In a casual conversation during Covid-19 lockdowns, Professor Wenyao Xu of the University of Buffalo posed a question that would become the focus of Gabe Guo’s three-year study at Columbia University: Are fingerprints truly unique? Traditional wisdom held that each fingerprint is distinct, forming the basis of forensic identification for centuries.
Guo, now an undergraduate senior in Columbia’s department of computer science, spearheaded a study that deployed a cutting-edge artificial intelligence model, a deep contrastive network commonly used for facial recognition. Their research, published in Science Advances, analyzed a database of 60,000 fingerprints, challenging the notion that fingerprints are unequivocally unique.
The study faced initial rejection from forensic experts, who considered the uniqueness of fingerprints a well-established fact. However, the team persisted, refining their methodology and increasing accuracy until their evidence became incontrovertible.
The AI’s Surprising Findings
Using a deep contrastive network, the team fed the AI system pairs of fingerprints, some belonging to the same person (from different fingers) and others to different individuals. Contrary to the accepted belief, the AI system found strong similarities between fingerprints from different fingers of the same person. The accuracy for determining whether fingerprints belonged to the same individual peaked at 77%.
Insights and Implications
The study highlights the role of angles and curvatures at the center of fingerprints, challenging the traditional focus on minutiae – branchings and endpoints in fingerprint ridges. While the authors acknowledge potential biases in the data, they argue that their findings can enhance criminal investigations.
Guo emphasized the immediate application of the research in generating new leads for cold cases. In scenarios where crime scene fingerprints differ from those on file, the study suggests potential benefits in identifying previously unnoticed correlations. This could not only aid in catching more criminals but also prevent unnecessary investigations into innocent individuals.
Skepticism and Response
Forensic experts, including Christophe Champod, a professor of forensic science in Switzerland, expressed skepticism, stating that the correlation between fingerprints from different fingers of the same person has been known for years. Guo responded, asserting that their study systematically quantified and used the similarities to a degree not done before.
Simon Cole, a professor at the University of California, Irvine, acknowledged the study’s interest but questioned its practical utility. He pointed out that law enforcement typically records all 10 fingers during fingerprinting, making scenarios where only some fingerprints are on record rare.
The team has open-sourced the AI code for transparency and further scrutiny. Guo sees the study’s significance extending beyond forensics, emphasizing the power of AI to reveal hidden patterns. He envisions a cascade of discoveries as people use AI to unveil aspects that have been hiding in plain sight, illustrating the transformative potential of artificial intelligence.
This groundbreaking research challenges established beliefs, sparking a broader conversation about the intersection of technology and forensic science. As AI continues to evolve, its capacity to uncover hidden insights in various domains may reshape our understanding of the world around us.