An artificial intelligence has just broken a belief rooted for decades in the field of forensic sciences: that each fingerprint is unique and random, including those from different fingers of the same person. An analysis based on neural networks has detected that the digital imprints of an individual present many more similarities between them than previously thought because they were being compared incorrectly or, to be more precise, because traditional analyzes did not take into account. Consider the characteristics that allow them to be linked.
The authors of the study, engineers at Columbia University, demonstrate with 99.99% confidence that fingerprints from any two fingers of the same person are extremely similar when looking at the orientation of the ridges near the center of the print. , a discovery that, they say, could greatly improve the efficiency of forensic investigations by being able to link fingerprints from different fingers of the same person in different crime scenes, for example.
The results of the study, which will be published on Friday in an article in the journal Science Advances, so challenge established beliefs in the forensic community that a couple of specialized journals in the field previously rejected their publication. But the Columbia team did not give up, feeding the artificial intelligence system with more data and improving it to offer even more precision in detecting when seemingly unique fingerprints belonged to the same person and when they did not.
One of the things they delved into was to see what marker the AI was using to find similarities that had not been taken into account during decades of forensic analysis.
And after careful visualizations of the system's decision process, they found that "the AI did not use minutiae - the branches and end points in the ridges of the fingerprints - which are the patterns used in traditional fingerprint comparison," explains in a statement the engineer Gabe Guo, who initiated the study in 2021, while he was still a student at Columbia. He notes that, instead, the AI "used something related to the angles and curvatures of the eddies and loops in the center of the fingerprint.
“It is a revolutionary discovery because it breaks with what is known until now; For more than a century we have considered that the prints we have on each finger are random and different from one to another, and now in this study with 60,000 samples they see that the trace that is in the central part of a finger is not independent of the one there will be in the other fingers of the same person,” says Manuel Gené, professor of Legal and Forensic Medicine and of Occupational Medicine at the UB, who has not participated in the study.
After analyzing the study, he emphasizes that it is a notable contribution to science and that in the future it may be of great importance for forensic expert activity if a certain mathematical probability (a likelihood index) can be calculated that two different fingerprints found at two different crime scenes belong to the fingers of the same person.
However, he points out that for now the finding will have little impact on the practice of forensic medicine and police investigation because to transfer the results of research studies to the judicial field it is necessary for the scientific community of the affected area, in this case the forensic sciences, accepts and supports them. In addition to this acceptance, adds Gené, “it would be necessary for this technique to have a known error rate, there are standards for its control that allow a counter-examination to be carried out” for it to be judicially accepted.
The research now published was promoted by a team of engineering students led by Gabe Guo who, without prior knowledge of forensic science, questioned the presumption that fingerprints are unique and therefore incomparable. Guo found a public US government database with about 60,000 fingerprints and entered them in pairs into an artificial intelligence system specialized in comparing data.
They expanded the data and improved the AI system until they extracted a representation of fingerprint vectors from 525,000 images, being able to detect with increasing precision when a pair of them that seemed unique belonged to the same person.
The team admits that using their technique in police or judicial practice will require careful validation of the system using larger data sets. However, they emphasize that their work is a clear example of how even a fairly simple AI, with a fairly simple data set that scientists have had at their disposal for years, can provide insights that experts have overlooked for decades.
“Even more exciting is the fact that a college student, without any background in forensic science, can use AI to successfully challenge a widespread belief across an entire professional field; "We are about to experience an explosion of AI-driven scientific discoveries by non-experts, and the expert community, including academia, must prepare," said Hod Lipson, director of the Creative Machines Laboratory at the University of Columbia, when announcing the discovery.
