The cost of AI is someone else's time
The article argues that while AI appears to save time and effort for users, the true cost is often hidden in the labor of human workers in regions like the Global South, who are paid low wages to label data, filter harmful content, and fine-tune models. This outsourced "ghost work" underpins AI's seamless experience, raising ethical questions about fair compensation and working conditions.
Background
- AI tools such as ChatGPT and Midjourney rely on large language models (LLMs) trained on vast amounts of data scraped from the internet, including copyrighted images, text, and code.
- This article highlights that the cost savings AI offers to companies is often externalized — shifted onto artists, writers, and developers whose unpaid or uncompensated labor (their creative works, bug fixes, forum posts) becomes training data.
- A key concept here is "data labeling" or "annotation": low-paid gig workers in the Global South manually tag and label data to train AI systems, a hidden human cost behind the "automated" product.
- The piece argues this is a form of wage theft or exploitation, making AI's efficiency a social cost rather than a pure technological gain. It connects to larger debates about fair use, copyright law, and the ethics of generative AI training data.