Lessons from Facebook (Meta) for Ending Human Trafficking with Analytics

Recent revelations from the Facebook Papers have led many to ask if the company Facebook (recently renamed Meta Platforms Inc.) is doing enough to correct the role it has played in the facilitation of global human trafficking across its platforms (Facebook, Instagram, Messenger and WhatsApp).

Human trafficking is modern-day slavery and involves the use of force, fraud or coercion to obtain some type of labor or commercial sex act. It takes the freedom away from the victim, creates a false reality for the buyer, and brings out the worst in the trafficker who enslaves the victim. Each person involved loses a sense of humanity. Human trafficking, however, is not the same as smuggling, which is based on movement across borders. The term “trafficking” in “human trafficking” is based on the exploitation of humans and involves the trade of humans.

The Facebook Papers are a set of internal documents that were provided to the U.S. Congress in redacted form by the legal counsel of whistleblower Frances Haugen as proof that the leaders of Facebook repeatedly and knowingly chose maximum user engagement for advertising profits and company image over public good. A consortium of news organizations reviewed the documents in recent weeks and have reported findings. Conclusions related to analytics and human trafficking indicate that the platform design decisions and algorithms provided vulnerable communities around the world with large amounts of dangerous and inflammatory content. For example, from 2017-2020, Facebook’s algorithm gave a high weight to content that received an “angry” emoji and a low weight for “likes,” which exacerbated the spread of disturbing and dehumanizing content. In addition, the Facebook and Instagram platforms were used to run ads for human trafficking. The apps WhatsApp and Messenger were used to facilitate victimization.

In 2018, Facebook partnered with Article One, who conducted the 2018 Human Rights Impact Assessment (HRIA) of the Facebook platform on Indonesia and Sri Lanka. Facebook received several analytics-related recommendations to improve its impact on human rights.

Since Apple considered removing the Facebook and Instagram apps from its App Store in 2019, Facebook (Meta) has invested in the targeted use of analytics to remove content related to specific instances of human trafficking from its platforms.

It is unclear why in 2020, Facebook deactivated a software tool to automatically detect content violating their slavery (“domestic servitude”) policy. As of the last update on the Facebook transparency page in July 2021, internal reviewers label violating content using three categories: hate speech, violence or nudity. There is not currently a category for content related to human exploitation.

As Facebook has discovered its role in global human trafficking, there are a few lessons that the analytics community can learn about efforts to end human trafficking with analytics.

Lesson 1: Monitoring content is more important than you think

The United States and Canada account for roughly 10% of monthly Facebook users and about 85% of the investment to monitor content. In contrast, nearly 100% of new users of Meta apps are in developing countries, where the apps are the main online communication channel and source for news. In 2020, 13% of the investment to monitor content was used for developing countries.

Although CEO Mark Zuckerberg has downplayed the role of the platform in social actions, when faced with being pulled from the Apple App Store in 2019 for being used to facilitate human trafficking, he responded by removing accounts, pages and hashtags, and developing technology to proactively find and take action on content related to human trafficking disguised as domestic servitude in Arabic-speaking countries. In one week, Facebook removed 130,000 pieces of Arabic content and launched proactive detection tools in Arabic and English to disrupt the use of its platform for slavery, disguised as domestic servitude content.

Lesson 2: Algorithms need to speak multiple languages and adapt to diverse laws and customs

Despite more than 40,000 employees and contractors around the world working on content review, according to the Facebook Papers, analyzing comments in various languages and dialects, and routing questionable content for review are key analytics challenges for Meta. Although artificial intelligence is the backbone of Meta’s policy enforcement, language is a barrier for algorithms, which are currently not capable of handling different dialects. This allows inflammatory content that violates community standards to thrive on the platform in developing countries.

In addition, the laws and customs of some developing countries do not support a hard stand against human trafficking when it is discovered. The leadership and law enforcement practices can complicate efforts to end human trafficking. 

Lesson 3: Analytics can have a significant disruptive impact in a short period of time

Some Facebook responses have indicated a desire for users to report content that violates community standards, but Facebook can use analytics to have a more significant impact than user reports in regulating its platform.

For example, from April to June 2021, Facebook proactively found and flagged 97% of more than 26 million pieces of violating content (e.g., photos, videos, comments and posts) related to the endangerment, nudity, abuse or sexual exploitation of children while users reported less than 3% of violating content. Of the flagged content found, almost none of it was appealed or restored. The significant disruptive impact of removing this content in a short period of time indicates what is possible with analytics.

Lesson 4: Suppliers are resilient to satisfy demand

The primary action taken by Facebook when faced with violating content is to remove pages, accounts and content that violate community standards. When justice is common in a country, Facebook involves the local police department. In general, as content is removed, it can be replaced as suppliers show their resiliency to satisfy demand. Single exercises to remove supplied content are insufficient to eliminate demand.

In a February 2021 report, Facebook acknowledged that opportunities remain in improving prevention, detection and enforcement efforts regarding human trafficking. Since 2012, INFORMS community members have performed research in these areas, including using satellite imagery and various technologies to uncover patterns in human trafficking, using deep learning, fuzzy logic, natural language processing, machine learning, spatial analysis, dialectic synthesis, Bayesian networks and neural networks to find victims and traffickers. This research uses data outside Meta platforms.

Approaches to ending global human trafficking for labor differ from those to end commercial sexual exploitation. The exploitation of children and adults in the labor market is reduced by institutional changes. The backdrop of structural motivators for demand include poverty, discriminatory immigration measures, unregulated labor markets, unsafe working conditions, lack of labor and workers’ rights protections, awareness among investors and consumers of how they can reduce demand, and legal migration channels for migrant workers.

In contrast, Facebook and the INFORMS research community has the potential to greatly influence demand for commercial sexual exploitation of children with analytics, particularly at the intersection of the individual buyer’s interaction with the internet and various technologies. In the market for commercial sexual exploitation of children – where children as young as 12 years old are sold to 10-15 buyers per night – without demand, supply would end. Buyers of commercial sex are rarely prosecuted and those who are tend to not become repeat offenders. Analytics is needed to disrupt buyer access, analyze databases of buyer information to characterize buyers, and employ technology to redirect buyers to identified alternatives.

The updated INFORMS Analytics Collection on “Ending Human Trafficking with Analytics” [1] includes articles, TED Talks, videos, research papers and other links to databases, websites and software tools regarding recent efforts to reduce both supply and demand of human trafficking in labor and commercial sex markets. We hope your review of this collection inspires you to lend your support in this effort.

Reference

1. https://pubsonline.informs.org/editorscut/trafficking