UN survey of 3,500 people in 75 countries using microtask platforms like Mechanical Turk found workers made an average of $6.54/hr in US and $3.31/hr worldwide
Megan Geuss / Ars Technica : Tweets: @turkernational Tweets: Kristy Milland / @turkernational : @Techmeme @MeganGeuss And that was a survey just asking for data. In our study which automatically collected payment and time data, the figure was actually $2.
Context & Ripple Effects
The UN survey is the first global benchmark for a workforce previously documented piecemeal: TechRepublic reported back in 2016 that crowd workers earn pennies to train AI, and Kristy Milland's study — which logged payments and time automatically rather than relying on recall — put average pay closer to $2/hour than the UN's self-reported figures. The numbers land after Amazon had already doubled its commission on each task from 10% to 20%, taking a bigger cut of the same low-value gigs.
Why it matters now: accessible crowd work such as photo tagging and surveys saw a boom as jobless people turned to online piecework during COVID-19, and a follow-up survey found many Mechanical Turk workers feel they have no recourse when requesters underpay them or tasks fail on technical errors.
First-order effects
- Mechanical Turk workers now have an official cross-country pay benchmark — $6.54/hour in the US, $3.31 worldwide — that sits far below the $15–17/hour Amazon advertises for its own warehouse hires, sharpening the wage contrast inside a single company.
- Requesters posting tasks face a credibility problem: the gap between the UN's self-reported averages and Milland's instrumented $2/hour finding suggests worker-recall surveys overstate what crowd work actually pays.
Second-order effects
- The pandemic-era influx of new crowd workers expands the labor supply exactly when pay is under scrutiny, easing pressure on requesters to raise per-task rates even as the UN data circulates.
- With workers reporting no effective channel to contest underpayment, scrutiny shifts upward from individual requesters to Amazon's platform rules and its doubled fee structure — the party positioned to fix disputes is also the one profiting per gig.
Third-order effects
- Researchers estimate 33–46% of Mechanical Turk workers already used LLMs on a text-summarization task, pointing toward a microtask economy where humans increasingly verify machine output — repricing the work around cost per verified task rather than raw completions.
- If measured pay keeps running below minimum-wage norms across dozens of countries, microtask platforms become a candidate for labor-law classification fights, though how any regulator would draw the employer line for global, anonymous piecework remains genuinely unsettled.
The trend: AI's training-data supply chain is industrializing around a global, low-paid microtask workforce whose real earnings are only now being measured — just as language models begin absorbing the tasks themselves.