AI Peril #3 – Data Swamps By Alec Hill

The Danger and Damage of AI

A local contractor hired Mophat Okinyi to do AI “content moderation” for $2.20 an hour; it seemed like a gift from heaven. Mophat Okinyi grew up in an impoverished Kenyan village where jobs were scarce. Meta and OpenAI hired the contractor to scrub bad files from their databases.

The gift quickly turned into a nightmare. Mophat’s job was to review 15,000 videos and texts a month. He spent nine hours a day weeding out images of rape, suicide, child abuse, beheading, incest, and bestiality.

Over time, the graphic content caused vicarious trauma, emotionally breaking him. Dreams became so awful that he couldn’t sleep. Withdrawing into himself, he shut out loved ones. In his wife’s departure note, she wrote: “You’ve changed. You’re not the same man I married. I don’t understand you anymore.” She was pregnant with their first child.

Documented Dangers

According to research by faculty at the University of Minnesota, more than half of African content moderators suffer from clinical depression. One in four uses drugs (legal or illegal) to cope with mental health issues.  

By outsourcing content removal to people in developing nations like Kenya, India, and the Philippines – even to refugee camps in Lebanon and Colombia – tech companies hide the ugly underbelly of their business model from customers, employees, and investors.

Context and Background

Before the launch of ChatGPT 3.5 by OpenAI in 2022, Google held a decided technological lead over its rival. But due to concerns about unsavory content in its training data, Google leadership delayed moving forward until its internal ethics board approved. OpenAI placed no such filtering guardrails and won the race to produce the breakthrough program.

As the AI frenzy ensued, ethics boards disappeared. With huge financial implications at stake, imperfect new products were rolled out as quickly as possible. Moral concerns, it was rationalized, could be fixed later.

Garbage In, Garbage Out

AI reflects whatever data it is fed. Large Language Models require an incredible amount of raw information. The resulting “data swamp” contains falsehoods, biases, and graphic content. DALL-E, an image program, scraped 250 million photos and content items from datasets replete with graphic sexual content, ranting political blogs, racist slurs, and the entire email collection of Enron (a bankrupt company famous for committing fraud).

Hence, it is no coincidence that within two years of the launch of ChatGPT 3.5, the U.S. National Center for Missing and Exploited Children reported a 1,500% spike in online child sexual abuse materials. Amazon’s figures rocketed from about 5,000 reported incidents to 1 million. Observes a data scientist: “If you hoover up a ton of the internet, you’re going to get child sexual abuse material.” Talk about a moral cesspool.

Our Consciences

The reporting on this topic by Karen Hao, formerly of the Wall Street Journal, is deeply disturbing.

As we learn more about the global poor taking on AI’s nasty piece work, Scripture should be screaming in our ears.

The prophet Jeremiah condemns those who “become fat and sleek” on the backs of the poor (5:28) while Isaiah promises judgment on those who live well off the “spoil” of the disenfranchised (5:28).

In the New Testament, Jesus pronounces woe on those who cause the powerless (like Mophat Okinyi) to stumble (Luke 17:1-2). The apostle James lambasts those who live high on the hog at the expense of the less fortunate (5:1-6).

As consumers, we should avoid AI providers with the worst track records on dirty content. We must do all we can to become less complicit in harming others. 

Alec Hill is President Emeritus of InterVarsity Christian Fellowship USA. This essay is an adaptation from the forthcoming fourth edition of his book Just Business: Christian Ethics in the Marketplace (IVP). 

Don’t miss the complete AI Peril Series:

AI Peril #1: Flattering Chatbots By Alec Hill

AI Peril #2: Diminished Critical Thinking and Your Brain By Alec Hill

Now is the time to reserve your seat for the Outcomes Conference Global Digital Experience, September 1 – October 31, 2026.

Don’t miss the 2026 Leader2Leader Podcast series, the Multiplication Factor, sponsored by FaithSearch Partners!