In 2012, Snip Snap had a problem. Their coupon-scanning app had just been featured in the Apple App Store, delivering tens of thousands of users overnight. The problem: they didn't have any coupon-reading software. The company had been reading each coupon and inputting the data by hand. That was easy when it was just a few dozen users and a dedicated team, but now they needed hundreds of thousands of labor-hours to keep up with new users. And they needed it fast.
The average wage on mTurk is $2 an hour
The answer was Mechanical Turk, Amazon's cloud labor platform, which lets requesters call in anonymous workers from around the globe for what the service calls "human intelligence tasks." Using the service, Snip Snap was able to line up an army of coupon taggers in a matter of hours. The app kept running, and the users never knew the difference.
In the eight years since launch, Amazon’s micro-labor platform, also known as mTurk, has become something of a secret weapon for startups. It’s big business too: Analysts estimate the "cloud labor" sector, as it's called, is worth $1.2 billion in annual revenue. But it's also come under fire as exploitative. Researchers have estimated the average wage on Mechanical Turk is just $2 an hour, and some claim that’s an overestimate. Craigslist-style scams are common, in which requesters ask for up-front payments in exchange for later rewards, then disappear. If employers decide a completed task is unsatisfactory, they can decline to pay and still keep the resulting work. As a result, workers complain that many requesters decline work simply to get out of paying.
Experts estimate Mechanical Turk sees as much as $400,000 worth of transactions every day, but despite the money, Amazon has kept a hands-off attitude to the marketplace. Workers are left to fend for themselves.
Giving workers a chance to rate employers
But a new tool may give Turkers a secret weapon of their own. It's called Turkopticon, a browser plug-in that aims to turn the tables on requesters by giving workers a chance to rate employers by reliability. If a requester frequently backs out of paying workers, he'll get a bad rating and workers will know to stay away. Outright scams will be flagged early and the accounts behind them will be seen as untrustworthy. Already, 7,000 workers have installed the plug-in. If they’re going to help make apps like Snip Snap successful, they want to be able to tell who they’re working for.
It seems like a simple feature, one Amazon might have added years ago, but as Amazon sees it, the mTurk community already has this problem solved. Reached for comment, an Amazon Web Services representative said, "There are many tools and forums available that allow workers to share information so they can make informed choices."
"This is not just a way of evening out information. This is a tool to spark awareness."
Still, for the developers involved, Turkopticon means a fundamental shift in the way workers and employers interact, forcing requesters to maintain a good reputation. Mechanical Turk is built on the idea of dispensable labor, which appears when it's needed and disappears when it's not. That's useful for developers, who often need to test out ideas or burn through one-time data sets. If a task can't be achieved through coding alone, they can build Mechanical Turk requests into the code and summon workers on request. It's a unique system, and one that puts the workers at a disadvantage. Requesters have all the power, and workers often have to struggle just to be paid for what they've done. Larger programs like Houdini and CrowdFlower make things a little better, singling out particular skills and aligning the best workers with the best requesters for higher pay, but they do little to change the basic dynamic.
"As work changes, the organizing has to change."
The Turkopticon developers want to change the balance of power more permanently, and the plug-in is just the beginning. "This is not just a way of evening out information," says Lilly Irani, a former Google UI designer who co-designed the project. "This is a tool to spark awareness." The plug-in arose out of research into a Turker's Bill of Rights, outlining the basic worker protections most important to Mechanical Turk's users. The biggest request turned out to be a way to rate employers. More traditional worker protections, like a minimum wage, were much less popular, as Turkers worried they would cut down on the number of available jobs. Actual unions were even more controversial. "A lot of the workers were specifically not interested in unions, although worker-run forums like Turker Nation will do certain kinds of collective actions like boycotting bad requesters," Irani says. "It may be that the word ‘union’ has become too politically charged."
The larger problem is how distributed the Mechanical Turk workforce has become. Amazon makes a point of keeping workers anonymous, only distinguishing between workers that are in the US or those who are not. The only place Turkers can talk collectively is on forums, and most labor organizing tactics start by getting all the relevant parties into the same room. But while technology may have changed the landscape, Irani and her team are eager to adapt. "As work changes, the organizing has to change."