For twenty years, a quiet corner of Amazon paid people a few cents at a time to do the small jobs software could not handle. That marketplace is now closing its doors to newcomers, and the reasons behind it say a lot about where Malaysia sits in the global scramble to build AI.
Amazon will stop accepting new customers for its Mechanical Turk service on 30 July 2026, according to a notice on the platform's website first reported by TechCrunch. Amazon Web Services said the decision came after "careful consideration," adding that existing customers can keep using the service as normal while it funds security and reliability but adds no new features. The Register described the platform as being wound down rather than switched off outright. Launched in 2005, Mechanical Turk let companies pay people tiny sums for tasks that resisted automation, from solving CAPTCHAs to judging the sentiment of a sentence. From 2018, Amazon also pitched it as a way to label the data used to train AI models.
Editor
Kai T chevron_right
Table of Contents
The Hidden Workforce Behind Every AI
Mechanical Turk was only ever the visible tip of a much larger economy. Behind almost every AI product sits a workforce of people who label images, transcribe audio, sort text and check machine output so the models have clean examples to learn from. The World Bank estimates that between 150 million and 430 million people around the world do this kind of data work, much of it in the Global South. Southeast Asia has become one of the busiest hubs, partly on cost: annotation specialists in the region deliver comparable accuracy for roughly USD 15 to USD 25 an hour, against USD 50 to USD 80 in Western markets.
Why Malaysia's Data Workers Should Care
Malaysia is firmly inside this supply chain. Local job boards carry dozens of listings for caption annotators, audio transcription specialists and voice talent producing AI training data, and regional firms route labeling contracts through Malaysian teams. The government has leaned in too: MDEC has run HRD Corp-claimable AI skills training through April 2026, while the country courts tens of billions of dollars in data-centre investment alongside Singapore and Thailand. The people doing the labeling are part of the same AI build-out as the server halls in Johor.
Mechanical Turk's slow fade is a signal for that workforce, not a catastrophe. The platform had already lost ground to managed providers such as Appen and Scale AI, which promise quality-checked data instead of an open marketplace crawling with bots. A 2023 analysis cited by TechCrunch found that between 33 and 46 percent of Turk workers were themselves using AI chatbots to finish their tasks, which raised an awkward question about whether the humans were still adding value at the bottom of the market.
The Work Is Changing, Not Disappearing
For Malaysian data workers, the takeaway is that the lowest-skill microtasks are the most exposed. As models get better at labeling their own training data, the pay-per-click piecework that Mechanical Turk pioneered is drying up. What is growing instead is higher-skill, human-in-the-loop work: evaluating and ranking model answers, red-teaming systems for safety, and specialist annotation in Malay and other local languages that global models still handle poorly. That is the rung worth climbing toward, and it pays far better than a few cents per click.
The first Mechanical Turk, built in the 18th century, was a chess-playing machine with a person hidden inside. Two centuries on, its namesake is fading just as the industry it helped feed is booming. For Malaysia, the opening is to move up the value chain before the cheapest tasks disappear.
Images courtesy of Julio Lopez and Kelly Sikkema on Unsplash.