Summary in Plain Language
This article reports that Amazon’s legendary manual data annotation platform, MTurk, which has been in operation for 21 years, has officially announced its closure in 2026. MTurk can be considered the “pioneer infrastructure” of the AI industry. At its peak, it had over 500,000 registered users worldwide. It broke down complex tasks into small, easy-to-manage tasks that could be completed for just a few cents each. This platform not only helped researchers like Li Feifei to annotate 14 million ImageNet training images in just two years (which would have taken 19 years if done by hiring undergraduate students) but also supported the training of almost all early AI technologies, including the initial version of ChatGPT. It even contributed to the emergence of the “gig economy” concept.
However, due to a long period of unregulated operation and a lack of quality control, MTurk could no longer meet the high-quality data requirements of modern large-scale AI models. Eventually, there were absurd situations where workers used large AI models to cheat and complete tasks. The end of MTurk does not indicate a decline in the data annotation industry; rather, it marks a complete transformation: low-level tasks that anyone can do are becoming fewer and fewer, while the value of highly skilled professionals with specialized knowledge has skyrocketed. The industry has evolved from a labor-intensive business that exploited wage differences to a high-end service that relies on organizing professional expertise.
---
Detailed Explanation
1. MTurk: The “Invisible Teacher” of the AI Industry
Many people don’t realize that the AI systems we use today, which can chat and draw, were once as primitive as kindergarten children, unable to even distinguish between cats and dogs and unable to work independently. Amazon’s initial goal in creating MTurk was simple: its e-commerce platform had millions of products that needed to be categorized and corrected. Since existing algorithms couldn’t handle such tasks, Amazon came up with a solution: it divided the work into small, quick-to-complete tasks (such as labeling products, drawing frames around people in images, or converting audio to text) and posted them online for workers around the world to complete. The results were made to look as if they were generated by AI itself.
MTurk unexpectedly became the foundation for the revival of the AI industry. It helped create the ImageNet dataset, which transformed the field of computer vision, and was also used for speech recognition, autonomous driving, content moderation, and even in the development of ChatGPT by OpenAI in 2022. The term “gig economy” was coined by scholars studying MTurk’s operational model, reflecting the underlying logic of modern delivery and ride-hailing services, which all originated from MTurk.
2. The Rise and Fall of MTurk
MTurk’s success was based on its laissez-faire approach: no need for labor contracts, social security, or training. Anyone with a computer could take on tasks, keeping costs low. However, this freedom turned into a flaw as the platform neither screened workers’ skills nor ensured quality control. Workers often worked for hours only to have their work rejected by employers with no compensation. Ironically, before its closure in 2023, researchers found that 30%-50% of users were already using large AI models to complete tasks. MTurk, which was meant to allow humans to mimic machines, had become a place where humans used machines to deceive others, resulting in low-quality data that didn’t meet AI model training requirements.
2. Different Fates for Similar Companies
The end of MTurk is not a sign of a shrinking industry but rather a signal of a major transformation. Companies in the data annotation sector are facing vastly different destinies:
- Companies that adapted to the new trends have thrived. For example, Scale AI in the US initially focused on general data annotation but quickly shifted to more specialized tasks and even generated synthetic data for AI. After being acquired by Meta, its value soared to $29 billion. Mercor, founded by three millennials, grew to a $20 billion valuation in just three years by leveraging expert knowledge to provide advanced evaluations and custom training for AI models.
- Companies that clung to the old MTurk model suffered greatly. Appen, an Australian company once seen as a “larger version of MTurk,” relied on cheap workers worldwide to handle Google’s annotation tasks. Its stock price plummeted to 1% after Google terminated its contract, nearly leading to bankruptcy.
3. Data Annotation: No Longer a Low-Value Labor Task
The situation in the data annotation industry has changed rapidly. In China, data annotation jobs that used to require low-skilled workers now attract professionals with high salaries. Large companies have rebranded these roles as “data alchemists” or “AI question experts,” with recruitment requirements increasing significantly (e.g., a Mandarin proficiency level of CET-8 for certain tasks or a master’s degree for AI model evaluations). The government has set a target of a 20% annual growth rate for the industry by 2027, with a focus on AI-preannotated data followed by expert review. The need is for professionals who can teach AI, not just for routine tasks.
4. A Heart-Wrenching Metaphor
The name MTurk is a poignant reminder of a famous 18th-century scam: a “automated Turkish machine” that could play chess, which deceived Europe for decades before it was revealed to be just a human operator hiding inside. Amazon’s choice of name was a play on the idea that MTurk was essentially “human AI hidden behind a machine.” Twenty-one years later, this metaphor has come true: the data annotated by MTurk has made AI smarter, and now AI can perform those simple tasks on its own, eliminating the need for human workers. This doesn’t mean AI will replace all annotation jobs, but rather that humans will move on to more complex, knowledge-intensive tasks that AI cannot handle.
In summary, MTurk’s closure marks a significant shift in the data annotation industry, from a labor-intensive sector to one that values professional expertise.