AI discovers that not all fingerprints are unique

From “Law and Order” to “CSI” to real life, investigators have used fingerprints as a reference to connect criminals to a crime. But if a perpetrator leaves different fingerprints at two different crime scenes, those scenes are very difficult to connect and the trace may become nonexistent.

It is a well-accepted fact in the forensic community that fingerprints from different fingers of the same person – “intra-person fingerprints” – are unique and, therefore, incomparable.

A team led by Columbia Engineering undergraduate Gabe Guo challenged this widely held presumption. Guo, who had no prior knowledge of forensics, found a public U.S. government database containing some 60,000 fingerprints and fed them in pairs into an artificial intelligence-based system known as a network deep contrastive. Sometimes the pairs belonged to the same person (but with different fingers), and sometimes they belonged to different people.

Over time, the AI ​​system, which the team designed by modifying a state-of-the-art framework, became better at determining when seemingly unique fingerprints belonged to the same person and when they did not. same person. The accuracy for a single pair reaches 77%. When multiple pairs were presented, accuracy was significantly higher, potentially increasing current forensic efficiency by more than ten times.

The project, a collaboration between Hod Lipson’s Creative Machines Lab at Columbia Engineering and Wenyao Xu’s Integrated Sensors and Computing Lab at the University at Buffalo, SUNY, was published today in Scientists progress.

Study results challenge – and surprise – the forensic community

Once the team verified their results, they quickly sent the results to a well-established forensic journal, only to receive a rejection a few months later. The anonymous expert and editor concluded that “it is well known that each fingerprint is unique” and therefore it would not be possible to detect similarities even if the fingerprints came from the same person.

The team didn’t give up. They doubled their lead, fed their AI system with even more data, and the system continued to improve. Aware of the skepticism of the forensic community, the team chose to submit their manuscript to a wider audience. The paper was rejected again, but Lipson, the James and Sally Scapa Professor of Innovation in the Department of Mechanical Engineering and co-director of the Makerspace Facility, appealed.

“Usually I don’t challenge editorial decisions, but this finding was too important to ignore,” he said. “If this information tips the scales, then I imagine cold cases could be restarted and even innocent people could be acquitted.”

Although the system’s accuracy is insufficient to officially decide a case, it can help prioritize leads in ambiguous situations. After numerous discussions, the article was finally accepted for publication by Scientists progress.

A new type of forensic marker to accurately capture fingerprints

One sticking point was the question: What alternative information was the AI ​​actually using that had escaped decades of forensic analysis? After carefully visualizing the AI ​​system’s decision process, the team concluded that the AI ​​was using a new forensic marker.

“The AI ​​did not use ‘minutiae,’ which are the branches and ends of fingerprint ridges, the patterns used in traditional fingerprint comparison,” said Guo, who began the study in as a freshman at Columbia Engineering in 2021. “Instead, it used something else, related to the angles and curvatures of the swirls and loops in the center of the fingerprint.”

Columbia engineering senior Aniv Ray, a Ph.D. student Judah Goldfeder, who helped analyze the data, noted that their results are just the beginning. “Imagine how well this will work once it is trained on millions of fingerprints instead of thousands,” Ray said.

The team is aware of potential biases in the data. The authors present evidence that indicates that AI performs similarly across genders and races for which samples were available. However, they note that more careful validation needs to be carried out using datasets with wider coverage if this technique is to be used in practice.

Transformative potential of AI in a well-established field

This discovery is an example of the more surprising things that AI could do, notes Lipson: “A lot of people think that AI can’t really make new discoveries, that it just regurgitates knowledge,” he said. he declared. “But this research is an example of how even fairly simple AI, given a fairly simple data set that the research community has had for years, can provide insights that have eluded experts for decades.”

He added: “Even more exciting is the fact that an undergraduate student, without any forensic training, can use AI to successfully challenge a widely held belief in a field entire. We are on the verge of an explosion of AI-driven scientific discoveries. by non-experts, and the expert community, including academia, must prepare. »