Artificial intelligence is not as artificial as you might think

Data workers in precarious conditions are behind every AI system

Behind every AI system is a huge amount of human effort. Most of this work is not done by well-paid engineers, but by millions of precariously employed workers. The World Bank estimates that there are between 150 and 430 million of such workers worldwide. An exact number is hard to determine because many of these workers are freelancers without formal employment contracts. That’s not a coincidence. It’s by design.

Without these workers, AI wouldn’t exist.

Data workers are creating data that AI systems need. Without huge amounts of human-created data, AI systems wouldn’t exist. Examples include collecting data; labelling data (saying what an image has in it) or organising data so that a machine can process it. You might have seen this kind of work in CAPTCHA systems – imagine doing that kind of thing, for 10+ hours a day under enormous time pressure and surveillance. But data work expands beyond labeling and moderation. It also includes data generation, verification or impersonation.

AI or… humans?

Tech companies want us to believe that AI can do anything – and they’ll lie to try to convince us. Sometimes, an AI system that has been rolled out to the public actually just has people operating it behind the scenes – like robots in grocery stores in Japan controlled by remote workers in the Philippines, or Amazon’s ‘AI-powered’ cashier-free shops in the US, which actually relied on data workers in India inputting items (just like cashiers, but invisible).

This is manipulative on many levels. Lying like this completely ignores and invisibilises the human labour. It feeds into the (fake) narrative that AI systems are more advanced than they are. As a result, people might choose to invest or use such systems in their work, without realising they’ve been tricked. The technology behind AI is still imperfect, so humans have to step in to maintain the illusion of smart machines and seamless automation. In some places, human labor is so inexpensive that it’s actually cheaper than developing a fully automated system.

Why is data work so important?

AI systems can’t ‘improve’ without the quality of data they rely on also being more accurate, and without there being more of that data. The best way to get accurate data is to get humans to create that data. This work is also vital because everyone – from companies to governments – is chasing the AI hype, constantly pushing for bigger, faster, and smarter models. All of this comes at a huge cost, both in terms of human labor and ecological resources. And those doing the work are often the ones who benefit the least.

AI systems perpetuate systems of colonial exploitation

AI companies are some of the wealthiest companies on the planet, and they rely on the work of data workers – yet they fail to pay workers a fair wage, and treat them as disposable. Companies like OpenAI, Google, Deepseek and Anthropic scrape the web indiscriminately, collecting everything, from everyday content to pornography, violence, and illegal activities. Humans are needed to filter out the unwanted material and teach the machines what they’re seeing. By labeling images and other content, workers are training AI systems.

Without that, AI systems wouldn’t be able to filter out harmful or violent images. The violence in these systems doesn’t go away; it’s just outsourced to other people – and it’s no coincidence that much of that labour takes place in the Global South.

Joan Kinuya’s story

Joan is a former data worker. She holds a university degree in business and began data labeling because it was advertised as a pathway for women to enter the tech industry. Over time, she has annotated nearly everything: street images for self-driving cars, photos of homes taken by robot vacuums, even scans of passports. The tasks were so fragmented that Joan often had no idea which company or product she was contributing to. She worked from home, using a computer she had to buy herself, along with the broadband internet connection she needed for the job.

At her busiest, she spent up to 20 hours a day in front of her computer at home. Working as a freelancer, she had to be available around the clock just to secure enough tasks. The pay was low, the workload highly unpredictable, and some of the content deeply disturbing. At times, she even had to annotate images depicting extreme violence or dead bodies.

Then, a couple of years into her work as a data worker, the company simply deleted her profile overnight, suddenly leaving Joan without a job or any income.

Data workers fighting back

After that, Joan joined with other data workers to advocate for data workers’ rights – through a new organisation they set up, called the Data Labelers’ Association. This organisation is just one of many pushing for the rights of data workers around the world.

Together, they fight for:

• Fair wages and social protections for everyone doing data work.
• Mental health support for those exposed to disturbing content.
• Transparent and fair terms of employment.
• The right to organize, speak out, and assemble freely.
• Tech companies to take responsibility for the safety and well-being of their data workers.

Read more about data workers’ demands, visit SUPPER’s digital magazine edition about the unheard stories of data workers, and visit Data Workers Inquiry, a global, radically participatory research project by Milagros Miceli.

Contribution by: Julia Kloiber (she/her), Co-Founder and Managing Director & Zara Rahman (she/her) Managing Director, EDRi member,  SUPERRR Lab
This article was first published here by EDRi member,  SUPERRR Lab