I refused to train the AI that could replace me

I recently completed a Ph.D. on artificial intelligence's impact on agriculture in South Africa and anticipated embarking on an academic career. However, my initial job offer came from a recruiter, who sought someone to train an AI system to create assessments, teach students, and grade papers. In essence, I was being asked to transfer my decades of knowledge to artificial intelligence.

This prompted me to accept an interview. The question arises: why would anyone agree to train an AI in the very skills they have invested years to attain? The answer lies in South Africa's economic landscape, where the national minimum wage is $2 an hour and youth unemployment is at 47.4% in the second quarter of 2026. When the prospect of earning a living clashes with one's principles about preventing technology from potentially replacing their profession, what course of action should be taken?

I am not the only one grappling with this dilemma; many highly skilled professionals find themselves compelled to train AI systems that may eventually replace their work. Writers are paid meager wages to refine AI-generated texts, while waste sorters and welders in India don head-mounted devices to capture their daily activities for $2.60 an hour, the data then used to train humanoid robots to replace their jobs.

Despite the stark contrast in pay scales, we share a common task: transferring human knowledge and judgment to machines. If this becomes a defining characteristic of the AI economy, Africa will face significant challenges. The continent boasts one of the lowest AI adoption rates globally, with most countries below the global average of approximately 18% among the working-age population.

South Africa fares better, with an adoption rate of 23%, yet the continent contends with a youthful population and a growing number of highly educated professionals facing unemployment, low wages, and sluggish economic growth. AI companies exploit this scenario by offering higher-paying AI jobs to well-educated Africans compared to data labeling and annotation jobs commonly outsourced to regions like Kenya and Nigeria.

These roles require a unique skill set, as they involve not just transferring knowledge but also judgment gained from years of classroom learning and teaching undergraduate students. Over countless hours, I have honed my ability to discern what concepts truly matter in an undergraduate course, evaluate essays, distinguish between memorization and genuine understanding, and determine whether a student deserves a 75% or a 60% grade on a paper.

The AI would not merely learn what I knew; it would learn how I make decisions. White-collar professions, such as law, medicine, and engineering, rely heavily on discretion. Lawyers employ legal judgment, doctors rely on clinical judgment, and teachers utilize pedagogical judgment. These skills are often deemed challenging, if not impossible, to automate due to their contextual interpretation requirements.

Consequently, doctors, lawyers, and engineers are being actively recruited by AI platforms like Outlier, Mercor, and Surge. During my recruitment process, I underwent a 45-minute interview conducted by an AI, which subsequently emailed me feedback on my strengths and weaknesses. Traditionally, such decisions were made by humans, but now AI plays a role, determining who gets hired.

The AI that interviewed me recommended I retake part of the assessment. I declined, and ultimately, I opted not to pursue the position. While I can't definitively attribute my decision to principle, I remain uncertain about the source of my discomfort regarding the prospect of entrusting judgment to a machine. Moreover, I lack answers to the questions I have been contemplating.

How does handing over the judgment that defines our character and professions to a machine impact us? What kind of judgment are we imparting to these systems? Are we teaching them to be fair, ethical, and capable of recognizing context? My uncertainty persists, but I continue searching for an academic position that allows me to teach students, not AI.