Most Mechanical Turkers Are Young, College-Educated and Making Less Than $5 an Hour

Moshe Z. Marvit July 15, 2016

A new study lifts the veil on the Amazon-owned crowdworking company (Arne Krueger / Flickr)

Since 2005, a dis­persed group of sub-min­i­mum wage work­ers has been per­form­ing online tasks for pen­nies through an Ama­zon-con­trolled mar­ket­place called Mechan­i­cal Turk. These work­ers tag pho­tos, tran­scribe audio, take sur­veys, and do what­ev­er cur­rent com­put­er tech­nol­o­gy can­not. Their work-prod­uct is lit­tered across the Inter­net, and through aca­d­e­m­ic pub­li­ca­tions, but they have large­ly remained invis­i­ble. Var­i­ous stud­ies have attempt­ed to take a clos­er look, but any giv­en study is lim­it­ed when it must rely on self-report­ing from an anony­mous workforce.

This week, the Pew Research Cen­ter released a major study that fills in some of the gaps in under­stand­ing who the Turk­ers” are.

The report, titled Research in the Crowd­sourc­ing Age,” reveals some­thing sur­pris­ing about these cogs in the dig­i­tal machine: Most of them are young and col­lege-edu­cat­ed. The major­i­ty of them are also mak­ing less than $5 an hour per­form­ing short, repet­i­tive micro­tasks that paid 10 cents or less and could be com­plet­ed in a few minutes.”

In order to track down Turk­ers, researchers sur­veyed thou­sands of Mechan­i­cal Turk work­ers dur­ing var­i­ous times each day dur­ing a sam­ple peri­od. They offered Turk­ers var­i­ous amounts, between 5 cents and $2, to ensure that they attract­ed peo­ple who would accept a wide vari­ety of compensation.

Paul Hitlin, the study’s author, tells In These Times that he was sur­prised by the find­ings. Because it is so low pay­ing,” he says, we hypoth­e­sized that the peo­ple doing it would have much less edu­ca­tion.” The results seem to con­firm that in the gig econ­o­my, the old rules — of edu­ca­tion and expe­ri­ence lead­ing to greater com­pen­sa­tion — do not apply.

The study revealed anoth­er unex­pect­ed facet of crowd­work­ing plat­forms. While for-prof­it com­pa­nies use Mechan­i­cal Turk to per­form tasks rang­ing from image recog­ni­tion to tran­scrip­tion, aca­d­e­mics have also found a use for this pool of sub-min­i­mum wage labor. The report found that 36 per­cent of the unique requesters” (Mechan­i­cal Turk speak for employ­ers”) came from the acad­e­my, com­pared to 31 per­cent from business.

Exploita­tion is academic

In the last sev­er­al years, it seems the allure of cheap labor has led to a grow­ing reliance on Mechan­i­cal Turk for research. In 2015, more than 800 peer-reviewed stud­ies across a wide range of dis­ci­plines were pub­lished using data obtained from Turk­ers, accord­ing to the study. Of the jobs post­ed by aca­d­e­mics on Mechan­i­cal Turk, 89 per­cent were relat­ed to sur­veys used for research studies.

The use of Mechan­i­cal Turk by uni­ver­si­ties has become so com­mon that many uni­ver­si­ties pro­vide guid­ance for researchers using the site. (See here, here, here). Still, pay­ing sub­jects pen­nies on the dol­lar to com­plete online sur­veys has raised eth­i­cal con­cerns. Most of these con­cerns, how­ev­er, are byprod­ucts of an aca­d­e­m­ic sys­tem that relies on cheap labor, such as poor work prod­uct, pro­fes­sion­al par­tic­i­pants and lack of work­er pro­tec­tions. The focus so far has not been on pay. It’s been on accu­ra­cy and privacy.

In 2013, a group of promi­nent com­put­er sci­ence researchers pub­lished a paper with the unusu­al­ly sim­ple title, Mechan­i­cal Turk is Not Anony­mous.” Ama­zon pro­motes its plat­form as offer­ing anonymi­ty to Turk­ers, and researchers often must offer such anonymi­ty to their sub­jects, so this built-in fea­ture was ide­al for many in the acad­e­my. How­ev­er, the com­put­er sci­en­tists describe how their team acci­den­tal­ly dis­cov­ered that the same 14-char­ac­ter alphanu­mer­ic string to unique­ly iden­ti­fy a work­er for AMT is also used to unique­ly iden­ti­fy Ama­zon cus­tomers across all Ama­zon properties.”

This means that if one were to do an Inter­net search of a Turker’s 14-char­ac­ter iden­ti­fi­er, one would find her Ama­zon pro­file, includ­ing wish lists, cus­tomer reviews, and often real names and pho­tos. Though many uni­ver­si­ties and researchers had already become reg­u­lar users of Mechan­i­cal Turk for their research by this point, none was aware of this major vul­ner­a­bil­i­ty. They chron­i­cled the reac­tion in the paper, writ­ing, the ini­tial reac­tion was one of audi­ble gasps and dis­be­lief, fol­lowed by stunned silence. Of all the 30-some high­ly edu­cat­ed researchers in the room, many of whom had used AMT reg­u­lar­ly for years, no one had known.”

Ama­zon also col­lects Turk­ers’ IP address­es and has access to their sur­vey respons­es. Because of this, the Uni­ver­si­ty of Cal­i­for­nia Berkeley’s Com­mit­tee for Pro­tec­tion of Human Sub­jects sug­gests that researchers should only use Mechan­i­cal Turk as a recruit­ment tool, pro­vid­ing a link to anoth­er secure site for the actu­al research.

Fur­ther­more, even though Mechan­i­cal Turk has become the pri­ma­ry source for online sub­jects, many have ques­tioned the accu­ra­cy of the data col­lect­ed when using Turk­ers for aca­d­e­m­ic research because many of the Turk­ers will work as sub­jects across a wide vari­ety of relat­ed stud­ies. There­fore researchers are not using ran­dom sub­jects from a het­ero­ge­neous pool of work­ers, but rather pro­fes­sion­al par­tic­i­pants that may answer based on their pre­vi­ous expe­ri­ences as research subjects.

Despite these prob­lems, the appeal to uni­ver­si­ties of uti­liz­ing cheap labor is too great. There’s lit­tle to sug­gest the eco­nom­ics of this cal­cu­lus will change any­time soon.

Moshe Z. Mar­vit is an attor­ney and fel­low with The Cen­tu­ry Foun­da­tion and the co-author (with Richard Kahlen­berg) of the book Why Labor Orga­niz­ing Should be a Civ­il Right.

Limited Time:

SUBSCRIBE TO IN THESE TIMES MAGAZINE FOR JUST $1 A MONTH