The AI boom runs on hidden labour. Behind every chatbot response and algorithmic prediction are thousands of workers labeling data—often for pennies per task, in conditions rarely seen or acknowledged by the tech giants they serve.
By Stuart Kerr, Technology Correspondent
Published: 17 July 2025
Last Updated: 17 July 2025
Contact: liveaiwire@gmail.com | Twitter: @LiveAIWire
The Invisible Backbone of Machine Intelligence
Every AI system needs training. Before an algorithm can recognise sarcasm, detect cancer, or generate coherent text, it must first learn from labelled data. That data doesn’t label itself.
Enter the clickworker—individuals employed through digital platforms to annotate images, transcribe speech, flag hate content, or classify text snippets. This is the digital underclass of AI: essential, expendable, and almost always underpaid.
In a landmark 60 Minutes report, workers in Kenya and India described spending hours reviewing graphic content for content moderation algorithms. Their pay? Often less than $2 an hour. These jobs, marketed as flexible gig work, in practice resemble modern-day piecework—high-pressure, psychologically taxing, and hidden from the glossy press releases that celebrate AI breakthroughs.
Tech’s Hidden Supply Chain
The structure of data-labelling platforms mirrors that of global manufacturing: decentralised, opaque, and difficult to regulate. A revealing article by Ludditus highlights how workers are bound by NDAs, stripped of rights, and often blacklisted for low-quality scores—assessed by the very algorithms they’re helping to train.
The parallels to exploitative garment or electronics supply chains are striking. While tech companies speak of responsible AI, their models are trained on human toil with little oversight or accountability.
In LiveAIWire’s previous analysis of AI-driven wellness systems, we explored how personal profiling can subtly coerce users. In the context of clickwork, profiling takes on a more brutal edge—affecting livelihoods through opaque rating systems.
Ethics on Autopilot
A recent PDF study from the Weizenbaum Institute lays bare the ethical vacuum in this digital labour market. It notes that many AI firms outsource ethical risk to third-party contractors, shielding themselves from accountability while reaping the benefits.
This is compounded by structural inequality. Clickworkers are overwhelmingly based in the Global South. They face high rejection rates, non-payment for incomplete tasks, and lack any form of collective bargaining. As the arXiv paper “Digital Labor” outlines, this new form of work exists in regulatory limbo: not quite employment, not truly freelance, and rarely protected.
The stakes are not just financial. Content moderation tasks, particularly those involving violence or exploitation, have been linked to long-term psychological trauma. And yet, companies continue to treat this labour as disposable—a troubling contrast to the billions being invested in AI startups.
Human Effort, Machine Glory
AI achievements dominate headlines. Whether it’s diagnosing disease or generating realistic images, the praise goes to the model—not the humans who fed it.
In LiveAIWire’s emotional intelligence feature, we noted how simulated empathy is built from real human expressions. Behind that empathy is someone who manually tagged thousands of facial images for emotion classification. And behind that someone is likely a person earning below minimum wage.
The AI industry has mastered a kind of magical thinking: attributing success to algorithms while ignoring the labour required to make them function. As Time Magazine revealed, even OpenAI relied on outsourced workers in Nairobi to filter violent and sexual content used to train ChatGPT—again for a few dollars an hour.
Reclaiming Recognition and Rights
Ethical AI cannot be built on unethical foundations. Transparency in data pipelines must include not just the provenance of the data, but the conditions under which it was labelled. Just as “Fair Trade” labels changed coffee sourcing, it’s time to consider what “Fair Data” might mean.
There are signs of change. Activist groups are beginning to unionise clickworkers. Legal scholars are proposing frameworks for gig data labour rights. But for now, the vast majority of these workers remain unseen.
As explored in LiveAIWire’s mental health coverage, trust in AI systems is eroded when the human cost is ignored. If we value the outputs of AI, we must value the inputs—including the people.
About the Author
Stuart Kerr is the Technology Correspondent for LiveAIWire. He writes about artificial intelligence, ethics, and how technology is reshaping everyday life. Read more