The hidden humans powering the AI economy

Tina Lynn Wilson of Hamilton, Ontario has been working as a freelancer for DataAnnotation since January.

The 45-year-old says she enjoys the job, which involves checking the artificial intelligence model's answers for grammar, accuracy and creativity. It requires analytical skills and attention to detail, and she also has some interesting projects, such as choosing the best of two poetry samples.

“Because this is a creative answer, no fact checking will be required. You will have to indicate… which answer is better and why.”

The work Wilson does is part of a huge but little-known network of workers in the emerging artificial intelligence economy. Companies like Outlier AI and Handshake AI hire them as “AI trainers,” contracting with large AI platforms to help them train their models.

Some data annotation work is poorly paid—even exploitative in other parts of the world—but there is a wide range of work to train, maintain, and patch AI. It seems the tech giants prefer not to talk about this work. And as models evolve, they will require more specialized training, meaning companies may soon no longer need many of the very people who helped make them what they are today.

Human experience

We often hear that modern generative AI is trained on massive amounts of data to teach it how human ideas typically fit together. This is sometimes called pre-training and is just the first step. To ensure that these systems provide accurate, useful, and non-offensive answers, they need to be refined, especially if they are going to work in narrow areas of the real world.

This is called fine-tuning and depends on human experience. Essentially, this is a part-time job: done on a task-by-task basis, with no guaranteed hours. Canadian AI instructors we spoke with earned around $20 an hour, although more specialized work can pay around $40. However, inconsistency can be a problem.

“You can’t rely on this as your main source of income,” Wilson said, calling her job a generalist. According to her, many other commentators also consider it a part-time job.

Reinforcement learning based on human feedback this is a type of fine tuning it depends on the people evaluating the AI ​​results.

Wilson's work involves assessing how “human” an AI's response sounds.

“This is especially true for voice responses,” she said. “Is this what the person would like to hear?”

So, when ChatGPT or Claude sounds strangely human, it's because humans have taught it to be so.

“It's still a software product,” said Brian Merchant, a tech journalist who specializes in workforce and digital technologies. “You need quality assurance for a commercial software product.”

Outlier AI has more than 250,000 active members in 50 countries, according to Fiorella Riccobono, a spokeswoman for Scale AI, its parent company. Eighty-one percent have at least a bachelor's degree, she said. The company was unable to provide data for Canada.

Possibly changing market

There are signs that the market for this work is changing, with less demand for generalist work like that of people like Wilson. Scale AI recently laid off generalists in Dallas. according to Business Insiderin a shift towards more technical training. Meanwhile, newer, more advanced models have appeared, such as the model from the Chinese company DeepSeek. automated the reinforcement process.

“Demand the number of participants with specialized knowledge and advanced degrees has grown significantly as artificial intelligence systems become more complex,” Riccobono said.

Eric Zhou, 26, was one of these specialized workers. After studying materials and nanoscience at the University of Waterloo, he freelanced for Outlier AI for about a year. There, he evaluated AI clues and answers in undergraduate-level physics and chemistry and adjusted the answers.

“It’s a very exciting job if you’re just solving scientific questions,” he said. “So that problem-solving part really appealed to me.”

However, he found that tasks could take longer to complete than the company's allotted time, so a job that cost $20 per hour of work could take longer without additional pay.

There seems to be no shortage of Canadians working in specialized fields looking to supplement their income by improving artificial intelligence, including a number of Zhou's friends.

That means workers feel they can be constantly replaced, he said.

“Digital Sweatshops”

However, AI in general relies on a global supply chain throughout the learning process, much of which is outsourced to workers in lower-wage countries. This may mean fine-tuning the data, but much of the work involves labeling the data, which can be exhausting.

The number of people employed in this area is in the millions. The companies have been accused of using lenient labor laws in regions such as East Africa and Southeast Asia.

“There are a lot of what you might call digital sweatshops, anywhere from the Philippines to Kenya, where workers are essentially transforming these data sets into products that AI can use,” said James Muldoon, co-author of the book. Feeding machineabout human labor and the hidden costs of feeding AI.

He says the tasks can be brutal, as he discovered during fieldwork in Kenya and Uganda, where people worked up to 70 hours a week for little more than a dollar an hour, in conditions he called “truly horrendous.”

While many commentators had ambitions of more meaningful work in the technology sector, he said they were stuck doing “really boring and painful” tasks.

AI companies typically don't focus on the human labor that powers the automation. Merchant, the tech journalist, says these firms usually want to show that their product “feels magical, feels powerful, feels like it's the future.”

“It’s very rare that a job is completely occupied by technology, especially in an industrial setting,” he said.

“Usually you have technology that can help you do that.”