Dept. of Human Obsolescence/Journalism edition

In 1770, Wolfgang von Kempelen amazed the court by creating a turbaned mechanical man--a "Turk"--that could play chess. The automaton, a kind of 18th century Deep Blue, beat the likes of Napoleon and Ben Franklin.  But it turned out to be an elaborate hoax. Instead of an early example of artificial intelligence, it was just a glorified parlor trick with a human being crouched inside, manipulating the chess board.

In 2005 Jeff Bezos created Amazon Mechanical Turk (MTurk),  an online marketplace in which "requesters" pay anonymous "Turkers" to complete mind-numbing tasks like identifying objects in photos, verifying search results or filling out surveys. A few tasks, such as translating Urdu, are presumably more challenging, but they are equally unprofitable. Each  "Human Intelligence Task," or Hit, may pay as little as $.01, and they are all tasks that humans can do either better or more cheaply  than computers. (It's like having hundreds of little men inside the machine doing the boring work computers don't want to do. Or as Salon's Katharine Mieszkowski put it, "the tag line for the site could be: dirt-cheap artificial artificial intelligence.")

And who are these Turkers? If you guessed that they are impoverished Indians or poorly educated slackers, you'd be only partly right. A study from the University of California at Irvine found that 86% have at least some college, while 42% have a bachelor's degree. And while 32% are from India, the majority (57%) are from the U.S. (The respondents were paid to take part in the survey and the study did not verify the results.)

The survey found that while some do the piecework for the money, others apparently do it just to pass the time. That's pretty interesting, since the Hits offer none of what we would presume to be the benefits of work. There is little social interaction, little room for creativity, and the fragmentation of the tasks means there is none of the satisfaction of seeing a finished product, much less the satisfaction of putting one's individual stamp on a finished product. The good news is that Turkers can wear their pajamas and work whenever they want, the bad news is that employers can unilaterally choose not to pay them if they don't like the quality of the work.

So instead of computers doing the grunt work for college grads,  the grads are doing the grunt work for computers.  Of course even these jobs are endangered, because one of the main uses of MTurk is gathering data for natural language processing and machine learning. By collecting these vast quantities of data, as in the GWAP (Games with a Purpose) activities I described in a previous post, researchers will eventually teach computers to communicate and process language as well as human beings do. And then even the lowly Turkers will be obsolete.

Curiously, another sector employing--or exploiting--the Turkers, is journalism. In a post on a Poynter Institute blog, Megan Taylor suggests that journalists can use MTurk to manage "tedious, repetitive, time-consuming and expensive to outsource" tasks such as transcribing interviews. Pro Publica, the non-profit dedicated to "journalism in the public interest," has also  published a journalist's guide to using the service.

It's understandable--but more than a little ironic--that journalists are trying to save a few bucks by taking advantage of super-cheap labor on MTurk. As the profession struggles to find a viable business model, many writers, reporters, editors and photographers have found themselves out-sourced and down-sized, just like many Turkers, one guesses. As they compete with user generated content, aggregated data streams and bloggers who do it for free, journalists are in a race to the bottom of the pay scale. Sadly, fewer and fewer journalists are able to make a living wage at the profession they love. But what about the folks doing their grunt work? As one of the commenters on the Poynter site said:

I am one of the many people who complete Mechanical Turk tasks. Something to keep in mind is that we work very hard to make sure that we give you accurate answers or transcriptions. However, we are paid very little for our tasks. Many transcription tasks available on MTurk pay between 30 and 75 cents for about 5 minutes of transcription. We often have to listen to the same clip a few times to make sure everything is accurate before submitting the task. MTurk workers are already so undervalued, just like freelance writers. If you choose to use MTurk, please offer reasonable compensation.

Meanwhile, the Nieman Journalism Lab, which bills itself as a "collaborative attempt to figure out how quality journalism can survive and thrive in the Internet age," offered a piece on how journalists might use another crowd-sourcing gimmick for cheap copy-editing. That program, called Soylent,  is "an add-in for Microsoft Word that uses a distributed copy-editing system to perform tasks like proofreading and text-shortening, as well as a type of specialized edits its developers call The Human Macro'."

(If you are not up to speed on '70s science fiction, "Soylent Green" was a 1973 film in which society depends on a high-protein manufactured food called soylent green. Charlton Heston plays a police detective in the year 2022 who finds out, much to his horror, that soylent is actually human corpses processed into tasty green wafers. "Soylent Green is PEOPLE!!" he screams at the end of the film. Thus the sick humor of the slogan: Soylent, the word processor with a crowd inside.)

The idea that copy could be chopped up and farmed out to anonymous, untrained MTurk copy editors who would comb through and clean up snippets of stories seems ridiculous, and I don't know if any writers have found the model useful. But at least one commenter on the Nieman site  found it pretty silly.

As a lifelong copy editor I see lots of opportunities for extra work created by one person making one fix to a sentence in a paragraph and another person making a different kind of fix within the same paragraph, thus creating a new inconsistency. You don’t eliminate the problems between the lazy and the overeager–you potentially end up with both kinds of changes in the same document. Yes, you have other Turkers looking at this but you’ve had to build in redundancy to catch errors you introduced through the original design, to say nothing of the effect on the author’s original voice after you’ve run it through the mill twice.

 Hey, I know–let’s microsource articles by having individuals each write one sentence. @BarbChamberlain

If you believes that the art or craft of writing is something more than "word processing," then running your words through the Soylent factory probably isn't for you. And if the idea of a crowd of anonymous people inside a word processor seems a little too depressing, cheer up. We haven't gotten to this yet:

 

Comments

[lindsay]

And what if teachers start putting student papers up for some Mturk help instead of spending all those hours grading?

[Chris Cascarano]

Very interesting. My gut tells me that this isn't going to be a major part of our future, but I've been terribly wrong about things like this in the past. After all, it can be pretty damn efficient for mindless drudgery.