This article was originally published in English
The pattern on each of our fingers is unique, but a team used AI to find similarities between them.
If you're a fan of crime shows, crime novels, or crime podcasts, you've probably heard that fingerprints are unique.
Even identical twins, who have (almost) the same genes, do not have the same fingerprints.
This also means that forensic investigators will need fingerprints from the same finger if they want to link cases together, or that you must use the same finger to unlock your phone if you use this method on your device.
A team from Columbia University in the US says artificial intelligence (AI) can improve the accuracy of forensic science.
She developed a new artificial intelligence system to find similarities between fingerprints from different fingers of a person.
Hod Lipson, a professor of engineering at Columbia University, tells Euronews Next that one of his students stormed into his office a year ago and told him he wanted to challenge the idea that each finger's fingerprints are unique.
“So he introduced fingerprint pairs into a large AI system. Sometimes they were from the same person, sometimes they were from a different person,” he explains.
“After a little training, the AI learned that fingerprints from different fingers of the same person are actually very similar. You just have to look at them differently.”
When the researchers attempted to publish their findings in a reputable forensic science journal, they were rebuffed, but their work has since been published in Scientific advances.
They also examined how their model found similarities between fingerprints.
“It turns out that some of the curvature of the ridges is the most important element,” says Hod Lipson.
Not a new discovery
Christophe Champod, a professor of forensic science at the University of Lausanne in Switzerland and a global expert on fingerprints, said these similarities were already known.
“There is no discovery. The fact that what we call general patterns are correlated, that there are more relationships between the fingers of one person than between the fingers of different people, is known from the use of fingerprints for identification,” he tells Euronews Next.
He adds that the AI system used in the study – a so-called deep contrastive network – would not be useful for forensics compared to existing automated systems.
“They did not conduct tests based on partial, distorted and complicated traces. In this case, I am not at all convinced that the low correlations that exist would significantly improve the feeling of the effectiveness of the system,” affirms Christophe Champod.
The study authors acknowledged that this was a limitation.
“There could be issues with compromised data because fingerprints at a crime scene are typically deleted,” says Gabe Guo, the Columbia graduate student who led the research.
According to Christophe Champod, although the AI-based technique is interesting, he does not think it is particularly useful for forensics.
Hod Lipson, on the other hand, sees the system as an example of how AI can be used in the future.
“Many people think that AI can’t really make new discoveries, but that it just regurgitates knowledge.”
“But this research shows that even relatively simple AI, using a relatively simple data set that the research community has had for years, can provide insights that have eluded experts for decades,” he said.