Long before IBM booted up Deep Blue, there was a chess-playing automaton that toured Europe in the 18th century, beating such luminaries as Benjamin Franklin and Napoleon. The joke was that the mechanical man, called the Turk, wasn't a robot at all. A person was actually hiding inside the contraption, directing moves that seemed to observers to be made by a fully functioning machine.
So it is with Amazon's Mechanical Turk, or mTurk, service. Companies can take a task that a computer could almost do by itself but that still requires a bit of human finessing like transcribing audio into coherent phrases or clicking on photos that contain a particular object and farm it out to mTurk's hordes of online freelancers, who are eager for the work, however menial. The service, which launched in 2005, now has a pool of more than 200,000 workers who supply what Amazon CEO Jeff Bezos calls "artificial artificial intelligence."
By dividing a job into the smallest possible units known as human-intelligence tasks, or HITs companies can use mTurk to get low-level but labor-intensive work done quickly through crowdsourcing. And mTurk's workers can do the job cheaply; Turkers, as they're called, generally earn no more than a few cents a HIT, which usually takes a few minutes to complete.
I tried my hand at it, selecting a HIT posted by someone named SprintingTurk, who needed me to examine pairs of phrases and note whether they seemed to have any relationship to each other. The instructions encouraged me to use a search engine to augment my brainpower. When I read the first two terms, "black-eyed peas recipes" and "New Year's Eve plans," I answered yes after Google clued me in on the fact that black-eyed peas are a traditional New Year's dish in the South. After 14 more questions and approximately five minutes, I'd earned 2¢ and the right to call myself a Turker.
"For most of the people who do this, it's an interesting way to pick up a little cash," says Panagiotis Ipeirotis, a computer scientist at New York University who has studied mTurk.
The service is popular in academia and with organizations that want to do massive data collection on the cheap. Some social scientists rely on mTurk to quickly tap potential study subjects. ProPublica, a nonprofit investigative-news group, has used Turkers to gather data from government websites as part of its effort to track federal stimulus spending. Not only is using mTurk cheaper and often quicker than hiring a data analyst, but the sheer number of Turkers the service provides means it can often be more accurate than one bleary-eyed full-time worker. "It's a lifesaver for me," says Jennifer LaFleur, ProPublica's director of computer-assisted reporting.
There are, of course, seedier clients lurking on mTurk. Some may be using it to improve search terms for porn, and Ipeirotis has noticed that a large number of new requesters are hiring Turkers to goose ratings by writing positive reviews on Yelp or by downloading a free app. But if you're really down and out, at least you can get paid to be a ghost in the spam machine.
Still, mTurk is no worker's paradise. The system is heavily weighted toward employers, who are free to reject a Turker's work without explanation. And the pay scale is so low that some critics have called mTurk a digital sweatshop especially for the increasing number of Turkers in India, where the work can be their primary source of income. Plus, there's something alienating about the experience in the Marxist sense. The labor is anonymous and piecemeal; Turkers usually have no idea what they're working toward. To borrow a phrase from The Communist Manifesto, Turkers "become an appendage of the machine," doing the work the machine can't do. Or at least that the machine can't do on its own just yet.