Abstract:
Amazon announced that it will officially close its crowdsourcing work platform Amazon Mechanical Turk (MTurk) on September 30, 2026. Since its launch in 2005, this service has long provided companies and researchers with manual data processing and annotation capabilities. It was once described by Amazon founder Jeff Bezos as "artificial artificial intelligence."
Amazon stated in a notice posted on the MTurk website that the company will regularly evaluate its projects, tools and services and make adjustments accordingly. After completing the evaluation, Amazon decided to stop the AWS Mechanical Turk service, and the related business will be terminated on September 30.
Mechanical Turk was launched in 2005, mainly distributing a large number of small digital tasks to online workers for completion. Such tasks are called "human intelligence tasks", and the individual rewards are usually only a few cents. They include data annotation, audio or video transcription, questionnaire filling, etc. Its core positioning is to handle tasks that humans can easily complete but are still difficult for computers to solve efficiently.
The name of the service comes from the famous 18th century "Mechanical Turk" automatic chess-playing device. This seemingly autonomous chess-playing machine is actually controlled by a human chess player hidden inside. When Amazon first conceived MTurk, it also hoped to use distributed human labor to label massive data for its huge e-commerce website.
According to Amazon’s previous data, the MTurk platform served more than 500,000 workers at its peak, and these participants were also called “Turkers.” However, data worker Krista Pavlowski, organizer of the data worker rights group Turkopticon, said MTurk has been in decline in recent years.
With the rapid improvement of artificial intelligence model capabilities, the data service market specifically for AI training has become more crowded. Data startups such as Scale AI, Mercor and Prolific have entered the market one after another, recruiting human workers to provide services for model training, evaluation and data annotation. Amazon has also tried to grasp the demand for AI training and used MTurk as one of the data annotation sources for its SageMaker service.
Amazon Web Services previously stated in a page that Mechanical Turk workers can provide services around the clock, and human teams using the platform can usually complete tasks that require manual review and data annotation faster.
Pavlowski believes that while more data annotation service providers are emerging, Amazon does not seem to continue to invest enough resources to improve MTurk, causing many platform workers to turn to competitors. An Amazon representative did not immediately respond to a request for comment.
The quality of data produced by MTurk has also been questioned. A study released in 2023 by Swiss academic researchers found that as many as 46% of Mechanical Turk workers may use AI models to complete tasks on the platform. This phenomenon also highlights that in the artificial intelligence training chain, data tasks that rely on humans to complete are themselves increasingly affected by generative AI tools.
Amazon stopped accepting new Mechanical Turk customers last month. At the time, many workers took the move as a sign that the service was coming to an end. Now that the closure decision has been officially implemented, some insurance and travel companies that still rely on MTurk to hire data workers are busy looking for alternative platforms.
Pavloski herself started taking orders at MTurk during her maternity leave in 2008 to supplement her income; after being laid off in 2012 and needing to take care of her son with special needs, she turned to this type of work full-time. In 2019, she began to participate in the organization of Turkopticon. The organization initially focused on the rights of Mechanical Turk workers, and later expanded its focus to a broader group of data workers.
She said that Mechanical Turk and other data annotation platforms can provide meaningful income and flexible working hours. The model is similar to gig economy positions such as online ride-hailing and Amazon Flex package delivery, but its advantage is that it can be completed remotely and the work can be carried out using a smartphone. For those who still rely on MTurk as their main or even full-time source of income, the closure of the platform has brought obvious uncertainty and concern.
Comments