How intelligent computing can help win the war on crime

Law enforcement is leveraging deep machine learning to gain vital intelligence more quickly

By Laura Neitzel, PoliceOne BrandFocus Staff

The rise of the internet over the last 20 years has enhanced our lives in many ways. Online technologies have made it easier to find and buy products, learn new skills, get directions, book a hotel room, discover new music – even watch an entire season of a TV show in a single weekend and get recommendations on other shows we might like.

It’s also made it easier to commit the most heinous of crimes. The anonymity of the internet and digital currency make it possible to share, sell and trade illicit material from drugs to weapons to victims of human trafficking.

In 2004, the National Center for Missing and Exploited Children reviewed approximately 450,000 child sexual abuse files. By 2015, that number had risen to over 25 million. Most of the victims of child sexual abuse portrayed in the images and videos are under 8 years old.

While the internet and digital technology have aided the explosion of crimes like child pornography and sex trafficking, they can also be an essential tool in the effort to identify victims, accelerate search and rescue times, deter abusers and disrupt the digital platforms where these crimes proliferate.

Organizations like Thorn and technology providers like Veritone are building solutions that employ deep machine learning to fight crime by using it in ways like matching missing children photos to sexual exploitation images on the dark web, using traffic camera footage to locate a vehicle involved in an Amber or Silver alert and investigating a mass shooting incident by identifying key moments in bystander or security images.


Deep machine learning means that, instead of being programmed to perform specific tasks, a computer can learn independently and adapt its understanding based on exposure to new data. The more data it is exposed to, the better the computer can begin to recognize and identify objects based on complex pattern recognition, until it is finally able to independently and with a high degree of accuracy make predictions.  

You’ve probably helped with this learning endeavor by proving online that you’re not a robot. By identifying all the stop signs in a series of pictures, you’ve helped an artificial intelligence refine its ability to recognize what is or is not a stop sign.

Because computers can be trained with millions of labeled images, they can successfully identify images more quickly than a human can. The algorithms or “neural networks” that can classify what is or is not a picture of a stop sign can also be trained to match other types of images, like car models, license plates, weapons or missing persons.

A machine’s ability to “see” is similar to the way a barcode scanner “sees” the stripes in a uniform bar code. But computer vision – the ability to recognize and, more importantly, categorize and sort images – is what enables deep machine learning to have a wider application in police work. Preventing human trafficking, inhibiting child exploitation and reuniting missing children with their families are just a handful of use cases that some of these vision-based services can use in Law Enforcement.


According to the 2018 Technology Vision report from global management consulting firm Accenture:

“There are few crimes today that do not have at least some digital component; for example, with online threats from radicalization, the proliferation of indecent images and the use of social media for coercion and organization of disorder. That’s changing the way that the police need to prevent, detect and solve criminal activity.”

Fortunately, it’s never been easier or more cost-effective to collect data, store it, and build custom machine learning and deep-learning models. Combining machine learning algorithms with the computational power of the Intel Xeon Scalable processor enables developers to create intelligent and innovative new products – powered by machine learning – that can help law enforcement agencies fight and investigate crime. As cameras and tools become more and more accessible, machine learning will bring unmatched innovation in the way law enforcement agencies and their partners collect and analyze data.


Manually digging through a mountain of data to find relevant evidence is a virtually insurmountable task.  But the technology company Veritone has found a way to successfully use deep machine learning to help law enforcement officers extract mission-critical information from that mountain of data.

Tom Avery, Vice President, Veritone Government, explains how the proliferation of data has presented a challenge to law enforcement.

“One of the unintended consequences of bodycam video is that now all of a sudden the judicial system is just hammered with all this extra data that just a few years ago didn’t exist,” Avery said. “The problem with unstructured audio and video data is that if you want to examine it for relevance or factual evidence, you have to sit down and watch it. If you’re already resource-constrained, how do you deal with that?”

Veritone’s aiWARE solution can help solve the challenge by using artificial intelligence to search stored audio and video for objects, spoken words, logos, faces, and more.  Specific for law enforcement, Veritone’s IDentify employs AI to automatically compare known-offender and person-of-interest records with video and photographic evidence, enabling agencies to quickly identify potential suspects for further investigation.

“One of the biggest impacts we can have on law enforcement is actually making that data searchable,” Avery said. “Instead of watching it, you can take action against it using a suite of tools, just like you would use a search engine for data that you’re looking for an answer from.”

For example, he explains, if you had an incident where there was a shooting and you knew one of the officers was shouting “gun,” you could search for the word “gun” and it would take you to that point in the video.

“You can do that across multiple video sets for all the video you have for that particular incident,” said Avery. “Using artificial intelligence tools like aiWare to sort through the data is a force multiplier that helps law enforcement officers get through the issue quicker.”


Being able to get to the right information quickly is always important to law enforcement – especially so in a missing person case that involves a child, an Alzheimer’s patient or other vulnerable person who may be in danger.

Thorn is a nonprofit organization that works in partnership with technology companies, law enforcement, government agencies and other nonprofits to turn the tables on human traffickers and eliminate child sexual abuse material from the internet. Thorn’s tools leverage deep machine learning and artificial intelligence to scour the dark web for images of child sexual exploitation in order to narrow the search for likely matches so law enforcement can more quickly gain insight and act on relevant information.

“With technology, one of the great advantages we have is being able to sift through a lot of different information in a much shorter amount of time so you can aggregate information,” said Kristin Boorse, Thorn’s director of product management. “You can look for themes, trends and patterns to be able to identify who are the most vulnerable and connect them to resources.”

With the help of Thorn’s tools, law enforcement and investigators have been able to identify more than 5,700 child sex trafficking victims and rescue over 100 children from situations where their sexual abuse was recorded and distributed.

While deep machine learning can never replace human intelligence, it can be enormously helpful in reducing law enforcement’s burden of manually reviewing thousands of images or hours of footage to find a likely match or to pinpoint an exact moment. Thanks to organizations like Veritone and Thorn, law enforcement officials are gaining ground against the nefarious elements of the internet and using technology as a force for good.