The FBI wants artificial intelligence tools that can ID people with burnt, cut or otherwise altered fingerprints.
Any criminal worth their salt knows to wipe their crime scene for fingerprints, but some go a step further and try to erase the prints from their fingers entirely.
For decades, the practice of altering fingerprints has helped wrongdoers evade the law, but today the FBI thinks artificial intelligence could help catch those especially ambitious offenders.
The bureau on Friday asked the tech industry to weigh in on how AI tools could detect altered fingerprints and match them to their unaltered counterparts in the Next Generation Identification System, the FBI’s massive biometric database.
“The [Criminal Justice Information Services] Division has identified a growing trend in which criminals intentionally alter their fingerprints to defeat identification within the NGI System,” officials wrote in a request for information. “As those who seek to avoid identification continue to evolve their alteration techniques, it is critical that the NGI System maintain pace through the ability to learn in real time.”
The explanations behind altered fingerprints are as varied as the prints themselves, and not all are rooted in illegal intentions. Criminals might mask their identities using acid, surgery or a number of other techniques, but frequent contact with chemicals and rough surfaces, as well as certain diseases and medical treatments, could also unintentionally change fingerprints.
But whatever the cause, the FBI said it wants everyone to be identifiable in the NGI System.
Officials requested potential vendors detail the applications and real-time learning capabilities of their artificial intelligence platform and describe ways it could automate the biometric identification process. After assessing the different technologies, the FBI will select groups to prototype a system that can “distinguish [a] normal fingerprint pattern from an irregular pattern,” according to the RFI.
Ultimately, officials said they’re looking for a system that can automatically teach itself variations of any given fingerprint without explicit training.
Responses are due Oct. 12.