South African workers are being paid to teach AI the jobs they studied for years

# South African workers are being paid to teach AI the jobs they studied for years

A Rest of World essay describes a South African PhD graduate being recruited to train an AI system on the skills and judgments needed to design assessments, teach undergraduate students and mark essays. The episode captures a fast-growing corner of the AI economy: companies want experts to transfer not just knowledge, but the reasoning that underpins professional work.

The author says the offer arrived just after finishing a doctorate on how technologies such as artificial intelligence are reshaping agricultural work in South Africa. Instead of moving straight into a university post, the first call came from a recruiter. The pitch was unusual even by AI industry standards: help teach a model how to do work that normally takes years of academic training.

The pay explains part of the attraction. The role was offered at 600 rand, or about $37, an hour in a country where the national minimum wage is 30.23 rand, or roughly $2, an hour. The essay also points to a labor market under strain, with youth unemployment at 47.4% in the second quarter of 2026. In that setting, even a job that feels ethically awkward can look like a serious opportunity.

That tension is the heart of the piece. The author argues that people are increasingly being hired to train systems that may one day perform their own jobs. Some highly skilled writers are being paid to make AI-generated prose sound human. In India, the essay says, waste sorters and welders are wearing cameras or phones to record their work for low hourly pay so humanoid robots can learn from them. The pattern is familiar: labor that once sat at the center of a profession is being extracted and repackaged as model training data.

The risk, the author argues, is especially acute in Africa. The essay says the continent remains below the global average for AI adoption, but it has a young and growing pool of educated workers living with unemployment, low wages and modest economic growth. That combination makes expert labor relatively cheap for AI firms, and it may encourage companies to source judgment from African professionals without offering much long-term upside in return.

The article also describes how AI companies are moving beyond simple data labeling. Platforms including Outlier, Mercor and Surge are recruiting doctors, lawyers, engineers and teachers because their work depends on discretion. The value is not just in what those professionals know, but in how they decide. In the author’s case, an AI interviewed him for 45 minutes and then emailed feedback on strengths and weaknesses. When it suggested he retake part of the assessment, he walked away from the job.

The result is a sharp question for workers and employers alike: if AI systems are learning from the people whose livelihoods they may later undercut, what exactly is being handed over? The essay does not resolve that dilemma, but it shows how personal, economic and ethical pressures are becoming intertwined in AI hiring.

Event date: 2026-09-07