Are fingerprints unique? Not really, according to a study based on artificial intelligence.

(CNN) — “Do you think every fingerprint is truly unique?”

This is the question a professor asked Gabe Guo during a casual conversation while he was stuck at home during the Covid-19 quarantine, waiting to start his first year at Columbia University. “Little did I know that this conversation would shape the course of my life for the next three years,” Guo says.

Guo, now a senior in the computer science department at Columbia University, led the team that conducted the research on the topic, and Professor Wenyao Xu of the University at Buffalo was one of its co-authors. A study published this week in the journal Science Advances appears to challenge a long-accepted truth about fingerprints: Not all of them are unique, according to Guo and his colleagues.

In fact, several journals rejected the paper before the team appealed and got Science Advance to accept it. “There was a lot of opposition from the forensic community at first,” recalls Guo, who had no experience in the field before the study.

“The first version or two of our paper said that it was well known that no two fingerprints were the same. I believe this really helped improve our research because we kept introducing more data (and increasing accuracy) until, in the end, the evidence was overwhelming,” he explains.

A New Look at Footprints

To arrive at their surprising results, the team used an artificial intelligence model called a deep contrast network, which is commonly used for tasks such as facial recognition. The researchers put their own spin on this by downloading a US government database containing 60,000 paired fingerprints, which sometimes belonged to the same person (but from different fingers) and sometimes belonged to different people.

During the work, the artificial intelligence-based system discovered that fingerprints from different fingers of the same person were very similar, and was therefore able to determine when the fingerprints belonged to the same person and when they did not, with an accuracy of one pair. at 77%, which would seem to refute the claim that every fingerprint is “unique.”

“We found a strong explanation for why this is so: the angles and curvatures at the center of the fingerprint,” Guo said.

He added that over hundreds of years of forensic analysis, people have studied various features called “characteristic points,” the branches and end points on fingerprint ridges that are used as traditional markers for fingerprint identification. “They are great for matching fingerprints, but are not reliable for detecting correlations between fingerprints of the same person,” Guo explains. “And that’s the idea we had.”

The authors state that they are aware of the potential bias of the data. While they believe the AI ​​system performs very similarly across genders and races, more rigorous testing by analyzing a larger fingerprint database is required for the system to be used in real-world forensics, according to the study.

However, Guo is confident that this discovery can improve criminal investigations:

“The most immediate application is that it can help find new leads in cold cases where fingerprints left at a crime scene do not belong to the same fingers as those in the case,” he said. “But on the other hand, this will not only help catch more criminals. This will also help innocent people who will not have to continue to be investigated unnecessarily. And I think this is a victory for society.”

Exaggeration?

Using deep learning techniques on fingerprint images is an interesting topic, according to Christophe Champod, professor of criminology at the School of Criminal Justice at the University of Lausanne in Switzerland. But Champod, who was not involved in the study, said he didn’t think the work revealed anything new.

“Their argument that these shapes are somehow correlated between fingers has been known since the beginning of fingerprinting, when it was done by hand, and has been documented for many years,” he said. “I think they overstated their paper, in my opinion, due to a lack of knowledge. I’m glad they rediscovered something known, but it’s essentially an exaggeration.”

In response, Guo said that no one had quantified or systematically exploited the similarities between fingerprints from different fingers of the same person to the extent that the new study did.

“We are the first to directly indicate that the similarity is due to the orientation of the ridge at the center of the fingerprint,” Guo said. “We are also the first to attempt to match fingerprints from different fingers of the same person, at least using an automated system.”

Simon Cole, a professor in the department of criminology, law and society at the University of California, Irvine, agreed that the work was interesting but said its practical usefulness had been overstated. Cole was also not involved in the study.

“We weren’t wrong” about the fingerprints, he said, referring to forensic experts. “The unproven but intuitively true statement that no two fingerprints are “exactly alike” is not refuted by the discovery that fingerprints are similar. It has always been known that the fingerprints of different people, as well as those of the same person, are similar. “.

According to the article, the system could be useful at crime scenes where fingerprints are found on fingers other than those registered with the police, but Cole says this can only happen in exceptional cases because when fingerprinting takes all 10 fingers, and often all palms of the hands are regularly recorded. “I don’t understand how they think law enforcement will only have some of a person’s fingerprints on file, but not all of them,” he said.

The team behind the study is confident in the results and has open-sourced the AI ​​so others can test it—a decision that both Champod and Cole praised. But Guo says the study’s importance goes beyond fingerprints.

“We are talking not only about forensic science, but also about artificial intelligence. People have been looking at fingerprints for as long as we’ve existed, but no one noticed the similarities until we got our AI to analyze them. This shows the ability of AI to automatically recognize and extract relevant data. features,” he says.

“I think this study is just the first domino in a huge sequence of similar things. We’ll see people using AI to discover things that were literally hiding in plain sight, right in front of our eyes, like our fingers.”