Read it on Salon
British actress Olivia Williams with sabre fish.
A picture of a woman’s pink shoe floats on my computer screen. It’s a flat, a street version of a ballet shoe. My job is to categorize the shoe based on a list of basic colors: Is it red, blue, pink, purple, white, green, yellow, multicolored? A description next to it reads “Pink Lemonade Leather.” This is not exactly a brain-busting task; I’m doing it while talking to a friend on the phone. With the mouse, I check a box marked “pink.” In the next split second, a picture of a navy blue shirt appears. I check “blue.” Assuming my answers jibe with those of at least two other people being paid to scrutinize the same pictures, I’ve just earned 4 cents.
With my computer and Internet connection, I have become part of a new global workforce, one of the thousands of anonymous human hands pulling the strings inside of a Web site called Amazon Mechanical Turk. By color-coding the clothing sold by the online retailer, which helps customers to search for, well, pink shoes, I can now call myself a Mechanical Turker. In this new virtual workplace, everything is on a need-to-know basis, including who is doing the work, what the point of the work is and, in some cases, the very identity of the company soliciting the work.
Launched in November 2005, Amazon Mechanical Turk is named after a legendary automaton from the 18th century, “the Turk,” which could play chess. The wooden man, adorned with a turban, appeared to be powered by the machinery of a clock. He even check-mated Benjamin Franklin, a devotee of the game. The Turk was a sensation: a machine that seemed to think. Coinciding with the birth of the Industrial Revolution, the Turk heightened anxieties that machines would replace humans in the workplace. Of course, it turned out to be a fabulous hoax. The ghost in the machine was, as skeptics had suspected, all too human. A chess expert was hidden in the Turk, making all the right moves.
The 21st century twist on the Turk, conceived by Amazon CEO Jeff Bezos, doesn’t try to hide the people inside the machine. On the contrary, it celebrates the fact that we have become part of the machine. For fees ranging from dollars to single pennies per task, workers, who cheekily call themselves “turkers,” do tasks that may be rote, like matching a color to a photograph, but they can confound a computer. Conceived to help Amazon improve its own sites, Mturk.com is now a marketplace where many companies have solicited workers to do everything from transcribing podcasts for 19 cents a minute to writing blog posts for 50 cents. Amazon takes a cut from every task performed.
Amazon claims its virtual workplace provides “artificial artificial intelligence” — a catchy way of saying human thought. “From a philosophical perspective, it’s really turning the traditional computing paradigm on its head,” says Adam Selipsky, vice president of product management and developer relations for Amazon Web Services. “Usually people get help from computers to do tasks. In this case, it is computers getting help from people to do tasks.” As Tim O’Reilly, a computer book publisher and tech industry figure, puts it on his blog, old dreams of artificial intelligence are “being replaced by this new model, in which we are creating more intelligent systems by using humans as components of the application.”
So who wants to be the human component of a computer application? A lot of people, it turns out. Since last November, thousands of workers from the U.S. and more than 100 other countries have performed tasks on Mturk.com. The most dedicated turkers have even formed their own online communities, such as Turker Nation.
The companies are certainly happy. The ones I contacted remarked how stunningly little it costs them to get work done through Amazon Mechanical Turk. Divvying up projects to hundreds of people not only gets the job done more quickly than contracting it out to temps or consultants — much less an actual employee — it gets it done much more cheaply. The tag line for the site could be: dirt-cheap artificial artificial intelligence. One tech company that uses Mturk.com to answer troubleshooting questions brags that it pays tens of dollars on Mechanical Turk for work that would typically cost thousands. And, hey, if companies don’t like the quality of the work they get from turkers, they simply don’t pay them.
As soon as it launched, the Mechanical Turk site sparked a hue and cry in the blogosphere. “Amazon, you cheap bastard. Don’t you at least have the decency to pay minimum wage?” demanded one poster on a tech site. Another commentator sneered that it peddled “jobs even illegal aliens won’t do.” There is something a little disturbing about a billionaire like Bezos dreaming up new ways to get ordinary folk to do work for him for pennies. Is a cut-rate pittance the logical result of tapping into a global workforce of people with a computer, an Internet connection and an Amazon account? And, really, who are all these people working for a measly 1 cent?
The real human ingenuity of Mechanical Turk shines in the novel ways that companies and workers use it to get tasks done efficiently. CastingWords is a transcription service built entirely on the work of Mechanical Turk transcribers and editors. With a little code, plus the turkers, it has succeeded in basically automating the process. The company charges its customers from 42 cents a minute for podcast transcription to 75 cents a minute for other audio. CastingWords pays Mechanical Turk workers as little as 19 cents a minute for transcription. If a transcription job is posted on Mechanical Turk for a couple of hours at the rate of 19 cents a minute, and no worker has taken on the project, the software simply assumes the price is too low and starts raising it.
After a transcription assignment is accepted by a worker, and completed, it goes back out on Mturk.com for quality assurance, where another worker is paid a few cents to verify that it’s a faithful transcript of the audio. Then, the transcript goes back on Mturk.com a third time for editing, and even a fourth time for a quality assurance check. “It’s been terribly useful for us,” says Nathan McFarland of Seattle, one of the co-founders of CastingWords. Transcription is the type of relatively steady task that keeps turkers with good ears who are fast typists coming back. “There are people who have been with us for months, and they’re not leaving,” says McFarland.
One of those workers is Kristy Milland, 27, a mother of one who runs an at-home day care in Toronto, as well as a Web site called RealityBBQ about the reality TV show “Big Brother.” “I have a lot of free time basically sitting at the computer while the kids play,” she says. Among the work she does is editing and quality assurance for CastingWords, but not transcription, because she has tendinitis. When Mturk.com first began, Milland would churn through 3-cent HITs. (That’s “human intelligence tasks,” Turker lingo for jobs.) Amazon was paying turkers to make sure that photos of businesses used on its A9 site, a local search engine, matched the actual businesses listed, a task a computer can’t do. In an eight-hour day, when she didn’t have the kids to watch, Milland could go through 1,000 photos, making a cool $30.
Lately, she’s found a way to goose her earnings by competing for bonuses. A number of service companies use Mechanical Turk to do a “human augmented search.” Say you’re in a sports bar and having an argument about whether Roger Clemens has ever thrown a no-hitter. You can end the debate once and for all with a call to one of the services, which instantly posts the question on Mechanical Turk. Turkers then surf the Web and generally earn 2 cents for each answer.
Back in the sports bar, when you get the answer — “Clemens has never pitched a no-hitter” — you can rate the answer as great, good, lame or junk. Answers deemed “lame” or “junk” are rejected and the worker is not paid. If you don’t rate the answer at all, the worker is automatically paid his or her 2 cents after seven days. Turkers who get the most “great” votes in a week get bonuses of as much as $75. In a good week, Milland can answer 100 research questions, making all of $2, but scoring one of those lucrative performance bonuses, she says, makes the search worthwhile.
The trivia pursuits are so competitive that they’re snatched up by turkers within a minute of being posted. So Milland has set up software to notify her whenever a new question shows up on Mechanical Turk, so she can be the first to grab it. Plus, she’s armed her Web browser with links to her top 100 reference sites so she can answer the questions as efficiently and accurately as possible. Turkers can choose to be paid in Amazon credit, making it easy to shop at the company store. Just the other day, Milland ordered $600 worth of DVDs and books for her family, as well as prizes for contests on her RealityBBQ. “It still doesn’t add up to a lot of money per hour, but if I’m sitting there watching TV anyway, it’s more than I’d make just sitting there,” she says.
Milland’s main beef with Mturk.com is that there’s no way to complain if a company rips her off by refusing to pay for good, accurate work. “Amazon basically says, tough, they can reject what they want,” she says. “There’s no recourse.” (Word of bad-apple companies, however, spreads fast on turker forums.) Milland would also like to see more work and more companies on Mturk.com. When the site first launched there was more to do, she says. These days it feels as if there are fewer opportunities and too many workers competing for them.
Of course, for all its rhetoric about artificial intelligence, Amazon did not launch Mechanical Turk for the good of science. For every task a worker completes for another company, the retailer collects a 10 percent fee from that company. For cheap HITs that pay just a penny, Amazon charges the company half a cent per HIT. Companies need not know the real name, much less the address or Social Security number, of turkers. Unless a worker earns more than $600 from a given company, the business has no obligation to issue the worker a tax form, or report the earnings to the Internal Revenue Service. Few workers cross that $600 threshold with any one company. Yet workers are required to report the money they earn on Mturk.com to the IRS as income — yes, even the $1.45 I made — to be taxed at the high rates of the self-employed. There’s no chance that a worker might land a full-time job with a company through Mechanical Turk, since it’s expressly forbidden in the site’s “participation agreement,” which requires workers to submit all work through the site, and not directly to the requester.
To a labor activist like Marcus Courtney of WashTech, a tech workers union, the whole arrangement represents a dystopian vision of a virtual sweatshop. “What Amazon is trying to do is create the virtual day laborer hiring hall on the global scale to bid down wage rates to the advantage of the employer,” he says. “Here you have a major global corporation, based in the United States, that’s showing the dark side of globalization. If this is Jeff Bezos’ vision of the future of work, I think that’s a pretty scary vision, and we should be paying attention to that.”
Rebecca Smith, a lawyer for the National Employment Law Project, seconds that. “The creativity of business in avoiding its responsibility to workers never ceases to astound,” she says dryly. “It’s day labor in the virtual world.” Smith sees Mechanical Turk as just another scheme by companies to classify workers as independent contractors to avoid paying them minimum wage and overtime, complying with non-discrimination laws, and being forced to carry unemployment insurance and workers compensation. “It’s an example of cyberspace overtaking a country’s labor laws,” she says. Needless to say, the turkers don’t see it that way.
Curtis Taylor, 50, a corporate trainer in Clarksville, Ind., who has earned more than $345 on Mturk.com, doesn’t even think of turking as work. To him, it’s a way to kill time. “I’m not in it to make money, I’m in it to goof off,” he says. Taylor travels a lot for business and finds himself sitting around in hotel rooms at night. He doesn’t like to watch TV much, and says that turking beats playing free online poker. To him, it’s “mad money,” which he blows buying gifts on Amazon, like Bill Bennett’s “America, the Last Best Hope,” for his son, a junior in high school. “If I ever stop being entertained, I’ll stop doing it,” he says. “I’ll just quit.”
Yet what’s a happy diversion for Taylor is serious business for the companies on Amazon Mechanical Turk. Efficient Frontier, a search engine marketing firm, has used Mturk.com to accomplish tens of thousands of tasks since early 2006. Efficient Frontier helps companies figure out which keywords will bring Web surfers to their sites. With Mturk.com, Efficient Frontier can afford to pay three different people to look at each potential keyword, and vote whether those words are relevant to a given site. It costs the company just 4.5 cents to test each keyword, paying 1.5 cents to Amazon, and 1 cent each to three turkers.
Sherwood Stranieri of SkyPromote, another search engine marketing firm in Boston, says he now has a virtual staff of 120 workers on Mturk.com. “It’s like a giant human computer,” he says. “It’s like having an infinite attention span.” Stranieri pays qualified turkers to surf a Web directory and figure out exactly where a specific site should be listed. He can get 300 of these tasks done in just five or six hours, even if he posts them on Mturk.com in the middle of the night. He pays 5 cents a task. “Pricing is very low right now because there are so many more workers than tasks right now,” he says. “People are fishing around for work to do.” Why do people do it if the pay is so low? It’s a question Stranieri wonders about himself. “I think it’s something of a hybrid between trying to make money on the side and a diversion, a substitute for doing a crossword puzzle. It’s sort of a mental exercise.”
Eric Cranston, 18, who recently graduated from high school, got into turking because at the time he didn’t have anything better to do. “When it came out,” he says, “I had just broken my foot, so I was just at home doing nothing on the computer. So, why not?” He’s used the money he’s made answering survey questions and transcribing podcasts to buy a game controller and a computer monitor. He recently transferred $200 to his bank account. “I don’t think anyone could actually make a living off of Mturk. There isn’t enough work,” Cranston says. He is one turker, however, who is plotting how to move up the food chain. Currently, Cranston and a friend are working to launch a Web-based business altering photographs, called Image Den, using, naturally, Mechanical Turkers to treat the images.
In its earliest days, someone posted a request on Amazon Mechanical Turk, offering to pay 2 cents for a drawing of a sheep facing left. Peter Cohen, director of Amazon Mechanical Turk, says the company was “puzzled by” the request. The requester was Aaron Koblin, a student in UCLA’s Design/Media Arts program, who was writing his master’s thesis about the site. He was intrigued by Amazon’s effort to “establish a framework for the utilization of people as computers,” as he wrote in his thesis. “My project was very tongue-in-cheek,” he tells me. “On the one hand, it’s using the system the way it’s meant to be used. On the other hand, it’s asking them to do this ridiculous thing.”
The grad student invited turkers to draw up to five sheep at the rate of 2 cents apiece. Over 40 days and 40 nights, the sheep flooded in at a rate of 11 per hour. By the end, 7,599 turkers had participated. He collected 12,000 sheep and promptly put 10,000 of them up for sale at the rate of $20 for 20 sheep at the Sheep Market. This caused some consternation among the people who had drawn them. “They’re selling our sheep!!!” wailed one poster on a turker message board. Another wrote: “Does anyone remember signing over the rights to the drawings?” In fact, they had. To participate in Amazon Mechanical Turk, workers, in the legalese of the site, “agree that the work product of any Services you perform is deemed a ‘work made for hire’ for the benefit of the Requester, and all ownership rights, including worldwide intellectual property rights, will vest with the Requester immediately upon your performance of the Service.”
Why sheep? Koblin relished all the associations that sheep have from the biblical followers of the good shepherd to George Orwell’s “Animal Farm.” The term “cottage industry” comes from peasants’ setting up shop at home, when it wasn’t planting or harvesting season, often spinning wool. “The cottage industry, which would employ entire families from their houses, has notable similarities to Mechanical Turk, such as employing people for spare time, working from home, and relative anonymity,” he wrote in his thesis.
Koblin wanted his project to capture the creativity expressed by turkers, while drawing attention to the insignificant role each of them played in the process. He certainly succeeded in capturing their creativity. Even after he stopped accepting sheep, and started selling them by the lot on the Internet, more people wrote to him wanting to contribute sheep for free. They just wanted to see their sheep join the herd. “Most of these people clearly weren’t in it for the money,” Koblin says. “They weren’t doing it so they could get 2 cents. It was more about participating in something larger.”
Maybe so. But maybe the ultimate message is: Congrats, fellow humans, we’re not obsolete! The machines, they still need us! Only at a very sheepish price.
British actress Olivia Williams with sabre fish.
Gillian Anderson, aka Scully, with a conger eel.
British actor Nickolas Grace with a red mullet.
French actress Aure Atika with a parrotfish.
French-Portuguese actress Barbara Cabrita with a herring.
French actress Caroline Ducey with a barracuda.
French actor Emmanuel de Brantes with a barramundi.
British DJ Godlie with a redfish.
French/American actor Jean-Marc Barr with a mako shark.
BBC star Jeany Spark with a seabass.
Opera singer Joanna Bergin with a mackerel.
Japanese fashion designer Kenzo Takada with a bonito.
French actress Mélanie Bernier with a European eel.
British actor and director Serge Hazanavicius with a thicklip grey mullet.
French jazz guitarist Thomas Dutronc with a dusky grouper.