Identifying and then comparing fingerprints is the crime-fighting procedure we know best. The necessary technology was introduced in 1901 by Scotland Yard. But the original idea is older than that. In 1858, a British official in colonial India initiated precursors. Use fingerprints as a means of identification to prevent an employee from going back to the back of the line after receiving their paycheck to get a second package.
For decades we have believed that each of our 10 fingerprints is unique, and that they cannot necessarily be linked when registered separately. But does this statement still hold if we use the latest scientific techniques? A research team affiliated with the School of Engineering and Applied Sciences at Columbia University in the United States decided to test artificial intelligence.
Unlinkable?
The research material consists of a database of about 60 thousand fingerprints. The team fed this in pairs into a new self-designed AI-based network. Some pairs consist of fingerprints from the same person but from different fingers, and other pairs consist of fingerprints from two different people. research results He introduced nuance to the supposed uniqueness of fingerprints.
Suppose a criminal leaves a print of his left little finger during a burglary, and a week later leaves a print of his right little finger during a burglary. Hence, the definition of the problem indicates that the two publications and thus both cases cannot be linked to the same criminal. Yes, was the research team’s conclusion. Their AI system succeeded in 77 percent of comparative analyzes of separated pairs. Even more so when placing multiple pairs next to each other. According to researchers, this could increase the efficiency of forensic investigations tenfold.
Other criteria
All of this led to distrust and downright disbelief among forensic scientists, who ultimately raised the question that provided the key to the answer: What alternative or additional information, apparently ignored by people for decades, does AI use?
Other benchmarks showed an accurate depiction of the AI decision-making process. The system did not choose traditionally used branches of papillary lines or endpoints as a basis for comparison. However, the angles and curves are swirls and loops in the printing medium. Researchers say this discovery has huge potential. Because according to them, how efficient is this AI recognition method if the system is not “trained” with tens of thousands but with millions of fingerprints?
