The covert army specializing in ‘teaching’ AI.

Training AI models like ChatGPT is considered an important step, but the team behind it only receives a few dollars per hour.

Alexej Savreux, 34 years old, living in Kansas City (USA), stated that he has done various jobs for many years, ranging from selling bread and fast food to security and garbage collection. Now, he has joined a company as a trainer for AI models.

Savreux is part of the covert army – those who quietly teach AI systems how to recognize data and improve accuracy over time. Their task involves labeling and categorizing content in massive datasets, fine-tuning AI-generated text to be more suitable based on predefined criteria, thereby helping chatbots provide better responses.

People like Savreux play an indispensable role in the current generation of AI models. However, their position often goes unnoticed. The developers and publishers of products like ChatGPT receive recognition, while the silent team behind them remains unrecognized.

Savreux has been performing the task of data categorization and labeling for AI systems, including ChatGPT, for many years.

“We do the hard work. You can create a powerful AI system with the participation of top experts. But without the labeling team, there would be no ChatGPT or anything at all,” he said.

“There have been many discussions surrounding AI recently. But we are missing a crucial part of the story, which is that no matter how intelligent a machine is, it still relies on the human factor behind it,” said Sonam Jindal, head of the non-profit organization Partnership on AI in San Francisco.

The field of technology has been developing for many decades. However, at various times, businesses still relied on the labor force of thousands of workers, from punch card operators for computers in the 1950s to modern-day data entry personnel. Jobs like AI training are unstable and dependent on market demand. Individuals are mainly hired through third-party contractors and only work when there are orders. They are required to remain anonymous, and benefits such as health insurance are rare or non-existent.

In recent times, the booming development of AI text generation models has led to a significant increase in the demand for AI trainers. However, they are often paid very low wages. According to Time, OpenAI, the company behind ChatGPT, hired labor in Kenya for labeling and content moderation at a rate of less than $2 per hour. These individuals have to continuously review content on various topics such as child sexual abuse, violence, suicide, torture, self-harm, and incest to train AI in filtering information. In April 2022, Technology Review reported that self-driving car companies like Tesla hired workers in Venezuela for data labeling at an average rate of $0.9 per hour.

Some employees have begun demanding better benefits from companies rather than just hourly wages. In Nairobi, Kenya, over 150 people who are involved in data labeling for Facebook, TikTok, and ChatGPT have voted to establish a union due to low wages and mentally taxing work.

However, not everyone complains about training AI. Jatin Kumar, a 22-year-old from Austin, Texas, has been involved in data labeling for AI for over a year. According to him, the job has given him the opportunity to explore cutting-edge technology.

“It makes you start thinking about how to use pioneering technology before it hits the market,” said Kumar, a computer science graduate. “Initially, I thought I was just participating to learn about the technology. But now, I don’t want to leave this position.”