Machine learning systems often require human annotators to label or describe data before it can be used for training. For example, in the development of self-driving cars, human workers need to annotate dashcam videos by labelling cars, pedestrians, bicycles, etc. to teach the system to recognise different road elements.
This task is often delegated to contract labourers in the global South, who work in precarious jobs and receive wages that are barely above poverty level. In some cases, this work can be very stressful. For example, Kenyan workers had to view and flag content containing violence, explicit material and hate speech in order to train ChatGPT to avoid engaging with such topics.