Artificial intelligence today offers a relatively safe environment. It can automatically filter violence, block sensitive materials, and protect users from extreme content. In short, AI recognizes what is wrong and can respond to it quickly. However, a new investigative report by The Guardian has uncovered the darker side of training artificial intelligence. Its safety depends on thousands of women in India who watch extremely violent and sexually explicit content every day to help algorithms reliably identify this harmful material.
Artificial intelligence needs teachers
AI cannot determine on its own what is good and what is not. Someone always needs to show it and teach it to recognize the differences. And labeled data about violence, pornography, or abuse must be presented to it by ordinary people working as so-called content moderators and data annotators.
Currently, tens of thousands of women in India perform this work, with each worker viewing up to 800 videos and images depicting violent content every day. Their task is to carefully review the data, determine whether it violates the rules, and teach AI that such content is harmful. Without this human element, the safety filters of AI models would be virtually ineffective.
Mental health under pressure
Every content moderator watches videos of rape, abuse, physical violence, or accidents on a daily basis. According to reporters from The Guardian, many of them admit that processing such explicit content is extremely difficult.
This is also highlighted by psychologists – according to them, the work of data annotators and content moderators is a highly risky profession in terms of mental health. Research primarily points to anxiety, insomnia, recurring thoughts, and secondary trauma. However, the companies commissioning this work often do not see it as problematic and provide these women with no adequate care.
Women from rural areas are in demand
A large portion of this work is outsourced to India, and by around 2021, approximately 70,000 people were already working in the data annotation and moderation industry there. Employees from rural areas are in the highest demand. And the reason is simple: an available workforce and lower costs.
According to surveys, as many as 80% of these workers come from rural areas or the outskirts of cities. Moreover, women make up a substantial majority. They are generally considered more meticulous, and this type of employment allows them to earn a stable income without having to leave their families and homes.
Job advertisements do not tell the whole story
However, the most problematic aspect of the entire situation is that companies advertise the work in highly unspecific and vague terms. Many women therefore respond to an innocuous offer involving AI training or “data annotation,” which promises payment for viewing graphic content from home. And that is often how it actually begins. Only after signing the contract do they learn that they will not be dealing with harmless content, but with extreme material.
Employees are also bound by NDA clauses that prevent them from discussing their work with their families. One possible avenue is therefore psychological support, but this is the exception rather than the rule.
The dark side of the AI revolution
The stories of these women are opening up a broader debate about the ethics of artificial intelligence. While technology companies present AI as the pinnacle of innovation, part of its infrastructure relies on these “cheap invisible workers.” Better and safer AI therefore depends not only on data and computing power, but also on real people and hours spent watching violent content. Hopefully, however, improvements in automated filters will soon reduce the need for human review.
Source: The Guardian
