Amazon is shutting down Mechanical Turk on September 30, ending the 21-year-old marketplace that paid people pennies to do small digital tasks machines couldn’t handle. The tasks it specialized in, labeling images and transcribing audio, are precisely the work used to train the AI systems that can now do them.
What Mechanical Turk was
Launched in 2005, MTurk let workers complete “human intelligence tasks,” or HITs, such as labeling images, transcribing audio, filling out surveys, and sorting data, typically for a few cents each. At its peak the platform had more than 500,000 registered workers, who called themselves Turkers. Jeff Bezos described the service as “artificial artificial intelligence,” which was a joke about the name: the original Mechanical Turk was an 18th-century chess-playing automaton that appeared to think for itself and actually concealed a human operator inside the cabinet. Amazon’s version was the same trick, run at scale and in the open.
Why it’s closing
Amazon said the decision followed an internal assessment of its programs and services without giving a more specific reason. The commercial context is less ambiguous. MTurk had been losing ground to purpose-built data companies like Scale AI and Prolific, which offer vetted workers and quality controls that MTurk’s open marketplace never reliably provided. Amazon stopped accepting new MTurk customers on July 30, a move many workers read at the time as a signal the platform was being wound down.
Who this actually affects
Two groups, in different ways. The first is the Turkers themselves, many of whom used the platform for supplemental income precisely because it required no interview, no schedule, and no minimum commitment. There is no severance in a marketplace model, and no transition offer has been announced.
The second is academic research, which is the part that tends to get overlooked. MTurk became a default recruitment tool for social science, psychology, and economics studies, because it offered cheap access to a large pool of participants. A substantial body of published behavioral research over the past 15 years drew its subjects from this one platform. Its closure doesn’t invalidate that work, but it does remove a standard methodology, and researchers will have to move to paid alternatives that cost considerably more per participant.
The part worth sitting with
MTurk’s core business was humans doing the work computers couldn’t. Much of that work, image labeling and transcription in particular, went directly into training datasets for machine learning. The platform is closing because those systems now handle the tasks well enough that the marketplace isn’t needed. It is a fairly clean example of a workforce being paid, task by task and cent by cent, to build the thing that made the work obsolete.
Related from BeezLoop: The Bipartisan Rebellion Against Big Tech's Water-Gulping, Power-Grid-Squeezing AI Factories, Musk Is Suing Altman and An AI Company Says Its Model Now Leads a Quarter of the Work Building Its Next Model.
