After completing a Ph.D. earlier this year on technologies such as artificial intelligence reshaping agricultural work in South Africa, I was ready to start an academic career. I imagined that I would teach students, supervise research, and shape the next generation of scholars. But my first job offer did not come from a university. It was from a recruiter, who invited me to train an AI system to design assessments, teach undergraduate students, and mark essays. In short, I was being asked to transfer everything I had learned over the last decade to AI.
I agreed to an interview. You may ask: Why would anyone agree to train an AI model in the very skills they took years to acquire? The answer, in part, is the socioeconomic reality in South Africa. The role offered 600 rand ($37) an hour in a country where the national minimum wage is 30.23 rand ($2) an hour, and where youth unemployment stood at 47.4% in the second quarter of 2026. So when the opportunity to earn a livelihood collides with your principle on helping a technology that can compete with you one day, what do you do? I am not the only one who had to consider that trade-off.
When the opportunity to earn a livelihood collides with your principle on helping a technology that can compete with you, what do you do?”
People are increasingly being hired to train the very AI systems that will perform their jobs one day. Highly skilled writers are being paid low wages to “humanize” or edit AI-generated texts. In India, waste sorters and welders are strapping phones or cameras to their heads to capture their everyday activities for 250 rupees ($2.60) an hour. That data is being used to train humanoid robots that will replace the very jobs these workers rely on.
While a waste sorter sits at the opposite end of the pay scale from me, we are essentially doing the same job: transferring human knowledge and judgment to a machine. If this becomes a defining feature of the AI economy, Africa will be exposed. The continent has one of the lowest AI adoption rates in the world, with most countries below the global average of about 18% among the working-age population. South Africa fares slightly better, with an adoption rate of 23%.
Yet the continent has among the youngest populations, and a growing pool of highly educated professionals in an environment of widespread unemployment, low wages, and modest economic growth forecasts. For AI companies, there is an incentive to extract expert knowledge from African professionals at relatively lower costs than in Western societies.
The AI jobs on offer today for someone in Africa who is well-educated are different from the data labeling and data annotation jobs that have long been outsourced to Kenya, Nigeria, and elsewhere. The requirements are also different: I had been scouted not just to transfer knowledge but also my judgment that I have gained from years learning in the classroom, and teaching undergraduate students. Over thousands of hours, I have learned to judge what concepts truly matter in an undergraduate course, how to evaluate essays, how to tell the difference between memorization and genuine understanding, and why one student should get 75% instead of 60% on an essay. The AI would not simply learn what I knew; it would learn how I decided.
What does it mean for us to hand over the judgment that defines our character and our professions to a machine?”
Several white-collar professions are defined by discretion. Lawyers exercise legal judgment, doctors exercise clinical judgment, and teachers exercise pedagogical judgment. This kind of knowledge was considered difficult, if not impossible, to automate because it depends on interpreting context rather than simply applying rules. This is why doctors, lawyers, and engineers are being actively recruited now by AI platforms like Outlier, Mercor, and Surge.
In my own recruitment process, no human was involved. I was interviewed for 45 minutes by an AI, which then emailed me feedback on my strengths and weaknesses. Traditionally, this decision was made by humans, but now AI has taken over, and decides who gets hired. The AI that interviewed me suggested I retake part of the assessment. I didn’t, and stopped pursuing the job.
I wish I could tell you my choice was based on principle, that I refused because I feared helping to build the technology may replace the work I do. But I still don’t fully understand the source of my own discomfort, nor do I have answers to the questions I have been grappling with. What does it mean for us to hand over the judgment that defines our character and our professions to a machine? What kind of judgment are we teaching these systems? Are we teaching them to be fair, to be good, to be ethical, and to truly recognize context?
I still do not know why I walked away from the AI training job. Perhaps I will understand better in time. Meanwhile, I continue to look for an academic job teaching students, not AI.
