Forensic experts have always maintained that fingerprints from different fingers of the same individual, known as “intra-person fingerprints,” are distinct and can’t be matched.

A groundbreaking study by Gabe Guo, a senior undergraduate at Columbia Engineering, is challenging this conventional belief. Guo, who had no prior experience in forensics, undertook a research project whereby he was granted access to a public U.S. government database containing around 60,000 fingerprints. He used these fingerprints, fed in pairs, to train an AI system using a deep contrastive network. The pairs included fingerprints from either the same person (but different fingers) or different individuals.

This AI technology, which was adapted from an advanced framework, demonstrated increasing proficiency in identifying whether different fingerprints belonged to the same person. For a single pair, the system’s accuracy reached 77%. When multiple pairs were analyzed, the accuracy significantly improved, suggesting a potential to enhance forensic efficacy by over ten times. This project is a collaboration between Hod Lipson’s Creative Machines lab at Columbia Engineering and Wenyao Xu’s Embedded Sensors and Computing lab at the University at Buffalo, SUNY. The results were published in the journal Science Advances.

The research findings were initially met with skepticism and rejection from a well-known forensics journal, citing the longstanding belief in the uniqueness of each fingerprint. Undeterred, Guo’s team continued their research, improving the AI system further. Despite initial rejections, their paper was eventually published in Science Advances, thanks to Lipson’s persistence.

The study’s most intriguing aspect was identifying what unique markers the AI used, which had been overlooked by decades of forensic science. The AI didn’t rely on traditional minutiae, such as the branchings and endpoints in fingerprint ridges. Instead, it focused on the angles and curvatures of swirls and loops in the fingerprint’s center. This novel approach to forensic analysis could be further enhanced with more extensive training on larger datasets.

The team, including Columbia Engineering senior Aniv Ray and PhD student Judah Goldfeder, acknowledges the need for broader, more diverse datasets to validate this technique for practical use. They also noted that while the system’s current accuracy might not be sufficient for conclusive case decisions, it could prioritize leads in complex cases.

This research underscores the transformative potential of AI in established fields, as noted by Lipson. He emphasizes the ability of AI, even in its simpler forms, to uncover insights that have eluded experts for years. Moreover, the fact that an undergraduate student without a background in forensics could use AI to challenge a field’s long-held belief signifies a forthcoming surge in AI-driven scientific discoveries by non-experts. Lipson calls for the expert community, including academia, to prepare for this new era of discovery.”

Story Source:

Materials provided by Staffordshire University. Note: Content has been edited. 

Journal Reference:

1. Afsané Kruszelnicki, Jakob Schelker, Barbara Leoni, Veronica Nava, Jovan Kalem, Katrin Attermeyer, Claire Gwinnett. An investigation into the use of riverine mesocosms to analyse the effect of flow velocity and recipient textiles on forensic fibre persistence studies. Forensic Science International, 2023; 351: 111818 DOI: 10.1016/j.forsciint.2023.111818