AI System Challenges Uniqueness of Fingerprints, Matches Different Fingers to Same Person
-
AI system discovers new way to match fingerprints from different fingers of same person, challenging forensic presumption that all fingerprints are unique.
-
System trained on 60,000 fingerprint pairs, reaching 77% accuracy for individual pairs and over 90% for multiple pairs, potentially increasing forensic efficiency tenfold.
-
Team believes AI relies on swirl and loop patterns, not traditional minutiae used in fingerprint analysis.
-
Additional validation needed regarding potential gender and racial biases in training data.
-
Discovery demonstrates ability of even basic AI to generate new insights when trained on existing datasets, foreshadowing explosion of AI-led discovery by non-experts.