Fight Human Trafficking with Law and Technology

Project Traffic Light seeks to interview judicial professionals with knowledge of human trafficking in courts. We use data gathered to build awareness tools to help other professionals identify and help victims, who usually appear as criminal defendants. If you know what human trafficking looks like in your jurisdiction, please contact us at projecttrafficlight@gmail.com. The current tool can be accessed here: https://education.neotalogic.com/a/Traffic_Light
I was exploring projects with MS in Artificial Intelligence graduate students at the Northwestern Law School Innovation Lab when this especially important challenge came across our desks: Help court judges recognize when criminal defendants were actually victims of human trafficking. The project sponsors were both members of United Abolitionists, an anti-trafficking nonprofit out of Florida.
Orange County Senior Judge Wilfredo Martinez had partnered with longtime Thomson Reuters data scientist Brian Ulicny to bring a Legal Engineering design challenge straight to us – how can we (1) identify victims of human trafficking in courts, and (2) help less experienced judges know what to look for?

Judge Martinez had seen far too many victims in his court, and had an assortment of “victim stories” or fact patterns that were high indicators of human trafficking. The most common indicators in Orange County, Florida were:
Criminal Charges:
Sex trade crime (e.g. prostitution, unlicensed massage parlor, loitering)
ID Theft + Grand Theft Auto Charge Combination
Shoplifting for basic life necessities or hygiene products
Repeated driving while license suspended charges
Presence in Court:
Handler or overseer present
Confused or abnormal conduct in court, not aware of geographic location
Stockkeeping tattoos (bar scan codes or numbers), brands or injuries
Uncooperative with Public Defender
Some indicators might seem obvious, others less so (more on that below). With a team comprised of law students and computer scientists, and armed with critical inside knowledge, we began to analyze and digest the problem. We were certain that with help from Brian and his data science resources we could automate the process of identifying victims as they processed through the criminal justice system. The more difficult part (we thought) would be making judges aware of our findings.

Almost immediately we ran into problems – How can you tell from old court records if a specific defendant actually was a victim? With many courts still operating with paper documents, how can you analyze a current case for indications? If you already knew someone was a victim, will privacy laws prevent access to their court records? The list goes on. We didn’t have a data science problem on our hands (we lacked the 1000’s of hours it would take to research and build a working system), we had a design problem.
How can we use the time and resources we have (five graduate students and four months) to design and build a working toolset that could actually be deployed and of practical use?
With guidance from professors and the project sponsors, we devised a two-part tool that was realistically executable. The first part was a “Bench Card”: judges use these as quick-reference tools to access information during court proceedings. Those who work in medicine may be familiar with “Crash Cart Cards” affixed to the mobile carts in hospitals used to revive cardiac arrest patients. Information is laid out for simplicity and rapid use. An online tool (the second part) would not be useful if judges never used it. The bench card not only provided quick-glance reference, but served as a “real-world” reminder to judges to stay vigilant.

Expert knowledge extraction can be a difficult process. When professionals make decisions, they rely on years of experience and training. How do you interview an expert to catalog their knowledge? Asking 300 “if-then” scenario questions can become rather dull (and ineffective), so the process must be iterative and analytical. In our interviews, Judge Martinez would share stories, ideas and lessons learned. Brian connected us with Reuters customers in law enforcement to provide additional perspective. Our team would transcribe input, then meet to digest and categorize the information. Questions would arise about certain scenarios, and we would collect follow-up questions for the next interview. The end result of this process was a “decision tree” – a graphical representation of how an initial indicator should lead to follow-up questions, and through those to an assessment of whether or not a subject was a likely victim.

Translating expert knowledge into a logical set of questions is a great first step. However to build the online tool we needed to design the full user experience. Would we use secure logins or keep open access? We opted for a hybrid mode where the tool asked for a passphrase, but we only used it to record usage statistics across different areas. Each bench card would have its own unique passphrase, allowing maximum access with built-in options for future controls.
Would we ask for victim demographics? We opted for only gender and age. Should we ask which indicator drove the user to the tool, or start with a checklist to gather multiple data points? For this, we presented the indicators as a checklist but set the tool to automatically proceed when one box was checked, expediting the process flow. Only after deciding on user inputs and designing the Graphical User Interface (GUI) could our tech team to actually write the expert knowledge into code.

Four months of hard work got us to our goal, what tech folks call the Minimum Viable Product (MVP). Getting to MVP is a big step, but it is only the first milestone. Project Traffic Light was designed around the human trafficking patterns of Florida, and Judge Martinez made clear that local conditions cause patterns to vary from place to place. For example, one of his indicators, the ID Theft + Grand Theft Auto Charge Combination, was mysterious to us at first. As he explained:
“Florida has many large spectator sports events, including multiple Super Bowls over the years. Large numbers of (mostly) men come from all over the world to attend. Upon arrival, many of these men rent cars and at some point solicit a prostitute (women are trafficked in from other locations to satisfy increased demand). The prostitute directs her “john” to a certain hotel. Once inside the hotel room, she requests he take a shower. While in the shower, she takes his wallet and rental car keys. With a criminal enterprise waiting, she delivers the wallet to another room where hackers are set up to max out the john’s credit cards. The rental car keys go to runners who drive the cars to local “chop shops’ where the cars are disassembled and the parts sold off.”
“With the john required by the rental car company to file a police report,” Judge Martinez explained, “he helps identify the prostitute who took his belongings. She gets arrested, but he wants nothing else to do with the prosecution. He’s from out of town, and probably has a wife or girlfriend who he desperately wants to keep in the dark. This results in a situation more times than I care to count: A young woman is brought in on charges of identity theft and grand theft auto. The main ID theft victim is not present (he’s flown home but Visa is likely to also be a plaintiff), and the GTA victim is a rental car company. The young woman often bears strange marks on her body, and has a flashy phone or jewelry, but no other possessions. Last, many times she won’t be able to name what city she is in, and this is after looking behind her in court to get a nod from her handler, who is present in the gallery.”
This is just one example of local conditions creating unique fact patterns around criminal defendants as victims of human trafficking. Details can vary widely between regions depending on whether agriculture, construction, personal services (nail salons) or prostitution are the main venues. With such variation, more data are critical to fighting the problem.

The legal technology tool suite Neota Logic supports the current version our the tool, and hosts it here: https://education.neotalogic.com/a/Traffic_Light. In addition, the American Society of Legal Engineers supplies financial, logistical and technical support. We cannot continue the fight without outside help however, and donations can be made through the ASLE to support our cause.
Project Traffic Light needs more information from professionals in the justice system to enlarge the scope and improve the accuracy of the tool. Please contact us at projecttrafficlight@gmail.com so we can schedule an interview.