Data Labeling

Frequently, machine learning systems need human annotators to label or describe data before it can be utilized for training. For instance, in the development of self-driving cars, human workers must annotate dashcam videos by outlining cars, pedestrians, bicycles, etc., to teach the system to recognize different road elements. This task is often delegated to contract workers in the Global South, who may face unstable employment and receive barely-above poverty-level wages. In some cases, this work can be distressing, such as when Kenyan workers had to view and label content containing violence, explicit material, and hate speech to train ChatGPT to avoid engaging with such topics.