Unseen labour, exploitation: the hidden human cost of Artificial Intelligence, Pg11

Key Highlights:

• AI systems rely heavily on human labor for data annotation and training, often performed by workers in developing countries.
• Data annotators label images, audio, video, and text to train AI and Machine Learning (ML) models, enabling AI to understand and process raw data.
• Tech companies outsource data annotation work to countries like Kenya, India, and the Philippines, where workers face low wages and long hours.
• Human moderators are essential for filtering sensitive content on social media, which can lead to severe mental health issues due to exposure to graphic material.
• AI tech workers in Kenya reported poor working conditions, including exposure to harmful content and low pay, leading to accusations of modern-day slavery.

Detailed Insights:

• AI models require high-quality datasets, increasing the demand for human labor in data annotation to improve output accuracy.
• Large Language Models (LLMs) like ChatGPT undergo training in stages, with human annotators fine-tuning the data for accurate responses and error removal.
• Non-expert data labelers can contribute to errors in AI outputs, prompting companies to seek experts for specialized tasks like medical scan labeling.
• Social media platforms use human moderators to censor sensitive content, but this work can cause severe mental health issues like PTSD in workers.
• AI companies minimize costs by outsourcing work through digital platforms, leading to fragmented labor networks and a lack of transparency.
• Exploitation of AI tech workers includes low pay, strict deadlines, constant surveillance, and dismissal for failing to meet targets.
• Stricter laws and regulations are needed to ensure transparency, fair pay, and dignity for workers in the AI labor supply chains.

Key Concepts Involved:

• Data Annotation: The process of labeling raw data to train AI and Machine Learning models.
• Large Language Models (LLMs): AI models trained on vast datasets to generate human-like text.
• Machine Learning (ML): A type of AI that allows systems to learn from data without explicit programming.