The AI You Use is Created by Workers Earning $2 an Hour

You ask ChatGPT a question. You have Claude write a text. The AI returns answers in seconds. It’s convenient. It’s cheap. Unlimited use for just $20 a month.

However, it’s essential to understand the reality behind that “cheapness” at least once.

There is a company called Scale AI. They create training data for AI models. They employ tens of thousands of workers to label data for AI learning, evaluate quality, and provide feedback.

These workers earn a mere $2 an hour.

In Japanese yen, that’s about 300 yen—less than a third of what a part-time job at a convenience store pays. Our everyday AI services are built on this labor force.

The Reality of ‘Human-Powered AI’

AI is often thought to “learn automatically,” but the reality is different.

Training AI models requires a vast amount of “correct data.” Judgments like “this image is of a cat,” “this text is polite,” and “this answer is fact-based” are made by humans one by one. This process is known as “data annotation” or “RLHF (Reinforcement Learning from Human Feedback).”

Workers at Scale AI are responsible for this task, which includes the following:

• Evaluating the quality of AI responses on a five-point scale
• Labeling each object in an image
• Checking the grammatical correctness and factual accuracy of text
• In some cases, collecting and classifying personal social media posts or copyrighted content
• Transcribing pornographic audio

The last two tasks raise significant ethical concerns. However, the workers have no choice.

The case of Patrick Ciriello is emblematic. After a long professional career, he spent a year searching for a job before landing annotation work at Scale AI. He earns $2 an hour, performing “subcontracted work” for AI below minimum wage despite his expertise.

People like him are not exceptions. Scale AI has tens of thousands of workers, including older individuals who have been shut out of regular employment, workers from developing countries, and people looking to earn extra income.

Why Small Businesses Should Be Aware of This Issue

You might think, “That’s a Scale AI problem, and it doesn’t concern us.” However, for three reasons, it is not unrelated to small businesses.

Reason 1: Quality Risk
How seriously can $2-an-hour workers manage data quality? The answer is clear. In a low-wage labor environment, the accuracy of work declines. Motivation also drops. As a result, the quality of the data that AI models learn from becomes inconsistent.

AI outputs are directly linked to the quality of the training data. The principle of “Garbage In, Garbage Out” remains unchanged in the AI era. When small businesses use inexpensive AI services, the reliability of those outputs is influenced by such labor conditions.

Reason 2: Ethical Risks Become Brand Risks
In an era where ESG and supply chain transparency are demanded, large companies have already begun to scrutinize the ethical aspects of their partners’ AI usage. “What kind of labor conditions were the training data for the AI services you use created under?” Companies unable to answer this question risk being excluded from transactions.

Currently, small businesses may not often receive such inquiries. However, the situation is likely to change in the next two to three years.

Reason 3: Regulatory Risks
The EU’s AI regulation (AI Act) demands transparency regarding the training data for AI models. In Japan, guidelines concerning AI are also being developed. In the future, there may be an obligation to disclose “under what conditions the AI you are using was trained.”

The Line Between ‘Cheap’ and ‘Fair’

The price competition for AI services is intensifying. Services available for a few thousand yen a month are emerging one after another, expanding options for small businesses. This is a positive development.

However, there is a reason for “cheapness.” Whether that reason is “technological efficiency” or “exploitation of labor costs” is an entirely different matter.

Cost reductions through technological efficiency should be welcomed. Streamlining models, optimizing inference, and utilizing open-source resources are all healthy cost-saving measures.

On the other hand, cost reductions reliant on $2-an-hour labor are unsustainable. As workers’ awareness of their rights increases, wages will rise. If regulations tighten, costs will soar. The current “cheapness” may merely be a prepayment for future price increases.

So, What Should Small Businesses Do?

1. At a minimum, investigate the ‘backstory’ of the AI services you use.
Who trained the AI model? Is the method of data collection publicly available? Are there clear policies regarding labor conditions? You don’t need to conduct a perfect investigation, but you should avoid being in a state of “knowing nothing.”

2. Compare multiple services and consider the reasons for their pricing differences.
If there are services with significantly different prices for the same functionality, think about the reasons for that difference. Is it due to technological efficiency or labor cost compression?

3. Consider models fine-tuned with your own data.
Instead of relying on external large models, there is an option to fine-tune small models with your own data. In this case, you can control the quality of the training data yourself.

4. Prepare to answer when asked by partners.
“What are your ethical policies regarding AI usage?”—this question will likely come in the not-so-distant future. If you prepare now, you won’t have to panic later.

Choose with Knowledge

I do not intend to deny the benefits of AI. Just because there are workers earning $2 an hour doesn’t mean you shouldn’t use AI.

However, using AI without knowledge and using it with understanding are entirely different things.

Behind the “cheapness” of AI, there are people working for 300 yen. Knowing this fact, choose the AI services that are appropriate for your business. That is the “righteous choice” for small businesses in the upcoming AI era.