[Baking Typewriter] Humanity Exploited by the Drudgery Behind AI

Meta and Other AI Firms Outsource Content Moderation of Policy-Violating Videos
Ten-Hour Shifts in African Slums for Content Moderators
Watching Suicide, Torture, and Rape Videos Almost Daily
No Solutions in Sight for Harsh Working Conditions... Meta and Others Turn a Blind Eye

Mercy works as a content moderator for Meta at an outsourcing company in Nairobi, the capital of Kenya. She spends 10 hours a day checking a "flag" every 55 seconds. This time, the video is a graphic car accident. Someone filmed the scene and posted it on Facebook, and due to the disturbing nature of the footage, someone flagged it as inappropriate. Mercy's main job is to determine whether flagged posts violate Meta's content policies. As she stares at her computer screen as usual, Mercy suddenly freezes. The victim in the horrific accident is none other than her own grandfather. When she regains her composure and reports the incident to her supervisor, the response is shocking: "If you want to meet your quota for the day, you need to get back to your desk. You can take a day off tomorrow, but since you already came to work today, it's better to finish your shift."

Three authors, who research digital labor at the University of Oxford and the University of Essex in the United States, shed light on the hidden side of artificial intelligence that has swept the globe. They argue that AI is an "extraction machine" that absorbs human labor, creativity, and emotion. An extraction machine is a structural device that absorbs human emotion, creativity, and labor to generate data, processes it into algorithms, and then produces profit from it.

The term "extraction machine" highlights the inhumane labor of data annotators, content moderators, and logistics workers that underpins the convenience of AI in our daily lives. The authors report on what they witnessed firsthand over ten years across more than 30 countries. They focus on how AI deepens inequality and threatens democracy.

The core message of the book can be summarized as: "Machines grow by consuming humans. They devour our labor, our creativity, and our time. Then they convert it into data and statistics and feed it back to us." Content moderation center workers process about 1,000 flags per day. They are exposed to content involving suicide, torture, and rape almost every day. "The hardest part isn't the violent videos." "Sexually explicit and shocking content is even harder." "To do this job, you have to accept the abnormal as normal." Most testified that it was difficult to return to a peaceful state of mind after work. "Most of us are psychologically scarred. Some have attempted suicide, and some colleagues have had their spouses leave them."

What is even more difficult, they say, is the "company policy." Employees who left their desks after viewing shocking footage were reprimanded for violating company policy. If they left their desks without entering the required code into their computers before using the restroom, their productivity scores were penalized. Such workplaces are often set up in African slums, where 100 people sit in a row in front of computers, having their emotions suppressed to support AI algorithms.

However, the authors point out that Meta deliberately turns a blind eye to these circumstances. They criticize Meta's attitude of denying responsibility by claiming that these are the actions of outsourcing companies. While global AI companies continue to emphasize the convenience and possibilities brought by AI, the authors stress that behind this glamorous facade lies the sacrifice of humans forced to endure harsh working conditions and labor exploitation. "If content moderators did not constantly monitor posts, social media would quickly be flooded with violent or explicit content. If data annotators did not teach AI to distinguish between traffic lights and road signs, self-driving cars could not operate on the roads. Without workers tuning machine learning algorithms, AI tools like ChatGPT could not exist."

AI Eats Us Alive | Mark Graham et al. | Translated by Kim Dowan | 348 pages | 24,000 won