Asset 1
Nashville AI stops human trafficking
The Heart Inside the Machine: Using AI to Combat Modern Slavery

The incredible story of how artificial intelligence has connected a non-profit, governments, and the banking industry in the global fight against human trafficking.

“I’ve seen video content of a child that’s the same age as mine being raped by an American man that was a sex tourist in Cambodia. And this child was so conditioned by her environment that she thought she was engaging in play.”

With palpable emotion, Ashton Kutcher delivered these words to the Senate Foreign Relations Committee hearing on human trafficking and slavery. His impassioned testimony generated headlines around a world where, hidden from view, 27 million adults and 13 million children are victims of modern slavery. Any notion that slavery was disgrace of history is crushed by the realization that there are more slaves today than at any point in our past.

The dark side of the internet

Kutcher may be best known as an actor, but for anyone doubting his reason to speak on this issue he had a formidable rebuke. As chairman and co-founder of Thorn, he told the committee, his “day job” is to lead a non-profit organization committed to using technology to fight human trafficking and the sexual exploitation of children.

Thorn advances developments in technology to tackle the problem of modern slavery because, in common with many other industries, much of the market has moved online. Demand for prostitutes means almost 60 percent of trafficked humans are sold into the sex trade. Most are under the age of 24, but some (as Kutcher’s example underlines) are considerably younger. Online advertising is the contemporary equivalent of a seedy back street or phone box stickers, but reaches a much larger audience. When Thorn surveyed underage sex trafficking victims, 3 out of 4 said they had been advertised online.

The scale of the market is hard to fathom. Every day, more than 100,000 escort ads are posted to the internet in the United States alone. Law enforcement faces a massive challenge in identifying both victims and perpetrators within this enormous and ever-changing pile of data. Finding ads that indicate a trafficked victim among those which are legitimate requires trained officers. To say that the volume of the data has overwhelmed the available human resources is an egregious understatement.

Hidden in plain sight

Law enforcement had already adopted technology to help mitigate the task, applying keyword analytics to sift through escort ads and highlight dubious language. The problem with this technique is that human traffickers deliberately avoid using words or terms that might trigger an alert. Meanings are implied, spellings are distorted, and ads are continually edited and reposted to make it hard for officers to keep track.

Keyword analytics solves the problem of churning through huge volumes of data, but without any real understanding of meaning or context it frequently reports false positives while all too easily missing what is significant. This state of affairs leaves trafficked minors hidden in plain sight and law enforcement with no option but to commit to a complex and highly manual investigation task.

This type of computing problem had long irritated Tim Estes. Frustrated with the inability of computers to learn from data, he’d been inspired to build software that could understand the semantics of human language and how that are influenced by context. His company, Digital Reasoning, was formed with a mission to use this ability to help humans solve the world’s toughest problems.

Digital Reasoning found its first customers in the shadow of the 9/11 attacks on the United States. Working with the defense and intelligence communities, the company proved that machines could learn human communication and flexibly extract relevant details without relying on keywords or lexicons. Ongoing development led to software able to read vast amounts of communications data, learn what’s important, and cross-compare between multiple sources. These capabilities enabled it to report insights, patterns and relationships not apparent to the human eye.

From the battlefield to banks

Success in intelligence and on the battlefield attracted interest from other quarters. Rocked by scandals of market fixing and rogue trading, the financial services industry was keen to find better ways to identify corrupt employees. “Both Wall Street and the intelligence world want the same thing: to find unknown unknowns in the data,” said Roger Hockenberry, the former chief technology officer of the Central Intelligence Agency’s Directorate of Operations.

Pressure from regulators, billions in fines, and substantial reputational damage had caused banks to swell their compliance organizations. Despite this, finding the bad apples in businesses with many thousands of employees was a major challenge. Compliance officers knew that colluding employees used email and chat messages to plan and enact their crimes, but they faced similar challenges to their peers in law enforcement. Inferred meanings and coded terms made it difficult for keyword analytics to spot the messages that mattered. Compliance teams had little option but to manually review thousands of alerts, most of which would be false positives.

By this stage, Digital Reasoning had honed its technology into an artificial intelligence platform, Synthesys, that could be flexibly applied to analytics tasks where human communication was the main data source. Unlike keyword based solutions, Synthesys emulates a human-like ability to make sense of human communication in context. It learns as it works, allowing it to cope with nuance, aliases, coded terms, and ambiguity. What had proven to be effective for national security also worked for banks, with Synthesys piecing together insights that could not be extracted using conventional technology.

Artificial intelligence helped in two key ways. First, deeper insights and greater accuracy proved to be much more effective at disclosing which employees posed a threat, protecting the banks from future regulatory censure. Second, the technology was a huge time saver for compliance officers. Using the intelligent assistants built into Synthesys, insights were automatically resolved to specific entities (people, places, objects, events), focusing attention on the problem’s source and visualizing the hidden connections with other employees. This combined impact was just what compliance organizations needed. Within the space of two years, Synthesys became the most widely adopted artificial intelligence technology among leading investment banks.

Finding finance’s felons

Miscreant employees were not the only headache for compliance chiefs. Fines for failures to prevent money laundering were another source of huge cost and embarrassment for the industry. Virtually every major bank has been hit by millions, sometimes billions in fines. Many were quick to see the potential for using artificial intelligence to root out financial crime.

This was another situation where human communication in immense volumes was an impediment to progress. Despite acting in accordance with know-your-customer or KYC regulations that were created to prevent criminals exploiting the financial system, banks were struggling to confirm customers’ identities and intent, and to reveal any hidden relationships or undeclared interests. Inadequate knowledge allowed money launderers to circumvent banks’ defenses and disguise the source of their funds by using false identities, financial “smurfs” (people who use their account to make many small deposits and transfers on behalf of the money’s true owner), and by establishing shell corporations.

Financial crime and compliance chiefs believed that that data analytics might hold the answer. Barry Koch, formerly Senior Vice President and Chief Compliance Officer for The Western Union Company, described a process for using financial traits that indicate risk to identify potential criminality. While many of the attributes and characteristics he defined can be gleaned from transactional data, Koch highlights the importance of trawling public source information to improve the quality of the analysis. Useful sources for evidence of aliases and criminal interests include news sites, companies records, and court reports.

With a limited number of trained officers to call on, compliance organizations had focused their resources on customers deemed to be a higher risk. However, being based on flawed knowledge, risk scores were not reliable. Moreover, the time taken to complete investigations meant that illegal activity could continue for months, sometimes years before being spotted. KYC was a thoroughly broken process, but this time the banks could turn to artificial intelligence to solve the analytics problem.

The link to human trafficking

Human trafficking is typically understood as a human rights issue, but this assessment tends to overlook the financial incentives that drive it. Estimates of its profits range from $9 billion to $31.6 billion. A study in the Netherlands found that a pimp’s average earnings from a single sex slave were $250,000 per year. The Panama Papers exposed the magnitude of financial crime and the truth about how illegal funds are generated, among them the profits from human trafficking. Money launderers give these profits a veneer of legitimacy, allowing them to be moved into the global banking system.

Governments and intergovernmental organizations had enacted legislation to target human trafficking, but many nations have failed to implement these laws and among those which have the number of convictions remains low. Baroness Goudie, a member of the United Kingdom’s House of Lords and chair of the United Nation’s Women Leaders’ Council to Fight Human Trafficking, argues: “Ultimately, no number of declarations will end the business of human trafficking. We need real investment from governments, and the involvement of people who are trained not only to spot human trafficking but also to follow the money.”

Tougher legislation is starting to emerge. The UK Prime Minister, Theresa May, has promised to “lead the way in defeating modern slavery.” Focusing on the profits of human trafficking is a key part of the obligations set out in the country’s Modern Slavery Act. This stipulates that any firm with a turnover above £36 ($44) million must publish an annual statement of the steps it has taken to ensure that modern slavery is not occurring in its business. Crucially, this extends beyond the UK and includes all parts of its overseas supply chain. The threat of fines and reputational damage has sent bosses scrambling. As Edward Naish of PwC warns, “Imagine waking up to news reports that your business is taking no steps to combat slavery. It doesn’t bear thinking about.”

As Goudie points out, targeting the money is the most effective means of undermining modern slavery, but when that money is laundered it requires the analytical capabilities of artificial intelligence to expose the true identities of people and organizations linked to these crimes. Mounting pressure on businesses and banks to make more thorough checks on their customers and suppliers is helping to frustrate human traffickers, at least at the more senior levels of criminal networks. However, in a heartening twist of providence, the very artificial intelligence being used to root out financial crime in banks and business was about to fight human traffickers head on.

A solution for law enforcement

With Synthesys being used to defend national security and bring integrity back to banking, Digital Reasoning was making progress on its mission to use technology to help solve the world’s toughest problems. However, when the firm’s executives met Ashton Kutcher’s organization, Thorn, and learned how law enforcement was being engulfed by data volumes, they could not ignore the potential of Synthesys to deliver a solution.

Experiments with artificial intelligence were not in law enforcement budgets, but Digital Reasoning and Thorn formed a partnership to explore how Synthesys could be used to respond to the challenge. Supported by donated time and funds from sponsors, the team worked to create an intelligent assistant that could take on much of the rote work and complexity of analyzing escort ads.

The result was Spotlight: a web-based tool that provides law enforcement with intelligence about suspected human trafficking networks. Using the artificial intelligence capabilities of Synthesys, it uses text analytics and computer vision to automate the process of identifying which escort ads are for trafficked individuals.

An outsized impact

To date, Spotlight has assisted in more than 8,300 investigations conducted by 780 law enforcement agencies in all 50 US states. It has contributed to the identification of 6,625 victims and, by revealing hidden relationships in the data, has also help officers unmask 2,255 pimps. By bringing a human-like acuity to big data analytics, Spotlight has accelerated investigation times by 60%.

“Our mission matches the magnitude of the internet, but by working with Digital Reasoning we found a way to empower law enforcement to identify and rescue trafficked children by turning huge volumes of data into an asset,” said Julie Cordua, CEO of Thorn.

Working with banks, first to tackle internal rule breaking and later to identify and disrupt the activities of money launderers, has enabled Digital Reasoning to continue to support the development and maintenance of Spotlight. Now both Thorn and Digital Reasoning are eager to expand the use of Spotlight beyond the US. The UK government is looking at the technology and this would provide an anchor point to expand its use across Europe.

The fight to stop modern slavery and human trafficking will not be easily won, but artificial intelligence is empowering law enforcement, governments, and the banking industry to find the sources of these crimes and bring their perpetrators to justice.