Amazon`s Mechanical Turk: Where the 1700s Meet Web 2.0By Doug Bartholomew | Posted 2008-02-21 Email Print
Amazon’s Mechanical Turk is particularly well-suited for image processing tasks.
Imagine a computer that can harness the power of a million human brains working on different pieces of a monumental task and you’ve envisioned the Amazon Mechanical Turk.
The service is named for the infamous Mechanical Turk built in 1769 by Hungarian nobleman Wolfgang von Kempelen. His chess-playing “machine” astonished audiences throughout Europe before being exposed as a hoax.
The original Turk’s cabinet of intricate cogs and gears actually concealed a human chess master who defeated most opponents, including luminaries such as Napoleon Bonaparte and Benjamin Franklin.
With the Amazon Mechanical Turk, the true human brainpower also is hidden—in cyberspace. Using so-called “crowd sourcing,” AMT customers submit tasks that require human judgment and action. These human intelligence tasks (HITs) are performed by a community of “turkers” who are paid by the requesters, while Amazon gets a 10 percent commission.
The AMT was born when Amazon executives were aiming to ensure that photos of businesses taken for the online Yellow Pages on the company’s A9 search site corresponded to the correct listings. They set up a Web site and asked people to verify the pictures for a penny or two per photograph. It worked so well that in November 2005, the company launched the Amazon Mechanical Turk.
AMT is particularly well-suited for image processing tasks that require judgment, such as tagging objects found in an image, selecting the most effective images for advertising or checking images to rule out inappropriate content.
“Determining the tone of an article, for instance, is also a task that only a human can do,” explains AMT director Peter Cohen.
Although the Amazon Mechanical Turk continues to operate in beta, millions of turkers are currently working on jobs provided by online taskmasters. The individuals who assign the work define the tasks, including the output desired, the format, how the work items are displayed and how much they will pay for each task.
In addition, clients can require workers to pass a qualification test before tackling any tasks and can retrieve the results using Amazon’s Web services APIs.