ghost work

basicwidespread

Ghost work is the invisible, mostly poorly paid human labor without which fully automatic-looking AI systems would not function: labeling, verification, content filtering.

In plain terms

When an AI system recognizes a cat in a photo, filters out violent video, or politely answers your question, it seems the machine does it all alone. In reality, many people worked on it behind the scenes: labeling images, scoring answers, fixing errors. You do not see them, you do not know their names, so their work is called ghost work.

It is like a restaurant kitchen: the guest sees only the beautifully plated dish, not the dishwasher or the cook. The difference is that here the 'kitchen' is often on another continent, and the worker is not an employee but a casual laborer paid pennies per task.

The concept is important because it corrects the picture of AI: human knowledge and human work stand behind the system. It also raises the question of whether the way this work is paid and recognized is fair.

How it works

A company has countless small tasks that are still difficult for machines: labeling images, describing audio recordings, evaluating AI responses, filtering harmful content.
The task is broken down into small parts, called microtasks, and distributed via an online platform or through a subcontractor company.
People sitting anywhere in the world perform these tasks, typically for piece-rate pay, without employment contracts, paid leave, or social security.
Software collects and verifies the result, then the AI model learns from it, or this is used to fix its errors.
The end user sees only the finished, 'intelligent' product and is unaware of human involvement.

An everyday example

You ask a chatbot something and receive a precise, polite answer. What you do not see is that thousands of people previously scored similar answers and marked which ones were offensive, incorrect, or dangerous. The answer is partly the result of their work.

Who is affected

employeeTask-split, platform-based work is appearing in more professions and shows what work looks like without a safety net. → Work & HR

jobseeker, new graduateOnline microtasking seems like an easily accessible income opportunity, but the hourly wage is often very low and unpredictable. → Work & HR

entrepreneur, executiveIf you purchase an AI service or order data labeling, you are responsible for the working conditions in your supply chain, at least morally and regarding your reputation.

everyoneAnyone using AI indirectly consumes the results of this invisible work.

A common misconception

MythAI learns on its own and operates without humans.

In realityMost AI systems rely on massive human labor behind the scenes: data labeling, evaluation, and moderation. This work does not stop after the model is completed, as the system must be continuously monitored and improved.

Limits and risks

No one knows the exact number of ghost workers; global estimates range from several million to tens of millions, so treat specific figures with caution.
Content moderators are exposed daily to violent and disturbing material, which is documented to cause psychological harm.
Companies rarely reveal how much human work is behind their product, so as a buyer it is difficult to choose a fair provider.
The work takes place in a legal gray zone: workers are mostly not employees, so they receive little protection.

Similar, but not the same

szellemi munka – Mental work is the opposite of physical work, while ghostwork refers to the invisible background work behind AI, as if by a "ghost".
platformmunka – Platform work is a broader term that includes food couriers and drivers you see; ghost work is the hidden digital part.
ember a folyamatban (human-in-the-loop) – This is a design principle that openly and intentionally places a human beside the machine's decisions. Ghost work is not a principle, but the hidden, underpaid human labor of which the user is unaware.

Related concepts

Broader

platformmunka, gig economy, MI-etika

Narrower

data labeling, content moderation, mikromunka

On the edge

crowdsourcing, ember a folyamatban, training data, reinforcement learning from human feedback, algoritmikus menedzsment

A brief history

The term was popularized by anthropologist Mary L. Gray and computer scientist Siddharth Suri in their 2019 book, Ghost Work. The phenomenon itself is older: Amazon launched its Mechanical Turk platform in 2005, where anyone could post such microtasks.

Where things stand

At the end of 2026, mass human labeling and evaluation work, with a growing share of expert tasks, still underpins generative AI. EU member states must transpose the Platform Work Directive by 2 December 2026, but most of this work takes place outside the EU.

Status: 2026 November

Also known as

ghost work · szellemmunka · invisible work · mikromunka · microwork · crowdwork · clickwork · clickwork

Sources

Wikipedia: Ghost work

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.