An artificial intelligence (AI) has just shattered a decades-long belief in the field of forensic science: that every fingerprint is unique and random, including those from different fingers of the same person. An analysis using neural networks has found that an individual's fingerprints are much more similar than previously thought because they were incorrectly compared or, to be more precise, because traditional analyzes did not take into account the characteristics that allow them to be linked .
The authors of the study, engineers at Columbia University, show with 99.99% confidence that any two fingerprints from the same person are extremely similar if you look at the orientation of the ridges near the center , a discovery they say could greatly improve the efficiency of forensic investigations because prints from different fingers of the same person at different crime scenes can be linked, for example.
The results of the study, which are published today in Science Advances, so challenge established beliefs in the forensic community that a couple of specialized journals in this area previously refused to publish them. But the Columbia team didn't give up, feeding their system more data and improving it to offer even greater accuracy in detecting when seemingly unique fingerprints were from the same person and when they weren't. One of the aspects they delved into was to see which marker the AI used to find similarities that had been overlooked in decades of forensic analysis. And they verified that "the AI did not use minutiae - the ramifications and end points on the ridges of the fingerprints - which are the patterns used in the traditional comparison", explained in a statement the engineer Gabe Guo, who he was the one who started the research in 2021, when he was still a student at Columbia. He points out that, instead, the AI "used something related to the angles and curvatures of the swirls and loops in the center of the fingerprint."
"It is a revolutionary discovery because it breaks with what was 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 the other", says Manuel Gené, professor of Legal and Forensic Medicine at the UB, who has not participated in the study
After reviewing it, he emphasizes that it is a remarkable contribution to science and that in the future it may have great significance for forensic expert activity if a certain mathematical probability (an index of likelihood) can be calculated that two different prints found in two different places are of the same person's fingers. However, he warns that in the short term it will have little impact because in order to apply the results of research studies to the judicial sphere, the scientific community of the affected field, in this case the forensic sciences, must generally accept them. In addition to this support, adds Gené, "it is necessary that the technique has a known error rate and that there are rules to control it that allow a counter-expertise" in order for it to be accepted judicially.
The research was driven by a team of engineering students led by Gabe Guo who, with no prior knowledge of forensic science, questioned the presumption that fingerprints are unique and therefore incomparable, found a public database in the US with 60,000 fingerprints and entered them in pairs in an AI specialized in comparing data to look for connections. The team admits that using their technique in court practice will require careful validation of the system with larger datasets, but they note that it's a clear example of how even a fairly simple AI, with fairly simple data, can provide insights that have been overlooked by experts for decades.
“Even more exciting is that a university student, with no background in forensic science, can use AI to successfully challenge a widely held belief in an entire professional field; we are about to experience an explosion of AI-driven discoveries by non-experts,” Hod Lipson, a robotics expert at Columbia, said in announcing the discovery.
