Workers Reveal the Dark Side of Training AI: Devalued Jobs, Strange Mistakes, and a Looming Threat

Workers grappling with the rapid growth of artificial intelligence have expressed feeling devalued by the technology, warning of a downward trajectory in work quality. Recent analysis by the International Monetary Fund found AI could affect about 40% of jobs globally, with its head describing it as "like a tsunami hitting the labour market". Here are firsthand accounts from workers who have trained AI models to replace parts or all of their roles.

The Editor: Earning Less While Correcting AI's Strange Mistakes

Christie*, an editor for academics with English as a second language, was asked to train "assistant editors" without knowing they were AI programs. She assumed the company was training more people to handle the workload, but soon found herself correcting strange mistakes like unnecessary full stops or nonsensical country name changes. After months, the company revealed the assistants were AI, and her fee was reduced. "I now earn less money for correcting the mistakes of an AI, which takes me longer than editing from scratch," she says. Christie feels devalued, betrayed, and furious, trapped in a toxic cycle due to high work volume and financial needs.

The Palliative Care Consultant: AI Struggles with Human Nuances

Mark Taubert, a palliative care consultant, worked on a chatbot pilot to help patients with metastatic cancer. The AI was fed guidelines and patient questions but struggled with human pronunciation and errors, such as patients saying "morphium" instead of morphine. It got about 50% of responses spot-on but required adaptations for safety, including handling troubling questions. Taubert believes AI won't replace his role, as it relies on nuances like body language and facial expressions, but could enhance his work by reducing administrative duties.

The Translator: AI Leads to Decline in Quality

Philip*, a translator, has spent four years training AI-based translation engines that his supervisors aim to replace human translators with for cost savings. Despite improvements, the AI still produces formulaic and unreliable results, requiring word-by-word review. "It doesn't save time over directly translating the material myself," he says, noting a decline in quality. While AI is okay for rough translations, its unreliability means human oversight is still essential, with the threat of replacement looming.

The Marketing Writer: Training Your Robot Replacement

Joe*, an award-winning marketing writer, was laid off after spending months building AI workflows and documentation, believing he would oversee the system. "Training your robot replacement feels like digging your own digital grave," he says. His former workload is now handled by junior employees using his AI documentation. At 50, he faces uncertainty about future layoffs and is considering a career pivot into sales.

The Mathematician: AI Transforming Mundane Tasks

Filippo, an associate professor in mathematics, collaborates with startups on AI projects to prove theorems with minimal human input. While results are still limited, AI tools are getting stronger and more efficient daily. He predicts a mathematician's work will look completely different in 10 years or less, with AI replacing mundane tasks. However, he doesn't feel immediate job threat due to his teaching role and public institution employment, but worries for younger professionals.

*Names have been changed.