Ghost Workers of the AI Era

Artificial intelligence feels really magical.

You type a sentence, and it writes an essay. Upload a picture, and it identifies the objects. Ask a chatbot a question, and it understands you like a friend.

We talk about AI as if it thinks. As if it learns alone. As if it builds itself.

But behind every “smart” machine is this sort of invisible workforce: the ghost workers: hundreds of thousands of real people who label data, filter trauma, and perform the emotional labour that lets AI function.

We call it artificial intelligence, but its foundation is painfully human.

AI systems don’t come preloaded with knowledge; they learn from examples, millions and billions of them.

Every image labeled “cat,” every audio clip transcribed, every piece of content categorized as safe or unsafe, someone had to do that work.

This is called data annotation, and it is the backbone of all this new modern AI.

Companies often claim their algorithms are built on “massive datasets.”

But what they don’t say is that these datasets were labeled by human beings sitting in small rooms across Kenya, India, the Philippines, Pakistan, Venezuela, and other low-income countries, often earning less than $2–$4 per hour.

Ghost workers are frequently hired through outsourcing platforms such as:

1. Sama
2. Appen
3. Scale AI
4. Clickworker
5. Amazon Mechanical Turk

These workers are not seen as employees, they are “taskers,” “annotators,” “micro workers.”

Their job is fragmented into tiny tasks:

1. tagging emotions on faces
2. identifying hate speech
3. labeling medical scans
4. transcribing audio in dozens of accents
5. sorting user posts into categories
6. flagging violent or sexual content

Each task is paid in cents.

The workers must meet strict accuracy thresholds, respond fast, stay invisible, and accept that their labour will appear nowhere. Not in the final product, not in the research papers, not in the conference presentations. Absolutely nowhere.

If data annotation is invisible labour, content moderation is invisible trauma. AI systems cannot filter abuse, pornography, gore, racism, hate, or violence by themselves. Every piece of content flagged as “unsafe,” every hateful comment removed, every disturbing video that never reaches your feed, someone had to watch it, label it, categorize it.

In 2023, a TIME investigation revealed that OpenAI outsourced moderation to workers in Kenya earning $1.32 to $2.00 per hour, where they had to review violent assault, child abuse, and graphic murder footage to train AI to detect dangerous content.

Moderators report:

1. nightmares
2. PTSD symptoms
3. anxiety
4. depression
5. emotional numbness
6. desensitization

Their emotional wellbeing becomes the hidden cost of our digital cleanliness.

They’re cleaning the internet, and paying the psychological price. Because invisibility is the business model.

Tech companies market AI as:

1. fully autonomous
2. frictionless
3. magical
4. intelligent

Admitting that thousands of people sit behind screens labeling data disrupts the mythology. Admitting that the internet must be manually cleaned breaks the illusion of safety. Admitting that trauma is outsourced destroys the narrative of ethical AI.

There are movements fighting for:

1. better wages
2. mental health support
3. regulated working conditions
4. transparency about data labour
5. recognition in AI documentation
6. ethical sourcing of training datasets

Researchers argue for “data labour rights.” Labour unions are beginning to organize annotators. Whistleblowers are bringing moderator trauma to light. But we’re still far from acknowledging the human scaffolding behind AI.

The question is not whether AI will advance. It will.

The question is whether its progress will continue to depend on invisible human suffering. AI may be the future. But behind every algorithm, every model, every output,

there is a human being you’ll never meet: a ghost keeping the machine alive.