Joan Kinyua, 38, a former data labeller from Nairobi, Kenya, has revealed the disturbing nature of her work, which involved thinking about how to kill, skin, cook, and eat a human while her two-year-old daughter played in the next room. Joan, co-founder of the Data Labelers Association (DLA), which has nearly 900 members worldwide, told Metro that she was tasked with labelling gruesome content to train AI systems.
The hidden human cost of AI
Data labellers, also known as data annotators or AI tutors, tag text, images, and videos to teach algorithms. Joan's work included labelling surgical videos and responding to prompts about cannibalism, which left her unable to touch a laptop or use social media. 'These are things that I thought nobody ever thought about, especially publicly asking on laptops,' she said.
Michael Geoffrey Asia, secretary general of the DLA, tagged pornography for nearly a year, leading to insomnia, PTSD, and trouble having sex. He recounted witnessing a video of three men molesting a 13-year-old, which left him with lasting trauma. 'Having to witness three huge guys molest a 13-year-old, that remained with me,' Michael said. 'I have a daughter. It put me in a position where I couldn't trust any other man close to my child.'
Exploitation and lack of support
Many data labellers work through middlemen and sign non-disclosure agreements, leaving them without support. Michael only received therapy after caring for a cancer patient. Joan also trained companion bots, believing she was chatting with real people, which made her feel the humans were vulnerable.
A job board shown to Metro included listings like 'Inspector Genital' for selecting real or animated genitals, paying $0.30 for 1,000 tasks. Other gigs included naming films for $0.15 and determining controversial images for $1.50. Tasks are often timed, with workers expected to complete them within a minute.
Industry impact and calls for change
The Data Workers' Enquiry, representing workers from nine countries, found that AI tutoring can increase mental health issues and drug dependency. Camilla Salim Wagner, a research intern at the Distributed AI Research Institute, said tech giants look to African nations for data labelling to cut costs. The data collection and labelling market was valued at about £16 billion last year and is projected to reach £88 billion by 2034.
Joan and Michael are calling for better pay and mental health support. 'Everyone is so determined to train machines how to respond with empathy,' Michael added. 'But no one has ever trained people to love people like us, who have been working on this information all along.'
