AI Finds Fingerprints Less Unique Than Long Assumed, Raising Forensics Questions
• Columbia Engineering undergrads used AI to show fingerprints from different fingers of same person are actually similar, challenging forensics assumption they are unique.
• AI system trained on 60K fingerprints improved at matching "intra-person" prints; accuracy reached 77% for pairs, over 90% for multiple prints.
• AI uses new type of marker related to angles and curves in fingerprint swirls, not traditional minutiae.
• Study indicates AI performs similarly across genders and races, but broader validation on diverse datasets still needed.
• Shows AI can make surprising new discoveries, even on established datasets; expects explosion of non-expert AI-led breakthroughs, academia must prepare.