Archive · 9 Jun 2025
The Invisible Workforce: How Humans Train Artificial Intelligence
Imagine teaching a robot to understand the world.
Before the robot can do anything, people called annotators need to label data, like pictures of roads or sounds.
Many people around the world do this work, but their jobs are changing.
As AI becomes more advanced, it needs fewer basic instructions and more expert guidance.
Some people are concerned about the pay and conditions of these annotators, while developers are trying to create AI that learns more on its own.
Now, humans are needed for more complex tasks, like making sure the AI is safe and works well.
Think of it like the AI growing up from a child needing constant help to a teenager needing specialized advice.
Millions of workers, often in low-wage jobs, are involved in annotating data to train AI models.
AI development is shifting, reducing the demand for basic annotation and increasing the need for skilled specialists.
Annotation work is increasingly outsourced to developing nations like India.
The quality and ethics surrounding AI annotation practices are under scrutiny.
The role of humans is shifting toward evaluating and refining AI outputs as models become more complex.
- Who
- The article discusses data annotators, AI developers, AI companies (e.g., Scale AI, OpenAI, DeepSeek), and computer scientists (e.g., Fei Fei Li).
- What
- The article is about the hidden workforce behind AI, specifically the role of data annotators in training AI models and the changes occurring in this field.
- When
- The article discusses the evolution of AI and its workforce over time, mentioning specific developments since the mid-2000s and particularly since 2022.
- Where
- The article references AI annotation work being outsourced globally, with a focus on the US, Europe, China, and India.
- Why
- The article explores the changing landscape of AI development, highlighting the evolving role of human annotators and the shift towards more advanced AI models, as well as the economic and ethical implications of these changes.
Annotation Workforce
AI Developers
Human Input
Annotation Workforce
Human input, especially for tasks like evaluation, remains critical.
AI Developers
Prefer minimal human input, aiming for self-teaching AI.
Compensation and Work Conditions
Annotation Workforce
Concerns over wages and working conditions, including monitoring and low pay, are prevalent.
AI Developers
Focus on development and efficiency.
Future of Annotation
Annotation Workforce
Anticipate the need for skilled annotators to continue supporting AI development.
AI Developers
See a decline in the need for basic annotation as AI evolves.
- Start of ImageNet and AI Annotation
- Mid-2000s
- Images organized by Fei Fei Li
- 3.2 million
- Potential annual annotation revenue in India by 2030
- $7 billion
- Expected annotation jobs in India by 2030
- 1 million
- Decent pay for American annotators
- $10-20/hour
- Pay for annotators in poorer countries
- $4-8/hour
- Increase in AI startup funding in 2024
- 50%
- Time spent on data work in AI development
- 80%