The conversation around AI often overlooks the unseen human workforce that helps make the technology possible. Artificial intelligence research institute Partnership on AI noted in a blog post that the “data enrichment work” that assists in making machine learning models optimal is largely carried out in precarious conditions and by faraway workers. This includes tasks such as labeling images, providing feedback on potentially harmful or explicit material, and giving feedback to AI tools so they can provide accurate and up-to-date information.
One company that partners with AI providers in this regard is Sama, a US-based machine learning training institution that contracts remote staff from countries such as Kenya, Uganda, and India. In 2021, OpenAI, the makers of ChatGPT, used Sama for training purposes. The workers received between $1.32 and $2 per hour and many reported feeling mentally scarred by the sensitive nature of the work. OpenAI offered mental health support for the employees but it was reportedly not sufficient and the contract was cancelled after only eight months.
Another method for training AI is Amazon Mechanical Turk. This crowdsourcing marketplace has a network of 500,000 workers worldwide who bid for tasks such as surveys, data entry, and content moderation. However, the median earnings per hour for a worker is only $1.77 and the tasks can sometimes contain adult content. Amazon claims that the use of their platform is much more cost and time-effective than using a traditional temporary workforce.
The unseen human workforce behind AI, both those working for AI providers directly and those using Amazon Mechanical Turk, deserve recognition and appropriate compensation for their work. Companies like OpenAI and Amazon need to take greater responsibility in ensuring the mental wellbeing of workers and that wages are fair. This is the only way for machines to accurately understand the nuances of language, culture, and judgments needed to make AI truly useful and safe.
