The National Center for Missing and Exploited Children (NCMEC) looks to new artificial intelligence (AI) tech to solve and deter the horrors of human trafficking and child abduction.
From gathering leads that can help law enforcement find kids at risk to catching the criminals who exploit them, the National Center for Missing and Exploited Children (NCMEC) relies heavily on its CyberTipline.
In 2014, the line got roughly one million tips.
By 2016, that number shot up to eight million.
What once was a trickle of data has turned into a flood of images, written reports, IP addresses, email addresses and much more. NCMEC’s ambitions are to not only cope with the flood of information, but turn it into better, more actionable insights, more quickly.
The organization increasingly turns to technology, including emerging Artificial Intelligence (AI) capabilities — from finding patterns in reams of IP addresses to matching background items in a sea of grainy photographs.
“Our analysts have their eyes on those cases every day,” said John Clark, CEO of NCMEC. “But as the volume in the pipeline continues to rise, we have to be more and more efficient. Clearly technology has to be front and center.”
Too Many Leads
NCMEC is neither a law enforcement agency nor a government agency. It’s a nonprofit organization founded in 1984 by a group of concerned citizens including John Walsh, whose son Adam was abducted and killed in 1981.
NCMEC launched the CyberTipline in 1998 to extend its information-gathering ability over the internet.
The explosion of tips is due partly to growing awareness, and partly to improved online capabilities at partner organizations. A team of around 25 analysts works through these leads to identify the appropriate jurisdiction to refer and review the report.
A recent headline demonstrating the partnership between NCMEC’s analysts and law enforcement involved the discovery of a serial killer’s connection to 1980s kidnapping cases in California and New Hampshire.
Some cases are old, but many are new. Sometimes the center helps find runaway kids; sometimes it helps find kidnappers or child pornography distributors.
Today, depending on the volume of reporting, it could take up to 30 days to refer a CyberTipline report to law enforcement, said Michelle DeLaune, NCMEC’s senior vice president and chief operating officer.
“We don’t want to kick the can down the road to law enforcement,” said DeLaune, who was the first analyst hired by NCMEC at its inception and has watched the incoming volume of raw information grow exponentially over the years. “We need to provide them with better, more streamlined information.”
One of the center’s goals, then, is to automate as much of its analysts’ manual work as possible.
Automating Information Collection and Storage
Getting these capabilities into NCMEC’s workflow — as with any organization — requires a solid foundation of clean, consistent data and systems that work well together.
When John Clark joined NCMEC in late 2015, he found the organization’s systems weren’t ready. Many applications served the purpose they were built for, but there was little interoperability. Analysts frequently had to cut-and-paste information from one piece of software to the next.
That approach simply wouldn’t scale.
To put the right foundation in place, Clark brought aboard Mark Gianturco as chief technical officer.
“One of the first directions I gave our CTO was a whole needs and value assessment of all our technology,” said Clark. “We needed to create a roadmap, develop an action plan.”
NCMEC is working with a set of existing and new vendors to advance its technology, including Intel, Google, Microsoft and Palantir. Gianturco said the exchange of ideas across this ecosystem has been gratifying to watch.
One example is an enterprise data management project.
“If you think about what the center’s analysts need to do, the ability to extract information from images, video and audio is critical,” said Bob Rogers, Chief Data Officer for Intel.
“With the explosion of data, you want the ability to store it in its native state, so you don’t restrict the questions you want to ask that data,” he said.
Intel and its partners are helping NCMEC create a comprehensive plan to store massive volumes of data, query it in a variety of ways, and have that data shared consistently by all the applications they use. AI is automating and speeding up the process.
“Artificial intelligence is basically where machines make sense, learn, and interface with the external world without human beings having to specifically program them,” said Nidhi Chappell, director of machine learning at Intel.
AI is an umbrella term and under it Machine Learning (ML) is the set of techniques and tools that allow computers to “think” by creating mathematical algorithms based on accumulated data. Also under the umbrella, Deep Learning (DL) uses neural network models to do things like image recognition and language processing. AI perceives the world, using data to detect and recognize patterns then takes an action based on that recognition.
AI has become a vital tool for industries that are increasingly leveraging digital technologies, and it plays a big role in the creation of interactive games. Now it’s helping NCMEC automate its data curation and speed the creation of useful reports police and federal authorities use to find missing children.
Possibilities for AI and Facial Recognition
Gianturco is also optimistic that AI and other technologies will help find new ways to find kids.
“While automation is critical, the big technical leaps forward are even more exciting,” said Gianturco.
He said breakthroughs in machine learning allow the center to tackle many problems it couldn’t five years ago. For example, he pointed to IP addresses. Many leads and images are associated with one or more of these basic numeric addresses that are assigned to each location connected to the internet. A single IP address, though, may not be a reliable indicator of a geographic point of origin.
“It might be a proxy server that someone is going through to hide their actual location,” explained Rogers. A machine learning program can track and compare millions of IP addresses, the information that comes from them and identify useful clues.
Gianturco sees great promise in facial recognition technologies as a tool in the search for missing children. In a crowded place like a subway station, which has surveillance cameras in place, AI could help spot faces that match people in the missing persons database.
Further, he envisions more application of age progression — and regression — to facial images. Once a child has been missing for multiple years, a computer model could simulate how that person might look in the present day. While this is already a key strategy employed by NCMEC, advances in technology could help that process become more automated or even improve accuracy.
Conversely, he said, it will be possible to create a website that allows uploading of a current picture and then use the modeling process to look back in time, in order to see if a younger version of that face matches a missing person report.
These kinds of advances require careful thought about security and privacy as well as massive computing power, so the NCMEC’s modernization work has only just begun.
“We’re making sure we’re facing into the future, not just relying on where we are today,” said Clark.