In the midst of revising my ever-evolving essay on the Amazon Kindle e-reader, I stumbled upon the company's Mechanical Turk website. I was riveted.
The name “Mechanical Turk” pays homage to a faux automaton whose chess playing prowess captivated audiences throughout Europe in the late 18th century. Secretly, the robot’s skill derived not from any type of artificial intelligence but from a human chess master hiding inside the machine, who manipulated levers, pulleys, and magnets to create the illusion of self-directed game play. So too it is with Amazon Mechanical Turk, which the company refers to as “artificial artificial intelligence.”
The service is essentially a marketplace for 21st century piecework, the core of which are things called "human intelligence tasks." These are, in Amazon’s words, “questions that need an answer,” or rather data processing tasks that the present generation of computers is ill-equipped to handle (e.g. writing product reviews, performing rudimentary research, identifying and tagging images, and more). Collectively, Mechanical Turk workers comprise a flexible, on-demand labor force whose job it is to respond to these questions. Compensation depends on the complexity and duration of the task. Typically it consists of micro-payments ranging from a few pennies to a few dollars per job, paid for by the party who has issued a specific information request.
Today I decided to take Mechanical Turk out for a spin, having registered a few weeks ago as a Turk worker. I made seven cents in about five or ten minutes, having chosen four separate tasks that could be completed in two minutes or less. Either way you count in, that translates into less than a dollar per hour.
One of the jobs I initially accepted involved image tagging. Basically, I was asked to type in what I thought might be an appropriate search term for a photo appearing on screen. I got rather squeamish when I was presented with a shoulders-up image of what appeared to be a teenage girl looking coyly over her shoulder. I wondered then about what might be the end-result of my work and ended up rejecting the task.
I gather from the transcript of a recent Berkman Center forum that Jonathan Zittrain has already expressed similar concerns about Mechanical Turk. With traditional piecework or even assembly-line labor the worker, however estranged he or she may be from the end-product, nonetheless typically has at least some sense of the resulting whole. But as Zittrain has pointed out, the same doesn't generally apply to Mechanical Turk. There, laborers are so disaggregated that there's virtually no sense of what one's small contribution might ultimately result in.
I need to ponder this more carefully, as there seem to me some intriguing research applications were the Mechanical Turk service approached ethically. Without any type of ethical filter in place, however, I worry about its economic and political implications.
No comments:
Post a Comment