Ever since forensic science abandoned the Bertillion Method of identification in favor of fingerprints in the early 1900’s, we have been taught that every fingerprint is unique. Beginning formation as early as 12 weeks and continuing through the 19th week of gestation, fingerprints are influenced by a variety of factors in the uterus and fetal skin formation. Until recently this was thought to give each fingerprint on each finger of each person, completely unique characteristics. However, a recent application of Artificial Intelligence (AI) to fingerprint analysis may upend this nearly 125 year old idea.
A team of researchers at Columbia University, Gabe Guo, an engineering student, and Hod Lipson, a roboticist, fed approximately 60,000 fingerprint images into a computer. Some of the paired images were from the same person, but different fingers, and some of the paired images were from completely different people. Eventually the computer was able to identify, with 77% accuracy, fingerprint images from different fingers from the same person. Guo said “Our main discovery is that fingerprints from different fingers of the same person share strong similarities; these results hold across all combination of fingers, even from different hands of the same person”. It appears that AI was able to identify new patterns in the ridges of the center of the fingerprints that previously were not identified or understood.
What does this potentially mean for Forensic Sciences? While still too early to tell, one such benefit could be the ability to connect fingerprints left at different crime scenes together. For example, there is a left hand index-finger print that was left at a residential burglary, and a right hand middle-finger print left during a liquor store robbery. Both fingerprints are fed into the system, and the computer identifies that they came from the same person, tying the two crimes together. This would give law enforcement new means of examining cold cases and tying previously unsolved crimes together and to potential offenders.
The results of this discovery need to be further studied and replicated, and one of the shortfalls of AI applications such as this is referred to as the “black box” problem. The “black box” problem refers to the lack of transparency and interpretability of AI algorithms. In fact, computer programmers often struggle to understand how an AI system arrives at its conclusions or predictions. Meaning data is fed in, and a result is produced, but researchers struggle to understand how the AI came to that result. Ultimately it will be up to forensic scientists and the Courts (through Frye test hearings) to determine the applicability of this exciting discovery, however it illustrates the role of evolving technologies in law enforcement and potential for what the future holds.
For further information and references:
