Analysis of fingerprints is a reliable tool in resolving crime for over a century. Investigators rely on evidence of fingerprints to identify the suspects or connect them to specific crime scenes, believing that each impression offers a distinctive code.
However, a team of researchers found that the impressions of different fingers of the same person can sometimes be more alike.
This idea came from an artificial intelligence model that revealed surprising connections between the impressions.
Hod Lipson, by Columbia Engineering, stands out in this effort to question the mainly accepted forensic standards, in collaboration with Wenyao XU of the University of Buffalo.
AI takes the fingerprints
For decades, it has been required for granted that the fingerprints of different fingers of an individual do not correspond. A large part of this belief stems from the hypothesis that each finger displays ridges, curls and completely separated whirlpools.
An anonymous critic has even said: “It is well known that each fingerprint is unique”, faced with the work of researchers.
Despite such resistance, a undergraduate student at Columbia Engineering Appointed Gabe Guo led a study that contradicts this long -standing hypothesis.
Using a public government database with around 60,000 impressions, Guo has fed fingerprint pairs in a deep contrastive network. Some pairs belonged to the same person, while others came from different people.
The artificial intelligence system has become able to say when the impressions that seemed different were in fact an individual, reaching a precision of 77% for unique pairs.
In cases where several samples have been grouped, the precision has shown, offering the possibility of stimulating existing forensic methods of more than ten times.
Researchers rocked the boat
Although these results have promised new possibilities of connection of crime scenes, the researchers faced a difficult battle during the peer exam.
The project was rejected by a well-established forensic newspaper which did not accept the suggestion that different fingers could produce impressions with shared characteristics.
Without being discouraged, the group asked for a wider readership. The newspaper was again repressed, which prompted Lipson to challenge the decision.
“If this information leans in balance, then I imagine that cold cases could be relaunched, and even that the innocents could be acquitted,” noted Lipson, who co -edits Makerspace Facility in Columbia.
Determined not to go back a challenge, even if it meant disrupting more than 100 years of accepted practice, the team continued to refine their work.
Finally, their persistence has borne fruit because their study was finally recognized and published in the newspaper with a reading committee, Scientific advances.
AI gives new clues in the analysis of fingerprints
Traditional methods are based on minute, which refer to ramification models and termination points in the ridges.
“The AI did not use” minute “, which are the branches and parameters of fingerprint ridges – the patterns used in the traditional fingerprint comparison,” said Guo.
“Instead, he used something else, linked to the angles and curvatures of whirlpools and loops in the center of the fingerprint.”
His results suggest that experts may have neglected significant visual clues.
The collaboration included the graduate of Columbia Engineering, Aniv Ray and the doctoral student, Judah Goldfeder, who both indicated that the early success of the project could be strengthened with more important data sets.
“Imagine how it will happen once it is trained in millions of people, instead of thousands of fingerprints,” said Ray, suggesting that this approach could possibly refine how investigators hunt clues on several crime scenes.
Potential bias and following steps
Researchers are attentive to possible data gaps. They noted that their system presented similar performance in various demographic data, but underlined the need for larger and more diverse fingerprint collections.
They hope that an in -depth validation will respond to all the concerns concerning biases before anyone who adopts this technique in real surveys.
The long -term objective is to offer the police an additional tool that improves efficiency when cases seem tangled.
Although AI cannot officially conclude a legal case, it can help reduce the domain of suspects or connect distinct crime scenes depending on partial matches.
“Many people think that AI cannot really make new discoveries – that it simply regurgitates knowledge,” said Lipson, stressing a broader change in the way AI could support investigative work.
“But this research is an example of how even a fairly simple AI, given a fairly simple set of data that the research community has dragged for years, can provide information that has escaped experts for decades.”
IA, digital imprints and police
This study demonstrates that artificial intelligence can identify the models that traditional analysis methods could fail. It also highlights the value of open data sets that have been underused in many areas of research.
The results may encourage forensic experts to rethink certain procedures, especially when several impressions of the same suspect appear in different places.
Lipson sees a future where unexpected breakthroughs can come from new perspectives.
“Even more exciting is the fact that an undergraduate student, without experience in legal medicine, can use AI to successfully contest a largely owned belief in an entire field,” concluded Lipson.
“We are about to experience an explosion of the scientific discovery led by AI by non-experts, and the expert community, including the university world, must prepare.”
The complete study was published in the journal Scientific advances.
––
Do you like what you read? Subscribe to our newsletter to start articles, exclusive content and the latest updates.
Check us on Earthsnap, a free application provided to you by Eric Ralls and Earth.com.
––
