[ARTICLE START]
The workers who make artificial intelligence possible can earn as little as six Australian dollars a day. They draw boxes around images, annotate audio files, and sort through text and video so that AI models can learn from the data. Without them, the systems that power everything from self-driving cars to banking software would not function.
According to a report by Phys.org, a researcher has been interviewing data workers in China and Australia to document what this labor actually looks like. The fieldwork is ongoing, but findings from ten interviews so far reveal a workforce defined by low pay, physical strain, and almost no legal protections.
Before an AI model can process information, human workers must categorize, label, test, and moderate massive volumes of text, images, audio, and video to make the data usable for AI training. This work is performed by a growing global digital workforce that prepares datasets not only for large technology companies but also for banks, insurance companies, health care providers, and government agencies, including defense departments.
The inequality across the data labor market is sharp. Workers with doctoral-level qualifications and STEM credentials can sometimes access specialized tasks that pay between A$400 and A$800 per hour. But those positions are rare and difficult to obtain. Most workers perform general tasks, such as repeatedly drawing bounding boxes around objects in images used to train drones, self-driving vehicles, and automated vending machines. For that work, pay can fall to A$6 per day or less, an amount that workers say cannot cover basic daily expenses. Long hours at screens leave many with chronic eye strain and back pain.
One of the workers interviewed put the situation plainly: "We do the manual work so that they get the credit for the intelligence."
The structure of the work itself offers workers little recourse. Data laborers have no formal contracts and are not classified as employees. Platforms refer to them as "users" and assign tasks based on expertise and track record. The user agreements that govern their work are written primarily to protect the companies behind the outsourced tasks. Workers are typically required not to disclose any information they encounter, even though the datasets they handle are already anonymized. Many workers have no way to identify which company's data they are processing. They also do not know whether a human or an AI agent reviews their completed work, and they have minimal rights to appeal any assessment of their performance.
The researcher describes this arrangement as not unlike other poorly regulated jobs in the gig economy. Workers are called on when tasks are available and have no guarantee of steady work or income.
The report notes that precarious labor markets and marginalized social status have pushed digitally literate young workers into the data-labeling industry. Geographic location plays a significant role in determining pay and access to specialized work, with workers in the Global North generally earning more for comparable or equivalent tasks.
The data-labeling workforce supports industries far beyond consumer technology. Datasets prepared by these workers feed into high-stakes systems used in health care diagnostics, insurance risk modeling, financial services, and military applications. The scale and reach of the industry make the conditions these workers face a matter that extends well beyond any single technology company or platform.
The researcher's fieldwork in China and Australia is continuing, and additional interviews are expected to add to the picture of how this global labor force operates and is compensated.
