The perils of microtask work

by Megan Geuss, Ars Technica

Microtask platforms recruit humans to do the rating, tagging, review-writing, and poll-taking work that can’t quite be automated with an algorithm yet. In the US, the most common such platform is Amazon’s Mechanical Turk, but other platforms are prominent in other parts of the world.

Proponents of this kind of work say that these quick, simple tasks allow people flexible hours to make money, or help “fill in the gaps” for the un- and under-employed.

But a new study (PDF) from the United Nations’ International Labor Organization (ILO) questions whether these platforms are as good for society as the Silicon Valley investors and digital evangelists claim. The ILO surveyed 3,500 people across 75 countries who worked for Mechanical Turk, as well as Crowdflower, Clickworker, Prolific, and Microworker.

The work on these platforms is often menial and tedious, and the survey found that workers get paid startlingly little on all five platforms, especially when unpaid work is taken into account. The survey counted unpaid work as “time spent looking for tasks, earning qualifications, researching requesters through online forums, communicating with requesters or clients and leaving reviews, as well as unpaid/rejected tasks/tasks ultimately not submitted.”

In the US, accounting for all paid and unpaid work, Mechanical Turk workers made just $6.54 per hour on average, with a median of $5.63 per hour. That’s significantly lower than the minimum wage in most states, and lower than the federal minimum wage of $7.25 per hour.

And around the world, workers were invariably paid less than US microtask workers. The average hourly pay for all workers across the five platforms was just $3.31 (and that includes the US’ relatively “high” wages!) with a median hourly pay of $2.16. This doesn’t necessarily reflect a tendency for lower wages in poorer countries, either. The UN survey showed “that a substantial proportion of workers earn below their local minimum wage.”

Read more at Ars Technica →