Ghost Workers of AI: The Human Labour Behind Automation

Why in the News?

1. The world is rapidly moving towards an automated economy, where Artificial Intelligence (AI) systems are celebrated for being fast, efficient, and seemingly independent.
2. However, these systems are heavily dependent on human labour for data annotation, content moderation, and fine-tuning.
3. Reports have emerged, particularly from countries like Kenya, India, and the Philippines, highlighting labour exploitation, low wages, and mental health risks faced by invisible AI workers.

Key Highlights

1. The Promise of Automation
   1. AI systems are marketed as producing quick, efficient, and error-free outputs.
   2. The popular narrative suggests that AI functions autonomously, with minimal human input.
2. Hidden Human Contribution
   1. In reality, AI requires extensive human labour and energy resources.
   2. Data annotators label raw images, audio, video, and text, creating datasets that train AI and Machine Learning (ML)
   3. Example: Self-driving cars distinguish traffic signs from humans only because annotators labelled such features in training data.
3. Role in Training Large Language Models (LLMs)
   1. Training of LLMs like ChatGPT and Gemini involves three steps:
      1. Self-supervised learning – machine ingests large datasets.
      2. Supervised learning – annotators refine the data.

• Reinforcement learning with human feedback – annotators judge AI outputs, correct errors, and prevent misuse.

1. Hence, quality of human input directly shapes AI accuracy.

4. Global Outsourcing and Exploitation
   1. Tech companies in Silicon Valley outsource annotation tasks to low-cost labour markets in Kenya, India, Pakistan, China, and the Philippines.
   2. Workers face long hours, very low wages (sometimes <$2/hour), and strict deadlines.
   3. Many are unaware of the end-client (big tech firms) due to subcontracting via intermediary platforms.
5. Content Moderation and Mental Health
   1. Even so-called “automated” social media filters rely on moderators who label sensitive and harmful content.
   2. Workers often review graphic violence, pornography, and disturbing imagery, leading to PTSD, anxiety, and depression.
   3. A 2024 letter by Kenyan workers to Joe Biden compared their conditions to modern-day slavery.
6. Other Forms of Invisible Work
   1. Voice actors and performers are engaged to train AI in speech, singing, and movement recognition.
   2. Reports suggest even children have been involved in such tasks.
7. Labour Suppression and Lack of Rights
   1. Workers attempting to raise concerns faced job termination and union dismantling.
   2. The fragmented gig-work system, where labour is broken into “microtasks”, allows companies to avoid responsibility and accountability.
8. Call for Regulation
   1. The advancement of AI is powered by “ghost workers” — unrecognised, exploited human labour.
   2. Experts argue for stricter laws and regulations on AI companies, ensuring fair pay, transparency, and dignity at work in global AI supply chains.

Key Terms

1. Data Annotation
   1. The process of labelling raw data (text, images, audio, video) to make it understandable for AI models.
   2. Helps AI systems learn patterns (e.g., tagging “dog” in an image so AI can identify dogs).
   3. Can be simple (object tagging) or complex (medical scan interpretation).
   4. Directly impacts the accuracy of AI outputs.
   5. Labour-intensive, often outsourced to low-cost economies.
2. Large Language Models (LLMs)
   1. AI systems like ChatGPT trained on massive text datasets.
   2. Learn through self-supervised learning, supervised learning, and reinforcement learning with human feedback (RLHF).
   3. Can generate human-like text, answer queries, and perform language tasks.
   4. Dependent on continuous human fine-tuning for accuracy and safety.
   5. Criticised for being resource-intensive and reliant on hidden human labour.
3. Reinforcement Learning with Human Feedback (RLHF)
   1. A method where humans rank and correct AI outputs.
   2. Ensures AI avoids harmful or biased responses.
   3. Essential for making AI responses aligned with human values.
   4. Labour-intensive as workers must evaluate outputs one by one.
   5. Example: AI refusing harmful requests due to human-in-the-loop correction.
4. Gig Economy
   1. A labour market based on short-term contracts or freelance work.
   2. In AI, workers are paid per microtask (e.g., tagging 100 images).
   3. Provides flexible employment but often lacks job security, social protection, and fair wages.
   4. Enabled by digital platforms that connect big tech firms with cheap global labour.
   5. Raises regulatory challenges for labour rights enforcement.
5. Digital Colonialism
   1. A form of economic domination where data, digital services, and labour from developing countries benefit big corporations in developed nations.
   2. Developing nations provide cheap labour and raw digital inputs but do not capture the real value.
   3. Reinforces North-South inequality in technology.
   4. Seen in AI where annotators in Kenya or India train models profiting Silicon Valley companies.
   5. Calls for equitable participation in the digital economy.

Implications

1. For the AI Industry
   1. Reliance on low-cost human labour questions the ethics of AI development.
   2. Errors increase when non-experts are employed for specialised tasks (e.g., medical scans).
   3. Sustainability of AI systems depends on addressing labour exploitation risks.
2. For Labour Rights and Governance
   1. Raises concerns of labour informalisation, wage exploitation, and poor working conditions.
   2. Exposes gaps in international labour standards enforcement in the digital gig economy.
   3. Highlights need for collective bargaining rights and unionisation protections for digital workers.
3. For Developing Countries
   1. Countries like India, Kenya, and the Philippines are becoming labour hubs for AI.
   2. While providing employment opportunities, the model risks perpetuating digital colonialism — where value is captured by Western companies while workers remain underpaid.
   3. Exploitation could widen the North-South divide in technology economies.
4. For Society and Mental Health
   1. Exposure to graphic and violent content harms workers’ mental well-being.
   2. Creates a hidden public health issue in developing nations where psychological support systems are weak.
   3. Raises ethical concerns about outsourcing trauma from Global North to Global South.
5. For Policy and Regulation
   1. Need to extend labour laws into the digital economy, covering gig and microtask workers.
   2. International collaboration is necessary to enforce fair wage standards in the AI supply chain.
   3. Transparency in outsourcing contracts and accountability of big tech companies is critical.

Challenges and Way Forward

Conclusion

The so-called automated economy is sustained by a vast workforce of “ghost workers” — data annotators, moderators, and gig workers spread across the developing world. Their invisible contribution ensures AI accuracy, safety, and usability, but their labour is often underpaid, traumatising, and unprotected. For AI to be truly ethical and sustainable, labour justice, fair wages, and recognition must be integral to its governance. Otherwise, the celebrated AI revolution risks being built upon a foundation of digital exploitation.