Made in Pittsburgh: 5 Great Ideas
Modern-day creation in Pittsburgh doesn’t just involve physical products; we also have a knack for hatching new ideas that can solve problems in innovative, unexpected ways. These locally based thinkers are applying big thoughts to bigger problems.
(page 4 of 5)
The idea: To develop software that helps law enforcement fight sex trafficking in ways that humans alone cannot.
photo by erika gidle
The internet contains millions of advertisements for sex. In some cases, these are advertisements for at-will prostitution; more darkly, however, there is a $99 billion global industry of people trafficked, against their will, for sex. According to the National Center for Missing and Exploited Children, one in six endangered runaways is likely to have been swept into the tide of anonymous victims.
For decades, the only way to find these victims was for law enforcement — from local to state to federal — to manually look at one individual ad after another. In 2011, Emily Kennedy — then a senior humanities student at Carnegie Mellon University — wondered: What if there was a better way?
Kennedy had spent most of that year working on a related senior honors thesis, scouring those online ads for a better understanding of sex trafficking. She spent hours at a time learning to discern syntax and phrasing patterns that meant more than what they appeared to say. Originally, she says, “I wanted to better understand how technology and the internet have changed the way that victims are exploited — and have also affected law enforcement’s response to this crime.”
After Kennedy spoke with law enforcement and learned that they had no more sophisticated means of searching ads than she did, her thesis director suggested reaching out to Artur Dubrawski, the director of CMU’s Auton Lab. Dubrawski and the 45 people working in his lab had the requisite machine-learning background needed to advise on a more technical approach.
Kennedy told Dubrawski the statistics. What jumped out at him was a different number entirely: he thought of his love of Penguins games, and the fact that PPG Paints Arena can hold 18,000 fans. Then he thought of the five and a half arenas it would take to hold the 100,000 estimated victims of sex trafficking in the United States at any given time.
He brought in Kennedy’s project.
The Lab created a prototype tool that could sift through and make sense of truly massive amounts of data: code words, templates, locations, phone numbers and more.
“Detectives were spending hours and hours combing through data rather than going out and building investigations, using their expertise,” says Kennedy. “That’s why we wanted to bring computers into the situation to aid that.”
While the project was still in prototype, data from Pittsburgh sources revealed a cross-country trafficking ring. “Machine learning can pull together ads related to the same group,” Kennedy says, “and pull together these patterns that make sense when you see them — but a human would never be able to identify that pattern on their own.”
All of this happened as public awareness of sex trafficking grew; law enforcement and other agencies were eager for the assistance. Within six months, Kennedy was at the office of the California General Attorney, speaking about the hows of machine learning and artificial intelligence — and why, packaged as the program Traffic Jam, these things would be useful to the state in dismantling local sex trafficking.
After graduating, Kennedy joined Dubrawski’s lab to continue working on Traffic Jam. She started receiving funding, both locally from BNY Mellon as well as national grants from the National Science Foundation and the Defense Advanced Research Projects Agency. The algorithm grew, as did the client base. Every day, the algorithm parses hundreds of thousands of ads, adding to an existing database of tens of millions. In 2014, Kennedy formed Marinus Analytics to house Traffic Jam outside of, but still in connection with, the university.
Of the creation of Marinus Analytics and the continued growth of Traffic Jam, Dubrawski says, “This is our Pittsburgh dream. We get an ambitious young person, we train them, we show them how to make a difference and they go and make a difference.”
In June, the company released FaceSearch, a facial recognition algorithm that can comb advertisements for pictures that match those of missing people or known victims. The first few days alone identified two missing runaways sold online for sex.
Features like FaceSearch are often built from the demands and needs of law enforcement. Kennedy speaks often at conferences and meets face-to-face to learn what the immediate problems are and discuss long-term visions.
All told, “we’re finding on average about 15 victims a month and six traffickers a month,” Kennedy says. “We’ve done a lot with a relatively small amount of funding compared to what most startups get. And we’re really proud of that.”