But these seemingly ordinary photographs are becoming a new weapon in the fight against crime. Fedardo and his colleagues used these images to train machines to find clues in crime scene photos.
At crime scenes or suspects' homes, police often face a wealth of visual information. Everyday items they find in these places may contain crucial evidence that links the crime to a specific person.
Figalgo is a computer scientist at the University of Leon in northwestern Spain. He has been developing an evidence recognition tool that uses artificial intelligence to identify objects in police photographs and look for connections between them and other crimes. For example, police often photograph bedrooms at the scene of a sexual assault, capturing key information in the process.
- Criminal Psychology: The Strange Skills of Thieves
- Life experiences of villains, swindlers, and gangsters
- How police apprehend criminals using drones
- “These things in the bedroom—the toys, the material of the curtains, the texture of the floor—all of these,” Fedalgo said. “Assuming the toy is registered in the system, it can be retrieved if an accomplice used it in other crimes.”
-
This may not definitively link the suspect to past crimes, but it will certainly provide clues worth investigating. Otherwise, investigators (depending on who was at the crime scene) might miss these clues.
Figalgo and his colleagues have developed a prototype system capable of accomplishing this task. He hopes the Spanish police will try it out immediately. However, he mentioned that the police are already using other image recognition tools.
A small toolkit, like a series of toolboxes, is mounted on a laptop. This setup is intended for analyzing large amounts of photographs downloaded from the electronic devices of criminal suspects. The system can automatically identify known faces and estimate the age and gender of people in the photos. It can also discover images that may be related to child sexual abuse, eliminating the need for police to search entire image databases themselves.
This is just one way police forces around the world are using artificial intelligence to combat crime. AI is being used to analyze photos, CCTV footage, evidence documents, and criminal records to help police gain an edge in their fight against criminals trying to escape justice. With police budgets tightening in many countries, senior officials often hope to use AI to reduce departmental spending. In turn, the public perceives this as making communities safer.
- Workplace sexual harassment? Report it to an AI robot.
- The last remaining value of humanity in the age of artificial intelligence
- How long until your job is replaced by a machine?
-
Currently, the applications of such technologies are far wider than the public is aware. For example, Facebook recently revealed that it used artificial intelligence to discover nearly 9 million nude photos of children in its community in just three months, almost all of which had previously gone unreported. Facebook handed over details of the alleged crimes to the National Center for Missing and Exploited Children.
Nearly 200 law enforcement agencies across the United States are using an algorithm developed by the University of Southern California. This algorithm searches the internet for leads on victims of human trafficking and sex trafficking. It searches and interprets information contained in sexual advertisements on the open web and the so-called dark web to help investigators track down potential victims.
The algorithm has already searched 25 million pages. It has proven very successful, and the US Department of Defense is experimenting with it in investigating drugs, illegal arms sales, and counterfeit goods.
The real benefit of artificial intelligence for investigators lies in gathering complex sets of evidence. Digital forensics company Cellebrite has developed software that can automatically examine potential evidence on a suspect's phone, and British police are currently testing it. This software analyzes images and contact patterns, compares faces, and cross-references data from multiple devices to help police quickly and thoroughly understand interactions between suspects. Recently, the software helped identify Thai officials involved in a human trafficking case, including police officers, three politicians, and a military general.
The algorithm can also quickly query police data to identify various possible connections between criminal cases. William Wong, a professor of human-computer interaction at Middlesex University, said this is to help police realize that information about criminals, evidence, or crime patterns is readily available.
He participated in the development of a system called "Valcri (Visual Analytics for Sense-Making in Criminal Intelligence Analysis)".
"We don't give you an answer from a machine, but rather information that might be relevant," he said. "Let me know what cases in my various databases might appear to have a similar modus operandi to this crime."
In Europe, including after several years of basic trials by British police, Valcri is becoming an application tool.
"We're no longer experimenting with virtual, anonymized data; we're now experimenting with real data," William Wang said. He indicated that the system could potentially solve real crimes for the first time in the coming months.
Ruth Morgan, a forensics expert at University College London, said that the powerful capabilities of police databases, which are somewhat untapped, are waiting to be combined with artificial intelligence, and the prospects are promising.
She said, "The potential is absolutely amazing." But she pointed out that when using algorithms, it's not always possible to judge their decisions in court. Either the technology is patented and the company holding the patent doesn't want to hand over its secrets, or the system is too complex to prove how it reached its conclusions. These kinds of problems can hinder the application of the technology.
However, a number of incredible applications are still under research.
Morgan is developing an image analysis system that calculates the number of microparticles on the soles of a suspect's shoes, such as pollen particles or bullet residue. Based on the quantity of a particular type of microparticle on the shoe, police can estimate how long ago the shoe's owner was in a specific area.
Counting these particles could take human forensic experts weeks or even months. Morgan says that having a machine automatically photograph and count the particles would only take a few hours. After testing the particle counting system, she hopes to develop an algorithm that can also accurately identify the type of particle, such as which type of pollen it is.
In evidence gathering, even the smallest traces like these can have a profound impact. DNA analysis, introduced 30 years ago, has had a massive influence on criminal investigations. However, significant challenges remain. For instance, when samples are taken from clothing, objects, or the victim's body, the extracted DNA often comes from multiple sources. This could include genetic material from the victim, suspect, police officers, witnesses, or even pets. How do we differentiate these sources and calculate the "impact factor"?
Human DNA analysts have been doing this work for a long time, but it's time-consuming and error-prone. A study a few years ago found that 74 out of 108 forensic labs in the United States detected the DNA of three people in a test sample containing only the genetic material of two individuals. In real life, this could implicate an innocent person in a crime.
The chances of such errors are reduced if laboratories are more confident in the number of impact factors in their samples. Michael Marciano and Jonathan Adelman of the US Forensic & National Security Institute at Syracuse University in New York designed a system called Pace to assist DNA analysts. Pace stands for Probabilistic Assessment for Contributor Estimation.
They trained a machine learning algorithm using thousands of virtual samples containing DNA from multiple sources. The algorithm gradually learned to distinguish between samples containing DNA from two individuals and samples containing DNA from three different individuals. While Pace cannot determine the number of influencing factors with 100% certainty, Marciano and Adelman claim its accuracy is slightly higher than other analytical methods.
-
Sometimes, crime scenes leave behind things besides genetic material, but these remain puzzling. In the search for missing persons or murder victims, police occasionally find bone fragments during investigations. They may not be able to compare these fragments with DNA samples, but knowing what a person's face looks like can help in finding them. Currently, forensic anthropologists use materials such as clay to piece together skull fragments and construct facial tissue layers to reconstruct the face of the skull's owner.
This work is extremely challenging, and accuracy can vary depending on the forensic anthropologist performing the procedure. Xin Li, a computer scientist at Louisiana State University, believes that machines can help.
He is developing a system that can scan small fragments of the skull in 3D and then piece them together like a jigsaw puzzle. The system is trained on the shape and proportions of human skulls, and it knows how to digitally fill in the gaps while maintaining a reasonable level of accuracy.
Image captioningJust as advancements in DNA fingerprinting and ballistics have revolutionized criminal investigations, experts predict that artificial intelligence will also bring tremendous progress. (Credit: Alamy)
Image source,Alamy
But the following part is particularly ingenious. Li Xin also trained an algorithm using photos of faces to find the face that best matches the reconstructed skull. When faced with an unidentified skull, the system creates thousands of 3D reconstruction models and then searches for a matching face.
"We collected a lot of photos online and first tried to reconstruct a 3D face (for each photo)," he explained. "Then we did what's called overprinting, matching the 3D face with the skull."
The system can redraw any area in the 3D reconstructed face that does not completely correspond to the skull of the experimental subject. Through continuous modification, the 3D face can be made to look more like a potential victim.
"Given the ongoing aging process, it will be interesting to see how this approach works," Morgan said. "The face may change significantly over time, but the skull remains the same."
Li Xin said that the system he designed is currently running normally, and he hopes that forensic anthropology experts can try it out within a few months.
Image captioningBone fragments often make it difficult to identify their owners, but machine learning offers a way to reconstruct a victim's face using skull fragments (Credit: Getty Images).
Image source,Getty Images
In the long run, the accuracy of these technologies remains questionable. They may prove faster and more useful in small-scale trials, but the real test will come in actual cases. Police must demonstrate that adopting these systems not only brings tangible benefits but also complies with the law and ethics.
"Everything we use must be within the legal framework of criminal justice," said Nick Baker, deputy commissioner of Staffordshire Police. "The courts must approve it, and the public will approve it as well."
A century ago, courts around the world began accepting fingerprint evidence, but this didn't happen overnight. In 1902, fingerprint evidence was used for the first time to prove a suspect's guilt in a criminal trial in Britain. However, it wasn't until nine years later that courts in the United States accepted such evidence.
Most police AI applications are currently in the testing phase. Afterward, we will have a better understanding of their actual capabilities and operational effectiveness. But as Morgan points out, in the near future, tools that accelerate analysis and aggregate data for analysis could have a significant impact on criminal investigations.
"When we look back a few years from now and say, 'Can you believe we didn't have this five years ago?' it will be one of them," she said.
You must be logged in to post a comment.