Amazon is shutting down Mechanical Turk after 21 years, the platform Bezos called 'artificial artificial intelligence'

[Tel92]

Amazon confirmed this week that it's closing AWS Mechanical Turk on September 30, ending a 21 year run for the crowdsourced work platform that connected businesses with people willing to complete small digital tasks computers couldn't reliably handle. Launched in 2005, the service let companies and researchers pay workers small amounts, often just a few cents per task, to label images, transcribe audio, answer surveys, and review content, jobs collectively known inside the platform as Human Intelligence Tasks.

Jeff Bezos famously described Mechanical Turk as artificial artificial intelligence, a name borrowed from an 18th century chess playing machine that appeared to defeat human opponents through pure mechanical trickery, when in reality a skilled human chess player was hidden inside operating it the whole time. Amazon's internet age version flipped that concept into a real business, quietly putting humans behind the curtain wherever software wasn't yet capable of doing the job itself, and at its peak the platform counted more than 500,000 registered workers.

The platform's decline traces directly back to the same AI progress it originally existed to work around. Amazon tried repositioning Mechanical Turk as a human review and data annotation source for its SageMaker machine learning tools, but competing platforms including Scale AI, Mercor, and Prolific increasingly pulled workers away, and Amazon itself put fewer resources into improving the aging service over time. Signs of the coming shutdown became clear last month when Amazon stopped accepting new Mechanical Turk customers entirely, a move workers on Reddit correctly read as the beginning of the end.

The closure carries a truly strange irony worth sitting with. A 2023 academic study out of Switzerland found that up to 46 percent of surveyed Mechanical Turk workers may have already been using AI tools themselves to complete the very tasks meant to train AI in the first place, blurring the line between human and machine generated work long before Amazon's own shutdown decision. Krista Pawloski, a data worker and organizer with the advocacy group Turkopticon, told CNBC that some workers still relied on the platform full time for income, meaning the shutdown carries real practical consequences well beyond the symbolic end of one of the internet's oldest and strangest labor experiments

Amazon is closing Mechanical Turk, the human workforce it sold as AIthenextweb.com

[Reacher Quarry]

Bezos calling it artificial artificial intelligence is honestly one of the more self aware pieces of tech branding from that whole era. Most companies try to obscure how much human labor sits behind their supposedly automated systems rather than naming the trick outright in the product itself

[Oscar73]

The 46 percent figure of workers using AI to complete tasks meant to train AI is such a perfect, almost too neat closing of the loop on this whole story. The platform effectively got consumed from the inside by the exact technology it existed to work around in the first place

[Grace]

Curious where all those remaining active workers actually migrate to once the platform closes for good. Scale AI, Mercor, and Prolific presumably absorb some of them, but there's likely a meaningful chunk who just quietly stop doing this kind of work entirely rather than switching platforms

[Freddy]

This feels like a clean, symbolic bookend for an entire era of AI development that relied heavily on cheap, distributed human labor to produce the training data modern AI systems now generate and improve largely on their own. Worth remembering how much of today's AI progress was actually built on exactly this kind of underpaid, often invisible human work

[Callum28]

Real, actual people relying on this platform for full time income is the part that deserves more sympathy than most of the coverage seems to be giving it. Easy to treat this purely as a quirky tech history footnote, but there are genuine livelihoods disrupted by a shutdown like this one

[Sabu_ECW]

Watching newer platforms like Scale AI and Mercor pull workers away over time is a good illustration of how competitive even the human labor side of AI training has actually become. Data labeling and annotation turned into a really serious, well funded industry of its own, not just an Amazon side project anymore

[Charlotte]

Twenty one years is a actually long runtime for any internet platform, let alone one built around such a specific, narrow niche. Most services this old either get absorbed into a bigger product or just quietly die within a decade, so MTurk actually had a fairly impressive run before finally hitting its natural end