AI Challenges Long-Held Belief: Not Every Fingerprint Is Unique

 

New York, NY—January 12, 2024 — From fictional crime dramas like “Law and Order” to real-life criminal investigations, fingerprints have been regarded as the ultimate tool for linking individuals to crimes. However, a groundbreaking development by a team of engineers at Columbia University challenges a fundamental assumption in forensic science: the belief that every fingerprint, even from different fingers of the same person, is unique.

The team, led by Columbia Engineering undergraduate senior Gabe Guo, embarked on a journey that would ultimately redefine the way we understand fingerprint analysis. Despite having no prior knowledge of forensics, Guo stumbled upon a public U.S. government database containing over 60,000 fingerprints. He then fed these fingerprints into an artificial intelligence-based system, a deep contrastive network, to explore their similarities and differences.

AI Reveals Surprising Results

The initial premise was simple: Could the AI differentiate between fingerprints from the same person (but different fingers) and those from different individuals? As the AI system, which the team had modified from a state-of-the-art framework, processed more data, it started to excel in discerning seemingly unique fingerprints. The remarkable result was an accuracy rate of 77% for individual pairs. When multiple pairs were presented, the accuracy skyrocketed, potentially increasing current forensic efficiency by more than tenfold. This collaborative project 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, was published in Science Advances, marking a significant milestone in forensic science.

Challenging the Forensic Community

However, the team’s journey was far from straightforward. Their initial submission of the findings to a well-established forensics journal was met with rejection. The anonymous expert reviewer and editor asserted that “every fingerprint is unique,” dismissing the possibility of detecting similarities even when fingerprints came from the same person. Undeterred, the team doubled down on their research, feeding even more data into their AI system, which continued to improve.

Aware of the skepticism within the forensics community, the team chose to submit their manuscript to a broader audience. Despite another rejection, Professor Hod Lipson persisted, stating, “This finding was too important to ignore. If this information tips the balance, then I imagine that cold cases could be revived, and even innocent people could be acquitted.” After a series of discussions and revisions, the paper was eventually accepted for publication by Science Advances.

The Discovery of a New Forensic Marker

One of the key questions that arose was what alternative information the AI system was using to identify similarities that traditional fingerprint analysis had missed. After extensive visualizations of the AI’s decision-making process, the team concluded that it was utilizing a previously undiscovered forensic marker.

Instead of relying on “minutiae,” which are the branching and endpoints in fingerprint ridges commonly used in traditional comparisons, the AI focused on something entirely different. It honed in on the angles and curvatures of the swirls and loops in the center of the fingerprint, revealing a new dimension of fingerprint analysis that had eluded experts for decades.

Columbia Engineering senior Aniv Ray and PhD student Judah Goldfeder, who assisted in data analysis, believe that this is just the beginning. They envision even greater accuracy once the AI system is trained on millions of fingerprints, instead of thousands.

The Road Ahead

While this discovery is groundbreaking, the team acknowledges potential biases in the data used. They present evidence indicating that the AI performed similarly across genders and races when samples were available. However, they emphasize the need for more extensive validation using datasets with broader coverage if this technique is to be widely applied in practice.

Hod Lipson, the James and Sally Scapa Professor of Innovation in the Department of Mechanical Engineering and co-director of the Makerspace Facility, highlights the transformative potential of AI in well-established fields. He notes, “This research is an example of how even a fairly simple AI, given a fairly plain dataset that the research community has had lying around for years, can provide insights that have eluded experts for decades.”

He goes on to express excitement about the future, stating, “We are about to experience an explosion of AI-led scientific discovery by non-experts, and the expert community, including academia, needs to get ready.” In a world where technology continually challenges the status quo, the discovery that not every fingerprint is unique stands as a testament to the power of artificial intelligence to uncover hidden truths and reshape long-standing beliefs.

Share the Post: