Kenyan AI trainers faced poor working conditions amid AI investment boom — Report

Artificial Intelligence (AI) labellers in Kenya have reported unfavourable working conditions despite a boom in global AI investment.

According to reports, these labellers, also known as humans in the loop, (people sorting, labelling, and combing through reams of data to train and improve AI for companies like Meta, OpenAI, Microsoft, and Google) note that the job “needs to be done accurately and fast, and to do it cheaply, it’s often farmed out to places like Africa.”

“In the job, you are teaching robots or machines how to think like humans, to do things like humans,” said Naftali Wambalo, a labeller for videos and images, in a recent interview with 60 Minutes.

Wambalo and his colleagues spend eight hours a day in front of a screen studying photos and videos, drawing boxes around objects, labelling them, and teaching the AI algorithms to recognise them.

He added, “You’d label, let’s say, furniture in a house. And you say, “This is a TV. This is a microwave.” So, you are teaching the AI to identify these items. And then there was one for the faces of people. The color of the face. “If it looks like this, it is white. If it looks like this, it’s Black. This is Asian.” You’re teaching the AI to identify them automatically.”

Large language models (LLMs), like GPT-4, are advanced machine-learning models that mimic human-like understanding and content generation. Trained on extensive data, they excel at recognising complex language patterns and can perform diverse tasks.

Supervised learning comes next. It entails training the AI model against a labelled dataset, whose elements are labelled by human annotators, so that the model can learn ‘right’ responses from ‘wrong’ ones. This is supplemented by reinforcement learning from human feedback (RLHF). It involves human evaluators creating a supervised training dataset from specified prompts, responses, and human-evaluated rankings against which the model must retrain itself.

Nerima Wako-Ojiwa, a Kenyan civil rights activist, revealed that the role is like modern-day slavery.

She said, “I think that we’re so concerned with creating opportunities, but we’re not asking if they are good opportunities. The government has been courting tech giants like Microsoft, Google, Apple, and Intel to come here, promoting Kenya’s reputation as the Silicon Savannah.”

Wambalo noted that they are paid about $1.50 – $2 an hour, gross before tax.

Wambalo and his colleagues were hired by an American outsourcing company called SAMA. SAMA employs over 3,000 workers for Meta and OpenAI. According to documents, OpenAI agreed to pay SAMA $12.50 an hour per worker, much more than the $2 the workers got. However, SAMA notes that the $2 is a fair wage for the region.

Global investment in Artificial Intelligence (AI) is projected to reach $632 billion in 2028, according to the International Data Corporation (IDC). However, these investments do not represent the reality of the people.

When asked why he took the job, Nathan, another digital labeller disclosed, “I have a family to feed. And instead of staying home, let me just at least have something to do.”

The workers also noted that if they had finished their tasks before the stipulated period, they would forfeit their pay for the rest of the period and be rewarded with food.

Wambalo said, “They used to say thank you. They give you a bottle of soda and KFC chicken. Two pieces. And that is it.”

The workers also claimed that they were subjected to psychological harm through the content they interacted with for hours. They added that it affected their interactions with their families and friends alike.

“I looked at people being slaughtered, people engaging in sexual activity with animals. People abusing children physically, sexually. People committing suicide.”

Fasica noted that she was reviewing graphic and disturbing content. “I was watching dismembered bodies or drone attack victims. You name it. Whenever I talk about this, I still have flashbacks.”

When asked about labour laws, Wako-Ojiwa said, “Our labour law is about 20 years old. It doesn’t touch on digital labour. I do think that our labour laws need to recognise it, but not just in Kenya alone. Because what happens is when we start to push back in terms of protecting workers, a lot of these companies shut down, and they move to a neighbouring country.”