AI makes unexpected discovery: Fingerprints are not as unique as is commonly believed

Until recently, it was generally accepted that the pattern of lines on a person’s fingertips is unique and never repeated - not only on other people, but also on other fingers of the same person. All criminology is based on this assumption to this day. A group of researchers decided to test this assumption using a trained neural network and received unexpected conclusions: fingerprints are not at all as unique as is commonly believed.

This discovery, of course, is unlikely to force a re-examination of all criminal cases in which the main evidence was fingerprints left at the crime scene. But it may be useful for cases where there is not a complete set of fingerprints and more and more new ones pop up every now and then.

Scientists from Columbia University analyzed the fingerprints of 60,000 US citizens from an open government database using neural networks. After training, the neural network identified other human fingerprints using one of the fingerprints known to it with a 77% probability. It was enough for the network to see one fingerprint for it to immediately present all the others.

A new analysis of fingerprints using an improved algorithm has made it possible to find common points that are inherent in the prints of all the fingers of one person. These papillary line pattern features appear to be concentrated in the center of the pads. On each finger of one person, they have many of the same features - bends and turns.

Over time, the neural network became better at identifying two different fingerprints belonging to the same person. While each handprint is still unique, there are enough similarities between them for the AI to match them.