The dark side of AI exposed as underage workers are being exploited as they train AI models

The dark side of AI exposed as underage workers are being exploited as they train AI models

In a recent article by Wired, the dark side of the rapidly expanding artificial intelligence industry was exposed. It was found that thousands of underage workers from the global south are working for companies that provide AI data-labelling services, and in turn being exposed to traumatic content.

Wired found that many of the workers who are contributing to the development of machine learning algorithms are underage. They find work on various online crowdsourcing platforms such as Toloka, Appen and Clickworker. On these platforms, remote workers are connected with major tech companies, providing data labelling services. 

With machine learning, data labeling is the process of looking at raw data (images, text files, videos, etc) and adding labels to give context so a machine learning model can learn from it. For example an image recognition task could be distinguishing between cats and dogs in a dataset of images. The worker would need to give cat and dog labels to the appropriate image. This would then allow the AI model to categorize the animals accurately going forward. A worrying example that was given in the Wired article was one task of analysing pages of partially naked bodies (lingerie ads, sculptures, paintings, sexualised images) to help the algorithm distinguish between adult content, and acceptable forms of nudity. 

Although platforms require workers to be over 18, many motivated underage individuals are able to sidestep these restrictions by giving fake details, or even using relative's information. The global rush into AI is propelling the data labeling and collection industry to an estimated worth of over $17.1 billion by 2030. Companies like Amazon, Microsoft Azure, Google, and Adobe outsource micro-tasks to these platforms, which, in turn, employ a predominantly young workforce based in East Africa, Venezuela, Pakistan, India, and the Philippines. 

Workers are paid per task, with earnings ranging from a cent to a few dollars. Tasks include data annotation, content moderation, and solving captchas. The work often involves uploading personal content, such as photos of babies or children, for AI training. Some workers engage in explicit content moderation tasks that take tolls on their mental health.

The lack of oversight and distant nature of the work has led to an influx of underage workers, with approximately one-fifth of respondents in a survey indicating they were under 18. Researchers also highlight instances where workers share accounts within families, further complicating age verification.

The article highlights the exploitative nature of this digital work, with underage workers earning minimal wages for tasks that contribute significantly to the advancement of AI technology. The imbalance in benefits, coupled with concerns about digital colonialism, raises ethical questions about the industry's practices and its impact on vulnerable populations in the global south.