and disruption in traditional fields.
Sources: Forensic Mag, Science Advances
